cddl/contrib/opensolaris/lib/libzfs/common/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 <sys/types.h>
  29 #include <sys/stat.h>
  30 #include <ctype.h>
  31 #include <errno.h>
  32 #include <devid.h>
  33 #include <fcntl.h>
  34 #include <libintl.h>
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <strings.h>
  38 #include <unistd.h>
  39 #include <sys/zfs_ioctl.h>
  40 #include <dlfcn.h>
  41
  42 #include "zfs_namecheck.h"
  43 #include "zfs_prop.h"
  44 #include "libzfs_impl.h"
  45 #include "zfs_comutil.h"
  46
  47 static int read_efi_label(nvlist_t *config, diskaddr_t *sb);
  48
  49 #define DISK_ROOT       "/dev/dsk"
  50 #define RDISK_ROOT      "/dev/rdsk"
  51 #define BACKUP_SLICE    "s2"
  52
  53 typedef struct prop_flags {
  54         int create:1;   /* Validate property on creation */
  55         int import:1;   /* Validate property on import */
  56 } prop_flags_t;
  57
  58 /*
  59  * ====================================================================
  60  *   zpool property functions
  61  * ====================================================================
  62  */
  63
  64 static int
  65 zpool_get_all_props(zpool_handle_t *zhp)
  66 {
  67         zfs_cmd_t zc = { 0 };
  68         libzfs_handle_t *hdl = zhp->zpool_hdl;
  69
  70         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
  71
  72         if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
  73                 return (-1);
  74
  75         while (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
  76                 if (errno == ENOMEM) {
  77                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
  78                                 zcmd_free_nvlists(&zc);
  79                                 return (-1);
  80                         }
  81                 } else {
  82                         zcmd_free_nvlists(&zc);
  83                         return (-1);
  84                 }
  85         }
  86
  87         if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
  88                 zcmd_free_nvlists(&zc);
  89                 return (-1);
  90         }
  91
  92         zcmd_free_nvlists(&zc);
  93
  94         return (0);
  95 }
  96
  97 static int
  98 zpool_props_refresh(zpool_handle_t *zhp)
  99 {
 100         nvlist_t *old_props;
 101
 102         old_props = zhp->zpool_props;
 103
 104         if (zpool_get_all_props(zhp) != 0)
 105                 return (-1);
 106
 107         nvlist_free(old_props);
 108         return (0);
 109 }
 110
 111 static char *
 112 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
 113     zprop_source_t *src)
 114 {
 115         nvlist_t *nv, *nvl;
 116         uint64_t ival;
 117         char *value;
 118         zprop_source_t source;
 119
 120         nvl = zhp->zpool_props;
 121         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 122                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &ival) == 0);
 123                 source = ival;
 124                 verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
 125         } else {
 126                 source = ZPROP_SRC_DEFAULT;
 127                 if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
 128                         value = "-";
 129         }
 130
 131         if (src)
 132                 *src = source;
 133
 134         return (value);
 135 }
 136
 137 uint64_t
 138 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
 139 {
 140         nvlist_t *nv, *nvl;
 141         uint64_t value;
 142         zprop_source_t source;
 143
 144         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
 145                 /*
 146                  * zpool_get_all_props() has most likely failed because
 147                  * the pool is faulted, but if all we need is the top level
 148                  * vdev's guid then get it from the zhp config nvlist.
 149                  */
 150                 if ((prop == ZPOOL_PROP_GUID) &&
 151                     (nvlist_lookup_nvlist(zhp->zpool_config,
 152                     ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
 153                     (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
 154                     == 0)) {
 155                         return (value);
 156                 }
 157                 return (zpool_prop_default_numeric(prop));
 158         }
 159
 160         nvl = zhp->zpool_props;
 161         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 162                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &value) == 0);
 163                 source = value;
 164                 verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
 165         } else {
 166                 source = ZPROP_SRC_DEFAULT;
 167                 value = zpool_prop_default_numeric(prop);
 168         }
 169
 170         if (src)
 171                 *src = source;
 172
 173         return (value);
 174 }
 175
 176 /*
 177  * Map VDEV STATE to printed strings.
 178  */
 179 const char *
 180 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
 181 {
 182         switch (state) {
 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  * Map POOL STATE to printed strings.
 208  */
 209 const char *
 210 zpool_pool_state_to_name(pool_state_t state)
 211 {
 212         switch (state) {
 213         case POOL_STATE_ACTIVE:
 214                 return (gettext("ACTIVE"));
 215         case POOL_STATE_EXPORTED:
 216                 return (gettext("EXPORTED"));
 217         case POOL_STATE_DESTROYED:
 218                 return (gettext("DESTROYED"));
 219         case POOL_STATE_SPARE:
 220                 return (gettext("SPARE"));
 221         case POOL_STATE_L2CACHE:
 222                 return (gettext("L2CACHE"));
 223         case POOL_STATE_UNINITIALIZED:
 224                 return (gettext("UNINITIALIZED"));
 225         case POOL_STATE_UNAVAIL:
 226                 return (gettext("UNAVAIL"));
 227         case POOL_STATE_POTENTIALLY_ACTIVE:
 228                 return (gettext("POTENTIALLY_ACTIVE"));
 229         }
 230
 231         return (gettext("UNKNOWN"));
 232 }
 233
 234 /*
 235  * Get a zpool property value for 'prop' and return the value in
 236  * a pre-allocated buffer.
 237  */
 238 int
 239 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf, size_t len,
 240     zprop_source_t *srctype)
 241 {
 242         uint64_t intval;
 243         const char *strval;
 244         zprop_source_t src = ZPROP_SRC_NONE;
 245         nvlist_t *nvroot;
 246         vdev_stat_t *vs;
 247         uint_t vsc;
 248
 249         if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
 250                 switch (prop) {
 251                 case ZPOOL_PROP_NAME:
 252                         (void) strlcpy(buf, zpool_get_name(zhp), len);
 253                         break;
 254
 255                 case ZPOOL_PROP_HEALTH:
 256                         (void) strlcpy(buf, "FAULTED", len);
 257                         break;
 258
 259                 case ZPOOL_PROP_GUID:
 260                         intval = zpool_get_prop_int(zhp, prop, &src);
 261                         (void) snprintf(buf, len, "%llu", intval);
 262                         break;
 263
 264                 case ZPOOL_PROP_ALTROOT:
 265                 case ZPOOL_PROP_CACHEFILE:
 266                 case ZPOOL_PROP_COMMENT:
 267                         if (zhp->zpool_props != NULL ||
 268                             zpool_get_all_props(zhp) == 0) {
 269                                 (void) strlcpy(buf,
 270                                     zpool_get_prop_string(zhp, prop, &src),
 271                                     len);
 272                                 if (srctype != NULL)
 273                                         *srctype = src;
 274                                 return (0);
 275                         }
 276                         /* FALLTHROUGH */
 277                 default:
 278                         (void) strlcpy(buf, "-", len);
 279                         break;
 280                 }
 281
 282                 if (srctype != NULL)
 283                         *srctype = src;
 284                 return (0);
 285         }
 286
 287         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
 288             prop != ZPOOL_PROP_NAME)
 289                 return (-1);
 290
 291         switch (zpool_prop_get_type(prop)) {
 292         case PROP_TYPE_STRING:
 293                 (void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
 294                     len);
 295                 break;
 296
 297         case PROP_TYPE_NUMBER:
 298                 intval = zpool_get_prop_int(zhp, prop, &src);
 299
 300                 switch (prop) {
 301                 case ZPOOL_PROP_SIZE:
 302                 case ZPOOL_PROP_ALLOCATED:
 303                 case ZPOOL_PROP_FREE:
 304                 case ZPOOL_PROP_EXPANDSZ:
 305                         (void) zfs_nicenum(intval, buf, len);
 306                         break;
 307
 308                 case ZPOOL_PROP_CAPACITY:
 309                         (void) snprintf(buf, len, "%llu%%",
 310                             (u_longlong_t)intval);
 311                         break;
 312
 313                 case ZPOOL_PROP_DEDUPRATIO:
 314                         (void) snprintf(buf, len, "%llu.%02llux",
 315                             (u_longlong_t)(intval / 100),
 316                             (u_longlong_t)(intval % 100));
 317                         break;
 318
 319                 case ZPOOL_PROP_HEALTH:
 320                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 321                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 322                         verify(nvlist_lookup_uint64_array(nvroot,
 323                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
 324                             == 0);
 325
 326                         (void) strlcpy(buf, zpool_state_to_name(intval,
 327                             vs->vs_aux), len);
 328                         break;
 329                 default:
 330                         (void) snprintf(buf, len, "%llu", intval);
 331                 }
 332                 break;
 333
 334         case PROP_TYPE_INDEX:
 335                 intval = zpool_get_prop_int(zhp, prop, &src);
 336                 if (zpool_prop_index_to_string(prop, intval, &strval)
 337                     != 0)
 338                         return (-1);
 339                 (void) strlcpy(buf, strval, len);
 340                 break;
 341
 342         default:
 343                 abort();
 344         }
 345
 346         if (srctype)
 347                 *srctype = src;
 348
 349         return (0);
 350 }
 351
 352 /*
 353  * Check if the bootfs name has the same pool name as it is set to.
 354  * Assuming bootfs is a valid dataset name.
 355  */
 356 static boolean_t
 357 bootfs_name_valid(const char *pool, char *bootfs)
 358 {
 359         int len = strlen(pool);
 360
 361         if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
 362                 return (B_FALSE);
 363
 364         if (strncmp(pool, bootfs, len) == 0 &&
 365             (bootfs[len] == '/' || bootfs[len] == '\0'))
 366                 return (B_TRUE);
 367
 368         return (B_FALSE);
 369 }
 370
 371 /*
 372  * Inspect the configuration to determine if any of the devices contain
 373  * an EFI label.
 374  */
 375 static boolean_t
 376 pool_uses_efi(nvlist_t *config)
 377 {
 378 #ifdef sun
 379         nvlist_t **child;
 380         uint_t c, children;
 381
 382         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
 383             &child, &children) != 0)
 384                 return (read_efi_label(config, NULL) >= 0);
 385
 386         for (c = 0; c < children; c++) {
 387                 if (pool_uses_efi(child[c]))
 388                         return (B_TRUE);
 389         }
 390 #endif  /* sun */
 391         return (B_FALSE);
 392 }
 393
 394 boolean_t
 395 zpool_is_bootable(zpool_handle_t *zhp)
 396 {
 397         char bootfs[ZPOOL_MAXNAMELEN];
 398
 399         return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
 400             sizeof (bootfs), NULL) == 0 && strncmp(bootfs, "-",
 401             sizeof (bootfs)) != 0);
 402 }
 403
 404
 405 /*
 406  * Given an nvlist of zpool properties to be set, validate that they are
 407  * correct, and parse any numeric properties (index, boolean, etc) if they are
 408  * specified as strings.
 409  */
 410 static nvlist_t *
 411 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 412     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
 413 {
 414         nvpair_t *elem;
 415         nvlist_t *retprops;
 416         zpool_prop_t prop;
 417         char *strval;
 418         uint64_t intval;
 419         char *slash, *check;
 420         struct stat64 statbuf;
 421         zpool_handle_t *zhp;
 422         nvlist_t *nvroot;
 423
 424         if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
 425                 (void) no_memory(hdl);
 426                 return (NULL);
 427         }
 428
 429         elem = NULL;
 430         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 431                 const char *propname = nvpair_name(elem);
 432
 433                 /*
 434                  * Make sure this property is valid and applies to this type.
 435                  */
 436                 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL) {
 437                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 438                             "invalid property '%s'"), propname);
 439                         (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 440                         goto error;
 441                 }
 442
 443                 if (zpool_prop_readonly(prop)) {
 444                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 445                             "is readonly"), propname);
 446                         (void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
 447                         goto error;
 448                 }
 449
 450                 if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
 451                     &strval, &intval, errbuf) != 0)
 452                         goto error;
 453
 454                 /*
 455                  * Perform additional checking for specific properties.
 456                  */
 457                 switch (prop) {
 458                 case ZPOOL_PROP_VERSION:
 459                         if (intval < version || intval > SPA_VERSION) {
 460                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 461                                     "property '%s' number %d is invalid."),
 462                                     propname, intval);
 463                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 464                                 goto error;
 465                         }
 466                         break;
 467
 468                 case ZPOOL_PROP_BOOTFS:
 469                         if (flags.create || flags.import) {
 470                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 471                                     "property '%s' cannot be set at creation "
 472                                     "or import time"), propname);
 473                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 474                                 goto error;
 475                         }
 476
 477                         if (version < SPA_VERSION_BOOTFS) {
 478                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 479                                     "pool must be upgraded to support "
 480                                     "'%s' property"), propname);
 481                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 482                                 goto error;
 483                         }
 484
 485                         /*
 486                          * bootfs property value has to be a dataset name and
 487                          * the dataset has to be in the same pool as it sets to.
 488                          */
 489                         if (strval[0] != '\0' && !bootfs_name_valid(poolname,
 490                             strval)) {
 491                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 492                                     "is an invalid name"), strval);
 493                                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
 494                                 goto error;
 495                         }
 496
 497                         if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
 498                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 499                                     "could not open pool '%s'"), poolname);
 500                                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
 501                                 goto error;
 502                         }
 503                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 504                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 505
 506 #ifdef sun
 507                         /*
 508                          * bootfs property cannot be set on a disk which has
 509                          * been EFI labeled.
 510                          */
 511                         if (pool_uses_efi(nvroot)) {
 512                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 513                                     "property '%s' not supported on "
 514                                     "EFI labeled devices"), propname);
 515                                 (void) zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf);
 516                                 zpool_close(zhp);
 517                                 goto error;
 518                         }
 519 #endif  /* sun */
 520                         zpool_close(zhp);
 521                         break;
 522
 523                 case ZPOOL_PROP_ALTROOT:
 524                         if (!flags.create && !flags.import) {
 525                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 526                                     "property '%s' can only be set during pool "
 527                                     "creation or import"), propname);
 528                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 529                                 goto error;
 530                         }
 531
 532                         if (strval[0] != '/') {
 533                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 534                                     "bad alternate root '%s'"), strval);
 535                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 536                                 goto error;
 537                         }
 538                         break;
 539
 540                 case ZPOOL_PROP_CACHEFILE:
 541                         if (strval[0] == '\0')
 542                                 break;
 543
 544                         if (strcmp(strval, "none") == 0)
 545                                 break;
 546
 547                         if (strval[0] != '/') {
 548                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 549                                     "property '%s' must be empty, an "
 550                                     "absolute path, or 'none'"), propname);
 551                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 552                                 goto error;
 553                         }
 554
 555                         slash = strrchr(strval, '/');
 556
 557                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 558                             strcmp(slash, "/..") == 0) {
 559                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 560                                     "'%s' is not a valid file"), strval);
 561                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 562                                 goto error;
 563                         }
 564
 565                         *slash = '\0';
 566
 567                         if (strval[0] != '\0' &&
 568                             (stat64(strval, &statbuf) != 0 ||
 569                             !S_ISDIR(statbuf.st_mode))) {
 570                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 571                                     "'%s' is not a valid directory"),
 572                                     strval);
 573                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 574                                 goto error;
 575                         }
 576
 577                         *slash = '/';
 578                         break;
 579
 580                 case ZPOOL_PROP_COMMENT:
 581                         for (check = strval; *check != '\0'; check++) {
 582                                 if (!isprint(*check)) {
 583                                         zfs_error_aux(hdl,
 584                                             dgettext(TEXT_DOMAIN,
 585                                             "comment may only have printable "
 586                                             "characters"));
 587                                         (void) zfs_error(hdl, EZFS_BADPROP,
 588                                             errbuf);
 589                                         goto error;
 590                                 }
 591                         }
 592                         if (strlen(strval) > ZPROP_MAX_COMMENT) {
 593                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 594                                     "comment must not exceed %d characters"),
 595                                     ZPROP_MAX_COMMENT);
 596                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 597                                 goto error;
 598                         }
 599                         break;
 600                 case ZPOOL_PROP_READONLY:
 601                         if (!flags.import) {
 602                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 603                                     "property '%s' can only be set at "
 604                                     "import time"), propname);
 605                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 606                                 goto error;
 607                         }
 608                         break;
 609                 }
 610         }
 611
 612         return (retprops);
 613 error:
 614         nvlist_free(retprops);
 615         return (NULL);
 616 }
 617
 618 /*
 619  * Set zpool property : propname=propval.
 620  */
 621 int
 622 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
 623 {
 624         zfs_cmd_t zc = { 0 };
 625         int ret = -1;
 626         char errbuf[1024];
 627         nvlist_t *nvl = NULL;
 628         nvlist_t *realprops;
 629         uint64_t version;
 630         prop_flags_t flags = { 0 };
 631
 632         (void) snprintf(errbuf, sizeof (errbuf),
 633             dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
 634             zhp->zpool_name);
 635
 636         if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
 637                 return (no_memory(zhp->zpool_hdl));
 638
 639         if (nvlist_add_string(nvl, propname, propval) != 0) {
 640                 nvlist_free(nvl);
 641                 return (no_memory(zhp->zpool_hdl));
 642         }
 643
 644         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
 645         if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
 646             zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
 647                 nvlist_free(nvl);
 648                 return (-1);
 649         }
 650
 651         nvlist_free(nvl);
 652         nvl = realprops;
 653
 654         /*
 655          * Execute the corresponding ioctl() to set this property.
 656          */
 657         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 658
 659         if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
 660                 nvlist_free(nvl);
 661                 return (-1);
 662         }
 663
 664         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
 665
 666         zcmd_free_nvlists(&zc);
 667         nvlist_free(nvl);
 668
 669         if (ret)
 670                 (void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
 671         else
 672                 (void) zpool_props_refresh(zhp);
 673
 674         return (ret);
 675 }
 676
 677 int
 678 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
 679 {
 680         libzfs_handle_t *hdl = zhp->zpool_hdl;
 681         zprop_list_t *entry;
 682         char buf[ZFS_MAXPROPLEN];
 683
 684         if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
 685                 return (-1);
 686
 687         for (entry = *plp; entry != NULL; entry = entry->pl_next) {
 688
 689                 if (entry->pl_fixed)
 690                         continue;
 691
 692                 if (entry->pl_prop != ZPROP_INVAL &&
 693                     zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
 694                     NULL) == 0) {
 695                         if (strlen(buf) > entry->pl_width)
 696                                 entry->pl_width = strlen(buf);
 697                 }
 698         }
 699
 700         return (0);
 701 }
 702
 703
 704 /*
 705  * Don't start the slice at the default block of 34; many storage
 706  * devices will use a stripe width of 128k, so start there instead.
 707  */
 708 #define NEW_START_BLOCK 256
 709
 710 /*
 711  * Validate the given pool name, optionally putting an extended error message in
 712  * 'buf'.
 713  */
 714 boolean_t
 715 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
 716 {
 717         namecheck_err_t why;
 718         char what;
 719         int ret;
 720
 721         ret = pool_namecheck(pool, &why, &what);
 722
 723         /*
 724          * The rules for reserved pool names were extended at a later point.
 725          * But we need to support users with existing pools that may now be
 726          * invalid.  So we only check for this expanded set of names during a
 727          * create (or import), and only in userland.
 728          */
 729         if (ret == 0 && !isopen &&
 730             (strncmp(pool, "mirror", 6) == 0 ||
 731             strncmp(pool, "raidz", 5) == 0 ||
 732             strncmp(pool, "spare", 5) == 0 ||
 733             strcmp(pool, "log") == 0)) {
 734                 if (hdl != NULL)
 735                         zfs_error_aux(hdl,
 736                             dgettext(TEXT_DOMAIN, "name is reserved"));
 737                 return (B_FALSE);
 738         }
 739
 740
 741         if (ret != 0) {
 742                 if (hdl != NULL) {
 743                         switch (why) {
 744                         case NAME_ERR_TOOLONG:
 745                                 zfs_error_aux(hdl,
 746                                     dgettext(TEXT_DOMAIN, "name is too long"));
 747                                 break;
 748
 749                         case NAME_ERR_INVALCHAR:
 750                                 zfs_error_aux(hdl,
 751                                     dgettext(TEXT_DOMAIN, "invalid character "
 752                                     "'%c' in pool name"), what);
 753                                 break;
 754
 755                         case NAME_ERR_NOLETTER:
 756                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 757                                     "name must begin with a letter"));
 758                                 break;
 759
 760                         case NAME_ERR_RESERVED:
 761                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 762                                     "name is reserved"));
 763                                 break;
 764
 765                         case NAME_ERR_DISKLIKE:
 766                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 767                                     "pool name is reserved"));
 768                                 break;
 769
 770                         case NAME_ERR_LEADING_SLASH:
 771                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 772                                     "leading slash in name"));
 773                                 break;
 774
 775                         case NAME_ERR_EMPTY_COMPONENT:
 776                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 777                                     "empty component in name"));
 778                                 break;
 779
 780                         case NAME_ERR_TRAILING_SLASH:
 781                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 782                                     "trailing slash in name"));
 783                                 break;
 784
 785                         case NAME_ERR_MULTIPLE_AT:
 786                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 787                                     "multiple '@' delimiters in name"));
 788                                 break;
 789
 790                         }
 791                 }
 792                 return (B_FALSE);
 793         }
 794
 795         return (B_TRUE);
 796 }
 797
 798 /*
 799  * Open a handle to the given pool, even if the pool is currently in the FAULTED
 800  * state.
 801  */
 802 zpool_handle_t *
 803 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
 804 {
 805         zpool_handle_t *zhp;
 806         boolean_t missing;
 807
 808         /*
 809          * Make sure the pool name is valid.
 810          */
 811         if (!zpool_name_valid(hdl, B_TRUE, pool)) {
 812                 (void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
 813                     dgettext(TEXT_DOMAIN, "cannot open '%s'"),
 814                     pool);
 815                 return (NULL);
 816         }
 817
 818         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
 819                 return (NULL);
 820
 821         zhp->zpool_hdl = hdl;
 822         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
 823
 824         if (zpool_refresh_stats(zhp, &missing) != 0) {
 825                 zpool_close(zhp);
 826                 return (NULL);
 827         }
 828
 829         if (missing) {
 830                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
 831                 (void) zfs_error_fmt(hdl, EZFS_NOENT,
 832                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
 833                 zpool_close(zhp);
 834                 return (NULL);
 835         }
 836
 837         return (zhp);
 838 }
 839
 840 /*
 841  * Like the above, but silent on error.  Used when iterating over pools (because
 842  * the configuration cache may be out of date).
 843  */
 844 int
 845 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
 846 {
 847         zpool_handle_t *zhp;
 848         boolean_t missing;
 849
 850         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
 851                 return (-1);
 852
 853         zhp->zpool_hdl = hdl;
 854         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
 855
 856         if (zpool_refresh_stats(zhp, &missing) != 0) {
 857                 zpool_close(zhp);
 858                 return (-1);
 859         }
 860
 861         if (missing) {
 862                 zpool_close(zhp);
 863                 *ret = NULL;
 864                 return (0);
 865         }
 866
 867         *ret = zhp;
 868         return (0);
 869 }
 870
 871 /*
 872  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
 873  * state.
 874  */
 875 zpool_handle_t *
 876 zpool_open(libzfs_handle_t *hdl, const char *pool)
 877 {
 878         zpool_handle_t *zhp;
 879
 880         if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
 881                 return (NULL);
 882
 883         if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
 884                 (void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
 885                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
 886                 zpool_close(zhp);
 887                 return (NULL);
 888         }
 889
 890         return (zhp);
 891 }
 892
 893 /*
 894  * Close the handle.  Simply frees the memory associated with the handle.
 895  */
 896 void
 897 zpool_close(zpool_handle_t *zhp)
 898 {
 899         if (zhp->zpool_config)
 900                 nvlist_free(zhp->zpool_config);
 901         if (zhp->zpool_old_config)
 902                 nvlist_free(zhp->zpool_old_config);
 903         if (zhp->zpool_props)
 904                 nvlist_free(zhp->zpool_props);
 905         free(zhp);
 906 }
 907
 908 /*
 909  * Return the name of the pool.
 910  */
 911 const char *
 912 zpool_get_name(zpool_handle_t *zhp)
 913 {
 914         return (zhp->zpool_name);
 915 }
 916
 917
 918 /*
 919  * Return the state of the pool (ACTIVE or UNAVAILABLE)
 920  */
 921 int
 922 zpool_get_state(zpool_handle_t *zhp)
 923 {
 924         return (zhp->zpool_state);
 925 }
 926
 927 /*
 928  * Create the named pool, using the provided vdev list.  It is assumed
 929  * that the consumer has already validated the contents of the nvlist, so we
 930  * don't have to worry about error semantics.
 931  */
 932 int
 933 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
 934     nvlist_t *props, nvlist_t *fsprops)
 935 {
 936         zfs_cmd_t zc = { 0 };
 937         nvlist_t *zc_fsprops = NULL;
 938         nvlist_t *zc_props = NULL;
 939         char msg[1024];
 940         char *altroot;
 941         int ret = -1;
 942
 943         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
 944             "cannot create '%s'"), pool);
 945
 946         if (!zpool_name_valid(hdl, B_FALSE, pool))
 947                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
 948
 949         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
 950                 return (-1);
 951
 952         if (props) {
 953                 prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
 954
 955                 if ((zc_props = zpool_valid_proplist(hdl, pool, props,
 956                     SPA_VERSION_1, flags, msg)) == NULL) {
 957                         goto create_failed;
 958                 }
 959         }
 960
 961         if (fsprops) {
 962                 uint64_t zoned;
 963                 char *zonestr;
 964
 965                 zoned = ((nvlist_lookup_string(fsprops,
 966                     zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
 967                     strcmp(zonestr, "on") == 0);
 968
 969                 if ((zc_fsprops = zfs_valid_proplist(hdl,
 970                     ZFS_TYPE_FILESYSTEM, fsprops, zoned, NULL, msg)) == NULL) {
 971                         goto create_failed;
 972                 }
 973                 if (!zc_props &&
 974                     (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
 975                         goto create_failed;
 976                 }
 977                 if (nvlist_add_nvlist(zc_props,
 978                     ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
 979                         goto create_failed;
 980                 }
 981         }
 982
 983         if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
 984                 goto create_failed;
 985
 986         (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
 987
 988         if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
 989
 990                 zcmd_free_nvlists(&zc);
 991                 nvlist_free(zc_props);
 992                 nvlist_free(zc_fsprops);
 993
 994                 switch (errno) {
 995                 case EBUSY:
 996                         /*
 997                          * This can happen if the user has specified the same
 998                          * device multiple times.  We can't reliably detect this
 999                          * until we try to add it and see we already have a
1000                          * label.
1001                          */
1002                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1003                             "one or more vdevs refer to the same device"));
1004                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1005
1006                 case EOVERFLOW:
1007                         /*
1008                          * This occurs when one of the devices is below
1009                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1010                          * device was the problem device since there's no
1011                          * reliable way to determine device size from userland.
1012                          */
1013                         {
1014                                 char buf[64];
1015
1016                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1017
1018                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1019                                     "one or more devices is less than the "
1020                                     "minimum size (%s)"), buf);
1021                         }
1022                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1023
1024                 case ENOSPC:
1025                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1026                             "one or more devices is out of space"));
1027                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1028
1029                 case ENOTBLK:
1030                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1031                             "cache device must be a disk or disk slice"));
1032                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1033
1034                 default:
1035                         return (zpool_standard_error(hdl, errno, msg));
1036                 }
1037         }
1038
1039         /*
1040          * If this is an alternate root pool, then we automatically set the
1041          * mountpoint of the root dataset to be '/'.
1042          */
1043         if (nvlist_lookup_string(props, zpool_prop_to_name(ZPOOL_PROP_ALTROOT),
1044             &altroot) == 0) {
1045                 zfs_handle_t *zhp;
1046
1047                 verify((zhp = zfs_open(hdl, pool, ZFS_TYPE_DATASET)) != NULL);
1048                 verify(zfs_prop_set(zhp, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT),
1049                     "/") == 0);
1050
1051                 zfs_close(zhp);
1052         }
1053
1054 create_failed:
1055         zcmd_free_nvlists(&zc);
1056         nvlist_free(zc_props);
1057         nvlist_free(zc_fsprops);
1058         return (ret);
1059 }
1060
1061 /*
1062  * Destroy the given pool.  It is up to the caller to ensure that there are no
1063  * datasets left in the pool.
1064  */
1065 int
1066 zpool_destroy(zpool_handle_t *zhp)
1067 {
1068         zfs_cmd_t zc = { 0 };
1069         zfs_handle_t *zfp = NULL;
1070         libzfs_handle_t *hdl = zhp->zpool_hdl;
1071         char msg[1024];
1072
1073         if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1074             (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1075                 return (-1);
1076
1077         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1078
1079         if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1080                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1081                     "cannot destroy '%s'"), zhp->zpool_name);
1082
1083                 if (errno == EROFS) {
1084                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1085                             "one or more devices is read only"));
1086                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1087                 } else {
1088                         (void) zpool_standard_error(hdl, errno, msg);
1089                 }
1090
1091                 if (zfp)
1092                         zfs_close(zfp);
1093                 return (-1);
1094         }
1095
1096         if (zfp) {
1097                 remove_mountpoint(zfp);
1098                 zfs_close(zfp);
1099         }
1100
1101         return (0);
1102 }
1103
1104 /*
1105  * Add the given vdevs to the pool.  The caller must have already performed the
1106  * necessary verification to ensure that the vdev specification is well-formed.
1107  */
1108 int
1109 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1110 {
1111         zfs_cmd_t zc = { 0 };
1112         int ret;
1113         libzfs_handle_t *hdl = zhp->zpool_hdl;
1114         char msg[1024];
1115         nvlist_t **spares, **l2cache;
1116         uint_t nspares, nl2cache;
1117
1118         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1119             "cannot add to '%s'"), zhp->zpool_name);
1120
1121         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1122             SPA_VERSION_SPARES &&
1123             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1124             &spares, &nspares) == 0) {
1125                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1126                     "upgraded to add hot spares"));
1127                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1128         }
1129
1130         if (zpool_is_bootable(zhp) && nvlist_lookup_nvlist_array(nvroot,
1131             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) {
1132                 uint64_t s;
1133
1134                 for (s = 0; s < nspares; s++) {
1135                         char *path;
1136
1137                         if (nvlist_lookup_string(spares[s], ZPOOL_CONFIG_PATH,
1138                             &path) == 0 && pool_uses_efi(spares[s])) {
1139                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1140                                     "device '%s' contains an EFI label and "
1141                                     "cannot be used on root pools."),
1142                                     zpool_vdev_name(hdl, NULL, spares[s],
1143                                     B_FALSE));
1144                                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
1145                         }
1146                 }
1147         }
1148
1149         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1150             SPA_VERSION_L2CACHE &&
1151             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1152             &l2cache, &nl2cache) == 0) {
1153                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1154                     "upgraded to add cache devices"));
1155                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1156         }
1157
1158         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1159                 return (-1);
1160         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1161
1162         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1163                 switch (errno) {
1164                 case EBUSY:
1165                         /*
1166                          * This can happen if the user has specified the same
1167                          * device multiple times.  We can't reliably detect this
1168                          * until we try to add it and see we already have a
1169                          * label.
1170                          */
1171                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1172                             "one or more vdevs refer to the same device"));
1173                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1174                         break;
1175
1176                 case EOVERFLOW:
1177                         /*
1178                          * This occurrs when one of the devices is below
1179                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1180                          * device was the problem device since there's no
1181                          * reliable way to determine device size from userland.
1182                          */
1183                         {
1184                                 char buf[64];
1185
1186                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1187
1188                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1189                                     "device is less than the minimum "
1190                                     "size (%s)"), buf);
1191                         }
1192                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1193                         break;
1194
1195                 case ENOTSUP:
1196                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1197                             "pool must be upgraded to add these vdevs"));
1198                         (void) zfs_error(hdl, EZFS_BADVERSION, msg);
1199                         break;
1200
1201                 case EDOM:
1202                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1203                             "root pool can not have multiple vdevs"
1204                             " or separate logs"));
1205                         (void) zfs_error(hdl, EZFS_POOL_NOTSUP, msg);
1206                         break;
1207
1208                 case ENOTBLK:
1209                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1210                             "cache device must be a disk or disk slice"));
1211                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1212                         break;
1213
1214                 default:
1215                         (void) zpool_standard_error(hdl, errno, msg);
1216                 }
1217
1218                 ret = -1;
1219         } else {
1220                 ret = 0;
1221         }
1222
1223         zcmd_free_nvlists(&zc);
1224
1225         return (ret);
1226 }
1227
1228 /*
1229  * Exports the pool from the system.  The caller must ensure that there are no
1230  * mounted datasets in the pool.
1231  */
1232 int
1233 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce)
1234 {
1235         zfs_cmd_t zc = { 0 };
1236         char msg[1024];
1237
1238         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1239             "cannot export '%s'"), zhp->zpool_name);
1240
1241         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1242         zc.zc_cookie = force;
1243         zc.zc_guid = hardforce;
1244
1245         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1246                 switch (errno) {
1247                 case EXDEV:
1248                         zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1249                             "use '-f' to override the following errors:\n"
1250                             "'%s' has an active shared spare which could be"
1251                             " used by other pools once '%s' is exported."),
1252                             zhp->zpool_name, zhp->zpool_name);
1253                         return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1254                             msg));
1255                 default:
1256                         return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1257                             msg));
1258                 }
1259         }
1260
1261         return (0);
1262 }
1263
1264 int
1265 zpool_export(zpool_handle_t *zhp, boolean_t force)
1266 {
1267         return (zpool_export_common(zhp, force, B_FALSE));
1268 }
1269
1270 int
1271 zpool_export_force(zpool_handle_t *zhp)
1272 {
1273         return (zpool_export_common(zhp, B_TRUE, B_TRUE));
1274 }
1275
1276 static void
1277 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1278     nvlist_t *config)
1279 {
1280         nvlist_t *nv = NULL;
1281         uint64_t rewindto;
1282         int64_t loss = -1;
1283         struct tm t;
1284         char timestr[128];
1285
1286         if (!hdl->libzfs_printerr || config == NULL)
1287                 return;
1288
1289         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0)
1290                 return;
1291
1292         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1293                 return;
1294         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1295
1296         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1297             strftime(timestr, 128, 0, &t) != 0) {
1298                 if (dryrun) {
1299                         (void) printf(dgettext(TEXT_DOMAIN,
1300                             "Would be able to return %s "
1301                             "to its state as of %s.\n"),
1302                             name, timestr);
1303                 } else {
1304                         (void) printf(dgettext(TEXT_DOMAIN,
1305                             "Pool %s returned to its state as of %s.\n"),
1306                             name, timestr);
1307                 }
1308                 if (loss > 120) {
1309                         (void) printf(dgettext(TEXT_DOMAIN,
1310                             "%s approximately %lld "),
1311                             dryrun ? "Would discard" : "Discarded",
1312                             (loss + 30) / 60);
1313                         (void) printf(dgettext(TEXT_DOMAIN,
1314                             "minutes of transactions.\n"));
1315                 } else if (loss > 0) {
1316                         (void) printf(dgettext(TEXT_DOMAIN,
1317                             "%s approximately %lld "),
1318                             dryrun ? "Would discard" : "Discarded", loss);
1319                         (void) printf(dgettext(TEXT_DOMAIN,
1320                             "seconds of transactions.\n"));
1321                 }
1322         }
1323 }
1324
1325 void
1326 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1327     nvlist_t *config)
1328 {
1329         nvlist_t *nv = NULL;
1330         int64_t loss = -1;
1331         uint64_t edata = UINT64_MAX;
1332         uint64_t rewindto;
1333         struct tm t;
1334         char timestr[128];
1335
1336         if (!hdl->libzfs_printerr)
1337                 return;
1338
1339         if (reason >= 0)
1340                 (void) printf(dgettext(TEXT_DOMAIN, "action: "));
1341         else
1342                 (void) printf(dgettext(TEXT_DOMAIN, "\t"));
1343
1344         /* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1345         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1346             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1347                 goto no_info;
1348
1349         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1350         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1351             &edata);
1352
1353         (void) printf(dgettext(TEXT_DOMAIN,
1354             "Recovery is possible, but will result in some data loss.\n"));
1355
1356         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1357             strftime(timestr, 128, 0, &t) != 0) {
1358                 (void) printf(dgettext(TEXT_DOMAIN,
1359                     "\tReturning the pool to its state as of %s\n"
1360                     "\tshould correct the problem.  "),
1361                     timestr);
1362         } else {
1363                 (void) printf(dgettext(TEXT_DOMAIN,
1364                     "\tReverting the pool to an earlier state "
1365                     "should correct the problem.\n\t"));
1366         }
1367
1368         if (loss > 120) {
1369                 (void) printf(dgettext(TEXT_DOMAIN,
1370                     "Approximately %lld minutes of data\n"
1371                     "\tmust be discarded, irreversibly.  "), (loss + 30) / 60);
1372         } else if (loss > 0) {
1373                 (void) printf(dgettext(TEXT_DOMAIN,
1374                     "Approximately %lld seconds of data\n"
1375                     "\tmust be discarded, irreversibly.  "), loss);
1376         }
1377         if (edata != 0 && edata != UINT64_MAX) {
1378                 if (edata == 1) {
1379                         (void) printf(dgettext(TEXT_DOMAIN,
1380                             "After rewind, at least\n"
1381                             "\tone persistent user-data error will remain.  "));
1382                 } else {
1383                         (void) printf(dgettext(TEXT_DOMAIN,
1384                             "After rewind, several\n"
1385                             "\tpersistent user-data errors will remain.  "));
1386                 }
1387         }
1388         (void) printf(dgettext(TEXT_DOMAIN,
1389             "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1390             reason >= 0 ? "clear" : "import", name);
1391
1392         (void) printf(dgettext(TEXT_DOMAIN,
1393             "A scrub of the pool\n"
1394             "\tis strongly recommended after recovery.\n"));
1395         return;
1396
1397 no_info:
1398         (void) printf(dgettext(TEXT_DOMAIN,
1399             "Destroy and re-create the pool from\n\ta backup source.\n"));
1400 }
1401
1402 /*
1403  * zpool_import() is a contracted interface. Should be kept the same
1404  * if possible.
1405  *
1406  * Applications should use zpool_import_props() to import a pool with
1407  * new properties value to be set.
1408  */
1409 int
1410 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1411     char *altroot)
1412 {
1413         nvlist_t *props = NULL;
1414         int ret;
1415
1416         if (altroot != NULL) {
1417                 if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1418                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1419                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1420                             newname));
1421                 }
1422
1423                 if (nvlist_add_string(props,
1424                     zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1425                     nvlist_add_string(props,
1426                     zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1427                         nvlist_free(props);
1428                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1429                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1430                             newname));
1431                 }
1432         }
1433
1434         ret = zpool_import_props(hdl, config, newname, props,
1435             ZFS_IMPORT_NORMAL);
1436         if (props)
1437                 nvlist_free(props);
1438         return (ret);
1439 }
1440
1441 static void
1442 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1443     int indent)
1444 {
1445         nvlist_t **child;
1446         uint_t c, children;
1447         char *vname;
1448         uint64_t is_log = 0;
1449
1450         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1451             &is_log);
1452
1453         if (name != NULL)
1454                 (void) printf("\t%*s%s%s\n", indent, "", name,
1455                     is_log ? " [log]" : "");
1456
1457         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1458             &child, &children) != 0)
1459                 return;
1460
1461         for (c = 0; c < children; c++) {
1462                 vname = zpool_vdev_name(hdl, NULL, child[c], B_TRUE);
1463                 print_vdev_tree(hdl, vname, child[c], indent + 2);
1464                 free(vname);
1465         }
1466 }
1467
1468 /*
1469  * Import the given pool using the known configuration and a list of
1470  * properties to be set. The configuration should have come from
1471  * zpool_find_import(). The 'newname' parameters control whether the pool
1472  * is imported with a different name.
1473  */
1474 int
1475 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1476     nvlist_t *props, int flags)
1477 {
1478         zfs_cmd_t zc = { 0 };
1479         zpool_rewind_policy_t policy;
1480         nvlist_t *nv = NULL;
1481         nvlist_t *nvinfo = NULL;
1482         nvlist_t *missing = NULL;
1483         char *thename;
1484         char *origname;
1485         int ret;
1486         int error = 0;
1487         char errbuf[1024];
1488
1489         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1490             &origname) == 0);
1491
1492         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1493             "cannot import pool '%s'"), origname);
1494
1495         if (newname != NULL) {
1496                 if (!zpool_name_valid(hdl, B_FALSE, newname))
1497                         return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1498                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1499                             newname));
1500                 thename = (char *)newname;
1501         } else {
1502                 thename = origname;
1503         }
1504
1505         if (props) {
1506                 uint64_t version;
1507                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1508
1509                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1510                     &version) == 0);
1511
1512                 if ((props = zpool_valid_proplist(hdl, origname,
1513                     props, version, flags, errbuf)) == NULL) {
1514                         return (-1);
1515                 } else if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1516                         nvlist_free(props);
1517                         return (-1);
1518                 }
1519         }
1520
1521         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1522
1523         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1524             &zc.zc_guid) == 0);
1525
1526         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1527                 nvlist_free(props);
1528                 return (-1);
1529         }
1530         if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1531                 nvlist_free(props);
1532                 return (-1);
1533         }
1534
1535         zc.zc_cookie = flags;
1536         while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1537             errno == ENOMEM) {
1538                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1539                         zcmd_free_nvlists(&zc);
1540                         return (-1);
1541                 }
1542         }
1543         if (ret != 0)
1544                 error = errno;
1545
1546         (void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1547         zpool_get_rewind_policy(config, &policy);
1548
1549         if (error) {
1550                 char desc[1024];
1551
1552                 /*
1553                  * Dry-run failed, but we print out what success
1554                  * looks like if we found a best txg
1555                  */
1556                 if (policy.zrp_request & ZPOOL_TRY_REWIND) {
1557                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1558                             B_TRUE, nv);
1559                         nvlist_free(nv);
1560                         return (-1);
1561                 }
1562
1563                 if (newname == NULL)
1564                         (void) snprintf(desc, sizeof (desc),
1565                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1566                             thename);
1567                 else
1568                         (void) snprintf(desc, sizeof (desc),
1569                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1570                             origname, thename);
1571
1572                 switch (error) {
1573                 case ENOTSUP:
1574                         /*
1575                          * Unsupported version.
1576                          */
1577                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1578                         break;
1579
1580                 case EINVAL:
1581                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1582                         break;
1583
1584                 case EROFS:
1585                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1586                             "one or more devices is read only"));
1587                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1588                         break;
1589
1590                 case ENXIO:
1591                         if (nv && nvlist_lookup_nvlist(nv,
1592                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1593                             nvlist_lookup_nvlist(nvinfo,
1594                             ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1595                                 (void) printf(dgettext(TEXT_DOMAIN,
1596                                     "The devices below are missing, use "
1597                                     "'-m' to import the pool anyway:\n"));
1598                                 print_vdev_tree(hdl, NULL, missing, 2);
1599                                 (void) printf("\n");
1600                         }
1601                         (void) zpool_standard_error(hdl, error, desc);
1602                         break;
1603
1604                 case EEXIST:
1605                         (void) zpool_standard_error(hdl, error, desc);
1606                         break;
1607
1608                 default:
1609                         (void) zpool_standard_error(hdl, error, desc);
1610                         zpool_explain_recover(hdl,
1611                             newname ? origname : thename, -error, nv);
1612                         break;
1613                 }
1614
1615                 nvlist_free(nv);
1616                 ret = -1;
1617         } else {
1618                 zpool_handle_t *zhp;
1619
1620                 /*
1621                  * This should never fail, but play it safe anyway.
1622                  */
1623                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1624                         ret = -1;
1625                 else if (zhp != NULL)
1626                         zpool_close(zhp);
1627                 if (policy.zrp_request &
1628                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1629                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1630                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1631                 }
1632                 nvlist_free(nv);
1633                 return (0);
1634         }
1635
1636         zcmd_free_nvlists(&zc);
1637         nvlist_free(props);
1638
1639         return (ret);
1640 }
1641
1642 /*
1643  * Scan the pool.
1644  */
1645 int
1646 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1647 {
1648         zfs_cmd_t zc = { 0 };
1649         char msg[1024];
1650         libzfs_handle_t *hdl = zhp->zpool_hdl;
1651
1652         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1653         zc.zc_cookie = func;
1654
1655         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1656             (errno == ENOENT && func != POOL_SCAN_NONE))
1657                 return (0);
1658
1659         if (func == POOL_SCAN_SCRUB) {
1660                 (void) snprintf(msg, sizeof (msg),
1661                     dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1662         } else if (func == POOL_SCAN_NONE) {
1663                 (void) snprintf(msg, sizeof (msg),
1664                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1665                     zc.zc_name);
1666         } else {
1667                 assert(!"unexpected result");
1668         }
1669
1670         if (errno == EBUSY) {
1671                 nvlist_t *nvroot;
1672                 pool_scan_stat_t *ps = NULL;
1673                 uint_t psc;
1674
1675                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
1676                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1677                 (void) nvlist_lookup_uint64_array(nvroot,
1678                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1679                 if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1680                         return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1681                 else
1682                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
1683         } else if (errno == ENOENT) {
1684                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1685         } else {
1686                 return (zpool_standard_error(hdl, errno, msg));
1687         }
1688 }
1689
1690 /*
1691  * This provides a very minimal check whether a given string is likely a
1692  * c#t#d# style string.  Users of this are expected to do their own
1693  * verification of the s# part.
1694  */
1695 #define CTD_CHECK(str)  (str && str[0] == 'c' && isdigit(str[1]))
1696
1697 /*
1698  * More elaborate version for ones which may start with "/dev/dsk/"
1699  * and the like.
1700  */
1701 static int
1702 ctd_check_path(char *str) {
1703         /*
1704          * If it starts with a slash, check the last component.
1705          */
1706         if (str && str[0] == '/') {
1707                 char *tmp = strrchr(str, '/');
1708
1709                 /*
1710                  * If it ends in "/old", check the second-to-last
1711                  * component of the string instead.
1712                  */
1713                 if (tmp != str && strcmp(tmp, "/old") == 0) {
1714                         for (tmp--; *tmp != '/'; tmp--)
1715                                 ;
1716                 }
1717                 str = tmp + 1;
1718         }
1719         return (CTD_CHECK(str));
1720 }
1721
1722 /*
1723  * Find a vdev that matches the search criteria specified. We use the
1724  * the nvpair name to determine how we should look for the device.
1725  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1726  * spare; but FALSE if its an INUSE spare.
1727  */
1728 static nvlist_t *
1729 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1730     boolean_t *l2cache, boolean_t *log)
1731 {
1732         uint_t c, children;
1733         nvlist_t **child;
1734         nvlist_t *ret;
1735         uint64_t is_log;
1736         char *srchkey;
1737         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
1738
1739         /* Nothing to look for */
1740         if (search == NULL || pair == NULL)
1741                 return (NULL);
1742
1743         /* Obtain the key we will use to search */
1744         srchkey = nvpair_name(pair);
1745
1746         switch (nvpair_type(pair)) {
1747         case DATA_TYPE_UINT64:
1748                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
1749                         uint64_t srchval, theguid;
1750
1751                         verify(nvpair_value_uint64(pair, &srchval) == 0);
1752                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
1753                             &theguid) == 0);
1754                         if (theguid == srchval)
1755                                 return (nv);
1756                 }
1757                 break;
1758
1759         case DATA_TYPE_STRING: {
1760                 char *srchval, *val;
1761
1762                 verify(nvpair_value_string(pair, &srchval) == 0);
1763                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
1764                         break;
1765
1766                 /*
1767                  * Search for the requested value. Special cases:
1768                  *
1769                  * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
1770                  *   "s0" or "s0/old".  The "s0" part is hidden from the user,
1771                  *   but included in the string, so this matches around it.
1772                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
1773                  *
1774                  * Otherwise, all other searches are simple string compares.
1775                  */
1776                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0 &&
1777                     ctd_check_path(val)) {
1778                         uint64_t wholedisk = 0;
1779
1780                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1781                             &wholedisk);
1782                         if (wholedisk) {
1783                                 int slen = strlen(srchval);
1784                                 int vlen = strlen(val);
1785
1786                                 if (slen != vlen - 2)
1787                                         break;
1788
1789                                 /*
1790                                  * make_leaf_vdev() should only set
1791                                  * wholedisk for ZPOOL_CONFIG_PATHs which
1792                                  * will include "/dev/dsk/", giving plenty of
1793                                  * room for the indices used next.
1794                                  */
1795                                 ASSERT(vlen >= 6);
1796
1797                                 /*
1798                                  * strings identical except trailing "s0"
1799                                  */
1800                                 if (strcmp(&val[vlen - 2], "s0") == 0 &&
1801                                     strncmp(srchval, val, slen) == 0)
1802                                         return (nv);
1803
1804                                 /*
1805                                  * strings identical except trailing "s0/old"
1806                                  */
1807                                 if (strcmp(&val[vlen - 6], "s0/old") == 0 &&
1808                                     strcmp(&srchval[slen - 4], "/old") == 0 &&
1809                                     strncmp(srchval, val, slen - 4) == 0)
1810                                         return (nv);
1811
1812                                 break;
1813                         }
1814                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
1815                         char *type, *idx, *end, *p;
1816                         uint64_t id, vdev_id;
1817
1818                         /*
1819                          * Determine our vdev type, keeping in mind
1820                          * that the srchval is composed of a type and
1821                          * vdev id pair (i.e. mirror-4).
1822                          */
1823                         if ((type = strdup(srchval)) == NULL)
1824                                 return (NULL);
1825
1826                         if ((p = strrchr(type, '-')) == NULL) {
1827                                 free(type);
1828                                 break;
1829                         }
1830                         idx = p + 1;
1831                         *p = '\0';
1832
1833                         /*
1834                          * If the types don't match then keep looking.
1835                          */
1836                         if (strncmp(val, type, strlen(val)) != 0) {
1837                                 free(type);
1838                                 break;
1839                         }
1840
1841                         verify(strncmp(type, VDEV_TYPE_RAIDZ,
1842                             strlen(VDEV_TYPE_RAIDZ)) == 0 ||
1843                             strncmp(type, VDEV_TYPE_MIRROR,
1844                             strlen(VDEV_TYPE_MIRROR)) == 0);
1845                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
1846                             &id) == 0);
1847
1848                         errno = 0;
1849                         vdev_id = strtoull(idx, &end, 10);
1850
1851                         free(type);
1852                         if (errno != 0)
1853                                 return (NULL);
1854
1855                         /*
1856                          * Now verify that we have the correct vdev id.
1857                          */
1858                         if (vdev_id == id)
1859                                 return (nv);
1860                 }
1861
1862                 /*
1863                  * Common case
1864                  */
1865                 if (strcmp(srchval, val) == 0)
1866                         return (nv);
1867                 break;
1868         }
1869
1870         default:
1871                 break;
1872         }
1873
1874         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1875             &child, &children) != 0)
1876                 return (NULL);
1877
1878         for (c = 0; c < children; c++) {
1879                 if ((ret = vdev_to_nvlist_iter(child[c], search,
1880                     avail_spare, l2cache, NULL)) != NULL) {
1881                         /*
1882                          * The 'is_log' value is only set for the toplevel
1883                          * vdev, not the leaf vdevs.  So we always lookup the
1884                          * log device from the root of the vdev tree (where
1885                          * 'log' is non-NULL).
1886                          */
1887                         if (log != NULL &&
1888                             nvlist_lookup_uint64(child[c],
1889                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
1890                             is_log) {
1891                                 *log = B_TRUE;
1892                         }
1893                         return (ret);
1894                 }
1895         }
1896
1897         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1898             &child, &children) == 0) {
1899                 for (c = 0; c < children; c++) {
1900                         if ((ret = vdev_to_nvlist_iter(child[c], search,
1901                             avail_spare, l2cache, NULL)) != NULL) {
1902                                 *avail_spare = B_TRUE;
1903                                 return (ret);
1904                         }
1905                 }
1906         }
1907
1908         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1909             &child, &children) == 0) {
1910                 for (c = 0; c < children; c++) {
1911                         if ((ret = vdev_to_nvlist_iter(child[c], search,
1912                             avail_spare, l2cache, NULL)) != NULL) {
1913                                 *l2cache = B_TRUE;
1914                                 return (ret);
1915                         }
1916                 }
1917         }
1918
1919         return (NULL);
1920 }
1921
1922 /*
1923  * Given a physical path (minus the "/devices" prefix), find the
1924  * associated vdev.
1925  */
1926 nvlist_t *
1927 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
1928     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
1929 {
1930         nvlist_t *search, *nvroot, *ret;
1931
1932         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1933         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
1934
1935         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1936             &nvroot) == 0);
1937
1938         *avail_spare = B_FALSE;
1939         *l2cache = B_FALSE;
1940         if (log != NULL)
1941                 *log = B_FALSE;
1942         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
1943         nvlist_free(search);
1944
1945         return (ret);
1946 }
1947
1948 /*
1949  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
1950  */
1951 boolean_t
1952 zpool_vdev_is_interior(const char *name)
1953 {
1954         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
1955             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
1956                 return (B_TRUE);
1957         return (B_FALSE);
1958 }
1959
1960 nvlist_t *
1961 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
1962     boolean_t *l2cache, boolean_t *log)
1963 {
1964         char buf[MAXPATHLEN];
1965         char *end;
1966         nvlist_t *nvroot, *search, *ret;
1967         uint64_t guid;
1968
1969         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1970
1971         guid = strtoull(path, &end, 10);
1972         if (guid != 0 && *end == '\0') {
1973                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
1974         } else if (zpool_vdev_is_interior(path)) {
1975                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
1976         } else if (path[0] != '/') {
1977                 (void) snprintf(buf, sizeof (buf), "%s%s", _PATH_DEV, path);
1978                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, buf) == 0);
1979         } else {
1980                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
1981         }
1982
1983         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1984             &nvroot) == 0);
1985
1986         *avail_spare = B_FALSE;
1987         *l2cache = B_FALSE;
1988         if (log != NULL)
1989                 *log = B_FALSE;
1990         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
1991         nvlist_free(search);
1992
1993         return (ret);
1994 }
1995
1996 static int
1997 vdev_online(nvlist_t *nv)
1998 {
1999         uint64_t ival;
2000
2001         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2002             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2003             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2004                 return (0);
2005
2006         return (1);
2007 }
2008
2009 /*
2010  * Helper function for zpool_get_physpaths().
2011  */
2012 static int
2013 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2014     size_t *bytes_written)
2015 {
2016         size_t bytes_left, pos, rsz;
2017         char *tmppath;
2018         const char *format;
2019
2020         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2021             &tmppath) != 0)
2022                 return (EZFS_NODEVICE);
2023
2024         pos = *bytes_written;
2025         bytes_left = physpath_size - pos;
2026         format = (pos == 0) ? "%s" : " %s";
2027
2028         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2029         *bytes_written += rsz;
2030
2031         if (rsz >= bytes_left) {
2032                 /* if physpath was not copied properly, clear it */
2033                 if (bytes_left != 0) {
2034                         physpath[pos] = 0;
2035                 }
2036                 return (EZFS_NOSPC);
2037         }
2038         return (0);
2039 }
2040
2041 static int
2042 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2043     size_t *rsz, boolean_t is_spare)
2044 {
2045         char *type;
2046         int ret;
2047
2048         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2049                 return (EZFS_INVALCONFIG);
2050
2051         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2052                 /*
2053                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2054                  * For a spare vdev, we only want to boot from the active
2055                  * spare device.
2056                  */
2057                 if (is_spare) {
2058                         uint64_t spare = 0;
2059                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2060                             &spare);
2061                         if (!spare)
2062                                 return (EZFS_INVALCONFIG);
2063                 }
2064
2065                 if (vdev_online(nv)) {
2066                         if ((ret = vdev_get_one_physpath(nv, physpath,
2067                             phypath_size, rsz)) != 0)
2068                                 return (ret);
2069                 }
2070         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2071             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2072             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2073                 nvlist_t **child;
2074                 uint_t count;
2075                 int i, ret;
2076
2077                 if (nvlist_lookup_nvlist_array(nv,
2078                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2079                         return (EZFS_INVALCONFIG);
2080
2081                 for (i = 0; i < count; i++) {
2082                         ret = vdev_get_physpaths(child[i], physpath,
2083                             phypath_size, rsz, is_spare);
2084                         if (ret == EZFS_NOSPC)
2085                                 return (ret);
2086                 }
2087         }
2088
2089         return (EZFS_POOL_INVALARG);
2090 }
2091
2092 /*
2093  * Get phys_path for a root pool config.
2094  * Return 0 on success; non-zero on failure.
2095  */
2096 static int
2097 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2098 {
2099         size_t rsz;
2100         nvlist_t *vdev_root;
2101         nvlist_t **child;
2102         uint_t count;
2103         char *type;
2104
2105         rsz = 0;
2106
2107         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2108             &vdev_root) != 0)
2109                 return (EZFS_INVALCONFIG);
2110
2111         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2112             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2113             &child, &count) != 0)
2114                 return (EZFS_INVALCONFIG);
2115
2116         /*
2117          * root pool can not have EFI labeled disks and can only have
2118          * a single top-level vdev.
2119          */
2120         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1 ||
2121             pool_uses_efi(vdev_root))
2122                 return (EZFS_POOL_INVALARG);
2123
2124         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2125             B_FALSE);
2126
2127         /* No online devices */
2128         if (rsz == 0)
2129                 return (EZFS_NODEVICE);
2130
2131         return (0);
2132 }
2133
2134 /*
2135  * Get phys_path for a root pool
2136  * Return 0 on success; non-zero on failure.
2137  */
2138 int
2139 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2140 {
2141         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2142             phypath_size));
2143 }
2144
2145 /*
2146  * If the device has being dynamically expanded then we need to relabel
2147  * the disk to use the new unallocated space.
2148  */
2149 static int
2150 zpool_relabel_disk(libzfs_handle_t *hdl, const char *name)
2151 {
2152 #ifdef sun
2153         char path[MAXPATHLEN];
2154         char errbuf[1024];
2155         int fd, error;
2156         int (*_efi_use_whole_disk)(int);
2157
2158         if ((_efi_use_whole_disk = (int (*)(int))dlsym(RTLD_DEFAULT,
2159             "efi_use_whole_disk")) == NULL)
2160                 return (-1);
2161
2162         (void) snprintf(path, sizeof (path), "%s/%s", RDISK_ROOT, name);
2163
2164         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
2165                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2166                     "relabel '%s': unable to open device"), name);
2167                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
2168         }
2169
2170         /*
2171          * It's possible that we might encounter an error if the device
2172          * does not have any unallocated space left. If so, we simply
2173          * ignore that error and continue on.
2174          */
2175         error = _efi_use_whole_disk(fd);
2176         (void) close(fd);
2177         if (error && error != VT_ENOSPC) {
2178                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2179                     "relabel '%s': unable to read disk capacity"), name);
2180                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
2181         }
2182 #endif  /* sun */
2183         return (0);
2184 }
2185
2186 /*
2187  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2188  * ZFS_ONLINE_* flags.
2189  */
2190 int
2191 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2192     vdev_state_t *newstate)
2193 {
2194         zfs_cmd_t zc = { 0 };
2195         char msg[1024];
2196         nvlist_t *tgt;
2197         boolean_t avail_spare, l2cache, islog;
2198         libzfs_handle_t *hdl = zhp->zpool_hdl;
2199
2200         if (flags & ZFS_ONLINE_EXPAND) {
2201                 (void) snprintf(msg, sizeof (msg),
2202                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2203         } else {
2204                 (void) snprintf(msg, sizeof (msg),
2205                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2206         }
2207
2208         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2209         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2210             &islog)) == NULL)
2211                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2212
2213         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2214
2215         if (avail_spare)
2216                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2217
2218         if (flags & ZFS_ONLINE_EXPAND ||
2219             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2220                 char *pathname = NULL;
2221                 uint64_t wholedisk = 0;
2222
2223                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2224                     &wholedisk);
2225                 verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
2226                     &pathname) == 0);
2227
2228                 /*
2229                  * XXX - L2ARC 1.0 devices can't support expansion.
2230                  */
2231                 if (l2cache) {
2232                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2233                             "cannot expand cache devices"));
2234                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2235                 }
2236
2237                 if (wholedisk) {
2238                         pathname += strlen(DISK_ROOT) + 1;
2239                         (void) zpool_relabel_disk(hdl, pathname);
2240                 }
2241         }
2242
2243         zc.zc_cookie = VDEV_STATE_ONLINE;
2244         zc.zc_obj = flags;
2245
2246         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2247                 if (errno == EINVAL) {
2248                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2249                             "from this pool into a new one.  Use '%s' "
2250                             "instead"), "zpool detach");
2251                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2252                 }
2253                 return (zpool_standard_error(hdl, errno, msg));
2254         }
2255
2256         *newstate = zc.zc_cookie;
2257         return (0);
2258 }
2259
2260 /*
2261  * Take the specified vdev offline
2262  */
2263 int
2264 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2265 {
2266         zfs_cmd_t zc = { 0 };
2267         char msg[1024];
2268         nvlist_t *tgt;
2269         boolean_t avail_spare, l2cache;
2270         libzfs_handle_t *hdl = zhp->zpool_hdl;
2271
2272         (void) snprintf(msg, sizeof (msg),
2273             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2274
2275         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2276         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2277             NULL)) == NULL)
2278                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2279
2280         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2281
2282         if (avail_spare)
2283                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2284
2285         zc.zc_cookie = VDEV_STATE_OFFLINE;
2286         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2287
2288         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2289                 return (0);
2290
2291         switch (errno) {
2292         case EBUSY:
2293
2294                 /*
2295                  * There are no other replicas of this device.
2296                  */
2297                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2298
2299         case EEXIST:
2300                 /*
2301                  * The log device has unplayed logs
2302                  */
2303                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2304
2305         default:
2306                 return (zpool_standard_error(hdl, errno, msg));
2307         }
2308 }
2309
2310 /*
2311  * Mark the given vdev faulted.
2312  */
2313 int
2314 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2315 {
2316         zfs_cmd_t zc = { 0 };
2317         char msg[1024];
2318         libzfs_handle_t *hdl = zhp->zpool_hdl;
2319
2320         (void) snprintf(msg, sizeof (msg),
2321             dgettext(TEXT_DOMAIN, "cannot fault %llu"), guid);
2322
2323         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2324         zc.zc_guid = guid;
2325         zc.zc_cookie = VDEV_STATE_FAULTED;
2326         zc.zc_obj = aux;
2327
2328         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2329                 return (0);
2330
2331         switch (errno) {
2332         case EBUSY:
2333
2334                 /*
2335                  * There are no other replicas of this device.
2336                  */
2337                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2338
2339         default:
2340                 return (zpool_standard_error(hdl, errno, msg));
2341         }
2342
2343 }
2344
2345 /*
2346  * Mark the given vdev degraded.
2347  */
2348 int
2349 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2350 {
2351         zfs_cmd_t zc = { 0 };
2352         char msg[1024];
2353         libzfs_handle_t *hdl = zhp->zpool_hdl;
2354
2355         (void) snprintf(msg, sizeof (msg),
2356             dgettext(TEXT_DOMAIN, "cannot degrade %llu"), guid);
2357
2358         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2359         zc.zc_guid = guid;
2360         zc.zc_cookie = VDEV_STATE_DEGRADED;
2361         zc.zc_obj = aux;
2362
2363         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2364                 return (0);
2365
2366         return (zpool_standard_error(hdl, errno, msg));
2367 }
2368
2369 /*
2370  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2371  * a hot spare.
2372  */
2373 static boolean_t
2374 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2375 {
2376         nvlist_t **child;
2377         uint_t c, children;
2378         char *type;
2379
2380         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2381             &children) == 0) {
2382                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2383                     &type) == 0);
2384
2385                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2386                     children == 2 && child[which] == tgt)
2387                         return (B_TRUE);
2388
2389                 for (c = 0; c < children; c++)
2390                         if (is_replacing_spare(child[c], tgt, which))
2391                                 return (B_TRUE);
2392         }
2393
2394         return (B_FALSE);
2395 }
2396
2397 /*
2398  * Attach new_disk (fully described by nvroot) to old_disk.
2399  * If 'replacing' is specified, the new disk will replace the old one.
2400  */
2401 int
2402 zpool_vdev_attach(zpool_handle_t *zhp,
2403     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2404 {
2405         zfs_cmd_t zc = { 0 };
2406         char msg[1024];
2407         int ret;
2408         nvlist_t *tgt;
2409         boolean_t avail_spare, l2cache, islog;
2410         uint64_t val;
2411         char *newname;
2412         nvlist_t **child;
2413         uint_t children;
2414         nvlist_t *config_root;
2415         libzfs_handle_t *hdl = zhp->zpool_hdl;
2416         boolean_t rootpool = zpool_is_bootable(zhp);
2417
2418         if (replacing)
2419                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2420                     "cannot replace %s with %s"), old_disk, new_disk);
2421         else
2422                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2423                     "cannot attach %s to %s"), new_disk, old_disk);
2424
2425         /*
2426          * If this is a root pool, make sure that we're not attaching an
2427          * EFI labeled device.
2428          */
2429         if (rootpool && pool_uses_efi(nvroot)) {
2430                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2431                     "EFI labeled devices are not supported on root pools."));
2432                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
2433         }
2434
2435         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2436         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2437             &islog)) == 0)
2438                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2439
2440         if (avail_spare)
2441                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2442
2443         if (l2cache)
2444                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2445
2446         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2447         zc.zc_cookie = replacing;
2448
2449         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2450             &child, &children) != 0 || children != 1) {
2451                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2452                     "new device must be a single disk"));
2453                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2454         }
2455
2456         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2457             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2458
2459         if ((newname = zpool_vdev_name(NULL, NULL, child[0], B_FALSE)) == NULL)
2460                 return (-1);
2461
2462         /*
2463          * If the target is a hot spare that has been swapped in, we can only
2464          * replace it with another hot spare.
2465          */
2466         if (replacing &&
2467             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2468             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2469             NULL) == NULL || !avail_spare) &&
2470             is_replacing_spare(config_root, tgt, 1)) {
2471                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2472                     "can only be replaced by another hot spare"));
2473                 free(newname);
2474                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2475         }
2476
2477         free(newname);
2478
2479         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2480                 return (-1);
2481
2482         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2483
2484         zcmd_free_nvlists(&zc);
2485
2486         if (ret == 0) {
2487                 if (rootpool) {
2488                         /*
2489                          * XXX need a better way to prevent user from
2490                          * booting up a half-baked vdev.
2491                          */
2492                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2493                             "sure to wait until resilver is done "
2494                             "before rebooting.\n"));
2495                         (void) fprintf(stderr, "\n");
2496                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "If "
2497                             "you boot from pool '%s', you may need to update\n"
2498                             "boot code on newly attached disk '%s'.\n\n"
2499                             "Assuming you use GPT partitioning and 'da0' is "
2500                             "your new boot disk\n"
2501                             "you may use the following command:\n\n"
2502                             "\tgpart bootcode -b /boot/pmbr -p "
2503                             "/boot/gptzfsboot -i 1 da0\n\n"),
2504                             zhp->zpool_name, new_disk);
2505                 }
2506                 return (0);
2507         }
2508
2509         switch (errno) {
2510         case ENOTSUP:
2511                 /*
2512                  * Can't attach to or replace this type of vdev.
2513                  */
2514                 if (replacing) {
2515                         uint64_t version = zpool_get_prop_int(zhp,
2516                             ZPOOL_PROP_VERSION, NULL);
2517
2518                         if (islog)
2519                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2520                                     "cannot replace a log with a spare"));
2521                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2522                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2523                                     "already in replacing/spare config; wait "
2524                                     "for completion or use 'zpool detach'"));
2525                         else
2526                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2527                                     "cannot replace a replacing device"));
2528                 } else {
2529                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2530                             "can only attach to mirrors and top-level "
2531                             "disks"));
2532                 }
2533                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2534                 break;
2535
2536         case EINVAL:
2537                 /*
2538                  * The new device must be a single disk.
2539                  */
2540                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2541                     "new device must be a single disk"));
2542                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2543                 break;
2544
2545         case EBUSY:
2546                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2547                     new_disk);
2548                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2549                 break;
2550
2551         case EOVERFLOW:
2552                 /*
2553                  * The new device is too small.
2554                  */
2555                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2556                     "device is too small"));
2557                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2558                 break;
2559
2560         case EDOM:
2561                 /*
2562                  * The new device has a different alignment requirement.
2563                  */
2564                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2565                     "devices have different sector alignment"));
2566                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2567                 break;
2568
2569         case ENAMETOOLONG:
2570                 /*
2571                  * The resulting top-level vdev spec won't fit in the label.
2572                  */
2573                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2574                 break;
2575
2576         default:
2577                 (void) zpool_standard_error(hdl, errno, msg);
2578         }
2579
2580         return (-1);
2581 }
2582
2583 /*
2584  * Detach the specified device.
2585  */
2586 int
2587 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2588 {
2589         zfs_cmd_t zc = { 0 };
2590         char msg[1024];
2591         nvlist_t *tgt;
2592         boolean_t avail_spare, l2cache;
2593         libzfs_handle_t *hdl = zhp->zpool_hdl;
2594
2595         (void) snprintf(msg, sizeof (msg),
2596             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2597
2598         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2599         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2600             NULL)) == 0)
2601                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2602
2603         if (avail_spare)
2604                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2605
2606         if (l2cache)
2607                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2608
2609         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2610
2611         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2612                 return (0);
2613
2614         switch (errno) {
2615
2616         case ENOTSUP:
2617                 /*
2618                  * Can't detach from this type of vdev.
2619                  */
2620                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2621                     "applicable to mirror and replacing vdevs"));
2622                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2623                 break;
2624
2625         case EBUSY:
2626                 /*
2627                  * There are no other replicas of this device.
2628                  */
2629                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2630                 break;
2631
2632         default:
2633                 (void) zpool_standard_error(hdl, errno, msg);
2634         }
2635
2636         return (-1);
2637 }
2638
2639 /*
2640  * Find a mirror vdev in the source nvlist.
2641  *
2642  * The mchild array contains a list of disks in one of the top-level mirrors
2643  * of the source pool.  The schild array contains a list of disks that the
2644  * user specified on the command line.  We loop over the mchild array to
2645  * see if any entry in the schild array matches.
2646  *
2647  * If a disk in the mchild array is found in the schild array, we return
2648  * the index of that entry.  Otherwise we return -1.
2649  */
2650 static int
2651 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2652     nvlist_t **schild, uint_t schildren)
2653 {
2654         uint_t mc;
2655
2656         for (mc = 0; mc < mchildren; mc++) {
2657                 uint_t sc;
2658                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2659                     mchild[mc], B_FALSE);
2660
2661                 for (sc = 0; sc < schildren; sc++) {
2662                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2663                             schild[sc], B_FALSE);
2664                         boolean_t result = (strcmp(mpath, spath) == 0);
2665
2666                         free(spath);
2667                         if (result) {
2668                                 free(mpath);
2669                                 return (mc);
2670                         }
2671                 }
2672
2673                 free(mpath);
2674         }
2675
2676         return (-1);
2677 }
2678
2679 /*
2680  * Split a mirror pool.  If newroot points to null, then a new nvlist
2681  * is generated and it is the responsibility of the caller to free it.
2682  */
2683 int
2684 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2685     nvlist_t *props, splitflags_t flags)
2686 {
2687         zfs_cmd_t zc = { 0 };
2688         char msg[1024];
2689         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2690         nvlist_t **varray = NULL, *zc_props = NULL;
2691         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2692         libzfs_handle_t *hdl = zhp->zpool_hdl;
2693         uint64_t vers;
2694         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2695         int retval = 0;
2696
2697         (void) snprintf(msg, sizeof (msg),
2698             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2699
2700         if (!zpool_name_valid(hdl, B_FALSE, newname))
2701                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2702
2703         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2704                 (void) fprintf(stderr, gettext("Internal error: unable to "
2705                     "retrieve pool configuration\n"));
2706                 return (-1);
2707         }
2708
2709         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2710             == 0);
2711         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2712
2713         if (props) {
2714                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2715                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2716                     props, vers, flags, msg)) == NULL)
2717                         return (-1);
2718         }
2719
2720         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2721             &children) != 0) {
2722                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2723                     "Source pool is missing vdev tree"));
2724                 if (zc_props)
2725                         nvlist_free(zc_props);
2726                 return (-1);
2727         }
2728
2729         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2730         vcount = 0;
2731
2732         if (*newroot == NULL ||
2733             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2734             &newchild, &newchildren) != 0)
2735                 newchildren = 0;
2736
2737         for (c = 0; c < children; c++) {
2738                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2739                 char *type;
2740                 nvlist_t **mchild, *vdev;
2741                 uint_t mchildren;
2742                 int entry;
2743
2744                 /*
2745                  * Unlike cache & spares, slogs are stored in the
2746                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2747                  */
2748                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2749                     &is_log);
2750                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2751                     &is_hole);
2752                 if (is_log || is_hole) {
2753                         /*
2754                          * Create a hole vdev and put it in the config.
2755                          */
2756                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2757                                 goto out;
2758                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2759                             VDEV_TYPE_HOLE) != 0)
2760                                 goto out;
2761                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2762                             1) != 0)
2763                                 goto out;
2764                         if (lastlog == 0)
2765                                 lastlog = vcount;
2766                         varray[vcount++] = vdev;
2767                         continue;
2768                 }
2769                 lastlog = 0;
2770                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
2771                     == 0);
2772                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
2773                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2774                             "Source pool must be composed only of mirrors\n"));
2775                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2776                         goto out;
2777                 }
2778
2779                 verify(nvlist_lookup_nvlist_array(child[c],
2780                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2781
2782                 /* find or add an entry for this top-level vdev */
2783                 if (newchildren > 0 &&
2784                     (entry = find_vdev_entry(zhp, mchild, mchildren,
2785                     newchild, newchildren)) >= 0) {
2786                         /* We found a disk that the user specified. */
2787                         vdev = mchild[entry];
2788                         ++found;
2789                 } else {
2790                         /* User didn't specify a disk for this vdev. */
2791                         vdev = mchild[mchildren - 1];
2792                 }
2793
2794                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
2795                         goto out;
2796         }
2797
2798         /* did we find every disk the user specified? */
2799         if (found != newchildren) {
2800                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
2801                     "include at most one disk from each mirror"));
2802                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2803                 goto out;
2804         }
2805
2806         /* Prepare the nvlist for populating. */
2807         if (*newroot == NULL) {
2808                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
2809                         goto out;
2810                 freelist = B_TRUE;
2811                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
2812                     VDEV_TYPE_ROOT) != 0)
2813                         goto out;
2814         } else {
2815                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
2816         }
2817
2818         /* Add all the children we found */
2819         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
2820             lastlog == 0 ? vcount : lastlog) != 0)
2821                 goto out;
2822
2823         /*
2824          * If we're just doing a dry run, exit now with success.
2825          */
2826         if (flags.dryrun) {
2827                 memory_err = B_FALSE;
2828                 freelist = B_FALSE;
2829                 goto out;
2830         }
2831
2832         /* now build up the config list & call the ioctl */
2833         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
2834                 goto out;
2835
2836         if (nvlist_add_nvlist(newconfig,
2837             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
2838             nvlist_add_string(newconfig,
2839             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
2840             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
2841                 goto out;
2842
2843         /*
2844          * The new pool is automatically part of the namespace unless we
2845          * explicitly export it.
2846          */
2847         if (!flags.import)
2848                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
2849         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2850         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
2851         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
2852                 goto out;
2853         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
2854                 goto out;
2855
2856         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
2857                 retval = zpool_standard_error(hdl, errno, msg);
2858                 goto out;
2859         }
2860
2861         freelist = B_FALSE;
2862         memory_err = B_FALSE;
2863
2864 out:
2865         if (varray != NULL) {
2866                 int v;
2867
2868                 for (v = 0; v < vcount; v++)
2869                         nvlist_free(varray[v]);
2870                 free(varray);
2871         }
2872         zcmd_free_nvlists(&zc);
2873         if (zc_props)
2874                 nvlist_free(zc_props);
2875         if (newconfig)
2876                 nvlist_free(newconfig);
2877         if (freelist) {
2878                 nvlist_free(*newroot);
2879                 *newroot = NULL;
2880         }
2881
2882         if (retval != 0)
2883                 return (retval);
2884
2885         if (memory_err)
2886                 return (no_memory(hdl));
2887
2888         return (0);
2889 }
2890
2891 /*
2892  * Remove the given device.  Currently, this is supported only for hot spares
2893  * and level 2 cache devices.
2894  */
2895 int
2896 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
2897 {
2898         zfs_cmd_t zc = { 0 };
2899         char msg[1024];
2900         nvlist_t *tgt;
2901         boolean_t avail_spare, l2cache, islog;
2902         libzfs_handle_t *hdl = zhp->zpool_hdl;
2903         uint64_t version;
2904
2905         (void) snprintf(msg, sizeof (msg),
2906             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
2907
2908         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2909         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2910             &islog)) == 0)
2911                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2912         /*
2913          * XXX - this should just go away.
2914          */
2915         if (!avail_spare && !l2cache && !islog) {
2916                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2917                     "only inactive hot spares, cache, top-level, "
2918                     "or log devices can be removed"));
2919                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2920         }
2921
2922         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
2923         if (islog && version < SPA_VERSION_HOLES) {
2924                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2925                     "pool must be upgrade to support log removal"));
2926                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
2927         }
2928
2929         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2930
2931         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
2932                 return (0);
2933
2934         return (zpool_standard_error(hdl, errno, msg));
2935 }
2936
2937 /*
2938  * Clear the errors for the pool, or the particular device if specified.
2939  */
2940 int
2941 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
2942 {
2943         zfs_cmd_t zc = { 0 };
2944         char msg[1024];
2945         nvlist_t *tgt;
2946         zpool_rewind_policy_t policy;
2947         boolean_t avail_spare, l2cache;
2948         libzfs_handle_t *hdl = zhp->zpool_hdl;
2949         nvlist_t *nvi = NULL;
2950         int error;
2951
2952         if (path)
2953                 (void) snprintf(msg, sizeof (msg),
2954                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
2955                     path);
2956         else
2957                 (void) snprintf(msg, sizeof (msg),
2958                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
2959                     zhp->zpool_name);
2960
2961         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2962         if (path) {
2963                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
2964                     &l2cache, NULL)) == 0)
2965                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
2966
2967                 /*
2968                  * Don't allow error clearing for hot spares.  Do allow
2969                  * error clearing for l2cache devices.
2970                  */
2971                 if (avail_spare)
2972                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
2973
2974                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
2975                     &zc.zc_guid) == 0);
2976         }
2977
2978         zpool_get_rewind_policy(rewindnvl, &policy);
2979         zc.zc_cookie = policy.zrp_request;
2980
2981         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
2982                 return (-1);
2983
2984         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
2985                 return (-1);
2986
2987         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
2988             errno == ENOMEM) {
2989                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
2990                         zcmd_free_nvlists(&zc);
2991                         return (-1);
2992                 }
2993         }
2994
2995         if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
2996             errno != EPERM && errno != EACCES)) {
2997                 if (policy.zrp_request &
2998                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
2999                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3000                         zpool_rewind_exclaim(hdl, zc.zc_name,
3001                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
3002                             nvi);
3003                         nvlist_free(nvi);
3004                 }
3005                 zcmd_free_nvlists(&zc);
3006                 return (0);
3007         }
3008
3009         zcmd_free_nvlists(&zc);
3010         return (zpool_standard_error(hdl, errno, msg));
3011 }
3012
3013 /*
3014  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3015  */
3016 int
3017 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3018 {
3019         zfs_cmd_t zc = { 0 };
3020         char msg[1024];
3021         libzfs_handle_t *hdl = zhp->zpool_hdl;
3022
3023         (void) snprintf(msg, sizeof (msg),
3024             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3025             guid);
3026
3027         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3028         zc.zc_guid = guid;
3029         zc.zc_cookie = ZPOOL_NO_REWIND;
3030
3031         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
3032                 return (0);
3033
3034         return (zpool_standard_error(hdl, errno, msg));
3035 }
3036
3037 /*
3038  * Change the GUID for a pool.
3039  */
3040 int
3041 zpool_reguid(zpool_handle_t *zhp)
3042 {
3043         char msg[1024];
3044         libzfs_handle_t *hdl = zhp->zpool_hdl;
3045         zfs_cmd_t zc = { 0 };
3046
3047         (void) snprintf(msg, sizeof (msg),
3048             dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3049
3050         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3051         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3052                 return (0);
3053
3054         return (zpool_standard_error(hdl, errno, msg));
3055 }
3056
3057 /*
3058  * Reopen the pool.
3059  */
3060 int
3061 zpool_reopen(zpool_handle_t *zhp)
3062 {
3063         zfs_cmd_t zc = { 0 };
3064         char msg[1024];
3065         libzfs_handle_t *hdl = zhp->zpool_hdl;
3066
3067         (void) snprintf(msg, sizeof (msg),
3068             dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3069             zhp->zpool_name);
3070
3071         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3072         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3073                 return (0);
3074         return (zpool_standard_error(hdl, errno, msg));
3075 }
3076
3077 /*
3078  * Convert from a devid string to a path.
3079  */
3080 static char *
3081 devid_to_path(char *devid_str)
3082 {
3083         ddi_devid_t devid;
3084         char *minor;
3085         char *path;
3086         devid_nmlist_t *list = NULL;
3087         int ret;
3088
3089         if (devid_str_decode(devid_str, &devid, &minor) != 0)
3090                 return (NULL);
3091
3092         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3093
3094         devid_str_free(minor);
3095         devid_free(devid);
3096
3097         if (ret != 0)
3098                 return (NULL);
3099
3100         if ((path = strdup(list[0].devname)) == NULL)
3101                 return (NULL);
3102
3103         devid_free_nmlist(list);
3104
3105         return (path);
3106 }
3107
3108 /*
3109  * Convert from a path to a devid string.
3110  */
3111 static char *
3112 path_to_devid(const char *path)
3113 {
3114         int fd;
3115         ddi_devid_t devid;
3116         char *minor, *ret;
3117
3118         if ((fd = open(path, O_RDONLY)) < 0)
3119                 return (NULL);
3120
3121         minor = NULL;
3122         ret = NULL;
3123         if (devid_get(fd, &devid) == 0) {
3124                 if (devid_get_minor_name(fd, &minor) == 0)
3125                         ret = devid_str_encode(devid, minor);
3126                 if (minor != NULL)
3127                         devid_str_free(minor);
3128                 devid_free(devid);
3129         }
3130         (void) close(fd);
3131
3132         return (ret);
3133 }
3134
3135 /*
3136  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3137  * ignore any failure here, since a common case is for an unprivileged user to
3138  * type 'zpool status', and we'll display the correct information anyway.
3139  */
3140 static void
3141 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3142 {
3143         zfs_cmd_t zc = { 0 };
3144
3145         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3146         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3147         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3148             &zc.zc_guid) == 0);
3149
3150         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3151 }
3152
3153 /*
3154  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3155  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3156  * We also check if this is a whole disk, in which case we strip off the
3157  * trailing 's0' slice name.
3158  *
3159  * This routine is also responsible for identifying when disks have been
3160  * reconfigured in a new location.  The kernel will have opened the device by
3161  * devid, but the path will still refer to the old location.  To catch this, we
3162  * first do a path -> devid translation (which is fast for the common case).  If
3163  * the devid matches, we're done.  If not, we do a reverse devid -> path
3164  * translation and issue the appropriate ioctl() to update the path of the vdev.
3165  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3166  * of these checks.
3167  */
3168 char *
3169 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3170     boolean_t verbose)
3171 {
3172         char *path, *devid;
3173         uint64_t value;
3174         char buf[64];
3175         vdev_stat_t *vs;
3176         uint_t vsc;
3177         int have_stats;
3178         int have_path;
3179
3180         have_stats = nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3181             (uint64_t **)&vs, &vsc) == 0;
3182         have_path = nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0;
3183
3184         /*
3185          * If the device is not currently present, assume it will not
3186          * come back at the same device path.  Display the device by GUID.
3187          */
3188         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &value) == 0 ||
3189             have_path && have_stats && vs->vs_state <= VDEV_STATE_CANT_OPEN) {
3190                 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3191                     &value) == 0);
3192                 (void) snprintf(buf, sizeof (buf), "%llu",
3193                     (u_longlong_t)value);
3194                 path = buf;
3195         } else if (have_path) {
3196
3197                 /*
3198                  * If the device is dead (faulted, offline, etc) then don't
3199                  * bother opening it.  Otherwise we may be forcing the user to
3200                  * open a misbehaving device, which can have undesirable
3201                  * effects.
3202                  */
3203                 if ((have_stats == 0 ||
3204                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3205                     zhp != NULL &&
3206                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3207                         /*
3208                          * Determine if the current path is correct.
3209                          */
3210                         char *newdevid = path_to_devid(path);
3211
3212                         if (newdevid == NULL ||
3213                             strcmp(devid, newdevid) != 0) {
3214                                 char *newpath;
3215
3216                                 if ((newpath = devid_to_path(devid)) != NULL) {
3217                                         /*
3218                                          * Update the path appropriately.
3219                                          */
3220                                         set_path(zhp, nv, newpath);
3221                                         if (nvlist_add_string(nv,
3222                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3223                                                 verify(nvlist_lookup_string(nv,
3224                                                     ZPOOL_CONFIG_PATH,
3225                                                     &path) == 0);
3226                                         free(newpath);
3227                                 }
3228                         }
3229
3230                         if (newdevid)
3231                                 devid_str_free(newdevid);
3232                 }
3233
3234 #ifdef sun
3235                 if (strncmp(path, "/dev/dsk/", 9) == 0)
3236                         path += 9;
3237
3238                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
3239                     &value) == 0 && value) {
3240                         int pathlen = strlen(path);
3241                         char *tmp = zfs_strdup(hdl, path);
3242
3243                         /*
3244                          * If it starts with c#, and ends with "s0", chop
3245                          * the "s0" off, or if it ends with "s0/old", remove
3246                          * the "s0" from the middle.
3247                          */
3248                         if (CTD_CHECK(tmp)) {
3249                                 if (strcmp(&tmp[pathlen - 2], "s0") == 0) {
3250                                         tmp[pathlen - 2] = '\0';
3251                                 } else if (pathlen > 6 &&
3252                                     strcmp(&tmp[pathlen - 6], "s0/old") == 0) {
3253                                         (void) strcpy(&tmp[pathlen - 6],
3254                                             "/old");
3255                                 }
3256                         }
3257                         return (tmp);
3258                 }
3259 #else   /* !sun */
3260                 if (strncmp(path, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
3261                         path += sizeof(_PATH_DEV) - 1;
3262 #endif  /* !sun */
3263         } else {
3264                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3265
3266                 /*
3267                  * If it's a raidz device, we need to stick in the parity level.
3268                  */
3269                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3270                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3271                             &value) == 0);
3272                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3273                             (u_longlong_t)value);
3274                         path = buf;
3275                 }
3276
3277                 /*
3278                  * We identify each top-level vdev by using a <type-id>
3279                  * naming convention.
3280                  */
3281                 if (verbose) {
3282                         uint64_t id;
3283
3284                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3285                             &id) == 0);
3286                         (void) snprintf(buf, sizeof (buf), "%s-%llu", path,
3287                             (u_longlong_t)id);
3288                         path = buf;
3289                 }
3290         }
3291
3292         return (zfs_strdup(hdl, path));
3293 }
3294
3295 static int
3296 zbookmark_compare(const void *a, const void *b)
3297 {
3298         return (memcmp(a, b, sizeof (zbookmark_t)));
3299 }
3300
3301 /*
3302  * Retrieve the persistent error log, uniquify the members, and return to the
3303  * caller.
3304  */
3305 int
3306 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3307 {
3308         zfs_cmd_t zc = { 0 };
3309         uint64_t count;
3310         zbookmark_t *zb = NULL;
3311         int i;
3312
3313         /*
3314          * Retrieve the raw error list from the kernel.  If the number of errors
3315          * has increased, allocate more space and continue until we get the
3316          * entire list.
3317          */
3318         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3319             &count) == 0);
3320         if (count == 0)
3321                 return (0);
3322         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3323             count * sizeof (zbookmark_t))) == (uintptr_t)NULL)
3324                 return (-1);
3325         zc.zc_nvlist_dst_size = count;
3326         (void) strcpy(zc.zc_name, zhp->zpool_name);
3327         for (;;) {
3328                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3329                     &zc) != 0) {
3330                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3331                         if (errno == ENOMEM) {
3332                                 count = zc.zc_nvlist_dst_size;
3333                                 if ((zc.zc_nvlist_dst = (uintptr_t)
3334                                     zfs_alloc(zhp->zpool_hdl, count *
3335                                     sizeof (zbookmark_t))) == (uintptr_t)NULL)
3336                                         return (-1);
3337                         } else {
3338                                 return (-1);
3339                         }
3340                 } else {
3341                         break;
3342                 }
3343         }
3344
3345         /*
3346          * Sort the resulting bookmarks.  This is a little confusing due to the
3347          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3348          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3349          * _not_ copied as part of the process.  So we point the start of our
3350          * array appropriate and decrement the total number of elements.
3351          */
3352         zb = ((zbookmark_t *)(uintptr_t)zc.zc_nvlist_dst) +
3353             zc.zc_nvlist_dst_size;
3354         count -= zc.zc_nvlist_dst_size;
3355
3356         qsort(zb, count, sizeof (zbookmark_t), zbookmark_compare);
3357
3358         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3359
3360         /*
3361          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3362          */
3363         for (i = 0; i < count; i++) {
3364                 nvlist_t *nv;
3365
3366                 /* ignoring zb_blkid and zb_level for now */
3367                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3368                     zb[i-1].zb_object == zb[i].zb_object)
3369                         continue;
3370
3371                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3372                         goto nomem;
3373                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3374                     zb[i].zb_objset) != 0) {
3375                         nvlist_free(nv);
3376                         goto nomem;
3377                 }
3378                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3379                     zb[i].zb_object) != 0) {
3380                         nvlist_free(nv);
3381                         goto nomem;
3382                 }
3383                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3384                         nvlist_free(nv);
3385                         goto nomem;
3386                 }
3387                 nvlist_free(nv);
3388         }
3389
3390         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3391         return (0);
3392
3393 nomem:
3394         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3395         return (no_memory(zhp->zpool_hdl));
3396 }
3397
3398 /*
3399  * Upgrade a ZFS pool to the latest on-disk version.
3400  */
3401 int
3402 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3403 {
3404         zfs_cmd_t zc = { 0 };
3405         libzfs_handle_t *hdl = zhp->zpool_hdl;
3406
3407         (void) strcpy(zc.zc_name, zhp->zpool_name);
3408         zc.zc_cookie = new_version;
3409
3410         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3411                 return (zpool_standard_error_fmt(hdl, errno,
3412                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3413                     zhp->zpool_name));
3414         return (0);
3415 }
3416
3417 void
3418 zpool_set_history_str(const char *subcommand, int argc, char **argv,
3419     char *history_str)
3420 {
3421         int i;
3422
3423         (void) strlcpy(history_str, subcommand, HIS_MAX_RECORD_LEN);
3424         for (i = 1; i < argc; i++) {
3425                 if (strlen(history_str) + 1 + strlen(argv[i]) >
3426                     HIS_MAX_RECORD_LEN)
3427                         break;
3428                 (void) strlcat(history_str, " ", HIS_MAX_RECORD_LEN);
3429                 (void) strlcat(history_str, argv[i], HIS_MAX_RECORD_LEN);
3430         }
3431 }
3432
3433 /*
3434  * Stage command history for logging.
3435  */
3436 int
3437 zpool_stage_history(libzfs_handle_t *hdl, const char *history_str)
3438 {
3439         if (history_str == NULL)
3440                 return (EINVAL);
3441
3442         if (strlen(history_str) > HIS_MAX_RECORD_LEN)
3443                 return (EINVAL);
3444
3445         if (hdl->libzfs_log_str != NULL)
3446                 free(hdl->libzfs_log_str);
3447
3448         if ((hdl->libzfs_log_str = strdup(history_str)) == NULL)
3449                 return (no_memory(hdl));
3450
3451         return (0);
3452 }
3453
3454 /*
3455  * Perform ioctl to get some command history of a pool.
3456  *
3457  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3458  * logical offset of the history buffer to start reading from.
3459  *
3460  * Upon return, 'off' is the next logical offset to read from and
3461  * 'len' is the actual amount of bytes read into 'buf'.
3462  */
3463 static int
3464 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3465 {
3466         zfs_cmd_t zc = { 0 };
3467         libzfs_handle_t *hdl = zhp->zpool_hdl;
3468
3469         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3470
3471         zc.zc_history = (uint64_t)(uintptr_t)buf;
3472         zc.zc_history_len = *len;
3473         zc.zc_history_offset = *off;
3474
3475         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3476                 switch (errno) {
3477                 case EPERM:
3478                         return (zfs_error_fmt(hdl, EZFS_PERM,
3479                             dgettext(TEXT_DOMAIN,
3480                             "cannot show history for pool '%s'"),
3481                             zhp->zpool_name));
3482                 case ENOENT:
3483                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3484                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3485                             "'%s'"), zhp->zpool_name));
3486                 case ENOTSUP:
3487                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3488                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3489                             "'%s', pool must be upgraded"), zhp->zpool_name));
3490                 default:
3491                         return (zpool_standard_error_fmt(hdl, errno,
3492                             dgettext(TEXT_DOMAIN,
3493                             "cannot get history for '%s'"), zhp->zpool_name));
3494                 }
3495         }
3496
3497         *len = zc.zc_history_len;
3498         *off = zc.zc_history_offset;
3499
3500         return (0);
3501 }
3502
3503 /*
3504  * Process the buffer of nvlists, unpacking and storing each nvlist record
3505  * into 'records'.  'leftover' is set to the number of bytes that weren't
3506  * processed as there wasn't a complete record.
3507  */
3508 int
3509 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3510     nvlist_t ***records, uint_t *numrecords)
3511 {
3512         uint64_t reclen;
3513         nvlist_t *nv;
3514         int i;
3515
3516         while (bytes_read > sizeof (reclen)) {
3517
3518                 /* get length of packed record (stored as little endian) */
3519                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3520                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3521
3522                 if (bytes_read < sizeof (reclen) + reclen)
3523                         break;
3524
3525                 /* unpack record */
3526                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3527                         return (ENOMEM);
3528                 bytes_read -= sizeof (reclen) + reclen;
3529                 buf += sizeof (reclen) + reclen;
3530
3531                 /* add record to nvlist array */
3532                 (*numrecords)++;
3533                 if (ISP2(*numrecords + 1)) {
3534                         *records = realloc(*records,
3535                             *numrecords * 2 * sizeof (nvlist_t *));
3536                 }
3537                 (*records)[*numrecords - 1] = nv;
3538         }
3539
3540         *leftover = bytes_read;
3541         return (0);
3542 }
3543
3544 #define HIS_BUF_LEN     (128*1024)
3545
3546 /*
3547  * Retrieve the command history of a pool.
3548  */
3549 int
3550 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3551 {
3552         char buf[HIS_BUF_LEN];
3553         uint64_t off = 0;
3554         nvlist_t **records = NULL;
3555         uint_t numrecords = 0;
3556         int err, i;
3557
3558         do {
3559                 uint64_t bytes_read = sizeof (buf);
3560                 uint64_t leftover;
3561
3562                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3563                         break;
3564
3565                 /* if nothing else was read in, we're at EOF, just return */
3566                 if (!bytes_read)
3567                         break;
3568
3569                 if ((err = zpool_history_unpack(buf, bytes_read,
3570                     &leftover, &records, &numrecords)) != 0)
3571                         break;
3572                 off -= leftover;
3573
3574                 /* CONSTCOND */
3575         } while (1);
3576
3577         if (!err) {
3578                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3579                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3580                     records, numrecords) == 0);
3581         }
3582         for (i = 0; i < numrecords; i++)
3583                 nvlist_free(records[i]);
3584         free(records);
3585
3586         return (err);
3587 }
3588
3589 void
3590 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
3591     char *pathname, size_t len)
3592 {
3593         zfs_cmd_t zc = { 0 };
3594         boolean_t mounted = B_FALSE;
3595         char *mntpnt = NULL;
3596         char dsname[MAXNAMELEN];
3597
3598         if (dsobj == 0) {
3599                 /* special case for the MOS */
3600                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>", obj);
3601                 return;
3602         }
3603
3604         /* get the dataset's name */
3605         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3606         zc.zc_obj = dsobj;
3607         if (ioctl(zhp->zpool_hdl->libzfs_fd,
3608             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
3609                 /* just write out a path of two object numbers */
3610                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
3611                     dsobj, obj);
3612                 return;
3613         }
3614         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
3615
3616         /* find out if the dataset is mounted */
3617         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
3618
3619         /* get the corrupted object's path */
3620         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
3621         zc.zc_obj = obj;
3622         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
3623             &zc) == 0) {
3624                 if (mounted) {
3625                         (void) snprintf(pathname, len, "%s%s", mntpnt,
3626                             zc.zc_value);
3627                 } else {
3628                         (void) snprintf(pathname, len, "%s:%s",
3629                             dsname, zc.zc_value);
3630                 }
3631         } else {
3632                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname, obj);
3633         }
3634         free(mntpnt);
3635 }
3636
3637 #ifdef sun
3638 /*
3639  * Read the EFI label from the config, if a label does not exist then
3640  * pass back the error to the caller. If the caller has passed a non-NULL
3641  * diskaddr argument then we set it to the starting address of the EFI
3642  * partition.
3643  */
3644 static int
3645 read_efi_label(nvlist_t *config, diskaddr_t *sb)
3646 {
3647         char *path;
3648         int fd;
3649         char diskname[MAXPATHLEN];
3650         int err = -1;
3651
3652         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
3653                 return (err);
3654
3655         (void) snprintf(diskname, sizeof (diskname), "%s%s", RDISK_ROOT,
3656             strrchr(path, '/'));
3657         if ((fd = open(diskname, O_RDONLY|O_NDELAY)) >= 0) {
3658                 struct dk_gpt *vtoc;
3659
3660                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
3661                         if (sb != NULL)
3662                                 *sb = vtoc->efi_parts[0].p_start;
3663                         efi_free(vtoc);
3664                 }
3665                 (void) close(fd);
3666         }
3667         return (err);
3668 }
3669
3670 /*
3671  * determine where a partition starts on a disk in the current
3672  * configuration
3673  */
3674 static diskaddr_t
3675 find_start_block(nvlist_t *config)
3676 {
3677         nvlist_t **child;
3678         uint_t c, children;
3679         diskaddr_t sb = MAXOFFSET_T;
3680         uint64_t wholedisk;
3681
3682         if (nvlist_lookup_nvlist_array(config,
3683             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
3684                 if (nvlist_lookup_uint64(config,
3685                     ZPOOL_CONFIG_WHOLE_DISK,
3686                     &wholedisk) != 0 || !wholedisk) {
3687                         return (MAXOFFSET_T);
3688                 }
3689                 if (read_efi_label(config, &sb) < 0)
3690                         sb = MAXOFFSET_T;
3691                 return (sb);
3692         }
3693
3694         for (c = 0; c < children; c++) {
3695                 sb = find_start_block(child[c]);
3696                 if (sb != MAXOFFSET_T) {
3697                         return (sb);
3698                 }
3699         }
3700         return (MAXOFFSET_T);
3701 }
3702 #endif /* sun */
3703
3704 /*
3705  * Label an individual disk.  The name provided is the short name,
3706  * stripped of any leading /dev path.
3707  */
3708 int
3709 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, const char *name)
3710 {
3711 #ifdef sun
3712         char path[MAXPATHLEN];
3713         struct dk_gpt *vtoc;
3714         int fd;
3715         size_t resv = EFI_MIN_RESV_SIZE;
3716         uint64_t slice_size;
3717         diskaddr_t start_block;
3718         char errbuf[1024];
3719
3720         /* prepare an error message just in case */
3721         (void) snprintf(errbuf, sizeof (errbuf),
3722             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
3723
3724         if (zhp) {
3725                 nvlist_t *nvroot;
3726
3727                 if (zpool_is_bootable(zhp)) {
3728                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3729                             "EFI labeled devices are not supported on root "
3730                             "pools."));
3731                         return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
3732                 }
3733
3734                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
3735                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3736
3737                 if (zhp->zpool_start_block == 0)
3738                         start_block = find_start_block(nvroot);
3739                 else
3740                         start_block = zhp->zpool_start_block;
3741                 zhp->zpool_start_block = start_block;
3742         } else {
3743                 /* new pool */
3744                 start_block = NEW_START_BLOCK;
3745         }
3746
3747         (void) snprintf(path, sizeof (path), "%s/%s%s", RDISK_ROOT, name,
3748             BACKUP_SLICE);
3749
3750         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
3751                 /*
3752                  * This shouldn't happen.  We've long since verified that this
3753                  * is a valid device.
3754                  */
3755                 zfs_error_aux(hdl,
3756                     dgettext(TEXT_DOMAIN, "unable to open device"));
3757                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
3758         }
3759
3760         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
3761                 /*
3762                  * The only way this can fail is if we run out of memory, or we
3763                  * were unable to read the disk's capacity
3764                  */
3765                 if (errno == ENOMEM)
3766                         (void) no_memory(hdl);
3767
3768                 (void) close(fd);
3769                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3770                     "unable to read disk capacity"), name);
3771
3772                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
3773         }
3774
3775         slice_size = vtoc->efi_last_u_lba + 1;
3776         slice_size -= EFI_MIN_RESV_SIZE;
3777         if (start_block == MAXOFFSET_T)
3778                 start_block = NEW_START_BLOCK;
3779         slice_size -= start_block;
3780
3781         vtoc->efi_parts[0].p_start = start_block;
3782         vtoc->efi_parts[0].p_size = slice_size;
3783
3784         /*
3785          * Why we use V_USR: V_BACKUP confuses users, and is considered
3786          * disposable by some EFI utilities (since EFI doesn't have a backup
3787          * slice).  V_UNASSIGNED is supposed to be used only for zero size
3788          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
3789          * etc. were all pretty specific.  V_USR is as close to reality as we
3790          * can get, in the absence of V_OTHER.
3791          */
3792         vtoc->efi_parts[0].p_tag = V_USR;
3793         (void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
3794
3795         vtoc->efi_parts[8].p_start = slice_size + start_block;
3796         vtoc->efi_parts[8].p_size = resv;
3797         vtoc->efi_parts[8].p_tag = V_RESERVED;
3798
3799         if (efi_write(fd, vtoc) != 0) {
3800                 /*
3801                  * Some block drivers (like pcata) may not support EFI
3802                  * GPT labels.  Print out a helpful error message dir-
3803                  * ecting the user to manually label the disk and give
3804                  * a specific slice.
3805                  */
3806                 (void) close(fd);
3807                 efi_free(vtoc);
3808
3809                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3810                     "try using fdisk(1M) and then provide a specific slice"));
3811                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
3812         }
3813
3814         (void) close(fd);
3815         efi_free(vtoc);
3816 #endif /* sun */
3817         return (0);
3818 }
3819
3820 static boolean_t
3821 supported_dump_vdev_type(libzfs_handle_t *hdl, nvlist_t *config, char *errbuf)
3822 {
3823         char *type;
3824         nvlist_t **child;
3825         uint_t children, c;
3826
3827         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_TYPE, &type) == 0);
3828         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0 ||
3829             strcmp(type, VDEV_TYPE_FILE) == 0 ||
3830             strcmp(type, VDEV_TYPE_LOG) == 0 ||
3831             strcmp(type, VDEV_TYPE_HOLE) == 0 ||
3832             strcmp(type, VDEV_TYPE_MISSING) == 0) {
3833                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3834                     "vdev type '%s' is not supported"), type);
3835                 (void) zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf);
3836                 return (B_FALSE);
3837         }
3838         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
3839             &child, &children) == 0) {
3840                 for (c = 0; c < children; c++) {
3841                         if (!supported_dump_vdev_type(hdl, child[c], errbuf))
3842                                 return (B_FALSE);
3843                 }
3844         }
3845         return (B_TRUE);
3846 }
3847
3848 /*
3849  * check if this zvol is allowable for use as a dump device; zero if
3850  * it is, > 0 if it isn't, < 0 if it isn't a zvol
3851  */
3852 int
3853 zvol_check_dump_config(char *arg)
3854 {
3855         zpool_handle_t *zhp = NULL;
3856         nvlist_t *config, *nvroot;
3857         char *p, *volname;
3858         nvlist_t **top;
3859         uint_t toplevels;
3860         libzfs_handle_t *hdl;
3861         char errbuf[1024];
3862         char poolname[ZPOOL_MAXNAMELEN];
3863         int pathlen = strlen(ZVOL_FULL_DEV_DIR);
3864         int ret = 1;
3865
3866         if (strncmp(arg, ZVOL_FULL_DEV_DIR, pathlen)) {
3867                 return (-1);
3868         }
3869
3870         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3871             "dump is not supported on device '%s'"), arg);
3872
3873         if ((hdl = libzfs_init()) == NULL)
3874                 return (1);
3875         libzfs_print_on_error(hdl, B_TRUE);
3876
3877         volname = arg + pathlen;
3878
3879         /* check the configuration of the pool */
3880         if ((p = strchr(volname, '/')) == NULL) {
3881                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3882                     "malformed dataset name"));
3883                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
3884                 return (1);
3885         } else if (p - volname >= ZFS_MAXNAMELEN) {
3886                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3887                     "dataset name is too long"));
3888                 (void) zfs_error(hdl, EZFS_NAMETOOLONG, errbuf);
3889                 return (1);
3890         } else {
3891                 (void) strncpy(poolname, volname, p - volname);
3892                 poolname[p - volname] = '\0';
3893         }
3894
3895         if ((zhp = zpool_open(hdl, poolname)) == NULL) {
3896                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3897                     "could not open pool '%s'"), poolname);
3898                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
3899                 goto out;
3900         }
3901         config = zpool_get_config(zhp, NULL);
3902         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3903             &nvroot) != 0) {
3904                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3905                     "could not obtain vdev configuration for  '%s'"), poolname);
3906                 (void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
3907                 goto out;
3908         }
3909
3910         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3911             &top, &toplevels) == 0);
3912         if (toplevels != 1) {
3913                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3914                     "'%s' has multiple top level vdevs"), poolname);
3915                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, errbuf);
3916                 goto out;
3917         }
3918
3919         if (!supported_dump_vdev_type(hdl, top[0], errbuf)) {
3920                 goto out;
3921         }
3922         ret = 0;
3923
3924 out:
3925         if (zhp)
3926                 zpool_close(zhp);
3927         libzfs_fini(hdl);
3928         return (ret);
3929 }