lib/libzfs/libzfs_pool.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
  24  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  25  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
  26  */
  27
  28 #include <ctype.h>
  29 #include <errno.h>
  30 #include <devid.h>
  31 #include <fcntl.h>
  32 #include <libintl.h>
  33 #include <stdio.h>
  34 #include <stdlib.h>
  35 #include <strings.h>
  36 #include <unistd.h>
  37 #include <libgen.h>
  38 #include <zone.h>
  39 #include <sys/stat.h>
  40 #include <sys/efi_partition.h>
  41 #include <sys/vtoc.h>
  42 #include <sys/zfs_ioctl.h>
  43 #include <dlfcn.h>
  44
  45 #include "zfs_namecheck.h"
  46 #include "zfs_prop.h"
  47 #include "libzfs_impl.h"
  48 #include "zfs_comutil.h"
  49 #include "zfeature_common.h"
  50
  51 static int read_efi_label(nvlist_t *config, diskaddr_t *sb);
  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 char *
 180 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
 181 {
 182         switch (state) {
 183         default:
 184                 break;
 185         case VDEV_STATE_CLOSED:
 186         case VDEV_STATE_OFFLINE:
 187                 return (gettext("OFFLINE"));
 188         case VDEV_STATE_REMOVED:
 189                 return (gettext("REMOVED"));
 190         case VDEV_STATE_CANT_OPEN:
 191                 if (aux == VDEV_AUX_CORRUPT_DATA || aux == VDEV_AUX_BAD_LOG)
 192                         return (gettext("FAULTED"));
 193                 else if (aux == VDEV_AUX_SPLIT_POOL)
 194                         return (gettext("SPLIT"));
 195                 else
 196                         return (gettext("UNAVAIL"));
 197         case VDEV_STATE_FAULTED:
 198                 return (gettext("FAULTED"));
 199         case VDEV_STATE_DEGRADED:
 200                 return (gettext("DEGRADED"));
 201         case VDEV_STATE_HEALTHY:
 202                 return (gettext("ONLINE"));
 203         }
 204
 205         return (gettext("UNKNOWN"));
 206 }
 207
 208 /*
 209  * Map POOL STATE to printed strings.
 210  */
 211 const char *
 212 zpool_pool_state_to_name(pool_state_t state)
 213 {
 214         switch (state) {
 215         default:
 216                 break;
 217         case POOL_STATE_ACTIVE:
 218                 return (gettext("ACTIVE"));
 219         case POOL_STATE_EXPORTED:
 220                 return (gettext("EXPORTED"));
 221         case POOL_STATE_DESTROYED:
 222                 return (gettext("DESTROYED"));
 223         case POOL_STATE_SPARE:
 224                 return (gettext("SPARE"));
 225         case POOL_STATE_L2CACHE:
 226                 return (gettext("L2CACHE"));
 227         case POOL_STATE_UNINITIALIZED:
 228                 return (gettext("UNINITIALIZED"));
 229         case POOL_STATE_UNAVAIL:
 230                 return (gettext("UNAVAIL"));
 231         case POOL_STATE_POTENTIALLY_ACTIVE:
 232                 return (gettext("POTENTIALLY_ACTIVE"));
 233         }
 234
 235         return (gettext("UNKNOWN"));
 236 }
 237
 238 /*
 239  * Get a zpool property value for 'prop' and return the value in
 240  * a pre-allocated buffer.
 241  */
 242 int
 243 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf,
 244     size_t len, zprop_source_t *srctype, boolean_t literal)
 245 {
 246         uint64_t intval;
 247         const char *strval;
 248         zprop_source_t src = ZPROP_SRC_NONE;
 249         nvlist_t *nvroot;
 250         vdev_stat_t *vs;
 251         uint_t vsc;
 252
 253         if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
 254                 switch (prop) {
 255                 case ZPOOL_PROP_NAME:
 256                         (void) strlcpy(buf, zpool_get_name(zhp), len);
 257                         break;
 258
 259                 case ZPOOL_PROP_HEALTH:
 260                         (void) strlcpy(buf, "FAULTED", len);
 261                         break;
 262
 263                 case ZPOOL_PROP_GUID:
 264                         intval = zpool_get_prop_int(zhp, prop, &src);
 265                         (void) snprintf(buf, len, "%llu", (u_longlong_t)intval);
 266                         break;
 267
 268                 case ZPOOL_PROP_ALTROOT:
 269                 case ZPOOL_PROP_CACHEFILE:
 270                 case ZPOOL_PROP_COMMENT:
 271                         if (zhp->zpool_props != NULL ||
 272                             zpool_get_all_props(zhp) == 0) {
 273                                 (void) strlcpy(buf,
 274                                     zpool_get_prop_string(zhp, prop, &src),
 275                                     len);
 276                                 break;
 277                         }
 278                         /* FALLTHROUGH */
 279                 default:
 280                         (void) strlcpy(buf, "-", len);
 281                         break;
 282                 }
 283
 284                 if (srctype != NULL)
 285                         *srctype = src;
 286                 return (0);
 287         }
 288
 289         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
 290             prop != ZPOOL_PROP_NAME)
 291                 return (-1);
 292
 293         switch (zpool_prop_get_type(prop)) {
 294         case PROP_TYPE_STRING:
 295                 (void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
 296                     len);
 297                 break;
 298
 299         case PROP_TYPE_NUMBER:
 300                 intval = zpool_get_prop_int(zhp, prop, &src);
 301
 302                 switch (prop) {
 303                 case ZPOOL_PROP_SIZE:
 304                 case ZPOOL_PROP_ALLOCATED:
 305                 case ZPOOL_PROP_FREE:
 306                 case ZPOOL_PROP_FREEING:
 307                 case ZPOOL_PROP_LEAKED:
 308                 case ZPOOL_PROP_ASHIFT:
 309                         if (literal)
 310                                 (void) snprintf(buf, len, "%llu",
 311                                         (u_longlong_t)intval);
 312                         else
 313                                 (void) zfs_nicenum(intval, buf, len);
 314                         break;
 315
 316                 case ZPOOL_PROP_EXPANDSZ:
 317                         if (intval == 0) {
 318                                 (void) strlcpy(buf, "-", len);
 319                         } else if (literal) {
 320                                 (void) snprintf(buf, len, "%llu",
 321                                     (u_longlong_t)intval);
 322                         } else {
 323                                 (void) zfs_nicenum(intval, buf, len);
 324                         }
 325                         break;
 326
 327                 case ZPOOL_PROP_CAPACITY:
 328                         if (literal) {
 329                                 (void) snprintf(buf, len, "%llu",
 330                                     (u_longlong_t)intval);
 331                         } else {
 332                                 (void) snprintf(buf, len, "%llu%%",
 333                                     (u_longlong_t)intval);
 334                         }
 335                         break;
 336
 337                 case ZPOOL_PROP_FRAGMENTATION:
 338                         if (intval == UINT64_MAX) {
 339                                 (void) strlcpy(buf, "-", len);
 340                         } else if (literal) {
 341                                 (void) snprintf(buf, len, "%llu",
 342                                     (u_longlong_t)intval);
 343                         } else {
 344                                 (void) snprintf(buf, len, "%llu%%",
 345                                     (u_longlong_t)intval);
 346                         }
 347                         break;
 348
 349                 case ZPOOL_PROP_DEDUPRATIO:
 350                         if (literal)
 351                                 (void) snprintf(buf, len, "%llu.%02llu",
 352                                     (u_longlong_t)(intval / 100),
 353                                     (u_longlong_t)(intval % 100));
 354                         else
 355                                 (void) snprintf(buf, len, "%llu.%02llux",
 356                                     (u_longlong_t)(intval / 100),
 357                                     (u_longlong_t)(intval % 100));
 358                         break;
 359
 360                 case ZPOOL_PROP_HEALTH:
 361                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 362                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 363                         verify(nvlist_lookup_uint64_array(nvroot,
 364                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
 365                             == 0);
 366
 367                         (void) strlcpy(buf, zpool_state_to_name(intval,
 368                             vs->vs_aux), len);
 369                         break;
 370                 case ZPOOL_PROP_VERSION:
 371                         if (intval >= SPA_VERSION_FEATURES) {
 372                                 (void) snprintf(buf, len, "-");
 373                                 break;
 374                         }
 375                         /* FALLTHROUGH */
 376                 default:
 377                         (void) snprintf(buf, len, "%llu", (u_longlong_t)intval);
 378                 }
 379                 break;
 380
 381         case PROP_TYPE_INDEX:
 382                 intval = zpool_get_prop_int(zhp, prop, &src);
 383                 if (zpool_prop_index_to_string(prop, intval, &strval)
 384                     != 0)
 385                         return (-1);
 386                 (void) strlcpy(buf, strval, len);
 387                 break;
 388
 389         default:
 390                 abort();
 391         }
 392
 393         if (srctype)
 394                 *srctype = src;
 395
 396         return (0);
 397 }
 398
 399 /*
 400  * Check if the bootfs name has the same pool name as it is set to.
 401  * Assuming bootfs is a valid dataset name.
 402  */
 403 static boolean_t
 404 bootfs_name_valid(const char *pool, char *bootfs)
 405 {
 406         int len = strlen(pool);
 407
 408         if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
 409                 return (B_FALSE);
 410
 411         if (strncmp(pool, bootfs, len) == 0 &&
 412             (bootfs[len] == '/' || bootfs[len] == '\0'))
 413                 return (B_TRUE);
 414
 415         return (B_FALSE);
 416 }
 417
 418 #if defined(__sun__) || defined(__sun)
 419 /*
 420  * Inspect the configuration to determine if any of the devices contain
 421  * an EFI label.
 422  */
 423 static boolean_t
 424 pool_uses_efi(nvlist_t *config)
 425 {
 426         nvlist_t **child;
 427         uint_t c, children;
 428
 429         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
 430             &child, &children) != 0)
 431                 return (read_efi_label(config, NULL) >= 0);
 432
 433         for (c = 0; c < children; c++) {
 434                 if (pool_uses_efi(child[c]))
 435                         return (B_TRUE);
 436         }
 437         return (B_FALSE);
 438 }
 439 #endif
 440
 441 boolean_t
 442 zpool_is_bootable(zpool_handle_t *zhp)
 443 {
 444         char bootfs[ZFS_MAX_DATASET_NAME_LEN];
 445
 446         return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
 447             sizeof (bootfs), NULL, B_FALSE) == 0 && strncmp(bootfs, "-",
 448             sizeof (bootfs)) != 0);
 449 }
 450
 451
 452 /*
 453  * Given an nvlist of zpool properties to be set, validate that they are
 454  * correct, and parse any numeric properties (index, boolean, etc) if they are
 455  * specified as strings.
 456  */
 457 static nvlist_t *
 458 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 459     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
 460 {
 461         nvpair_t *elem;
 462         nvlist_t *retprops;
 463         zpool_prop_t prop;
 464         char *strval;
 465         uint64_t intval;
 466         char *slash, *check;
 467         struct stat64 statbuf;
 468         zpool_handle_t *zhp;
 469         nvlist_t *nvroot;
 470
 471         if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
 472                 (void) no_memory(hdl);
 473                 return (NULL);
 474         }
 475
 476         elem = NULL;
 477         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 478                 const char *propname = nvpair_name(elem);
 479
 480                 prop = zpool_name_to_prop(propname);
 481                 if (prop == ZPROP_INVAL && zpool_prop_feature(propname)) {
 482                         int err;
 483                         char *fname = strchr(propname, '@') + 1;
 484
 485                         err = zfeature_lookup_name(fname, NULL);
 486                         if (err != 0) {
 487                                 ASSERT3U(err, ==, ENOENT);
 488                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 489                                     "invalid feature '%s'"), fname);
 490                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 491                                 goto error;
 492                         }
 493
 494                         if (nvpair_type(elem) != DATA_TYPE_STRING) {
 495                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 496                                     "'%s' must be a string"), propname);
 497                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 498                                 goto error;
 499                         }
 500
 501                         (void) nvpair_value_string(elem, &strval);
 502                         if (strcmp(strval, ZFS_FEATURE_ENABLED) != 0) {
 503                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 504                                     "property '%s' can only be set to "
 505                                     "'enabled'"), propname);
 506                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 507                                 goto error;
 508                         }
 509
 510                         if (nvlist_add_uint64(retprops, propname, 0) != 0) {
 511                                 (void) no_memory(hdl);
 512                                 goto error;
 513                         }
 514                         continue;
 515                 }
 516
 517                 /*
 518                  * Make sure this property is valid and applies to this type.
 519                  */
 520                 if (prop == ZPROP_INVAL) {
 521                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 522                             "invalid property '%s'"), propname);
 523                         (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 524                         goto error;
 525                 }
 526
 527                 if (zpool_prop_readonly(prop)) {
 528                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 529                             "is readonly"), propname);
 530                         (void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
 531                         goto error;
 532                 }
 533
 534                 if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
 535                     &strval, &intval, errbuf) != 0)
 536                         goto error;
 537
 538                 /*
 539                  * Perform additional checking for specific properties.
 540                  */
 541                 switch (prop) {
 542                 default:
 543                         break;
 544                 case ZPOOL_PROP_VERSION:
 545                         if (intval < version ||
 546                             !SPA_VERSION_IS_SUPPORTED(intval)) {
 547                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 548                                     "property '%s' number %d is invalid."),
 549                                     propname, intval);
 550                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 551                                 goto error;
 552                         }
 553                         break;
 554
 555                 case ZPOOL_PROP_ASHIFT:
 556                         if (!flags.create) {
 557                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 558                                     "property '%s' can only be set at "
 559                                     "creation time"), propname);
 560                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 561                                 goto error;
 562                         }
 563
 564                         if (intval != 0 && (intval < 9 || intval > 13)) {
 565                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 566                                     "property '%s' number %d is invalid."),
 567                                     propname, intval);
 568                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 569                                 goto error;
 570                         }
 571                         break;
 572
 573                 case ZPOOL_PROP_BOOTFS:
 574                         if (flags.create || flags.import) {
 575                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 576                                     "property '%s' cannot be set at creation "
 577                                     "or import time"), propname);
 578                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 579                                 goto error;
 580                         }
 581
 582                         if (version < SPA_VERSION_BOOTFS) {
 583                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 584                                     "pool must be upgraded to support "
 585                                     "'%s' property"), propname);
 586                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 587                                 goto error;
 588                         }
 589
 590                         /*
 591                          * bootfs property value has to be a dataset name and
 592                          * the dataset has to be in the same pool as it sets to.
 593                          */
 594                         if (strval[0] != '\0' && !bootfs_name_valid(poolname,
 595                             strval)) {
 596                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 597                                     "is an invalid name"), strval);
 598                                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
 599                                 goto error;
 600                         }
 601
 602                         if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
 603                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 604                                     "could not open pool '%s'"), poolname);
 605                                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
 606                                 goto error;
 607                         }
 608                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 609                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 610
 611 #if defined(__sun__) || defined(__sun)
 612                         /*
 613                          * bootfs property cannot be set on a disk which has
 614                          * been EFI labeled.
 615                          */
 616                         if (pool_uses_efi(nvroot)) {
 617                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 618                                     "property '%s' not supported on "
 619                                     "EFI labeled devices"), propname);
 620                                 (void) zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf);
 621                                 zpool_close(zhp);
 622                                 goto error;
 623                         }
 624 #endif
 625                         zpool_close(zhp);
 626                         break;
 627
 628                 case ZPOOL_PROP_ALTROOT:
 629                         if (!flags.create && !flags.import) {
 630                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 631                                     "property '%s' can only be set during pool "
 632                                     "creation or import"), propname);
 633                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 634                                 goto error;
 635                         }
 636
 637                         if (strval[0] != '/') {
 638                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 639                                     "bad alternate root '%s'"), strval);
 640                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 641                                 goto error;
 642                         }
 643                         break;
 644
 645                 case ZPOOL_PROP_CACHEFILE:
 646                         if (strval[0] == '\0')
 647                                 break;
 648
 649                         if (strcmp(strval, "none") == 0)
 650                                 break;
 651
 652                         if (strval[0] != '/') {
 653                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 654                                     "property '%s' must be empty, an "
 655                                     "absolute path, or 'none'"), propname);
 656                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 657                                 goto error;
 658                         }
 659
 660                         slash = strrchr(strval, '/');
 661
 662                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 663                             strcmp(slash, "/..") == 0) {
 664                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 665                                     "'%s' is not a valid file"), strval);
 666                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 667                                 goto error;
 668                         }
 669
 670                         *slash = '\0';
 671
 672                         if (strval[0] != '\0' &&
 673                             (stat64(strval, &statbuf) != 0 ||
 674                             !S_ISDIR(statbuf.st_mode))) {
 675                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 676                                     "'%s' is not a valid directory"),
 677                                     strval);
 678                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 679                                 goto error;
 680                         }
 681
 682                         *slash = '/';
 683                         break;
 684
 685                 case ZPOOL_PROP_COMMENT:
 686                         for (check = strval; *check != '\0'; check++) {
 687                                 if (!isprint(*check)) {
 688                                         zfs_error_aux(hdl,
 689                                             dgettext(TEXT_DOMAIN,
 690                                             "comment may only have printable "
 691                                             "characters"));
 692                                         (void) zfs_error(hdl, EZFS_BADPROP,
 693                                             errbuf);
 694                                         goto error;
 695                                 }
 696                         }
 697                         if (strlen(strval) > ZPROP_MAX_COMMENT) {
 698                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 699                                     "comment must not exceed %d characters"),
 700                                     ZPROP_MAX_COMMENT);
 701                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 702                                 goto error;
 703                         }
 704                         break;
 705                 case ZPOOL_PROP_READONLY:
 706                         if (!flags.import) {
 707                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 708                                     "property '%s' can only be set at "
 709                                     "import time"), propname);
 710                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 711                                 goto error;
 712                         }
 713                         break;
 714                 case ZPOOL_PROP_TNAME:
 715                         if (!flags.create) {
 716                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 717                                     "property '%s' can only be set at "
 718                                     "creation time"), propname);
 719                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 720                                 goto error;
 721                         }
 722                         break;
 723                 }
 724         }
 725
 726         return (retprops);
 727 error:
 728         nvlist_free(retprops);
 729         return (NULL);
 730 }
 731
 732 /*
 733  * Set zpool property : propname=propval.
 734  */
 735 int
 736 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
 737 {
 738         zfs_cmd_t zc = {"\0"};
 739         int ret = -1;
 740         char errbuf[1024];
 741         nvlist_t *nvl = NULL;
 742         nvlist_t *realprops;
 743         uint64_t version;
 744         prop_flags_t flags = { 0 };
 745
 746         (void) snprintf(errbuf, sizeof (errbuf),
 747             dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
 748             zhp->zpool_name);
 749
 750         if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
 751                 return (no_memory(zhp->zpool_hdl));
 752
 753         if (nvlist_add_string(nvl, propname, propval) != 0) {
 754                 nvlist_free(nvl);
 755                 return (no_memory(zhp->zpool_hdl));
 756         }
 757
 758         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
 759         if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
 760             zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
 761                 nvlist_free(nvl);
 762                 return (-1);
 763         }
 764
 765         nvlist_free(nvl);
 766         nvl = realprops;
 767
 768         /*
 769          * Execute the corresponding ioctl() to set this property.
 770          */
 771         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 772
 773         if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
 774                 nvlist_free(nvl);
 775                 return (-1);
 776         }
 777
 778         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
 779
 780         zcmd_free_nvlists(&zc);
 781         nvlist_free(nvl);
 782
 783         if (ret)
 784                 (void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
 785         else
 786                 (void) zpool_props_refresh(zhp);
 787
 788         return (ret);
 789 }
 790
 791 int
 792 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
 793 {
 794         libzfs_handle_t *hdl = zhp->zpool_hdl;
 795         zprop_list_t *entry;
 796         char buf[ZFS_MAXPROPLEN];
 797         nvlist_t *features = NULL;
 798         nvpair_t *nvp;
 799         zprop_list_t **last;
 800         boolean_t firstexpand = (NULL == *plp);
 801         int i;
 802
 803         if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
 804                 return (-1);
 805
 806         last = plp;
 807         while (*last != NULL)
 808                 last = &(*last)->pl_next;
 809
 810         if ((*plp)->pl_all)
 811                 features = zpool_get_features(zhp);
 812
 813         if ((*plp)->pl_all && firstexpand) {
 814                 for (i = 0; i < SPA_FEATURES; i++) {
 815                         zprop_list_t *entry = zfs_alloc(hdl,
 816                             sizeof (zprop_list_t));
 817                         entry->pl_prop = ZPROP_INVAL;
 818                         entry->pl_user_prop = zfs_asprintf(hdl, "feature@%s",
 819                             spa_feature_table[i].fi_uname);
 820                         entry->pl_width = strlen(entry->pl_user_prop);
 821                         entry->pl_all = B_TRUE;
 822
 823                         *last = entry;
 824                         last = &entry->pl_next;
 825                 }
 826         }
 827
 828         /* add any unsupported features */
 829         for (nvp = nvlist_next_nvpair(features, NULL);
 830             nvp != NULL; nvp = nvlist_next_nvpair(features, nvp)) {
 831                 char *propname;
 832                 boolean_t found;
 833                 zprop_list_t *entry;
 834
 835                 if (zfeature_is_supported(nvpair_name(nvp)))
 836                         continue;
 837
 838                 propname = zfs_asprintf(hdl, "unsupported@%s",
 839                     nvpair_name(nvp));
 840
 841                 /*
 842                  * Before adding the property to the list make sure that no
 843                  * other pool already added the same property.
 844                  */
 845                 found = B_FALSE;
 846                 entry = *plp;
 847                 while (entry != NULL) {
 848                         if (entry->pl_user_prop != NULL &&
 849                             strcmp(propname, entry->pl_user_prop) == 0) {
 850                                 found = B_TRUE;
 851                                 break;
 852                         }
 853                         entry = entry->pl_next;
 854                 }
 855                 if (found) {
 856                         free(propname);
 857                         continue;
 858                 }
 859
 860                 entry = zfs_alloc(hdl, sizeof (zprop_list_t));
 861                 entry->pl_prop = ZPROP_INVAL;
 862                 entry->pl_user_prop = propname;
 863                 entry->pl_width = strlen(entry->pl_user_prop);
 864                 entry->pl_all = B_TRUE;
 865
 866                 *last = entry;
 867                 last = &entry->pl_next;
 868         }
 869
 870         for (entry = *plp; entry != NULL; entry = entry->pl_next) {
 871
 872                 if (entry->pl_fixed)
 873                         continue;
 874
 875                 if (entry->pl_prop != ZPROP_INVAL &&
 876                     zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
 877                     NULL, B_FALSE) == 0) {
 878                         if (strlen(buf) > entry->pl_width)
 879                                 entry->pl_width = strlen(buf);
 880                 }
 881         }
 882
 883         return (0);
 884 }
 885
 886 /*
 887  * Get the state for the given feature on the given ZFS pool.
 888  */
 889 int
 890 zpool_prop_get_feature(zpool_handle_t *zhp, const char *propname, char *buf,
 891     size_t len)
 892 {
 893         uint64_t refcount;
 894         boolean_t found = B_FALSE;
 895         nvlist_t *features = zpool_get_features(zhp);
 896         boolean_t supported;
 897         const char *feature = strchr(propname, '@') + 1;
 898
 899         supported = zpool_prop_feature(propname);
 900         ASSERT(supported || zpool_prop_unsupported(propname));
 901
 902         /*
 903          * Convert from feature name to feature guid. This conversion is
 904          * unecessary for unsupported@... properties because they already
 905          * use guids.
 906          */
 907         if (supported) {
 908                 int ret;
 909                 spa_feature_t fid;
 910
 911                 ret = zfeature_lookup_name(feature, &fid);
 912                 if (ret != 0) {
 913                         (void) strlcpy(buf, "-", len);
 914                         return (ENOTSUP);
 915                 }
 916                 feature = spa_feature_table[fid].fi_guid;
 917         }
 918
 919         if (nvlist_lookup_uint64(features, feature, &refcount) == 0)
 920                 found = B_TRUE;
 921
 922         if (supported) {
 923                 if (!found) {
 924                         (void) strlcpy(buf, ZFS_FEATURE_DISABLED, len);
 925                 } else  {
 926                         if (refcount == 0)
 927                                 (void) strlcpy(buf, ZFS_FEATURE_ENABLED, len);
 928                         else
 929                                 (void) strlcpy(buf, ZFS_FEATURE_ACTIVE, len);
 930                 }
 931         } else {
 932                 if (found) {
 933                         if (refcount == 0) {
 934                                 (void) strcpy(buf, ZFS_UNSUPPORTED_INACTIVE);
 935                         } else {
 936                                 (void) strcpy(buf, ZFS_UNSUPPORTED_READONLY);
 937                         }
 938                 } else {
 939                         (void) strlcpy(buf, "-", len);
 940                         return (ENOTSUP);
 941                 }
 942         }
 943
 944         return (0);
 945 }
 946
 947 /*
 948  * Don't start the slice at the default block of 34; many storage
 949  * devices will use a stripe width of 128k, other vendors prefer a 1m
 950  * alignment.  It is best to play it safe and ensure a 1m alignment
 951  * given 512B blocks.  When the block size is larger by a power of 2
 952  * we will still be 1m aligned.  Some devices are sensitive to the
 953  * partition ending alignment as well.
 954  */
 955 #define NEW_START_BLOCK         2048
 956 #define PARTITION_END_ALIGNMENT 2048
 957
 958 /*
 959  * Validate the given pool name, optionally putting an extended error message in
 960  * 'buf'.
 961  */
 962 boolean_t
 963 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
 964 {
 965         namecheck_err_t why;
 966         char what;
 967         int ret;
 968
 969         ret = pool_namecheck(pool, &why, &what);
 970
 971         /*
 972          * The rules for reserved pool names were extended at a later point.
 973          * But we need to support users with existing pools that may now be
 974          * invalid.  So we only check for this expanded set of names during a
 975          * create (or import), and only in userland.
 976          */
 977         if (ret == 0 && !isopen &&
 978             (strncmp(pool, "mirror", 6) == 0 ||
 979             strncmp(pool, "raidz", 5) == 0 ||
 980             strncmp(pool, "spare", 5) == 0 ||
 981             strcmp(pool, "log") == 0)) {
 982                 if (hdl != NULL)
 983                         zfs_error_aux(hdl,
 984                             dgettext(TEXT_DOMAIN, "name is reserved"));
 985                 return (B_FALSE);
 986         }
 987
 988
 989         if (ret != 0) {
 990                 if (hdl != NULL) {
 991                         switch (why) {
 992                         case NAME_ERR_TOOLONG:
 993                                 zfs_error_aux(hdl,
 994                                     dgettext(TEXT_DOMAIN, "name is too long"));
 995                                 break;
 996
 997                         case NAME_ERR_INVALCHAR:
 998                                 zfs_error_aux(hdl,
 999                                     dgettext(TEXT_DOMAIN, "invalid character "
1000                                     "'%c' in pool name"), what);
1001                                 break;
1002
1003                         case NAME_ERR_NOLETTER:
1004                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1005                                     "name must begin with a letter"));
1006                                 break;
1007
1008                         case NAME_ERR_RESERVED:
1009                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1010                                     "name is reserved"));
1011                                 break;
1012
1013                         case NAME_ERR_DISKLIKE:
1014                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1015                                     "pool name is reserved"));
1016                                 break;
1017
1018                         case NAME_ERR_LEADING_SLASH:
1019                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1020                                     "leading slash in name"));
1021                                 break;
1022
1023                         case NAME_ERR_EMPTY_COMPONENT:
1024                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1025                                     "empty component in name"));
1026                                 break;
1027
1028                         case NAME_ERR_TRAILING_SLASH:
1029                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1030                                     "trailing slash in name"));
1031                                 break;
1032
1033                         case NAME_ERR_MULTIPLE_AT:
1034                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1035                                     "multiple '@' delimiters in name"));
1036                                 break;
1037                         case NAME_ERR_NO_AT:
1038                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1039                                     "permission set is missing '@'"));
1040                                 break;
1041                         }
1042                 }
1043                 return (B_FALSE);
1044         }
1045
1046         return (B_TRUE);
1047 }
1048
1049 /*
1050  * Open a handle to the given pool, even if the pool is currently in the FAULTED
1051  * state.
1052  */
1053 zpool_handle_t *
1054 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
1055 {
1056         zpool_handle_t *zhp;
1057         boolean_t missing;
1058
1059         /*
1060          * Make sure the pool name is valid.
1061          */
1062         if (!zpool_name_valid(hdl, B_TRUE, pool)) {
1063                 (void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1064                     dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1065                     pool);
1066                 return (NULL);
1067         }
1068
1069         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1070                 return (NULL);
1071
1072         zhp->zpool_hdl = hdl;
1073         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1074
1075         if (zpool_refresh_stats(zhp, &missing) != 0) {
1076                 zpool_close(zhp);
1077                 return (NULL);
1078         }
1079
1080         if (missing) {
1081                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
1082                 (void) zfs_error_fmt(hdl, EZFS_NOENT,
1083                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
1084                 zpool_close(zhp);
1085                 return (NULL);
1086         }
1087
1088         return (zhp);
1089 }
1090
1091 /*
1092  * Like the above, but silent on error.  Used when iterating over pools (because
1093  * the configuration cache may be out of date).
1094  */
1095 int
1096 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
1097 {
1098         zpool_handle_t *zhp;
1099         boolean_t missing;
1100
1101         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1102                 return (-1);
1103
1104         zhp->zpool_hdl = hdl;
1105         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1106
1107         if (zpool_refresh_stats(zhp, &missing) != 0) {
1108                 zpool_close(zhp);
1109                 return (-1);
1110         }
1111
1112         if (missing) {
1113                 zpool_close(zhp);
1114                 *ret = NULL;
1115                 return (0);
1116         }
1117
1118         *ret = zhp;
1119         return (0);
1120 }
1121
1122 /*
1123  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
1124  * state.
1125  */
1126 zpool_handle_t *
1127 zpool_open(libzfs_handle_t *hdl, const char *pool)
1128 {
1129         zpool_handle_t *zhp;
1130
1131         if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
1132                 return (NULL);
1133
1134         if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
1135                 (void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
1136                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
1137                 zpool_close(zhp);
1138                 return (NULL);
1139         }
1140
1141         return (zhp);
1142 }
1143
1144 /*
1145  * Close the handle.  Simply frees the memory associated with the handle.
1146  */
1147 void
1148 zpool_close(zpool_handle_t *zhp)
1149 {
1150         nvlist_free(zhp->zpool_config);
1151         nvlist_free(zhp->zpool_old_config);
1152         nvlist_free(zhp->zpool_props);
1153         free(zhp);
1154 }
1155
1156 /*
1157  * Return the name of the pool.
1158  */
1159 const char *
1160 zpool_get_name(zpool_handle_t *zhp)
1161 {
1162         return (zhp->zpool_name);
1163 }
1164
1165
1166 /*
1167  * Return the state of the pool (ACTIVE or UNAVAILABLE)
1168  */
1169 int
1170 zpool_get_state(zpool_handle_t *zhp)
1171 {
1172         return (zhp->zpool_state);
1173 }
1174
1175 /*
1176  * Create the named pool, using the provided vdev list.  It is assumed
1177  * that the consumer has already validated the contents of the nvlist, so we
1178  * don't have to worry about error semantics.
1179  */
1180 int
1181 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
1182     nvlist_t *props, nvlist_t *fsprops)
1183 {
1184         zfs_cmd_t zc = {"\0"};
1185         nvlist_t *zc_fsprops = NULL;
1186         nvlist_t *zc_props = NULL;
1187         char msg[1024];
1188         int ret = -1;
1189
1190         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1191             "cannot create '%s'"), pool);
1192
1193         if (!zpool_name_valid(hdl, B_FALSE, pool))
1194                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
1195
1196         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1197                 return (-1);
1198
1199         if (props) {
1200                 prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
1201
1202                 if ((zc_props = zpool_valid_proplist(hdl, pool, props,
1203                     SPA_VERSION_1, flags, msg)) == NULL) {
1204                         goto create_failed;
1205                 }
1206         }
1207
1208         if (fsprops) {
1209                 uint64_t zoned;
1210                 char *zonestr;
1211
1212                 zoned = ((nvlist_lookup_string(fsprops,
1213                     zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
1214                     strcmp(zonestr, "on") == 0);
1215
1216                 if ((zc_fsprops = zfs_valid_proplist(hdl, ZFS_TYPE_FILESYSTEM,
1217                     fsprops, zoned, NULL, NULL, msg)) == NULL) {
1218                         goto create_failed;
1219                 }
1220                 if (!zc_props &&
1221                     (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
1222                         goto create_failed;
1223                 }
1224                 if (nvlist_add_nvlist(zc_props,
1225                     ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
1226                         goto create_failed;
1227                 }
1228         }
1229
1230         if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
1231                 goto create_failed;
1232
1233         (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
1234
1235         if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
1236
1237                 zcmd_free_nvlists(&zc);
1238                 nvlist_free(zc_props);
1239                 nvlist_free(zc_fsprops);
1240
1241                 switch (errno) {
1242                 case EBUSY:
1243                         /*
1244                          * This can happen if the user has specified the same
1245                          * device multiple times.  We can't reliably detect this
1246                          * until we try to add it and see we already have a
1247                          * label.  This can also happen under if the device is
1248                          * part of an active md or lvm device.
1249                          */
1250                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1251                             "one or more vdevs refer to the same device, or "
1252                             "one of\nthe devices is part of an active md or "
1253                             "lvm device"));
1254                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1255
1256                 case ERANGE:
1257                         /*
1258                          * This happens if the record size is smaller or larger
1259                          * than the allowed size range, or not a power of 2.
1260                          *
1261                          * NOTE: although zfs_valid_proplist is called earlier,
1262                          * this case may have slipped through since the
1263                          * pool does not exist yet and it is therefore
1264                          * impossible to read properties e.g. max blocksize
1265                          * from the pool.
1266                          */
1267                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1268                             "record size invalid"));
1269                         return (zfs_error(hdl, EZFS_BADPROP, msg));
1270
1271                 case EOVERFLOW:
1272                         /*
1273                          * This occurs when one of the devices is below
1274                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1275                          * device was the problem device since there's no
1276                          * reliable way to determine device size from userland.
1277                          */
1278                         {
1279                                 char buf[64];
1280
1281                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1282
1283                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1284                                     "one or more devices is less than the "
1285                                     "minimum size (%s)"), buf);
1286                         }
1287                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1288
1289                 case ENOSPC:
1290                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1291                             "one or more devices is out of space"));
1292                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1293
1294                 case ENOTBLK:
1295                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1296                             "cache device must be a disk or disk slice"));
1297                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1298
1299                 default:
1300                         return (zpool_standard_error(hdl, errno, msg));
1301                 }
1302         }
1303
1304 create_failed:
1305         zcmd_free_nvlists(&zc);
1306         nvlist_free(zc_props);
1307         nvlist_free(zc_fsprops);
1308         return (ret);
1309 }
1310
1311 /*
1312  * Destroy the given pool.  It is up to the caller to ensure that there are no
1313  * datasets left in the pool.
1314  */
1315 int
1316 zpool_destroy(zpool_handle_t *zhp, const char *log_str)
1317 {
1318         zfs_cmd_t zc = {"\0"};
1319         zfs_handle_t *zfp = NULL;
1320         libzfs_handle_t *hdl = zhp->zpool_hdl;
1321         char msg[1024];
1322
1323         if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1324             (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1325                 return (-1);
1326
1327         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1328         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1329
1330         if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1331                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1332                     "cannot destroy '%s'"), zhp->zpool_name);
1333
1334                 if (errno == EROFS) {
1335                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1336                             "one or more devices is read only"));
1337                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1338                 } else {
1339                         (void) zpool_standard_error(hdl, errno, msg);
1340                 }
1341
1342                 if (zfp)
1343                         zfs_close(zfp);
1344                 return (-1);
1345         }
1346
1347         if (zfp) {
1348                 remove_mountpoint(zfp);
1349                 zfs_close(zfp);
1350         }
1351
1352         return (0);
1353 }
1354
1355 /*
1356  * Add the given vdevs to the pool.  The caller must have already performed the
1357  * necessary verification to ensure that the vdev specification is well-formed.
1358  */
1359 int
1360 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1361 {
1362         zfs_cmd_t zc = {"\0"};
1363         int ret;
1364         libzfs_handle_t *hdl = zhp->zpool_hdl;
1365         char msg[1024];
1366         nvlist_t **spares, **l2cache;
1367         uint_t nspares, nl2cache;
1368
1369         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1370             "cannot add to '%s'"), zhp->zpool_name);
1371
1372         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1373             SPA_VERSION_SPARES &&
1374             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1375             &spares, &nspares) == 0) {
1376                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1377                     "upgraded to add hot spares"));
1378                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1379         }
1380
1381 #if defined(__sun__) || defined(__sun)
1382         if (zpool_is_bootable(zhp) && nvlist_lookup_nvlist_array(nvroot,
1383             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) {
1384                 uint64_t s;
1385
1386                 for (s = 0; s < nspares; s++) {
1387                         char *path;
1388
1389                         if (nvlist_lookup_string(spares[s], ZPOOL_CONFIG_PATH,
1390                             &path) == 0 && pool_uses_efi(spares[s])) {
1391                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1392                                     "device '%s' contains an EFI label and "
1393                                     "cannot be used on root pools."),
1394                                     zpool_vdev_name(hdl, NULL, spares[s], 0));
1395                                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
1396                         }
1397                 }
1398         }
1399 #endif
1400
1401         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1402             SPA_VERSION_L2CACHE &&
1403             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1404             &l2cache, &nl2cache) == 0) {
1405                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1406                     "upgraded to add cache devices"));
1407                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1408         }
1409
1410         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1411                 return (-1);
1412         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1413
1414         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1415                 switch (errno) {
1416                 case EBUSY:
1417                         /*
1418                          * This can happen if the user has specified the same
1419                          * device multiple times.  We can't reliably detect this
1420                          * until we try to add it and see we already have a
1421                          * label.
1422                          */
1423                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1424                             "one or more vdevs refer to the same device"));
1425                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1426                         break;
1427
1428                 case EOVERFLOW:
1429                         /*
1430                          * This occurrs when one of the devices is below
1431                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1432                          * device was the problem device since there's no
1433                          * reliable way to determine device size from userland.
1434                          */
1435                         {
1436                                 char buf[64];
1437
1438                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1439
1440                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1441                                     "device is less than the minimum "
1442                                     "size (%s)"), buf);
1443                         }
1444                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1445                         break;
1446
1447                 case ENOTSUP:
1448                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1449                             "pool must be upgraded to add these vdevs"));
1450                         (void) zfs_error(hdl, EZFS_BADVERSION, msg);
1451                         break;
1452
1453                 case ENOTBLK:
1454                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1455                             "cache device must be a disk or disk slice"));
1456                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1457                         break;
1458
1459                 default:
1460                         (void) zpool_standard_error(hdl, errno, msg);
1461                 }
1462
1463                 ret = -1;
1464         } else {
1465                 ret = 0;
1466         }
1467
1468         zcmd_free_nvlists(&zc);
1469
1470         return (ret);
1471 }
1472
1473 /*
1474  * Exports the pool from the system.  The caller must ensure that there are no
1475  * mounted datasets in the pool.
1476  */
1477 static int
1478 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce,
1479     const char *log_str)
1480 {
1481         zfs_cmd_t zc = {"\0"};
1482         char msg[1024];
1483
1484         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1485             "cannot export '%s'"), zhp->zpool_name);
1486
1487         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1488         zc.zc_cookie = force;
1489         zc.zc_guid = hardforce;
1490         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1491
1492         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1493                 switch (errno) {
1494                 case EXDEV:
1495                         zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1496                             "use '-f' to override the following errors:\n"
1497                             "'%s' has an active shared spare which could be"
1498                             " used by other pools once '%s' is exported."),
1499                             zhp->zpool_name, zhp->zpool_name);
1500                         return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1501                             msg));
1502                 default:
1503                         return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1504                             msg));
1505                 }
1506         }
1507
1508         return (0);
1509 }
1510
1511 int
1512 zpool_export(zpool_handle_t *zhp, boolean_t force, const char *log_str)
1513 {
1514         return (zpool_export_common(zhp, force, B_FALSE, log_str));
1515 }
1516
1517 int
1518 zpool_export_force(zpool_handle_t *zhp, const char *log_str)
1519 {
1520         return (zpool_export_common(zhp, B_TRUE, B_TRUE, log_str));
1521 }
1522
1523 static void
1524 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1525     nvlist_t *config)
1526 {
1527         nvlist_t *nv = NULL;
1528         uint64_t rewindto;
1529         int64_t loss = -1;
1530         struct tm t;
1531         char timestr[128];
1532
1533         if (!hdl->libzfs_printerr || config == NULL)
1534                 return;
1535
1536         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1537             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0) {
1538                 return;
1539         }
1540
1541         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1542                 return;
1543         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1544
1545         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1546             strftime(timestr, 128, "%c", &t) != 0) {
1547                 if (dryrun) {
1548                         (void) printf(dgettext(TEXT_DOMAIN,
1549                             "Would be able to return %s "
1550                             "to its state as of %s.\n"),
1551                             name, timestr);
1552                 } else {
1553                         (void) printf(dgettext(TEXT_DOMAIN,
1554                             "Pool %s returned to its state as of %s.\n"),
1555                             name, timestr);
1556                 }
1557                 if (loss > 120) {
1558                         (void) printf(dgettext(TEXT_DOMAIN,
1559                             "%s approximately %lld "),
1560                             dryrun ? "Would discard" : "Discarded",
1561                             ((longlong_t)loss + 30) / 60);
1562                         (void) printf(dgettext(TEXT_DOMAIN,
1563                             "minutes of transactions.\n"));
1564                 } else if (loss > 0) {
1565                         (void) printf(dgettext(TEXT_DOMAIN,
1566                             "%s approximately %lld "),
1567                             dryrun ? "Would discard" : "Discarded",
1568                             (longlong_t)loss);
1569                         (void) printf(dgettext(TEXT_DOMAIN,
1570                             "seconds of transactions.\n"));
1571                 }
1572         }
1573 }
1574
1575 void
1576 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1577     nvlist_t *config)
1578 {
1579         nvlist_t *nv = NULL;
1580         int64_t loss = -1;
1581         uint64_t edata = UINT64_MAX;
1582         uint64_t rewindto;
1583         struct tm t;
1584         char timestr[128];
1585
1586         if (!hdl->libzfs_printerr)
1587                 return;
1588
1589         if (reason >= 0)
1590                 (void) printf(dgettext(TEXT_DOMAIN, "action: "));
1591         else
1592                 (void) printf(dgettext(TEXT_DOMAIN, "\t"));
1593
1594         /* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1595         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1596             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0 ||
1597             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1598                 goto no_info;
1599
1600         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1601         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1602             &edata);
1603
1604         (void) printf(dgettext(TEXT_DOMAIN,
1605             "Recovery is possible, but will result in some data loss.\n"));
1606
1607         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1608             strftime(timestr, 128, "%c", &t) != 0) {
1609                 (void) printf(dgettext(TEXT_DOMAIN,
1610                     "\tReturning the pool to its state as of %s\n"
1611                     "\tshould correct the problem.  "),
1612                     timestr);
1613         } else {
1614                 (void) printf(dgettext(TEXT_DOMAIN,
1615                     "\tReverting the pool to an earlier state "
1616                     "should correct the problem.\n\t"));
1617         }
1618
1619         if (loss > 120) {
1620                 (void) printf(dgettext(TEXT_DOMAIN,
1621                     "Approximately %lld minutes of data\n"
1622                     "\tmust be discarded, irreversibly.  "),
1623                     ((longlong_t)loss + 30) / 60);
1624         } else if (loss > 0) {
1625                 (void) printf(dgettext(TEXT_DOMAIN,
1626                     "Approximately %lld seconds of data\n"
1627                     "\tmust be discarded, irreversibly.  "),
1628                     (longlong_t)loss);
1629         }
1630         if (edata != 0 && edata != UINT64_MAX) {
1631                 if (edata == 1) {
1632                         (void) printf(dgettext(TEXT_DOMAIN,
1633                             "After rewind, at least\n"
1634                             "\tone persistent user-data error will remain.  "));
1635                 } else {
1636                         (void) printf(dgettext(TEXT_DOMAIN,
1637                             "After rewind, several\n"
1638                             "\tpersistent user-data errors will remain.  "));
1639                 }
1640         }
1641         (void) printf(dgettext(TEXT_DOMAIN,
1642             "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1643             reason >= 0 ? "clear" : "import", name);
1644
1645         (void) printf(dgettext(TEXT_DOMAIN,
1646             "A scrub of the pool\n"
1647             "\tis strongly recommended after recovery.\n"));
1648         return;
1649
1650 no_info:
1651         (void) printf(dgettext(TEXT_DOMAIN,
1652             "Destroy and re-create the pool from\n\ta backup source.\n"));
1653 }
1654
1655 /*
1656  * zpool_import() is a contracted interface. Should be kept the same
1657  * if possible.
1658  *
1659  * Applications should use zpool_import_props() to import a pool with
1660  * new properties value to be set.
1661  */
1662 int
1663 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1664     char *altroot)
1665 {
1666         nvlist_t *props = NULL;
1667         int ret;
1668
1669         if (altroot != NULL) {
1670                 if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1671                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1672                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1673                             newname));
1674                 }
1675
1676                 if (nvlist_add_string(props,
1677                     zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1678                     nvlist_add_string(props,
1679                     zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1680                         nvlist_free(props);
1681                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1682                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1683                             newname));
1684                 }
1685         }
1686
1687         ret = zpool_import_props(hdl, config, newname, props,
1688             ZFS_IMPORT_NORMAL);
1689         nvlist_free(props);
1690         return (ret);
1691 }
1692
1693 static void
1694 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1695     int indent)
1696 {
1697         nvlist_t **child;
1698         uint_t c, children;
1699         char *vname;
1700         uint64_t is_log = 0;
1701
1702         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1703             &is_log);
1704
1705         if (name != NULL)
1706                 (void) printf("\t%*s%s%s\n", indent, "", name,
1707                     is_log ? " [log]" : "");
1708
1709         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1710             &child, &children) != 0)
1711                 return;
1712
1713         for (c = 0; c < children; c++) {
1714                 vname = zpool_vdev_name(hdl, NULL, child[c], VDEV_NAME_TYPE_ID);
1715                 print_vdev_tree(hdl, vname, child[c], indent + 2);
1716                 free(vname);
1717         }
1718 }
1719
1720 void
1721 zpool_print_unsup_feat(nvlist_t *config)
1722 {
1723         nvlist_t *nvinfo, *unsup_feat;
1724         nvpair_t *nvp;
1725
1726         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nvinfo) ==
1727             0);
1728         verify(nvlist_lookup_nvlist(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT,
1729             &unsup_feat) == 0);
1730
1731         for (nvp = nvlist_next_nvpair(unsup_feat, NULL); nvp != NULL;
1732             nvp = nvlist_next_nvpair(unsup_feat, nvp)) {
1733                 char *desc;
1734
1735                 verify(nvpair_type(nvp) == DATA_TYPE_STRING);
1736                 verify(nvpair_value_string(nvp, &desc) == 0);
1737
1738                 if (strlen(desc) > 0)
1739                         (void) printf("\t%s (%s)\n", nvpair_name(nvp), desc);
1740                 else
1741                         (void) printf("\t%s\n", nvpair_name(nvp));
1742         }
1743 }
1744
1745 /*
1746  * Import the given pool using the known configuration and a list of
1747  * properties to be set. The configuration should have come from
1748  * zpool_find_import(). The 'newname' parameters control whether the pool
1749  * is imported with a different name.
1750  */
1751 int
1752 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1753     nvlist_t *props, int flags)
1754 {
1755         zfs_cmd_t zc = {"\0"};
1756         zpool_rewind_policy_t policy;
1757         nvlist_t *nv = NULL;
1758         nvlist_t *nvinfo = NULL;
1759         nvlist_t *missing = NULL;
1760         char *thename;
1761         char *origname;
1762         int ret;
1763         int error = 0;
1764         char errbuf[1024];
1765
1766         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1767             &origname) == 0);
1768
1769         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1770             "cannot import pool '%s'"), origname);
1771
1772         if (newname != NULL) {
1773                 if (!zpool_name_valid(hdl, B_FALSE, newname))
1774                         return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1775                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1776                             newname));
1777                 thename = (char *)newname;
1778         } else {
1779                 thename = origname;
1780         }
1781
1782         if (props != NULL) {
1783                 uint64_t version;
1784                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1785
1786                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1787                     &version) == 0);
1788
1789                 if ((props = zpool_valid_proplist(hdl, origname,
1790                     props, version, flags, errbuf)) == NULL)
1791                         return (-1);
1792                 if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1793                         nvlist_free(props);
1794                         return (-1);
1795                 }
1796                 nvlist_free(props);
1797         }
1798
1799         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1800
1801         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1802             &zc.zc_guid) == 0);
1803
1804         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1805                 zcmd_free_nvlists(&zc);
1806                 return (-1);
1807         }
1808         if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1809                 zcmd_free_nvlists(&zc);
1810                 return (-1);
1811         }
1812
1813         zc.zc_cookie = flags;
1814         while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1815             errno == ENOMEM) {
1816                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1817                         zcmd_free_nvlists(&zc);
1818                         return (-1);
1819                 }
1820         }
1821         if (ret != 0)
1822                 error = errno;
1823
1824         (void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1825
1826         zcmd_free_nvlists(&zc);
1827
1828         zpool_get_rewind_policy(config, &policy);
1829
1830         if (error) {
1831                 char desc[1024];
1832
1833                 /*
1834                  * Dry-run failed, but we print out what success
1835                  * looks like if we found a best txg
1836                  */
1837                 if (policy.zrp_request & ZPOOL_TRY_REWIND) {
1838                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1839                             B_TRUE, nv);
1840                         nvlist_free(nv);
1841                         return (-1);
1842                 }
1843
1844                 if (newname == NULL)
1845                         (void) snprintf(desc, sizeof (desc),
1846                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1847                             thename);
1848                 else
1849                         (void) snprintf(desc, sizeof (desc),
1850                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1851                             origname, thename);
1852
1853                 switch (error) {
1854                 case ENOTSUP:
1855                         if (nv != NULL && nvlist_lookup_nvlist(nv,
1856                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1857                             nvlist_exists(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT)) {
1858                                 (void) printf(dgettext(TEXT_DOMAIN, "This "
1859                                     "pool uses the following feature(s) not "
1860                                     "supported by this system:\n"));
1861                                 zpool_print_unsup_feat(nv);
1862                                 if (nvlist_exists(nvinfo,
1863                                     ZPOOL_CONFIG_CAN_RDONLY)) {
1864                                         (void) printf(dgettext(TEXT_DOMAIN,
1865                                             "All unsupported features are only "
1866                                             "required for writing to the pool."
1867                                             "\nThe pool can be imported using "
1868                                             "'-o readonly=on'.\n"));
1869                                 }
1870                         }
1871                         /*
1872                          * Unsupported version.
1873                          */
1874                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1875                         break;
1876
1877                 case EINVAL:
1878                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1879                         break;
1880
1881                 case EROFS:
1882                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1883                             "one or more devices is read only"));
1884                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1885                         break;
1886
1887                 case ENXIO:
1888                         if (nv && nvlist_lookup_nvlist(nv,
1889                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1890                             nvlist_lookup_nvlist(nvinfo,
1891                             ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1892                                 (void) printf(dgettext(TEXT_DOMAIN,
1893                                     "The devices below are missing, use "
1894                                     "'-m' to import the pool anyway:\n"));
1895                                 print_vdev_tree(hdl, NULL, missing, 2);
1896                                 (void) printf("\n");
1897                         }
1898                         (void) zpool_standard_error(hdl, error, desc);
1899                         break;
1900
1901                 case EEXIST:
1902                         (void) zpool_standard_error(hdl, error, desc);
1903                         break;
1904
1905                 case EBUSY:
1906                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1907                             "one or more devices are already in use\n"));
1908                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1909                         break;
1910                 case ENAMETOOLONG:
1911                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1912                             "new name of at least one dataset is longer than "
1913                             "the maximum allowable length"));
1914                         (void) zfs_error(hdl, EZFS_NAMETOOLONG, desc);
1915                         break;
1916                 default:
1917                         (void) zpool_standard_error(hdl, error, desc);
1918                         zpool_explain_recover(hdl,
1919                             newname ? origname : thename, -error, nv);
1920                         break;
1921                 }
1922
1923                 nvlist_free(nv);
1924                 ret = -1;
1925         } else {
1926                 zpool_handle_t *zhp;
1927
1928                 /*
1929                  * This should never fail, but play it safe anyway.
1930                  */
1931                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1932                         ret = -1;
1933                 else if (zhp != NULL)
1934                         zpool_close(zhp);
1935                 if (policy.zrp_request &
1936                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1937                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1938                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1939                 }
1940                 nvlist_free(nv);
1941                 return (0);
1942         }
1943
1944         return (ret);
1945 }
1946
1947 /*
1948  * Scan the pool.
1949  */
1950 int
1951 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1952 {
1953         zfs_cmd_t zc = {"\0"};
1954         char msg[1024];
1955         libzfs_handle_t *hdl = zhp->zpool_hdl;
1956
1957         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1958         zc.zc_cookie = func;
1959
1960         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1961             (errno == ENOENT && func != POOL_SCAN_NONE))
1962                 return (0);
1963
1964         if (func == POOL_SCAN_SCRUB) {
1965                 (void) snprintf(msg, sizeof (msg),
1966                     dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1967         } else if (func == POOL_SCAN_NONE) {
1968                 (void) snprintf(msg, sizeof (msg),
1969                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1970                     zc.zc_name);
1971         } else {
1972                 assert(!"unexpected result");
1973         }
1974
1975         if (errno == EBUSY) {
1976                 nvlist_t *nvroot;
1977                 pool_scan_stat_t *ps = NULL;
1978                 uint_t psc;
1979
1980                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
1981                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1982                 (void) nvlist_lookup_uint64_array(nvroot,
1983                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1984                 if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1985                         return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1986                 else
1987                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
1988         } else if (errno == ENOENT) {
1989                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1990         } else {
1991                 return (zpool_standard_error(hdl, errno, msg));
1992         }
1993 }
1994
1995 /*
1996  * Find a vdev that matches the search criteria specified. We use the
1997  * the nvpair name to determine how we should look for the device.
1998  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1999  * spare; but FALSE if its an INUSE spare.
2000  */
2001 static nvlist_t *
2002 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
2003     boolean_t *l2cache, boolean_t *log)
2004 {
2005         uint_t c, children;
2006         nvlist_t **child;
2007         nvlist_t *ret;
2008         uint64_t is_log;
2009         char *srchkey;
2010         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
2011
2012         /* Nothing to look for */
2013         if (search == NULL || pair == NULL)
2014                 return (NULL);
2015
2016         /* Obtain the key we will use to search */
2017         srchkey = nvpair_name(pair);
2018
2019         switch (nvpair_type(pair)) {
2020         case DATA_TYPE_UINT64:
2021                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
2022                         uint64_t srchval, theguid;
2023
2024                         verify(nvpair_value_uint64(pair, &srchval) == 0);
2025                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
2026                             &theguid) == 0);
2027                         if (theguid == srchval)
2028                                 return (nv);
2029                 }
2030                 break;
2031
2032         case DATA_TYPE_STRING: {
2033                 char *srchval, *val;
2034
2035                 verify(nvpair_value_string(pair, &srchval) == 0);
2036                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
2037                         break;
2038
2039                 /*
2040                  * Search for the requested value. Special cases:
2041                  *
2042                  * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
2043                  *   "-part1", or "p1".  The suffix is hidden from the user,
2044                  *   but included in the string, so this matches around it.
2045                  * - ZPOOL_CONFIG_PATH for short names zfs_strcmp_shortname()
2046                  *   is used to check all possible expanded paths.
2047                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
2048                  *
2049                  * Otherwise, all other searches are simple string compares.
2050                  */
2051                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0) {
2052                         uint64_t wholedisk = 0;
2053
2054                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
2055                             &wholedisk);
2056                         if (zfs_strcmp_pathname(srchval, val, wholedisk) == 0)
2057                                 return (nv);
2058
2059                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2060                         char *type, *idx, *end, *p;
2061                         uint64_t id, vdev_id;
2062
2063                         /*
2064                          * Determine our vdev type, keeping in mind
2065                          * that the srchval is composed of a type and
2066                          * vdev id pair (i.e. mirror-4).
2067                          */
2068                         if ((type = strdup(srchval)) == NULL)
2069                                 return (NULL);
2070
2071                         if ((p = strrchr(type, '-')) == NULL) {
2072                                 free(type);
2073                                 break;
2074                         }
2075                         idx = p + 1;
2076                         *p = '\0';
2077
2078                         /*
2079                          * If the types don't match then keep looking.
2080                          */
2081                         if (strncmp(val, type, strlen(val)) != 0) {
2082                                 free(type);
2083                                 break;
2084                         }
2085
2086                         verify(strncmp(type, VDEV_TYPE_RAIDZ,
2087                             strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2088                             strncmp(type, VDEV_TYPE_MIRROR,
2089                             strlen(VDEV_TYPE_MIRROR)) == 0);
2090                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
2091                             &id) == 0);
2092
2093                         errno = 0;
2094                         vdev_id = strtoull(idx, &end, 10);
2095
2096                         free(type);
2097                         if (errno != 0)
2098                                 return (NULL);
2099
2100                         /*
2101                          * Now verify that we have the correct vdev id.
2102                          */
2103                         if (vdev_id == id)
2104                                 return (nv);
2105                 }
2106
2107                 /*
2108                  * Common case
2109                  */
2110                 if (strcmp(srchval, val) == 0)
2111                         return (nv);
2112                 break;
2113         }
2114
2115         default:
2116                 break;
2117         }
2118
2119         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2120             &child, &children) != 0)
2121                 return (NULL);
2122
2123         for (c = 0; c < children; c++) {
2124                 if ((ret = vdev_to_nvlist_iter(child[c], search,
2125                     avail_spare, l2cache, NULL)) != NULL) {
2126                         /*
2127                          * The 'is_log' value is only set for the toplevel
2128                          * vdev, not the leaf vdevs.  So we always lookup the
2129                          * log device from the root of the vdev tree (where
2130                          * 'log' is non-NULL).
2131                          */
2132                         if (log != NULL &&
2133                             nvlist_lookup_uint64(child[c],
2134                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2135                             is_log) {
2136                                 *log = B_TRUE;
2137                         }
2138                         return (ret);
2139                 }
2140         }
2141
2142         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2143             &child, &children) == 0) {
2144                 for (c = 0; c < children; c++) {
2145                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2146                             avail_spare, l2cache, NULL)) != NULL) {
2147                                 *avail_spare = B_TRUE;
2148                                 return (ret);
2149                         }
2150                 }
2151         }
2152
2153         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2154             &child, &children) == 0) {
2155                 for (c = 0; c < children; c++) {
2156                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2157                             avail_spare, l2cache, NULL)) != NULL) {
2158                                 *l2cache = B_TRUE;
2159                                 return (ret);
2160                         }
2161                 }
2162         }
2163
2164         return (NULL);
2165 }
2166
2167 /*
2168  * Given a physical path (minus the "/devices" prefix), find the
2169  * associated vdev.
2170  */
2171 nvlist_t *
2172 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2173     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2174 {
2175         nvlist_t *search, *nvroot, *ret;
2176
2177         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2178         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
2179
2180         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2181             &nvroot) == 0);
2182
2183         *avail_spare = B_FALSE;
2184         *l2cache = B_FALSE;
2185         if (log != NULL)
2186                 *log = B_FALSE;
2187         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2188         nvlist_free(search);
2189
2190         return (ret);
2191 }
2192
2193 /*
2194  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2195  */
2196 boolean_t
2197 zpool_vdev_is_interior(const char *name)
2198 {
2199         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2200             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2201                 return (B_TRUE);
2202         return (B_FALSE);
2203 }
2204
2205 nvlist_t *
2206 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2207     boolean_t *l2cache, boolean_t *log)
2208 {
2209         char *end;
2210         nvlist_t *nvroot, *search, *ret;
2211         uint64_t guid;
2212
2213         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2214
2215         guid = strtoull(path, &end, 0);
2216         if (guid != 0 && *end == '\0') {
2217                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
2218         } else if (zpool_vdev_is_interior(path)) {
2219                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
2220         } else {
2221                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
2222         }
2223
2224         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2225             &nvroot) == 0);
2226
2227         *avail_spare = B_FALSE;
2228         *l2cache = B_FALSE;
2229         if (log != NULL)
2230                 *log = B_FALSE;
2231         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2232         nvlist_free(search);
2233
2234         return (ret);
2235 }
2236
2237 static int
2238 vdev_online(nvlist_t *nv)
2239 {
2240         uint64_t ival;
2241
2242         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2243             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2244             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2245                 return (0);
2246
2247         return (1);
2248 }
2249
2250 /*
2251  * Helper function for zpool_get_physpaths().
2252  */
2253 static int
2254 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2255     size_t *bytes_written)
2256 {
2257         size_t bytes_left, pos, rsz;
2258         char *tmppath;
2259         const char *format;
2260
2261         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2262             &tmppath) != 0)
2263                 return (EZFS_NODEVICE);
2264
2265         pos = *bytes_written;
2266         bytes_left = physpath_size - pos;
2267         format = (pos == 0) ? "%s" : " %s";
2268
2269         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2270         *bytes_written += rsz;
2271
2272         if (rsz >= bytes_left) {
2273                 /* if physpath was not copied properly, clear it */
2274                 if (bytes_left != 0) {
2275                         physpath[pos] = 0;
2276                 }
2277                 return (EZFS_NOSPC);
2278         }
2279         return (0);
2280 }
2281
2282 static int
2283 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2284     size_t *rsz, boolean_t is_spare)
2285 {
2286         char *type;
2287         int ret;
2288
2289         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2290                 return (EZFS_INVALCONFIG);
2291
2292         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2293                 /*
2294                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2295                  * For a spare vdev, we only want to boot from the active
2296                  * spare device.
2297                  */
2298                 if (is_spare) {
2299                         uint64_t spare = 0;
2300                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2301                             &spare);
2302                         if (!spare)
2303                                 return (EZFS_INVALCONFIG);
2304                 }
2305
2306                 if (vdev_online(nv)) {
2307                         if ((ret = vdev_get_one_physpath(nv, physpath,
2308                             phypath_size, rsz)) != 0)
2309                                 return (ret);
2310                 }
2311         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2312             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2313             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2314                 nvlist_t **child;
2315                 uint_t count;
2316                 int i, ret;
2317
2318                 if (nvlist_lookup_nvlist_array(nv,
2319                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2320                         return (EZFS_INVALCONFIG);
2321
2322                 for (i = 0; i < count; i++) {
2323                         ret = vdev_get_physpaths(child[i], physpath,
2324                             phypath_size, rsz, is_spare);
2325                         if (ret == EZFS_NOSPC)
2326                                 return (ret);
2327                 }
2328         }
2329
2330         return (EZFS_POOL_INVALARG);
2331 }
2332
2333 /*
2334  * Get phys_path for a root pool config.
2335  * Return 0 on success; non-zero on failure.
2336  */
2337 static int
2338 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2339 {
2340         size_t rsz;
2341         nvlist_t *vdev_root;
2342         nvlist_t **child;
2343         uint_t count;
2344         char *type;
2345
2346         rsz = 0;
2347
2348         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2349             &vdev_root) != 0)
2350                 return (EZFS_INVALCONFIG);
2351
2352         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2353             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2354             &child, &count) != 0)
2355                 return (EZFS_INVALCONFIG);
2356
2357 #if defined(__sun__) || defined(__sun)
2358         /*
2359          * root pool can not have EFI labeled disks and can only have
2360          * a single top-level vdev.
2361          */
2362         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1 ||
2363             pool_uses_efi(vdev_root))
2364                 return (EZFS_POOL_INVALARG);
2365 #endif
2366
2367         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2368             B_FALSE);
2369
2370         /* No online devices */
2371         if (rsz == 0)
2372                 return (EZFS_NODEVICE);
2373
2374         return (0);
2375 }
2376
2377 /*
2378  * Get phys_path for a root pool
2379  * Return 0 on success; non-zero on failure.
2380  */
2381 int
2382 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2383 {
2384         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2385             phypath_size));
2386 }
2387
2388 /*
2389  * If the device has being dynamically expanded then we need to relabel
2390  * the disk to use the new unallocated space.
2391  */
2392 static int
2393 zpool_relabel_disk(libzfs_handle_t *hdl, const char *path, const char *msg)
2394 {
2395         int fd, error;
2396
2397         if ((fd = open(path, O_RDWR|O_DIRECT)) < 0) {
2398                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2399                     "relabel '%s': unable to open device: %d"), path, errno);
2400                 return (zfs_error(hdl, EZFS_OPENFAILED, msg));
2401         }
2402
2403         /*
2404          * It's possible that we might encounter an error if the device
2405          * does not have any unallocated space left. If so, we simply
2406          * ignore that error and continue on.
2407          *
2408          * Also, we don't call efi_rescan() - that would just return EBUSY.
2409          * The module will do it for us in vdev_disk_open().
2410          */
2411         error = efi_use_whole_disk(fd);
2412         (void) close(fd);
2413         if (error && error != VT_ENOSPC) {
2414                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2415                     "relabel '%s': unable to read disk capacity"), path);
2416                 return (zfs_error(hdl, EZFS_NOCAP, msg));
2417         }
2418         return (0);
2419 }
2420
2421 /*
2422  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2423  * ZFS_ONLINE_* flags.
2424  */
2425 int
2426 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2427     vdev_state_t *newstate)
2428 {
2429         zfs_cmd_t zc = {"\0"};
2430         char msg[1024];
2431         nvlist_t *tgt;
2432         boolean_t avail_spare, l2cache, islog;
2433         libzfs_handle_t *hdl = zhp->zpool_hdl;
2434         int error;
2435
2436         if (flags & ZFS_ONLINE_EXPAND) {
2437                 (void) snprintf(msg, sizeof (msg),
2438                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2439         } else {
2440                 (void) snprintf(msg, sizeof (msg),
2441                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2442         }
2443
2444         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2445         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2446             &islog)) == NULL)
2447                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2448
2449         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2450
2451         if (avail_spare)
2452                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2453
2454         if (flags & ZFS_ONLINE_EXPAND ||
2455             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2456                 uint64_t wholedisk = 0;
2457
2458                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2459                     &wholedisk);
2460
2461                 /*
2462                  * XXX - L2ARC 1.0 devices can't support expansion.
2463                  */
2464                 if (l2cache) {
2465                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2466                             "cannot expand cache devices"));
2467                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2468                 }
2469
2470                 if (wholedisk) {
2471                         const char *fullpath = path;
2472                         char buf[MAXPATHLEN];
2473
2474                         if (path[0] != '/') {
2475                                 error = zfs_resolve_shortname(path, buf,
2476                                     sizeof (buf));
2477                                 if (error != 0)
2478                                         return (zfs_error(hdl, EZFS_NODEVICE,
2479                                             msg));
2480
2481                                 fullpath = buf;
2482                         }
2483
2484                         error = zpool_relabel_disk(hdl, fullpath, msg);
2485                         if (error != 0)
2486                                 return (error);
2487                 }
2488         }
2489
2490         zc.zc_cookie = VDEV_STATE_ONLINE;
2491         zc.zc_obj = flags;
2492
2493         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2494                 if (errno == EINVAL) {
2495                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2496                             "from this pool into a new one.  Use '%s' "
2497                             "instead"), "zpool detach");
2498                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2499                 }
2500                 return (zpool_standard_error(hdl, errno, msg));
2501         }
2502
2503         *newstate = zc.zc_cookie;
2504         return (0);
2505 }
2506
2507 /*
2508  * Take the specified vdev offline
2509  */
2510 int
2511 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2512 {
2513         zfs_cmd_t zc = {"\0"};
2514         char msg[1024];
2515         nvlist_t *tgt;
2516         boolean_t avail_spare, l2cache;
2517         libzfs_handle_t *hdl = zhp->zpool_hdl;
2518
2519         (void) snprintf(msg, sizeof (msg),
2520             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2521
2522         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2523         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2524             NULL)) == NULL)
2525                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2526
2527         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2528
2529         if (avail_spare)
2530                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2531
2532         zc.zc_cookie = VDEV_STATE_OFFLINE;
2533         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2534
2535         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2536                 return (0);
2537
2538         switch (errno) {
2539         case EBUSY:
2540
2541                 /*
2542                  * There are no other replicas of this device.
2543                  */
2544                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2545
2546         case EEXIST:
2547                 /*
2548                  * The log device has unplayed logs
2549                  */
2550                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2551
2552         default:
2553                 return (zpool_standard_error(hdl, errno, msg));
2554         }
2555 }
2556
2557 /*
2558  * Mark the given vdev faulted.
2559  */
2560 int
2561 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2562 {
2563         zfs_cmd_t zc = {"\0"};
2564         char msg[1024];
2565         libzfs_handle_t *hdl = zhp->zpool_hdl;
2566
2567         (void) snprintf(msg, sizeof (msg),
2568             dgettext(TEXT_DOMAIN, "cannot fault %llu"), (u_longlong_t)guid);
2569
2570         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2571         zc.zc_guid = guid;
2572         zc.zc_cookie = VDEV_STATE_FAULTED;
2573         zc.zc_obj = aux;
2574
2575         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2576                 return (0);
2577
2578         switch (errno) {
2579         case EBUSY:
2580
2581                 /*
2582                  * There are no other replicas of this device.
2583                  */
2584                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2585
2586         default:
2587                 return (zpool_standard_error(hdl, errno, msg));
2588         }
2589
2590 }
2591
2592 /*
2593  * Mark the given vdev degraded.
2594  */
2595 int
2596 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2597 {
2598         zfs_cmd_t zc = {"\0"};
2599         char msg[1024];
2600         libzfs_handle_t *hdl = zhp->zpool_hdl;
2601
2602         (void) snprintf(msg, sizeof (msg),
2603             dgettext(TEXT_DOMAIN, "cannot degrade %llu"), (u_longlong_t)guid);
2604
2605         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2606         zc.zc_guid = guid;
2607         zc.zc_cookie = VDEV_STATE_DEGRADED;
2608         zc.zc_obj = aux;
2609
2610         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2611                 return (0);
2612
2613         return (zpool_standard_error(hdl, errno, msg));
2614 }
2615
2616 /*
2617  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2618  * a hot spare.
2619  */
2620 static boolean_t
2621 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2622 {
2623         nvlist_t **child;
2624         uint_t c, children;
2625         char *type;
2626
2627         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2628             &children) == 0) {
2629                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2630                     &type) == 0);
2631
2632                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2633                     children == 2 && child[which] == tgt)
2634                         return (B_TRUE);
2635
2636                 for (c = 0; c < children; c++)
2637                         if (is_replacing_spare(child[c], tgt, which))
2638                                 return (B_TRUE);
2639         }
2640
2641         return (B_FALSE);
2642 }
2643
2644 /*
2645  * Attach new_disk (fully described by nvroot) to old_disk.
2646  * If 'replacing' is specified, the new disk will replace the old one.
2647  */
2648 int
2649 zpool_vdev_attach(zpool_handle_t *zhp,
2650     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2651 {
2652         zfs_cmd_t zc = {"\0"};
2653         char msg[1024];
2654         int ret;
2655         nvlist_t *tgt;
2656         boolean_t avail_spare, l2cache, islog;
2657         uint64_t val;
2658         char *newname;
2659         nvlist_t **child;
2660         uint_t children;
2661         nvlist_t *config_root;
2662         libzfs_handle_t *hdl = zhp->zpool_hdl;
2663         boolean_t rootpool = zpool_is_bootable(zhp);
2664
2665         if (replacing)
2666                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2667                     "cannot replace %s with %s"), old_disk, new_disk);
2668         else
2669                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2670                     "cannot attach %s to %s"), new_disk, old_disk);
2671
2672 #if defined(__sun__) || defined(__sun)
2673         /*
2674          * If this is a root pool, make sure that we're not attaching an
2675          * EFI labeled device.
2676          */
2677         if (rootpool && pool_uses_efi(nvroot)) {
2678                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2679                     "EFI labeled devices are not supported on root pools."));
2680                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
2681         }
2682 #endif
2683
2684         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2685         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2686             &islog)) == 0)
2687                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2688
2689         if (avail_spare)
2690                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2691
2692         if (l2cache)
2693                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2694
2695         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2696         zc.zc_cookie = replacing;
2697
2698         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2699             &child, &children) != 0 || children != 1) {
2700                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2701                     "new device must be a single disk"));
2702                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2703         }
2704
2705         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2706             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2707
2708         if ((newname = zpool_vdev_name(NULL, NULL, child[0], 0)) == NULL)
2709                 return (-1);
2710
2711         /*
2712          * If the target is a hot spare that has been swapped in, we can only
2713          * replace it with another hot spare.
2714          */
2715         if (replacing &&
2716             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2717             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2718             NULL) == NULL || !avail_spare) &&
2719             is_replacing_spare(config_root, tgt, 1)) {
2720                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2721                     "can only be replaced by another hot spare"));
2722                 free(newname);
2723                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2724         }
2725
2726         free(newname);
2727
2728         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2729                 return (-1);
2730
2731         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2732
2733         zcmd_free_nvlists(&zc);
2734
2735         if (ret == 0) {
2736                 if (rootpool) {
2737                         /*
2738                          * XXX need a better way to prevent user from
2739                          * booting up a half-baked vdev.
2740                          */
2741                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2742                             "sure to wait until resilver is done "
2743                             "before rebooting.\n"));
2744                 }
2745                 return (0);
2746         }
2747
2748         switch (errno) {
2749         case ENOTSUP:
2750                 /*
2751                  * Can't attach to or replace this type of vdev.
2752                  */
2753                 if (replacing) {
2754                         uint64_t version = zpool_get_prop_int(zhp,
2755                             ZPOOL_PROP_VERSION, NULL);
2756
2757                         if (islog)
2758                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2759                                     "cannot replace a log with a spare"));
2760                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2761                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2762                                     "already in replacing/spare config; wait "
2763                                     "for completion or use 'zpool detach'"));
2764                         else
2765                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2766                                     "cannot replace a replacing device"));
2767                 } else {
2768                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2769                             "can only attach to mirrors and top-level "
2770                             "disks"));
2771                 }
2772                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2773                 break;
2774
2775         case EINVAL:
2776                 /*
2777                  * The new device must be a single disk.
2778                  */
2779                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2780                     "new device must be a single disk"));
2781                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2782                 break;
2783
2784         case EBUSY:
2785                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2786                     new_disk);
2787                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2788                 break;
2789
2790         case EOVERFLOW:
2791                 /*
2792                  * The new device is too small.
2793                  */
2794                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2795                     "device is too small"));
2796                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2797                 break;
2798
2799         case EDOM:
2800                 /*
2801                  * The new device has a different optimal sector size.
2802                  */
2803                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2804                     "new device has a different optimal sector size; use the "
2805                     "option '-o ashift=N' to override the optimal size"));
2806                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2807                 break;
2808
2809         case ENAMETOOLONG:
2810                 /*
2811                  * The resulting top-level vdev spec won't fit in the label.
2812                  */
2813                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2814                 break;
2815
2816         default:
2817                 (void) zpool_standard_error(hdl, errno, msg);
2818         }
2819
2820         return (-1);
2821 }
2822
2823 /*
2824  * Detach the specified device.
2825  */
2826 int
2827 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2828 {
2829         zfs_cmd_t zc = {"\0"};
2830         char msg[1024];
2831         nvlist_t *tgt;
2832         boolean_t avail_spare, l2cache;
2833         libzfs_handle_t *hdl = zhp->zpool_hdl;
2834
2835         (void) snprintf(msg, sizeof (msg),
2836             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2837
2838         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2839         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2840             NULL)) == 0)
2841                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2842
2843         if (avail_spare)
2844                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2845
2846         if (l2cache)
2847                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2848
2849         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2850
2851         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2852                 return (0);
2853
2854         switch (errno) {
2855
2856         case ENOTSUP:
2857                 /*
2858                  * Can't detach from this type of vdev.
2859                  */
2860                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2861                     "applicable to mirror and replacing vdevs"));
2862                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2863                 break;
2864
2865         case EBUSY:
2866                 /*
2867                  * There are no other replicas of this device.
2868                  */
2869                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2870                 break;
2871
2872         default:
2873                 (void) zpool_standard_error(hdl, errno, msg);
2874         }
2875
2876         return (-1);
2877 }
2878
2879 /*
2880  * Find a mirror vdev in the source nvlist.
2881  *
2882  * The mchild array contains a list of disks in one of the top-level mirrors
2883  * of the source pool.  The schild array contains a list of disks that the
2884  * user specified on the command line.  We loop over the mchild array to
2885  * see if any entry in the schild array matches.
2886  *
2887  * If a disk in the mchild array is found in the schild array, we return
2888  * the index of that entry.  Otherwise we return -1.
2889  */
2890 static int
2891 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2892     nvlist_t **schild, uint_t schildren)
2893 {
2894         uint_t mc;
2895
2896         for (mc = 0; mc < mchildren; mc++) {
2897                 uint_t sc;
2898                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2899                     mchild[mc], 0);
2900
2901                 for (sc = 0; sc < schildren; sc++) {
2902                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2903                             schild[sc], 0);
2904                         boolean_t result = (strcmp(mpath, spath) == 0);
2905
2906                         free(spath);
2907                         if (result) {
2908                                 free(mpath);
2909                                 return (mc);
2910                         }
2911                 }
2912
2913                 free(mpath);
2914         }
2915
2916         return (-1);
2917 }
2918
2919 /*
2920  * Split a mirror pool.  If newroot points to null, then a new nvlist
2921  * is generated and it is the responsibility of the caller to free it.
2922  */
2923 int
2924 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2925     nvlist_t *props, splitflags_t flags)
2926 {
2927         zfs_cmd_t zc = {"\0"};
2928         char msg[1024];
2929         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2930         nvlist_t **varray = NULL, *zc_props = NULL;
2931         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2932         libzfs_handle_t *hdl = zhp->zpool_hdl;
2933         uint64_t vers;
2934         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2935         int retval = 0;
2936
2937         (void) snprintf(msg, sizeof (msg),
2938             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2939
2940         if (!zpool_name_valid(hdl, B_FALSE, newname))
2941                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2942
2943         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2944                 (void) fprintf(stderr, gettext("Internal error: unable to "
2945                     "retrieve pool configuration\n"));
2946                 return (-1);
2947         }
2948
2949         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2950             == 0);
2951         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2952
2953         if (props) {
2954                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2955                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2956                     props, vers, flags, msg)) == NULL)
2957                         return (-1);
2958         }
2959
2960         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2961             &children) != 0) {
2962                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2963                     "Source pool is missing vdev tree"));
2964                 nvlist_free(zc_props);
2965                 return (-1);
2966         }
2967
2968         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2969         vcount = 0;
2970
2971         if (*newroot == NULL ||
2972             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2973             &newchild, &newchildren) != 0)
2974                 newchildren = 0;
2975
2976         for (c = 0; c < children; c++) {
2977                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2978                 char *type;
2979                 nvlist_t **mchild, *vdev;
2980                 uint_t mchildren;
2981                 int entry;
2982
2983                 /*
2984                  * Unlike cache & spares, slogs are stored in the
2985                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2986                  */
2987                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2988                     &is_log);
2989                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2990                     &is_hole);
2991                 if (is_log || is_hole) {
2992                         /*
2993                          * Create a hole vdev and put it in the config.
2994                          */
2995                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2996                                 goto out;
2997                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2998                             VDEV_TYPE_HOLE) != 0)
2999                                 goto out;
3000                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
3001                             1) != 0)
3002                                 goto out;
3003                         if (lastlog == 0)
3004                                 lastlog = vcount;
3005                         varray[vcount++] = vdev;
3006                         continue;
3007                 }
3008                 lastlog = 0;
3009                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
3010                     == 0);
3011                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
3012                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3013                             "Source pool must be composed only of mirrors\n"));
3014                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3015                         goto out;
3016                 }
3017
3018                 verify(nvlist_lookup_nvlist_array(child[c],
3019                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
3020
3021                 /* find or add an entry for this top-level vdev */
3022                 if (newchildren > 0 &&
3023                     (entry = find_vdev_entry(zhp, mchild, mchildren,
3024                     newchild, newchildren)) >= 0) {
3025                         /* We found a disk that the user specified. */
3026                         vdev = mchild[entry];
3027                         ++found;
3028                 } else {
3029                         /* User didn't specify a disk for this vdev. */
3030                         vdev = mchild[mchildren - 1];
3031                 }
3032
3033                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
3034                         goto out;
3035         }
3036
3037         /* did we find every disk the user specified? */
3038         if (found != newchildren) {
3039                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
3040                     "include at most one disk from each mirror"));
3041                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3042                 goto out;
3043         }
3044
3045         /* Prepare the nvlist for populating. */
3046         if (*newroot == NULL) {
3047                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
3048                         goto out;
3049                 freelist = B_TRUE;
3050                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
3051                     VDEV_TYPE_ROOT) != 0)
3052                         goto out;
3053         } else {
3054                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
3055         }
3056
3057         /* Add all the children we found */
3058         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
3059             lastlog == 0 ? vcount : lastlog) != 0)
3060                 goto out;
3061
3062         /*
3063          * If we're just doing a dry run, exit now with success.
3064          */
3065         if (flags.dryrun) {
3066                 memory_err = B_FALSE;
3067                 freelist = B_FALSE;
3068                 goto out;
3069         }
3070
3071         /* now build up the config list & call the ioctl */
3072         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3073                 goto out;
3074
3075         if (nvlist_add_nvlist(newconfig,
3076             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3077             nvlist_add_string(newconfig,
3078             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3079             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3080                 goto out;
3081
3082         /*
3083          * The new pool is automatically part of the namespace unless we
3084          * explicitly export it.
3085          */
3086         if (!flags.import)
3087                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3088         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3089         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3090         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
3091                 goto out;
3092         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
3093                 goto out;
3094
3095         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3096                 retval = zpool_standard_error(hdl, errno, msg);
3097                 goto out;
3098         }
3099
3100         freelist = B_FALSE;
3101         memory_err = B_FALSE;
3102
3103 out:
3104         if (varray != NULL) {
3105                 int v;
3106
3107                 for (v = 0; v < vcount; v++)
3108                         nvlist_free(varray[v]);
3109                 free(varray);
3110         }
3111         zcmd_free_nvlists(&zc);
3112         nvlist_free(zc_props);
3113         nvlist_free(newconfig);
3114         if (freelist) {
3115                 nvlist_free(*newroot);
3116                 *newroot = NULL;
3117         }
3118
3119         if (retval != 0)
3120                 return (retval);
3121
3122         if (memory_err)
3123                 return (no_memory(hdl));
3124
3125         return (0);
3126 }
3127
3128 /*
3129  * Remove the given device.  Currently, this is supported only for hot spares
3130  * and level 2 cache devices.
3131  */
3132 int
3133 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3134 {
3135         zfs_cmd_t zc = {"\0"};
3136         char msg[1024];
3137         nvlist_t *tgt;
3138         boolean_t avail_spare, l2cache, islog;
3139         libzfs_handle_t *hdl = zhp->zpool_hdl;
3140         uint64_t version;
3141
3142         (void) snprintf(msg, sizeof (msg),
3143             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3144
3145         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3146         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3147             &islog)) == 0)
3148                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3149         /*
3150          * XXX - this should just go away.
3151          */
3152         if (!avail_spare && !l2cache && !islog) {
3153                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3154                     "only inactive hot spares, cache, top-level, "
3155                     "or log devices can be removed"));
3156                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3157         }
3158
3159         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3160         if (islog && version < SPA_VERSION_HOLES) {
3161                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3162                     "pool must be upgrade to support log removal"));
3163                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
3164         }
3165
3166         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
3167
3168         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3169                 return (0);
3170
3171         return (zpool_standard_error(hdl, errno, msg));
3172 }
3173
3174 /*
3175  * Clear the errors for the pool, or the particular device if specified.
3176  */
3177 int
3178 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3179 {
3180         zfs_cmd_t zc = {"\0"};
3181         char msg[1024];
3182         nvlist_t *tgt;
3183         zpool_rewind_policy_t policy;
3184         boolean_t avail_spare, l2cache;
3185         libzfs_handle_t *hdl = zhp->zpool_hdl;
3186         nvlist_t *nvi = NULL;
3187         int error;
3188
3189         if (path)
3190                 (void) snprintf(msg, sizeof (msg),
3191                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3192                     path);
3193         else
3194                 (void) snprintf(msg, sizeof (msg),
3195                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3196                     zhp->zpool_name);
3197
3198         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3199         if (path) {
3200                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3201                     &l2cache, NULL)) == 0)
3202                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
3203
3204                 /*
3205                  * Don't allow error clearing for hot spares.  Do allow
3206                  * error clearing for l2cache devices.
3207                  */
3208                 if (avail_spare)
3209                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
3210
3211                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
3212                     &zc.zc_guid) == 0);
3213         }
3214
3215         zpool_get_rewind_policy(rewindnvl, &policy);
3216         zc.zc_cookie = policy.zrp_request;
3217
3218         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
3219                 return (-1);
3220
3221         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
3222                 return (-1);
3223
3224         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3225             errno == ENOMEM) {
3226                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3227                         zcmd_free_nvlists(&zc);
3228                         return (-1);
3229                 }
3230         }
3231
3232         if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
3233             errno != EPERM && errno != EACCES)) {
3234                 if (policy.zrp_request &
3235                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
3236                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3237                         zpool_rewind_exclaim(hdl, zc.zc_name,
3238                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
3239                             nvi);
3240                         nvlist_free(nvi);
3241                 }
3242                 zcmd_free_nvlists(&zc);
3243                 return (0);
3244         }
3245
3246         zcmd_free_nvlists(&zc);
3247         return (zpool_standard_error(hdl, errno, msg));
3248 }
3249
3250 /*
3251  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3252  */
3253 int
3254 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3255 {
3256         zfs_cmd_t zc = {"\0"};
3257         char msg[1024];
3258         libzfs_handle_t *hdl = zhp->zpool_hdl;
3259
3260         (void) snprintf(msg, sizeof (msg),
3261             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3262             (u_longlong_t)guid);
3263
3264         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3265         zc.zc_guid = guid;
3266         zc.zc_cookie = ZPOOL_NO_REWIND;
3267
3268         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
3269                 return (0);
3270
3271         return (zpool_standard_error(hdl, errno, msg));
3272 }
3273
3274 /*
3275  * Change the GUID for a pool.
3276  */
3277 int
3278 zpool_reguid(zpool_handle_t *zhp)
3279 {
3280         char msg[1024];
3281         libzfs_handle_t *hdl = zhp->zpool_hdl;
3282         zfs_cmd_t zc = {"\0"};
3283
3284         (void) snprintf(msg, sizeof (msg),
3285             dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3286
3287         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3288         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3289                 return (0);
3290
3291         return (zpool_standard_error(hdl, errno, msg));
3292 }
3293
3294 /*
3295  * Reopen the pool.
3296  */
3297 int
3298 zpool_reopen(zpool_handle_t *zhp)
3299 {
3300         zfs_cmd_t zc = {"\0"};
3301         char msg[1024];
3302         libzfs_handle_t *hdl = zhp->zpool_hdl;
3303
3304         (void) snprintf(msg, sizeof (msg),
3305             dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3306             zhp->zpool_name);
3307
3308         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3309         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3310                 return (0);
3311         return (zpool_standard_error(hdl, errno, msg));
3312 }
3313
3314 #if defined(__sun__) || defined(__sun)
3315 /*
3316  * Convert from a devid string to a path.
3317  */
3318 static char *
3319 devid_to_path(char *devid_str)
3320 {
3321         ddi_devid_t devid;
3322         char *minor;
3323         char *path;
3324         devid_nmlist_t *list = NULL;
3325         int ret;
3326
3327         if (devid_str_decode(devid_str, &devid, &minor) != 0)
3328                 return (NULL);
3329
3330         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3331
3332         devid_str_free(minor);
3333         devid_free(devid);
3334
3335         if (ret != 0)
3336                 return (NULL);
3337
3338         /*
3339          * In a case the strdup() fails, we will just return NULL below.
3340          */
3341         path = strdup(list[0].devname);
3342
3343         devid_free_nmlist(list);
3344
3345         return (path);
3346 }
3347
3348 /*
3349  * Convert from a path to a devid string.
3350  */
3351 static char *
3352 path_to_devid(const char *path)
3353 {
3354         int fd;
3355         ddi_devid_t devid;
3356         char *minor, *ret;
3357
3358         if ((fd = open(path, O_RDONLY)) < 0)
3359                 return (NULL);
3360
3361         minor = NULL;
3362         ret = NULL;
3363         if (devid_get(fd, &devid) == 0) {
3364                 if (devid_get_minor_name(fd, &minor) == 0)
3365                         ret = devid_str_encode(devid, minor);
3366                 if (minor != NULL)
3367                         devid_str_free(minor);
3368                 devid_free(devid);
3369         }
3370         (void) close(fd);
3371
3372         return (ret);
3373 }
3374
3375 /*
3376  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3377  * ignore any failure here, since a common case is for an unprivileged user to
3378  * type 'zpool status', and we'll display the correct information anyway.
3379  */
3380 static void
3381 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3382 {
3383         zfs_cmd_t zc = {"\0"};
3384
3385         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3386         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3387         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3388             &zc.zc_guid) == 0);
3389
3390         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3391 }
3392 #endif /* sun */
3393
3394 /*
3395  * Remove partition suffix from a vdev path.  Partition suffixes may take three
3396  * forms: "-partX", "pX", or "X", where X is a string of digits.  The second
3397  * case only occurs when the suffix is preceded by a digit, i.e. "md0p0" The
3398  * third case only occurs when preceded by a string matching the regular
3399  * expression "^([hsv]|xv)d[a-z]+", i.e. a scsi, ide, virtio or xen disk.
3400  *
3401  * caller must free the returned string
3402  */
3403 char *
3404 zfs_strip_partition(libzfs_handle_t *hdl, char *path)
3405 {
3406         char *tmp = zfs_strdup(hdl, path);
3407         char *part = NULL, *d = NULL;
3408
3409         if ((part = strstr(tmp, "-part")) && part != tmp) {
3410                 d = part + 5;
3411         } else if ((part = strrchr(tmp, 'p')) &&
3412             part > tmp + 1 && isdigit(*(part-1))) {
3413                 d = part + 1;
3414         } else if ((tmp[0] == 'h' || tmp[0] == 's' || tmp[0] == 'v') &&
3415             tmp[1] == 'd') {
3416                 for (d = &tmp[2]; isalpha(*d); part = ++d);
3417         } else if (strncmp("xvd", tmp, 3) == 0) {
3418                 for (d = &tmp[3]; isalpha(*d); part = ++d);
3419         }
3420         if (part && d && *d != '\0') {
3421                 for (; isdigit(*d); d++);
3422                 if (*d == '\0')
3423                         *part = '\0';
3424         }
3425         return (tmp);
3426 }
3427
3428 #define PATH_BUF_LEN    64
3429
3430 /*
3431  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3432  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3433  * We also check if this is a whole disk, in which case we strip off the
3434  * trailing 's0' slice name.
3435  *
3436  * This routine is also responsible for identifying when disks have been
3437  * reconfigured in a new location.  The kernel will have opened the device by
3438  * devid, but the path will still refer to the old location.  To catch this, we
3439  * first do a path -> devid translation (which is fast for the common case).  If
3440  * the devid matches, we're done.  If not, we do a reverse devid -> path
3441  * translation and issue the appropriate ioctl() to update the path of the vdev.
3442  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3443  * of these checks.
3444  */
3445 char *
3446 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3447     int name_flags)
3448 {
3449         char *path, *type, *env;
3450         uint64_t value;
3451         char buf[PATH_BUF_LEN];
3452         char tmpbuf[PATH_BUF_LEN];
3453
3454         env = getenv("ZPOOL_VDEV_NAME_PATH");
3455         if (env && (strtoul(env, NULL, 0) > 0 ||
3456             !strncasecmp(env, "YES", 3) || !strncasecmp(env, "ON", 2)))
3457                 name_flags |= VDEV_NAME_PATH;
3458
3459         env = getenv("ZPOOL_VDEV_NAME_GUID");
3460         if (env && (strtoul(env, NULL, 0) > 0 ||
3461             !strncasecmp(env, "YES", 3) || !strncasecmp(env, "ON", 2)))
3462                 name_flags |= VDEV_NAME_GUID;
3463
3464         env = getenv("ZPOOL_VDEV_NAME_FOLLOW_LINKS");
3465         if (env && (strtoul(env, NULL, 0) > 0 ||
3466             !strncasecmp(env, "YES", 3) || !strncasecmp(env, "ON", 2)))
3467                 name_flags |= VDEV_NAME_FOLLOW_LINKS;
3468
3469         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &value) == 0 ||
3470             name_flags & VDEV_NAME_GUID) {
3471                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value);
3472                 (void) snprintf(buf, sizeof (buf), "%llu", (u_longlong_t)value);
3473                 path = buf;
3474         } else if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0) {
3475 #if defined(__sun__) || defined(__sun)
3476                 /*
3477                  * Live VDEV path updates to a kernel VDEV during a
3478                  * zpool_vdev_name lookup are not supported on Linux.
3479                  */
3480                 char *devid;
3481                 vdev_stat_t *vs;
3482                 uint_t vsc;
3483
3484                 /*
3485                  * If the device is dead (faulted, offline, etc) then don't
3486                  * bother opening it.  Otherwise we may be forcing the user to
3487                  * open a misbehaving device, which can have undesirable
3488                  * effects.
3489                  */
3490                 if ((nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3491                     (uint64_t **)&vs, &vsc) != 0 ||
3492                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3493                     zhp != NULL &&
3494                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3495                         /*
3496                          * Determine if the current path is correct.
3497                          */
3498                         char *newdevid = path_to_devid(path);
3499
3500                         if (newdevid == NULL ||
3501                             strcmp(devid, newdevid) != 0) {
3502                                 char *newpath;
3503
3504                                 if ((newpath = devid_to_path(devid)) != NULL) {
3505                                         /*
3506                                          * Update the path appropriately.
3507                                          */
3508                                         set_path(zhp, nv, newpath);
3509                                         if (nvlist_add_string(nv,
3510                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3511                                                 verify(nvlist_lookup_string(nv,
3512                                                     ZPOOL_CONFIG_PATH,
3513                                                     &path) == 0);
3514                                         free(newpath);
3515                                 }
3516                         }
3517
3518                         if (newdevid)
3519                                 devid_str_free(newdevid);
3520                 }
3521 #endif /* sun */
3522
3523                 if (name_flags & VDEV_NAME_FOLLOW_LINKS) {
3524                         char *rp = realpath(path, NULL);
3525                         if (rp) {
3526                                 strlcpy(buf, rp, sizeof (buf));
3527                                 path = buf;
3528                                 free(rp);
3529                         }
3530                 }
3531
3532                 /*
3533                  * For a block device only use the name.
3534                  */
3535                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
3536                 if ((strcmp(type, VDEV_TYPE_DISK) == 0) &&
3537                     !(name_flags & VDEV_NAME_PATH)) {
3538                         path = strrchr(path, '/');
3539                         path++;
3540                 }
3541
3542                 /*
3543                  * Remove the partition from the path it this is a whole disk.
3544                  */
3545                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, &value)
3546                     == 0 && value && !(name_flags & VDEV_NAME_PATH)) {
3547                         return (zfs_strip_partition(hdl, path));
3548                 }
3549         } else {
3550                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3551
3552                 /*
3553                  * If it's a raidz device, we need to stick in the parity level.
3554                  */
3555                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3556                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3557                             &value) == 0);
3558                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3559                             (u_longlong_t)value);
3560                         path = buf;
3561                 }
3562
3563                 /*
3564                  * We identify each top-level vdev by using a <type-id>
3565                  * naming convention.
3566                  */
3567                 if (name_flags & VDEV_NAME_TYPE_ID) {
3568                         uint64_t id;
3569                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3570                             &id) == 0);
3571                         (void) snprintf(tmpbuf, sizeof (tmpbuf), "%s-%llu",
3572                             path, (u_longlong_t)id);
3573                         path = tmpbuf;
3574                 }
3575         }
3576
3577         return (zfs_strdup(hdl, path));
3578 }
3579
3580 static int
3581 zbookmark_mem_compare(const void *a, const void *b)
3582 {
3583         return (memcmp(a, b, sizeof (zbookmark_phys_t)));
3584 }
3585
3586 /*
3587  * Retrieve the persistent error log, uniquify the members, and return to the
3588  * caller.
3589  */
3590 int
3591 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3592 {
3593         zfs_cmd_t zc = {"\0"};
3594         uint64_t count;
3595         zbookmark_phys_t *zb = NULL;
3596         int i;
3597
3598         /*
3599          * Retrieve the raw error list from the kernel.  If the number of errors
3600          * has increased, allocate more space and continue until we get the
3601          * entire list.
3602          */
3603         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3604             &count) == 0);
3605         if (count == 0)
3606                 return (0);
3607         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3608             count * sizeof (zbookmark_phys_t))) == (uintptr_t)NULL)
3609                 return (-1);
3610         zc.zc_nvlist_dst_size = count;
3611         (void) strcpy(zc.zc_name, zhp->zpool_name);
3612         for (;;) {
3613                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3614                     &zc) != 0) {
3615                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3616                         if (errno == ENOMEM) {
3617                                 void *dst;
3618
3619                                 count = zc.zc_nvlist_dst_size;
3620                                 dst = zfs_alloc(zhp->zpool_hdl, count *
3621                                     sizeof (zbookmark_phys_t));
3622                                 if (dst == NULL)
3623                                         return (-1);
3624                                 zc.zc_nvlist_dst = (uintptr_t)dst;
3625                         } else {
3626                                 return (-1);
3627                         }
3628                 } else {
3629                         break;
3630                 }
3631         }
3632
3633         /*
3634          * Sort the resulting bookmarks.  This is a little confusing due to the
3635          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3636          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3637          * _not_ copied as part of the process.  So we point the start of our
3638          * array appropriate and decrement the total number of elements.
3639          */
3640         zb = ((zbookmark_phys_t *)(uintptr_t)zc.zc_nvlist_dst) +
3641             zc.zc_nvlist_dst_size;
3642         count -= zc.zc_nvlist_dst_size;
3643
3644         qsort(zb, count, sizeof (zbookmark_phys_t), zbookmark_mem_compare);
3645
3646         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3647
3648         /*
3649          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3650          */
3651         for (i = 0; i < count; i++) {
3652                 nvlist_t *nv;
3653
3654                 /* ignoring zb_blkid and zb_level for now */
3655                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3656                     zb[i-1].zb_object == zb[i].zb_object)
3657                         continue;
3658
3659                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3660                         goto nomem;
3661                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3662                     zb[i].zb_objset) != 0) {
3663                         nvlist_free(nv);
3664                         goto nomem;
3665                 }
3666                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3667                     zb[i].zb_object) != 0) {
3668                         nvlist_free(nv);
3669                         goto nomem;
3670                 }
3671                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3672                         nvlist_free(nv);
3673                         goto nomem;
3674                 }
3675                 nvlist_free(nv);
3676         }
3677
3678         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3679         return (0);
3680
3681 nomem:
3682         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3683         return (no_memory(zhp->zpool_hdl));
3684 }
3685
3686 /*
3687  * Upgrade a ZFS pool to the latest on-disk version.
3688  */
3689 int
3690 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3691 {
3692         zfs_cmd_t zc = {"\0"};
3693         libzfs_handle_t *hdl = zhp->zpool_hdl;
3694
3695         (void) strcpy(zc.zc_name, zhp->zpool_name);
3696         zc.zc_cookie = new_version;
3697
3698         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3699                 return (zpool_standard_error_fmt(hdl, errno,
3700                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3701                     zhp->zpool_name));
3702         return (0);
3703 }
3704
3705 void
3706 zfs_save_arguments(int argc, char **argv, char *string, int len)
3707 {
3708         int i;
3709
3710         (void) strlcpy(string, basename(argv[0]), len);
3711         for (i = 1; i < argc; i++) {
3712                 (void) strlcat(string, " ", len);
3713                 (void) strlcat(string, argv[i], len);
3714         }
3715 }
3716
3717 int
3718 zpool_log_history(libzfs_handle_t *hdl, const char *message)
3719 {
3720         zfs_cmd_t zc = {"\0"};
3721         nvlist_t *args;
3722         int err;
3723
3724         args = fnvlist_alloc();
3725         fnvlist_add_string(args, "message", message);
3726         err = zcmd_write_src_nvlist(hdl, &zc, args);
3727         if (err == 0)
3728                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_LOG_HISTORY, &zc);
3729         nvlist_free(args);
3730         zcmd_free_nvlists(&zc);
3731         return (err);
3732 }
3733
3734 /*
3735  * Perform ioctl to get some command history of a pool.
3736  *
3737  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3738  * logical offset of the history buffer to start reading from.
3739  *
3740  * Upon return, 'off' is the next logical offset to read from and
3741  * 'len' is the actual amount of bytes read into 'buf'.
3742  */
3743 static int
3744 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3745 {
3746         zfs_cmd_t zc = {"\0"};
3747         libzfs_handle_t *hdl = zhp->zpool_hdl;
3748
3749         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3750
3751         zc.zc_history = (uint64_t)(uintptr_t)buf;
3752         zc.zc_history_len = *len;
3753         zc.zc_history_offset = *off;
3754
3755         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3756                 switch (errno) {
3757                 case EPERM:
3758                         return (zfs_error_fmt(hdl, EZFS_PERM,
3759                             dgettext(TEXT_DOMAIN,
3760                             "cannot show history for pool '%s'"),
3761                             zhp->zpool_name));
3762                 case ENOENT:
3763                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3764                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3765                             "'%s'"), zhp->zpool_name));
3766                 case ENOTSUP:
3767                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3768                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3769                             "'%s', pool must be upgraded"), zhp->zpool_name));
3770                 default:
3771                         return (zpool_standard_error_fmt(hdl, errno,
3772                             dgettext(TEXT_DOMAIN,
3773                             "cannot get history for '%s'"), zhp->zpool_name));
3774                 }
3775         }
3776
3777         *len = zc.zc_history_len;
3778         *off = zc.zc_history_offset;
3779
3780         return (0);
3781 }
3782
3783 /*
3784  * Process the buffer of nvlists, unpacking and storing each nvlist record
3785  * into 'records'.  'leftover' is set to the number of bytes that weren't
3786  * processed as there wasn't a complete record.
3787  */
3788 int
3789 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3790     nvlist_t ***records, uint_t *numrecords)
3791 {
3792         uint64_t reclen;
3793         nvlist_t *nv;
3794         int i;
3795
3796         while (bytes_read > sizeof (reclen)) {
3797
3798                 /* get length of packed record (stored as little endian) */
3799                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3800                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3801
3802                 if (bytes_read < sizeof (reclen) + reclen)
3803                         break;
3804
3805                 /* unpack record */
3806                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3807                         return (ENOMEM);
3808                 bytes_read -= sizeof (reclen) + reclen;
3809                 buf += sizeof (reclen) + reclen;
3810
3811                 /* add record to nvlist array */
3812                 (*numrecords)++;
3813                 if (ISP2(*numrecords + 1)) {
3814                         *records = realloc(*records,
3815                             *numrecords * 2 * sizeof (nvlist_t *));
3816                 }
3817                 (*records)[*numrecords - 1] = nv;
3818         }
3819
3820         *leftover = bytes_read;
3821         return (0);
3822 }
3823
3824 /*
3825  * Retrieve the command history of a pool.
3826  */
3827 int
3828 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3829 {
3830         char *buf;
3831         int buflen = 128 * 1024;
3832         uint64_t off = 0;
3833         nvlist_t **records = NULL;
3834         uint_t numrecords = 0;
3835         int err, i;
3836
3837         buf = malloc(buflen);
3838         if (buf == NULL)
3839                 return (ENOMEM);
3840         do {
3841                 uint64_t bytes_read = buflen;
3842                 uint64_t leftover;
3843
3844                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3845                         break;
3846
3847                 /* if nothing else was read in, we're at EOF, just return */
3848                 if (!bytes_read)
3849                         break;
3850
3851                 if ((err = zpool_history_unpack(buf, bytes_read,
3852                     &leftover, &records, &numrecords)) != 0)
3853                         break;
3854                 off -= leftover;
3855                 if (leftover == bytes_read) {
3856                         /*
3857                          * no progress made, because buffer is not big enough
3858                          * to hold this record; resize and retry.
3859                          */
3860                         buflen *= 2;
3861                         free(buf);
3862                         buf = malloc(buflen);
3863                         if (buf == NULL)
3864                                 return (ENOMEM);
3865                 }
3866
3867                 /* CONSTCOND */
3868         } while (1);
3869
3870         free(buf);
3871
3872         if (!err) {
3873                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3874                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3875                     records, numrecords) == 0);
3876         }
3877         for (i = 0; i < numrecords; i++)
3878                 nvlist_free(records[i]);
3879         free(records);
3880
3881         return (err);
3882 }
3883
3884 /*
3885  * Retrieve the next event given the passed 'zevent_fd' file descriptor.
3886  * If there is a new event available 'nvp' will contain a newly allocated
3887  * nvlist and 'dropped' will be set to the number of missed events since
3888  * the last call to this function.  When 'nvp' is set to NULL it indicates
3889  * no new events are available.  In either case the function returns 0 and
3890  * it is up to the caller to free 'nvp'.  In the case of a fatal error the
3891  * function will return a non-zero value.  When the function is called in
3892  * blocking mode (the default, unless the ZEVENT_NONBLOCK flag is passed),
3893  * it will not return until a new event is available.
3894  */
3895 int
3896 zpool_events_next(libzfs_handle_t *hdl, nvlist_t **nvp,
3897     int *dropped, unsigned flags, int zevent_fd)
3898 {
3899         zfs_cmd_t zc = {"\0"};
3900         int error = 0;
3901
3902         *nvp = NULL;
3903         *dropped = 0;
3904         zc.zc_cleanup_fd = zevent_fd;
3905
3906         if (flags & ZEVENT_NONBLOCK)
3907                 zc.zc_guid = ZEVENT_NONBLOCK;
3908
3909         if (zcmd_alloc_dst_nvlist(hdl, &zc, ZEVENT_SIZE) != 0)
3910                 return (-1);
3911
3912 retry:
3913         if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_NEXT, &zc) != 0) {
3914                 switch (errno) {
3915                 case ESHUTDOWN:
3916                         error = zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
3917                             dgettext(TEXT_DOMAIN, "zfs shutdown"));
3918                         goto out;
3919                 case ENOENT:
3920                         /* Blocking error case should not occur */
3921                         if (!(flags & ZEVENT_NONBLOCK))
3922                                 error = zpool_standard_error_fmt(hdl, errno,
3923                                     dgettext(TEXT_DOMAIN, "cannot get event"));
3924
3925                         goto out;
3926                 case ENOMEM:
3927                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3928                                 error = zfs_error_fmt(hdl, EZFS_NOMEM,
3929                                     dgettext(TEXT_DOMAIN, "cannot get event"));
3930                                 goto out;
3931                         } else {
3932                                 goto retry;
3933                         }
3934                 default:
3935                         error = zpool_standard_error_fmt(hdl, errno,
3936                             dgettext(TEXT_DOMAIN, "cannot get event"));
3937                         goto out;
3938                 }
3939         }
3940
3941         error = zcmd_read_dst_nvlist(hdl, &zc, nvp);
3942         if (error != 0)
3943                 goto out;
3944
3945         *dropped = (int)zc.zc_cookie;
3946 out:
3947         zcmd_free_nvlists(&zc);
3948
3949         return (error);
3950 }
3951
3952 /*
3953  * Clear all events.
3954  */
3955 int
3956 zpool_events_clear(libzfs_handle_t *hdl, int *count)
3957 {
3958         zfs_cmd_t zc = {"\0"};
3959         char msg[1024];
3960
3961         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
3962             "cannot clear events"));
3963
3964         if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_CLEAR, &zc) != 0)
3965                 return (zpool_standard_error_fmt(hdl, errno, msg));
3966
3967         if (count != NULL)
3968                 *count = (int)zc.zc_cookie; /* # of events cleared */
3969
3970         return (0);
3971 }
3972
3973 /*
3974  * Seek to a specific EID, ZEVENT_SEEK_START, or ZEVENT_SEEK_END for
3975  * the passed zevent_fd file handle.  On success zero is returned,
3976  * otherwise -1 is returned and hdl->libzfs_error is set to the errno.
3977  */
3978 int
3979 zpool_events_seek(libzfs_handle_t *hdl, uint64_t eid, int zevent_fd)
3980 {
3981         zfs_cmd_t zc = {"\0"};
3982         int error = 0;
3983
3984         zc.zc_guid = eid;
3985         zc.zc_cleanup_fd = zevent_fd;
3986
3987         if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_SEEK, &zc) != 0) {
3988                 switch (errno) {
3989                 case ENOENT:
3990                         error = zfs_error_fmt(hdl, EZFS_NOENT,
3991                             dgettext(TEXT_DOMAIN, "cannot get event"));
3992                         break;
3993
3994                 case ENOMEM:
3995                         error = zfs_error_fmt(hdl, EZFS_NOMEM,
3996                             dgettext(TEXT_DOMAIN, "cannot get event"));
3997                         break;
3998
3999                 default:
4000                         error = zpool_standard_error_fmt(hdl, errno,
4001                             dgettext(TEXT_DOMAIN, "cannot get event"));
4002                         break;
4003                 }
4004         }
4005
4006         return (error);
4007 }
4008
4009 void
4010 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
4011     char *pathname, size_t len)
4012 {
4013         zfs_cmd_t zc = {"\0"};
4014         boolean_t mounted = B_FALSE;
4015         char *mntpnt = NULL;
4016         char dsname[ZFS_MAX_DATASET_NAME_LEN];
4017
4018         if (dsobj == 0) {
4019                 /* special case for the MOS */
4020                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>",
4021                     (longlong_t)obj);
4022                 return;
4023         }
4024
4025         /* get the dataset's name */
4026         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
4027         zc.zc_obj = dsobj;
4028         if (ioctl(zhp->zpool_hdl->libzfs_fd,
4029             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
4030                 /* just write out a path of two object numbers */
4031                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
4032                     (longlong_t)dsobj, (longlong_t)obj);
4033                 return;
4034         }
4035         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
4036
4037         /* find out if the dataset is mounted */
4038         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
4039
4040         /* get the corrupted object's path */
4041         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
4042         zc.zc_obj = obj;
4043         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
4044             &zc) == 0) {
4045                 if (mounted) {
4046                         (void) snprintf(pathname, len, "%s%s", mntpnt,
4047                             zc.zc_value);
4048                 } else {
4049                         (void) snprintf(pathname, len, "%s:%s",
4050                             dsname, zc.zc_value);
4051                 }
4052         } else {
4053                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname,
4054                     (longlong_t)obj);
4055         }
4056         free(mntpnt);
4057 }
4058
4059 /*
4060  * Read the EFI label from the config, if a label does not exist then
4061  * pass back the error to the caller. If the caller has passed a non-NULL
4062  * diskaddr argument then we set it to the starting address of the EFI
4063  * partition.
4064  */
4065 static int
4066 read_efi_label(nvlist_t *config, diskaddr_t *sb)
4067 {
4068         char *path;
4069         int fd;
4070         char diskname[MAXPATHLEN];
4071         int err = -1;
4072
4073         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
4074                 return (err);
4075
4076         (void) snprintf(diskname, sizeof (diskname), "%s%s", DISK_ROOT,
4077             strrchr(path, '/'));
4078         if ((fd = open(diskname, O_RDWR|O_DIRECT)) >= 0) {
4079                 struct dk_gpt *vtoc;
4080
4081                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
4082                         if (sb != NULL)
4083                                 *sb = vtoc->efi_parts[0].p_start;
4084                         efi_free(vtoc);
4085                 }
4086                 (void) close(fd);
4087         }
4088         return (err);
4089 }
4090
4091 /*
4092  * determine where a partition starts on a disk in the current
4093  * configuration
4094  */
4095 static diskaddr_t
4096 find_start_block(nvlist_t *config)
4097 {
4098         nvlist_t **child;
4099         uint_t c, children;
4100         diskaddr_t sb = MAXOFFSET_T;
4101         uint64_t wholedisk;
4102
4103         if (nvlist_lookup_nvlist_array(config,
4104             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
4105                 if (nvlist_lookup_uint64(config,
4106                     ZPOOL_CONFIG_WHOLE_DISK,
4107                     &wholedisk) != 0 || !wholedisk) {
4108                         return (MAXOFFSET_T);
4109                 }
4110                 if (read_efi_label(config, &sb) < 0)
4111                         sb = MAXOFFSET_T;
4112                 return (sb);
4113         }
4114
4115         for (c = 0; c < children; c++) {
4116                 sb = find_start_block(child[c]);
4117                 if (sb != MAXOFFSET_T) {
4118                         return (sb);
4119                 }
4120         }
4121         return (MAXOFFSET_T);
4122 }
4123
4124 static int
4125 zpool_label_disk_check(char *path)
4126 {
4127         struct dk_gpt *vtoc;
4128         int fd, err;
4129
4130         if ((fd = open(path, O_RDWR|O_DIRECT)) < 0)
4131                 return (errno);
4132
4133         if ((err = efi_alloc_and_read(fd, &vtoc)) != 0) {
4134                 (void) close(fd);
4135                 return (err);
4136         }
4137
4138         if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
4139                 efi_free(vtoc);
4140                 (void) close(fd);
4141                 return (EIDRM);
4142         }
4143
4144         efi_free(vtoc);
4145         (void) close(fd);
4146         return (0);
4147 }
4148
4149 /*
4150  * Generate a unique partition name for the ZFS member.  Partitions must
4151  * have unique names to ensure udev will be able to create symlinks under
4152  * /dev/disk/by-partlabel/ for all pool members.  The partition names are
4153  * of the form <pool>-<unique-id>.
4154  */
4155 static void
4156 zpool_label_name(char *label_name, int label_size)
4157 {
4158         uint64_t id = 0;
4159         int fd;
4160
4161         fd = open("/dev/urandom", O_RDONLY);
4162         if (fd > 0) {
4163                 if (read(fd, &id, sizeof (id)) != sizeof (id))
4164                         id = 0;
4165
4166                 close(fd);
4167         }
4168
4169         if (id == 0)
4170                 id = (((uint64_t)rand()) << 32) | (uint64_t)rand();
4171
4172         snprintf(label_name, label_size, "zfs-%016llx", (u_longlong_t) id);
4173 }
4174
4175 /*
4176  * Label an individual disk.  The name provided is the short name,
4177  * stripped of any leading /dev path.
4178  */
4179 int
4180 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, char *name)
4181 {
4182         char path[MAXPATHLEN];
4183         struct dk_gpt *vtoc;
4184         int rval, fd;
4185         size_t resv = EFI_MIN_RESV_SIZE;
4186         uint64_t slice_size;
4187         diskaddr_t start_block;
4188         char errbuf[1024];
4189
4190         /* prepare an error message just in case */
4191         (void) snprintf(errbuf, sizeof (errbuf),
4192             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
4193
4194         if (zhp) {
4195                 nvlist_t *nvroot;
4196
4197 #if defined(__sun__) || defined(__sun)
4198                 if (zpool_is_bootable(zhp)) {
4199                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4200                             "EFI labeled devices are not supported on root "
4201                             "pools."));
4202                         return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
4203                 }
4204 #endif
4205
4206                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
4207                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4208
4209                 if (zhp->zpool_start_block == 0)
4210                         start_block = find_start_block(nvroot);
4211                 else
4212                         start_block = zhp->zpool_start_block;
4213                 zhp->zpool_start_block = start_block;
4214         } else {
4215                 /* new pool */
4216                 start_block = NEW_START_BLOCK;
4217         }
4218
4219         (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT, name);
4220
4221         if ((fd = open(path, O_RDWR|O_DIRECT|O_EXCL)) < 0) {
4222                 /*
4223                  * This shouldn't happen.  We've long since verified that this
4224                  * is a valid device.
4225                  */
4226                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
4227                     "label '%s': unable to open device: %d"), path, errno);
4228                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
4229         }
4230
4231         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
4232                 /*
4233                  * The only way this can fail is if we run out of memory, or we
4234                  * were unable to read the disk's capacity
4235                  */
4236                 if (errno == ENOMEM)
4237                         (void) no_memory(hdl);
4238
4239                 (void) close(fd);
4240                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
4241                     "label '%s': unable to read disk capacity"), path);
4242
4243                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
4244         }
4245
4246         slice_size = vtoc->efi_last_u_lba + 1;
4247         slice_size -= EFI_MIN_RESV_SIZE;
4248         if (start_block == MAXOFFSET_T)
4249                 start_block = NEW_START_BLOCK;
4250         slice_size -= start_block;
4251         slice_size = P2ALIGN(slice_size, PARTITION_END_ALIGNMENT);
4252
4253         vtoc->efi_parts[0].p_start = start_block;
4254         vtoc->efi_parts[0].p_size = slice_size;
4255
4256         /*
4257          * Why we use V_USR: V_BACKUP confuses users, and is considered
4258          * disposable by some EFI utilities (since EFI doesn't have a backup
4259          * slice).  V_UNASSIGNED is supposed to be used only for zero size
4260          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
4261          * etc. were all pretty specific.  V_USR is as close to reality as we
4262          * can get, in the absence of V_OTHER.
4263          */
4264         vtoc->efi_parts[0].p_tag = V_USR;
4265         zpool_label_name(vtoc->efi_parts[0].p_name, EFI_PART_NAME_LEN);
4266
4267         vtoc->efi_parts[8].p_start = slice_size + start_block;
4268         vtoc->efi_parts[8].p_size = resv;
4269         vtoc->efi_parts[8].p_tag = V_RESERVED;
4270
4271         if ((rval = efi_write(fd, vtoc)) != 0 || (rval = efi_rescan(fd)) != 0) {
4272                 /*
4273                  * Some block drivers (like pcata) may not support EFI
4274                  * GPT labels.  Print out a helpful error message dir-
4275                  * ecting the user to manually label the disk and give
4276                  * a specific slice.
4277                  */
4278                 (void) close(fd);
4279                 efi_free(vtoc);
4280
4281                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "try using "
4282                     "parted(8) and then provide a specific slice: %d"), rval);
4283                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4284         }
4285
4286         (void) close(fd);
4287         efi_free(vtoc);
4288
4289         (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT, name);
4290         (void) zfs_append_partition(path, MAXPATHLEN);
4291
4292         /* Wait to udev to signal use the device has settled. */
4293         rval = zpool_label_disk_wait(path, DISK_LABEL_WAIT);
4294         if (rval) {
4295                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "failed to "
4296                     "detect device partitions on '%s': %d"), path, rval);
4297                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4298         }
4299
4300         /* We can't be to paranoid.  Read the label back and verify it. */
4301         (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT, name);
4302         rval = zpool_label_disk_check(path);
4303         if (rval) {
4304                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "freshly written "
4305                     "EFI label on '%s' is damaged.  Ensure\nthis device "
4306                     "is not in in use, and is functioning properly: %d"),
4307                     path, rval);
4308                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4309         }
4310
4311         return (0);
4312 }