granicus.if.org Git - zfs/blob - 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[ZPOOL_MAXNAMELEN];
 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
1911                 default:
1912                         (void) zpool_standard_error(hdl, error, desc);
1913                         zpool_explain_recover(hdl,
1914                             newname ? origname : thename, -error, nv);
1915                         break;
1916                 }
1917
1918                 nvlist_free(nv);
1919                 ret = -1;
1920         } else {
1921                 zpool_handle_t *zhp;
1922
1923                 /*
1924                  * This should never fail, but play it safe anyway.
1925                  */
1926                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1927                         ret = -1;
1928                 else if (zhp != NULL)
1929                         zpool_close(zhp);
1930                 if (policy.zrp_request &
1931                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1932                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1933                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1934                 }
1935                 nvlist_free(nv);
1936                 return (0);
1937         }
1938
1939         return (ret);
1940 }
1941
1942 /*
1943  * Scan the pool.
1944  */
1945 int
1946 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1947 {
1948         zfs_cmd_t zc = {"\0"};
1949         char msg[1024];
1950         libzfs_handle_t *hdl = zhp->zpool_hdl;
1951
1952         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1953         zc.zc_cookie = func;
1954
1955         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1956             (errno == ENOENT && func != POOL_SCAN_NONE))
1957                 return (0);
1958
1959         if (func == POOL_SCAN_SCRUB) {
1960                 (void) snprintf(msg, sizeof (msg),
1961                     dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1962         } else if (func == POOL_SCAN_NONE) {
1963                 (void) snprintf(msg, sizeof (msg),
1964                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1965                     zc.zc_name);
1966         } else {
1967                 assert(!"unexpected result");
1968         }
1969
1970         if (errno == EBUSY) {
1971                 nvlist_t *nvroot;
1972                 pool_scan_stat_t *ps = NULL;
1973                 uint_t psc;
1974
1975                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
1976                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1977                 (void) nvlist_lookup_uint64_array(nvroot,
1978                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1979                 if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1980                         return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1981                 else
1982                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
1983         } else if (errno == ENOENT) {
1984                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1985         } else {
1986                 return (zpool_standard_error(hdl, errno, msg));
1987         }
1988 }
1989
1990 /*
1991  * Find a vdev that matches the search criteria specified. We use the
1992  * the nvpair name to determine how we should look for the device.
1993  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1994  * spare; but FALSE if its an INUSE spare.
1995  */
1996 static nvlist_t *
1997 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1998     boolean_t *l2cache, boolean_t *log)
1999 {
2000         uint_t c, children;
2001         nvlist_t **child;
2002         nvlist_t *ret;
2003         uint64_t is_log;
2004         char *srchkey;
2005         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
2006
2007         /* Nothing to look for */
2008         if (search == NULL || pair == NULL)
2009                 return (NULL);
2010
2011         /* Obtain the key we will use to search */
2012         srchkey = nvpair_name(pair);
2013
2014         switch (nvpair_type(pair)) {
2015         case DATA_TYPE_UINT64:
2016                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
2017                         uint64_t srchval, theguid;
2018
2019                         verify(nvpair_value_uint64(pair, &srchval) == 0);
2020                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
2021                             &theguid) == 0);
2022                         if (theguid == srchval)
2023                                 return (nv);
2024                 }
2025                 break;
2026
2027         case DATA_TYPE_STRING: {
2028                 char *srchval, *val;
2029
2030                 verify(nvpair_value_string(pair, &srchval) == 0);
2031                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
2032                         break;
2033
2034                 /*
2035                  * Search for the requested value. Special cases:
2036                  *
2037                  * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
2038                  *   "-part1", or "p1".  The suffix is hidden from the user,
2039                  *   but included in the string, so this matches around it.
2040                  * - ZPOOL_CONFIG_PATH for short names zfs_strcmp_shortname()
2041                  *   is used to check all possible expanded paths.
2042                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
2043                  *
2044                  * Otherwise, all other searches are simple string compares.
2045                  */
2046                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0) {
2047                         uint64_t wholedisk = 0;
2048
2049                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
2050                             &wholedisk);
2051                         if (zfs_strcmp_pathname(srchval, val, wholedisk) == 0)
2052                                 return (nv);
2053
2054                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2055                         char *type, *idx, *end, *p;
2056                         uint64_t id, vdev_id;
2057
2058                         /*
2059                          * Determine our vdev type, keeping in mind
2060                          * that the srchval is composed of a type and
2061                          * vdev id pair (i.e. mirror-4).
2062                          */
2063                         if ((type = strdup(srchval)) == NULL)
2064                                 return (NULL);
2065
2066                         if ((p = strrchr(type, '-')) == NULL) {
2067                                 free(type);
2068                                 break;
2069                         }
2070                         idx = p + 1;
2071                         *p = '\0';
2072
2073                         /*
2074                          * If the types don't match then keep looking.
2075                          */
2076                         if (strncmp(val, type, strlen(val)) != 0) {
2077                                 free(type);
2078                                 break;
2079                         }
2080
2081                         verify(strncmp(type, VDEV_TYPE_RAIDZ,
2082                             strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2083                             strncmp(type, VDEV_TYPE_MIRROR,
2084                             strlen(VDEV_TYPE_MIRROR)) == 0);
2085                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
2086                             &id) == 0);
2087
2088                         errno = 0;
2089                         vdev_id = strtoull(idx, &end, 10);
2090
2091                         free(type);
2092                         if (errno != 0)
2093                                 return (NULL);
2094
2095                         /*
2096                          * Now verify that we have the correct vdev id.
2097                          */
2098                         if (vdev_id == id)
2099                                 return (nv);
2100                 }
2101
2102                 /*
2103                  * Common case
2104                  */
2105                 if (strcmp(srchval, val) == 0)
2106                         return (nv);
2107                 break;
2108         }
2109
2110         default:
2111                 break;
2112         }
2113
2114         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2115             &child, &children) != 0)
2116                 return (NULL);
2117
2118         for (c = 0; c < children; c++) {
2119                 if ((ret = vdev_to_nvlist_iter(child[c], search,
2120                     avail_spare, l2cache, NULL)) != NULL) {
2121                         /*
2122                          * The 'is_log' value is only set for the toplevel
2123                          * vdev, not the leaf vdevs.  So we always lookup the
2124                          * log device from the root of the vdev tree (where
2125                          * 'log' is non-NULL).
2126                          */
2127                         if (log != NULL &&
2128                             nvlist_lookup_uint64(child[c],
2129                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2130                             is_log) {
2131                                 *log = B_TRUE;
2132                         }
2133                         return (ret);
2134                 }
2135         }
2136
2137         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2138             &child, &children) == 0) {
2139                 for (c = 0; c < children; c++) {
2140                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2141                             avail_spare, l2cache, NULL)) != NULL) {
2142                                 *avail_spare = B_TRUE;
2143                                 return (ret);
2144                         }
2145                 }
2146         }
2147
2148         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2149             &child, &children) == 0) {
2150                 for (c = 0; c < children; c++) {
2151                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2152                             avail_spare, l2cache, NULL)) != NULL) {
2153                                 *l2cache = B_TRUE;
2154                                 return (ret);
2155                         }
2156                 }
2157         }
2158
2159         return (NULL);
2160 }
2161
2162 /*
2163  * Given a physical path (minus the "/devices" prefix), find the
2164  * associated vdev.
2165  */
2166 nvlist_t *
2167 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2168     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2169 {
2170         nvlist_t *search, *nvroot, *ret;
2171
2172         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2173         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
2174
2175         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2176             &nvroot) == 0);
2177
2178         *avail_spare = B_FALSE;
2179         *l2cache = B_FALSE;
2180         if (log != NULL)
2181                 *log = B_FALSE;
2182         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2183         nvlist_free(search);
2184
2185         return (ret);
2186 }
2187
2188 /*
2189  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2190  */
2191 boolean_t
2192 zpool_vdev_is_interior(const char *name)
2193 {
2194         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2195             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2196                 return (B_TRUE);
2197         return (B_FALSE);
2198 }
2199
2200 nvlist_t *
2201 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2202     boolean_t *l2cache, boolean_t *log)
2203 {
2204         char *end;
2205         nvlist_t *nvroot, *search, *ret;
2206         uint64_t guid;
2207
2208         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2209
2210         guid = strtoull(path, &end, 0);
2211         if (guid != 0 && *end == '\0') {
2212                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
2213         } else if (zpool_vdev_is_interior(path)) {
2214                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
2215         } else {
2216                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
2217         }
2218
2219         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2220             &nvroot) == 0);
2221
2222         *avail_spare = B_FALSE;
2223         *l2cache = B_FALSE;
2224         if (log != NULL)
2225                 *log = B_FALSE;
2226         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2227         nvlist_free(search);
2228
2229         return (ret);
2230 }
2231
2232 static int
2233 vdev_online(nvlist_t *nv)
2234 {
2235         uint64_t ival;
2236
2237         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2238             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2239             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2240                 return (0);
2241
2242         return (1);
2243 }
2244
2245 /*
2246  * Helper function for zpool_get_physpaths().
2247  */
2248 static int
2249 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2250     size_t *bytes_written)
2251 {
2252         size_t bytes_left, pos, rsz;
2253         char *tmppath;
2254         const char *format;
2255
2256         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2257             &tmppath) != 0)
2258                 return (EZFS_NODEVICE);
2259
2260         pos = *bytes_written;
2261         bytes_left = physpath_size - pos;
2262         format = (pos == 0) ? "%s" : " %s";
2263
2264         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2265         *bytes_written += rsz;
2266
2267         if (rsz >= bytes_left) {
2268                 /* if physpath was not copied properly, clear it */
2269                 if (bytes_left != 0) {
2270                         physpath[pos] = 0;
2271                 }
2272                 return (EZFS_NOSPC);
2273         }
2274         return (0);
2275 }
2276
2277 static int
2278 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2279     size_t *rsz, boolean_t is_spare)
2280 {
2281         char *type;
2282         int ret;
2283
2284         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2285                 return (EZFS_INVALCONFIG);
2286
2287         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2288                 /*
2289                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2290                  * For a spare vdev, we only want to boot from the active
2291                  * spare device.
2292                  */
2293                 if (is_spare) {
2294                         uint64_t spare = 0;
2295                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2296                             &spare);
2297                         if (!spare)
2298                                 return (EZFS_INVALCONFIG);
2299                 }
2300
2301                 if (vdev_online(nv)) {
2302                         if ((ret = vdev_get_one_physpath(nv, physpath,
2303                             phypath_size, rsz)) != 0)
2304                                 return (ret);
2305                 }
2306         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2307             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2308             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2309                 nvlist_t **child;
2310                 uint_t count;
2311                 int i, ret;
2312
2313                 if (nvlist_lookup_nvlist_array(nv,
2314                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2315                         return (EZFS_INVALCONFIG);
2316
2317                 for (i = 0; i < count; i++) {
2318                         ret = vdev_get_physpaths(child[i], physpath,
2319                             phypath_size, rsz, is_spare);
2320                         if (ret == EZFS_NOSPC)
2321                                 return (ret);
2322                 }
2323         }
2324
2325         return (EZFS_POOL_INVALARG);
2326 }
2327
2328 /*
2329  * Get phys_path for a root pool config.
2330  * Return 0 on success; non-zero on failure.
2331  */
2332 static int
2333 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2334 {
2335         size_t rsz;
2336         nvlist_t *vdev_root;
2337         nvlist_t **child;
2338         uint_t count;
2339         char *type;
2340
2341         rsz = 0;
2342
2343         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2344             &vdev_root) != 0)
2345                 return (EZFS_INVALCONFIG);
2346
2347         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2348             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2349             &child, &count) != 0)
2350                 return (EZFS_INVALCONFIG);
2351
2352 #if defined(__sun__) || defined(__sun)
2353         /*
2354          * root pool can not have EFI labeled disks and can only have
2355          * a single top-level vdev.
2356          */
2357         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1 ||
2358             pool_uses_efi(vdev_root))
2359                 return (EZFS_POOL_INVALARG);
2360 #endif
2361
2362         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2363             B_FALSE);
2364
2365         /* No online devices */
2366         if (rsz == 0)
2367                 return (EZFS_NODEVICE);
2368
2369         return (0);
2370 }
2371
2372 /*
2373  * Get phys_path for a root pool
2374  * Return 0 on success; non-zero on failure.
2375  */
2376 int
2377 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2378 {
2379         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2380             phypath_size));
2381 }
2382
2383 /*
2384  * If the device has being dynamically expanded then we need to relabel
2385  * the disk to use the new unallocated space.
2386  */
2387 static int
2388 zpool_relabel_disk(libzfs_handle_t *hdl, const char *path, const char *msg)
2389 {
2390         int fd, error;
2391
2392         if ((fd = open(path, O_RDWR|O_DIRECT)) < 0) {
2393                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2394                     "relabel '%s': unable to open device: %d"), path, errno);
2395                 return (zfs_error(hdl, EZFS_OPENFAILED, msg));
2396         }
2397
2398         /*
2399          * It's possible that we might encounter an error if the device
2400          * does not have any unallocated space left. If so, we simply
2401          * ignore that error and continue on.
2402          *
2403          * Also, we don't call efi_rescan() - that would just return EBUSY.
2404          * The module will do it for us in vdev_disk_open().
2405          */
2406         error = efi_use_whole_disk(fd);
2407         (void) close(fd);
2408         if (error && error != VT_ENOSPC) {
2409                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2410                     "relabel '%s': unable to read disk capacity"), path);
2411                 return (zfs_error(hdl, EZFS_NOCAP, msg));
2412         }
2413         return (0);
2414 }
2415
2416 /*
2417  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2418  * ZFS_ONLINE_* flags.
2419  */
2420 int
2421 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2422     vdev_state_t *newstate)
2423 {
2424         zfs_cmd_t zc = {"\0"};
2425         char msg[1024];
2426         nvlist_t *tgt;
2427         boolean_t avail_spare, l2cache, islog;
2428         libzfs_handle_t *hdl = zhp->zpool_hdl;
2429         int error;
2430
2431         if (flags & ZFS_ONLINE_EXPAND) {
2432                 (void) snprintf(msg, sizeof (msg),
2433                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2434         } else {
2435                 (void) snprintf(msg, sizeof (msg),
2436                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2437         }
2438
2439         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2440         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2441             &islog)) == NULL)
2442                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2443
2444         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2445
2446         if (avail_spare)
2447                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2448
2449         if (flags & ZFS_ONLINE_EXPAND ||
2450             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2451                 uint64_t wholedisk = 0;
2452
2453                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2454                     &wholedisk);
2455
2456                 /*
2457                  * XXX - L2ARC 1.0 devices can't support expansion.
2458                  */
2459                 if (l2cache) {
2460                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2461                             "cannot expand cache devices"));
2462                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2463                 }
2464
2465                 if (wholedisk) {
2466                         const char *fullpath = path;
2467                         char buf[MAXPATHLEN];
2468
2469                         if (path[0] != '/') {
2470                                 error = zfs_resolve_shortname(path, buf,
2471                                     sizeof (buf));
2472                                 if (error != 0)
2473                                         return (zfs_error(hdl, EZFS_NODEVICE,
2474                                             msg));
2475
2476                                 fullpath = buf;
2477                         }
2478
2479                         error = zpool_relabel_disk(hdl, fullpath, msg);
2480                         if (error != 0)
2481                                 return (error);
2482                 }
2483         }
2484
2485         zc.zc_cookie = VDEV_STATE_ONLINE;
2486         zc.zc_obj = flags;
2487
2488         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2489                 if (errno == EINVAL) {
2490                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2491                             "from this pool into a new one.  Use '%s' "
2492                             "instead"), "zpool detach");
2493                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2494                 }
2495                 return (zpool_standard_error(hdl, errno, msg));
2496         }
2497
2498         *newstate = zc.zc_cookie;
2499         return (0);
2500 }
2501
2502 /*
2503  * Take the specified vdev offline
2504  */
2505 int
2506 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2507 {
2508         zfs_cmd_t zc = {"\0"};
2509         char msg[1024];
2510         nvlist_t *tgt;
2511         boolean_t avail_spare, l2cache;
2512         libzfs_handle_t *hdl = zhp->zpool_hdl;
2513
2514         (void) snprintf(msg, sizeof (msg),
2515             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2516
2517         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2518         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2519             NULL)) == NULL)
2520                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2521
2522         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2523
2524         if (avail_spare)
2525                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2526
2527         zc.zc_cookie = VDEV_STATE_OFFLINE;
2528         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2529
2530         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2531                 return (0);
2532
2533         switch (errno) {
2534         case EBUSY:
2535
2536                 /*
2537                  * There are no other replicas of this device.
2538                  */
2539                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2540
2541         case EEXIST:
2542                 /*
2543                  * The log device has unplayed logs
2544                  */
2545                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2546
2547         default:
2548                 return (zpool_standard_error(hdl, errno, msg));
2549         }
2550 }
2551
2552 /*
2553  * Mark the given vdev faulted.
2554  */
2555 int
2556 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2557 {
2558         zfs_cmd_t zc = {"\0"};
2559         char msg[1024];
2560         libzfs_handle_t *hdl = zhp->zpool_hdl;
2561
2562         (void) snprintf(msg, sizeof (msg),
2563             dgettext(TEXT_DOMAIN, "cannot fault %llu"), (u_longlong_t)guid);
2564
2565         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2566         zc.zc_guid = guid;
2567         zc.zc_cookie = VDEV_STATE_FAULTED;
2568         zc.zc_obj = aux;
2569
2570         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2571                 return (0);
2572
2573         switch (errno) {
2574         case EBUSY:
2575
2576                 /*
2577                  * There are no other replicas of this device.
2578                  */
2579                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2580
2581         default:
2582                 return (zpool_standard_error(hdl, errno, msg));
2583         }
2584
2585 }
2586
2587 /*
2588  * Mark the given vdev degraded.
2589  */
2590 int
2591 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2592 {
2593         zfs_cmd_t zc = {"\0"};
2594         char msg[1024];
2595         libzfs_handle_t *hdl = zhp->zpool_hdl;
2596
2597         (void) snprintf(msg, sizeof (msg),
2598             dgettext(TEXT_DOMAIN, "cannot degrade %llu"), (u_longlong_t)guid);
2599
2600         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2601         zc.zc_guid = guid;
2602         zc.zc_cookie = VDEV_STATE_DEGRADED;
2603         zc.zc_obj = aux;
2604
2605         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2606                 return (0);
2607
2608         return (zpool_standard_error(hdl, errno, msg));
2609 }
2610
2611 /*
2612  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2613  * a hot spare.
2614  */
2615 static boolean_t
2616 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2617 {
2618         nvlist_t **child;
2619         uint_t c, children;
2620         char *type;
2621
2622         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2623             &children) == 0) {
2624                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2625                     &type) == 0);
2626
2627                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2628                     children == 2 && child[which] == tgt)
2629                         return (B_TRUE);
2630
2631                 for (c = 0; c < children; c++)
2632                         if (is_replacing_spare(child[c], tgt, which))
2633                                 return (B_TRUE);
2634         }
2635
2636         return (B_FALSE);
2637 }
2638
2639 /*
2640  * Attach new_disk (fully described by nvroot) to old_disk.
2641  * If 'replacing' is specified, the new disk will replace the old one.
2642  */
2643 int
2644 zpool_vdev_attach(zpool_handle_t *zhp,
2645     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2646 {
2647         zfs_cmd_t zc = {"\0"};
2648         char msg[1024];
2649         int ret;
2650         nvlist_t *tgt;
2651         boolean_t avail_spare, l2cache, islog;
2652         uint64_t val;
2653         char *newname;
2654         nvlist_t **child;
2655         uint_t children;
2656         nvlist_t *config_root;
2657         libzfs_handle_t *hdl = zhp->zpool_hdl;
2658         boolean_t rootpool = zpool_is_bootable(zhp);
2659
2660         if (replacing)
2661                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2662                     "cannot replace %s with %s"), old_disk, new_disk);
2663         else
2664                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2665                     "cannot attach %s to %s"), new_disk, old_disk);
2666
2667 #if defined(__sun__) || defined(__sun)
2668         /*
2669          * If this is a root pool, make sure that we're not attaching an
2670          * EFI labeled device.
2671          */
2672         if (rootpool && pool_uses_efi(nvroot)) {
2673                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2674                     "EFI labeled devices are not supported on root pools."));
2675                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
2676         }
2677 #endif
2678
2679         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2680         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2681             &islog)) == 0)
2682                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2683
2684         if (avail_spare)
2685                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2686
2687         if (l2cache)
2688                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2689
2690         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2691         zc.zc_cookie = replacing;
2692
2693         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2694             &child, &children) != 0 || children != 1) {
2695                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2696                     "new device must be a single disk"));
2697                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2698         }
2699
2700         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2701             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2702
2703         if ((newname = zpool_vdev_name(NULL, NULL, child[0], 0)) == NULL)
2704                 return (-1);
2705
2706         /*
2707          * If the target is a hot spare that has been swapped in, we can only
2708          * replace it with another hot spare.
2709          */
2710         if (replacing &&
2711             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2712             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2713             NULL) == NULL || !avail_spare) &&
2714             is_replacing_spare(config_root, tgt, 1)) {
2715                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2716                     "can only be replaced by another hot spare"));
2717                 free(newname);
2718                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2719         }
2720
2721         free(newname);
2722
2723         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2724                 return (-1);
2725
2726         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2727
2728         zcmd_free_nvlists(&zc);
2729
2730         if (ret == 0) {
2731                 if (rootpool) {
2732                         /*
2733                          * XXX need a better way to prevent user from
2734                          * booting up a half-baked vdev.
2735                          */
2736                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2737                             "sure to wait until resilver is done "
2738                             "before rebooting.\n"));
2739                 }
2740                 return (0);
2741         }
2742
2743         switch (errno) {
2744         case ENOTSUP:
2745                 /*
2746                  * Can't attach to or replace this type of vdev.
2747                  */
2748                 if (replacing) {
2749                         uint64_t version = zpool_get_prop_int(zhp,
2750                             ZPOOL_PROP_VERSION, NULL);
2751
2752                         if (islog)
2753                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2754                                     "cannot replace a log with a spare"));
2755                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2756                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2757                                     "already in replacing/spare config; wait "
2758                                     "for completion or use 'zpool detach'"));
2759                         else
2760                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2761                                     "cannot replace a replacing device"));
2762                 } else {
2763                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2764                             "can only attach to mirrors and top-level "
2765                             "disks"));
2766                 }
2767                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2768                 break;
2769
2770         case EINVAL:
2771                 /*
2772                  * The new device must be a single disk.
2773                  */
2774                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2775                     "new device must be a single disk"));
2776                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2777                 break;
2778
2779         case EBUSY:
2780                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2781                     new_disk);
2782                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2783                 break;
2784
2785         case EOVERFLOW:
2786                 /*
2787                  * The new device is too small.
2788                  */
2789                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2790                     "device is too small"));
2791                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2792                 break;
2793
2794         case EDOM:
2795                 /*
2796                  * The new device has a different optimal sector size.
2797                  */
2798                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2799                     "new device has a different optimal sector size; use the "
2800                     "option '-o ashift=N' to override the optimal size"));
2801                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2802                 break;
2803
2804         case ENAMETOOLONG:
2805                 /*
2806                  * The resulting top-level vdev spec won't fit in the label.
2807                  */
2808                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2809                 break;
2810
2811         default:
2812                 (void) zpool_standard_error(hdl, errno, msg);
2813         }
2814
2815         return (-1);
2816 }
2817
2818 /*
2819  * Detach the specified device.
2820  */
2821 int
2822 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2823 {
2824         zfs_cmd_t zc = {"\0"};
2825         char msg[1024];
2826         nvlist_t *tgt;
2827         boolean_t avail_spare, l2cache;
2828         libzfs_handle_t *hdl = zhp->zpool_hdl;
2829
2830         (void) snprintf(msg, sizeof (msg),
2831             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2832
2833         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2834         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2835             NULL)) == 0)
2836                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2837
2838         if (avail_spare)
2839                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2840
2841         if (l2cache)
2842                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2843
2844         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2845
2846         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2847                 return (0);
2848
2849         switch (errno) {
2850
2851         case ENOTSUP:
2852                 /*
2853                  * Can't detach from this type of vdev.
2854                  */
2855                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2856                     "applicable to mirror and replacing vdevs"));
2857                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2858                 break;
2859
2860         case EBUSY:
2861                 /*
2862                  * There are no other replicas of this device.
2863                  */
2864                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2865                 break;
2866
2867         default:
2868                 (void) zpool_standard_error(hdl, errno, msg);
2869         }
2870
2871         return (-1);
2872 }
2873
2874 /*
2875  * Find a mirror vdev in the source nvlist.
2876  *
2877  * The mchild array contains a list of disks in one of the top-level mirrors
2878  * of the source pool.  The schild array contains a list of disks that the
2879  * user specified on the command line.  We loop over the mchild array to
2880  * see if any entry in the schild array matches.
2881  *
2882  * If a disk in the mchild array is found in the schild array, we return
2883  * the index of that entry.  Otherwise we return -1.
2884  */
2885 static int
2886 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2887     nvlist_t **schild, uint_t schildren)
2888 {
2889         uint_t mc;
2890
2891         for (mc = 0; mc < mchildren; mc++) {
2892                 uint_t sc;
2893                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2894                     mchild[mc], 0);
2895
2896                 for (sc = 0; sc < schildren; sc++) {
2897                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2898                             schild[sc], 0);
2899                         boolean_t result = (strcmp(mpath, spath) == 0);
2900
2901                         free(spath);
2902                         if (result) {
2903                                 free(mpath);
2904                                 return (mc);
2905                         }
2906                 }
2907
2908                 free(mpath);
2909         }
2910
2911         return (-1);
2912 }
2913
2914 /*
2915  * Split a mirror pool.  If newroot points to null, then a new nvlist
2916  * is generated and it is the responsibility of the caller to free it.
2917  */
2918 int
2919 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2920     nvlist_t *props, splitflags_t flags)
2921 {
2922         zfs_cmd_t zc = {"\0"};
2923         char msg[1024];
2924         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2925         nvlist_t **varray = NULL, *zc_props = NULL;
2926         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2927         libzfs_handle_t *hdl = zhp->zpool_hdl;
2928         uint64_t vers;
2929         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2930         int retval = 0;
2931
2932         (void) snprintf(msg, sizeof (msg),
2933             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2934
2935         if (!zpool_name_valid(hdl, B_FALSE, newname))
2936                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2937
2938         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2939                 (void) fprintf(stderr, gettext("Internal error: unable to "
2940                     "retrieve pool configuration\n"));
2941                 return (-1);
2942         }
2943
2944         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2945             == 0);
2946         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2947
2948         if (props) {
2949                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2950                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2951                     props, vers, flags, msg)) == NULL)
2952                         return (-1);
2953         }
2954
2955         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2956             &children) != 0) {
2957                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2958                     "Source pool is missing vdev tree"));
2959                 nvlist_free(zc_props);
2960                 return (-1);
2961         }
2962
2963         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2964         vcount = 0;
2965
2966         if (*newroot == NULL ||
2967             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2968             &newchild, &newchildren) != 0)
2969                 newchildren = 0;
2970
2971         for (c = 0; c < children; c++) {
2972                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2973                 char *type;
2974                 nvlist_t **mchild, *vdev;
2975                 uint_t mchildren;
2976                 int entry;
2977
2978                 /*
2979                  * Unlike cache & spares, slogs are stored in the
2980                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2981                  */
2982                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2983                     &is_log);
2984                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2985                     &is_hole);
2986                 if (is_log || is_hole) {
2987                         /*
2988                          * Create a hole vdev and put it in the config.
2989                          */
2990                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2991                                 goto out;
2992                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2993                             VDEV_TYPE_HOLE) != 0)
2994                                 goto out;
2995                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2996                             1) != 0)
2997                                 goto out;
2998                         if (lastlog == 0)
2999                                 lastlog = vcount;
3000                         varray[vcount++] = vdev;
3001                         continue;
3002                 }
3003                 lastlog = 0;
3004                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
3005                     == 0);
3006                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
3007                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3008                             "Source pool must be composed only of mirrors\n"));
3009                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3010                         goto out;
3011                 }
3012
3013                 verify(nvlist_lookup_nvlist_array(child[c],
3014                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
3015
3016                 /* find or add an entry for this top-level vdev */
3017                 if (newchildren > 0 &&
3018                     (entry = find_vdev_entry(zhp, mchild, mchildren,
3019                     newchild, newchildren)) >= 0) {
3020                         /* We found a disk that the user specified. */
3021                         vdev = mchild[entry];
3022                         ++found;
3023                 } else {
3024                         /* User didn't specify a disk for this vdev. */
3025                         vdev = mchild[mchildren - 1];
3026                 }
3027
3028                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
3029                         goto out;
3030         }
3031
3032         /* did we find every disk the user specified? */
3033         if (found != newchildren) {
3034                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
3035                     "include at most one disk from each mirror"));
3036                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3037                 goto out;
3038         }
3039
3040         /* Prepare the nvlist for populating. */
3041         if (*newroot == NULL) {
3042                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
3043                         goto out;
3044                 freelist = B_TRUE;
3045                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
3046                     VDEV_TYPE_ROOT) != 0)
3047                         goto out;
3048         } else {
3049                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
3050         }
3051
3052         /* Add all the children we found */
3053         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
3054             lastlog == 0 ? vcount : lastlog) != 0)
3055                 goto out;
3056
3057         /*
3058          * If we're just doing a dry run, exit now with success.
3059          */
3060         if (flags.dryrun) {
3061                 memory_err = B_FALSE;
3062                 freelist = B_FALSE;
3063                 goto out;
3064         }
3065
3066         /* now build up the config list & call the ioctl */
3067         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3068                 goto out;
3069
3070         if (nvlist_add_nvlist(newconfig,
3071             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3072             nvlist_add_string(newconfig,
3073             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3074             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3075                 goto out;
3076
3077         /*
3078          * The new pool is automatically part of the namespace unless we
3079          * explicitly export it.
3080          */
3081         if (!flags.import)
3082                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3083         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3084         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3085         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
3086                 goto out;
3087         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
3088                 goto out;
3089
3090         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3091                 retval = zpool_standard_error(hdl, errno, msg);
3092                 goto out;
3093         }
3094
3095         freelist = B_FALSE;
3096         memory_err = B_FALSE;
3097
3098 out:
3099         if (varray != NULL) {
3100                 int v;
3101
3102                 for (v = 0; v < vcount; v++)
3103                         nvlist_free(varray[v]);
3104                 free(varray);
3105         }
3106         zcmd_free_nvlists(&zc);
3107         nvlist_free(zc_props);
3108         nvlist_free(newconfig);
3109         if (freelist) {
3110                 nvlist_free(*newroot);
3111                 *newroot = NULL;
3112         }
3113
3114         if (retval != 0)
3115                 return (retval);
3116
3117         if (memory_err)
3118                 return (no_memory(hdl));
3119
3120         return (0);
3121 }
3122
3123 /*
3124  * Remove the given device.  Currently, this is supported only for hot spares
3125  * and level 2 cache devices.
3126  */
3127 int
3128 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3129 {
3130         zfs_cmd_t zc = {"\0"};
3131         char msg[1024];
3132         nvlist_t *tgt;
3133         boolean_t avail_spare, l2cache, islog;
3134         libzfs_handle_t *hdl = zhp->zpool_hdl;
3135         uint64_t version;
3136
3137         (void) snprintf(msg, sizeof (msg),
3138             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3139
3140         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3141         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3142             &islog)) == 0)
3143                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3144         /*
3145          * XXX - this should just go away.
3146          */
3147         if (!avail_spare && !l2cache && !islog) {
3148                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3149                     "only inactive hot spares, cache, top-level, "
3150                     "or log devices can be removed"));
3151                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3152         }
3153
3154         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3155         if (islog && version < SPA_VERSION_HOLES) {
3156                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3157                     "pool must be upgrade to support log removal"));
3158                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
3159         }
3160
3161         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
3162
3163         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3164                 return (0);
3165
3166         return (zpool_standard_error(hdl, errno, msg));
3167 }
3168
3169 /*
3170  * Clear the errors for the pool, or the particular device if specified.
3171  */
3172 int
3173 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3174 {
3175         zfs_cmd_t zc = {"\0"};
3176         char msg[1024];
3177         nvlist_t *tgt;
3178         zpool_rewind_policy_t policy;
3179         boolean_t avail_spare, l2cache;
3180         libzfs_handle_t *hdl = zhp->zpool_hdl;
3181         nvlist_t *nvi = NULL;
3182         int error;
3183
3184         if (path)
3185                 (void) snprintf(msg, sizeof (msg),
3186                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3187                     path);
3188         else
3189                 (void) snprintf(msg, sizeof (msg),
3190                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3191                     zhp->zpool_name);
3192
3193         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3194         if (path) {
3195                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3196                     &l2cache, NULL)) == 0)
3197                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
3198
3199                 /*
3200                  * Don't allow error clearing for hot spares.  Do allow
3201                  * error clearing for l2cache devices.
3202                  */
3203                 if (avail_spare)
3204                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
3205
3206                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
3207                     &zc.zc_guid) == 0);
3208         }
3209
3210         zpool_get_rewind_policy(rewindnvl, &policy);
3211         zc.zc_cookie = policy.zrp_request;
3212
3213         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
3214                 return (-1);
3215
3216         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
3217                 return (-1);
3218
3219         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3220             errno == ENOMEM) {
3221                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3222                         zcmd_free_nvlists(&zc);
3223                         return (-1);
3224                 }
3225         }
3226
3227         if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
3228             errno != EPERM && errno != EACCES)) {
3229                 if (policy.zrp_request &
3230                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
3231                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3232                         zpool_rewind_exclaim(hdl, zc.zc_name,
3233                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
3234                             nvi);
3235                         nvlist_free(nvi);
3236                 }
3237                 zcmd_free_nvlists(&zc);
3238                 return (0);
3239         }
3240
3241         zcmd_free_nvlists(&zc);
3242         return (zpool_standard_error(hdl, errno, msg));
3243 }
3244
3245 /*
3246  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3247  */
3248 int
3249 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3250 {
3251         zfs_cmd_t zc = {"\0"};
3252         char msg[1024];
3253         libzfs_handle_t *hdl = zhp->zpool_hdl;
3254
3255         (void) snprintf(msg, sizeof (msg),
3256             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3257             (u_longlong_t)guid);
3258
3259         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3260         zc.zc_guid = guid;
3261         zc.zc_cookie = ZPOOL_NO_REWIND;
3262
3263         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
3264                 return (0);
3265
3266         return (zpool_standard_error(hdl, errno, msg));
3267 }
3268
3269 /*
3270  * Change the GUID for a pool.
3271  */
3272 int
3273 zpool_reguid(zpool_handle_t *zhp)
3274 {
3275         char msg[1024];
3276         libzfs_handle_t *hdl = zhp->zpool_hdl;
3277         zfs_cmd_t zc = {"\0"};
3278
3279         (void) snprintf(msg, sizeof (msg),
3280             dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3281
3282         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3283         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3284                 return (0);
3285
3286         return (zpool_standard_error(hdl, errno, msg));
3287 }
3288
3289 /*
3290  * Reopen the pool.
3291  */
3292 int
3293 zpool_reopen(zpool_handle_t *zhp)
3294 {
3295         zfs_cmd_t zc = {"\0"};
3296         char msg[1024];
3297         libzfs_handle_t *hdl = zhp->zpool_hdl;
3298
3299         (void) snprintf(msg, sizeof (msg),
3300             dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3301             zhp->zpool_name);
3302
3303         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3304         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3305                 return (0);
3306         return (zpool_standard_error(hdl, errno, msg));
3307 }
3308
3309 #if defined(__sun__) || defined(__sun)
3310 /*
3311  * Convert from a devid string to a path.
3312  */
3313 static char *
3314 devid_to_path(char *devid_str)
3315 {
3316         ddi_devid_t devid;
3317         char *minor;
3318         char *path;
3319         devid_nmlist_t *list = NULL;
3320         int ret;
3321
3322         if (devid_str_decode(devid_str, &devid, &minor) != 0)
3323                 return (NULL);
3324
3325         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3326
3327         devid_str_free(minor);
3328         devid_free(devid);
3329
3330         if (ret != 0)
3331                 return (NULL);
3332
3333         /*
3334          * In a case the strdup() fails, we will just return NULL below.
3335          */
3336         path = strdup(list[0].devname);
3337
3338         devid_free_nmlist(list);
3339
3340         return (path);
3341 }
3342
3343 /*
3344  * Convert from a path to a devid string.
3345  */
3346 static char *
3347 path_to_devid(const char *path)
3348 {
3349         int fd;
3350         ddi_devid_t devid;
3351         char *minor, *ret;
3352
3353         if ((fd = open(path, O_RDONLY)) < 0)
3354                 return (NULL);
3355
3356         minor = NULL;
3357         ret = NULL;
3358         if (devid_get(fd, &devid) == 0) {
3359                 if (devid_get_minor_name(fd, &minor) == 0)
3360                         ret = devid_str_encode(devid, minor);
3361                 if (minor != NULL)
3362                         devid_str_free(minor);
3363                 devid_free(devid);
3364         }
3365         (void) close(fd);
3366
3367         return (ret);
3368 }
3369
3370 /*
3371  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3372  * ignore any failure here, since a common case is for an unprivileged user to
3373  * type 'zpool status', and we'll display the correct information anyway.
3374  */
3375 static void
3376 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3377 {
3378         zfs_cmd_t zc = {"\0"};
3379
3380         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3381         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3382         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3383             &zc.zc_guid) == 0);
3384
3385         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3386 }
3387 #endif /* sun */
3388
3389 /*
3390  * Remove partition suffix from a vdev path.  Partition suffixes may take three
3391  * forms: "-partX", "pX", or "X", where X is a string of digits.  The second
3392  * case only occurs when the suffix is preceded by a digit, i.e. "md0p0" The
3393  * third case only occurs when preceded by a string matching the regular
3394  * expression "^([hsv]|xv)d[a-z]+", i.e. a scsi, ide, virtio or xen disk.
3395  */
3396 static char *
3397 strip_partition(libzfs_handle_t *hdl, char *path)
3398 {
3399         char *tmp = zfs_strdup(hdl, path);
3400         char *part = NULL, *d = NULL;
3401
3402         if ((part = strstr(tmp, "-part")) && part != tmp) {
3403                 d = part + 5;
3404         } else if ((part = strrchr(tmp, 'p')) &&
3405             part > tmp + 1 && isdigit(*(part-1))) {
3406                 d = part + 1;
3407         } else if ((tmp[0] == 'h' || tmp[0] == 's' || tmp[0] == 'v') &&
3408             tmp[1] == 'd') {
3409                 for (d = &tmp[2]; isalpha(*d); part = ++d);
3410         } else if (strncmp("xvd", tmp, 3) == 0) {
3411                 for (d = &tmp[3]; isalpha(*d); part = ++d);
3412         }
3413         if (part && d && *d != '\0') {
3414                 for (; isdigit(*d); d++);
3415                 if (*d == '\0')
3416                         *part = '\0';
3417         }
3418         return (tmp);
3419 }
3420
3421 #define PATH_BUF_LEN    64
3422
3423 /*
3424  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3425  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3426  * We also check if this is a whole disk, in which case we strip off the
3427  * trailing 's0' slice name.
3428  *
3429  * This routine is also responsible for identifying when disks have been
3430  * reconfigured in a new location.  The kernel will have opened the device by
3431  * devid, but the path will still refer to the old location.  To catch this, we
3432  * first do a path -> devid translation (which is fast for the common case).  If
3433  * the devid matches, we're done.  If not, we do a reverse devid -> path
3434  * translation and issue the appropriate ioctl() to update the path of the vdev.
3435  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3436  * of these checks.
3437  */
3438 char *
3439 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3440     int name_flags)
3441 {
3442         char *path, *type, *env;
3443         uint64_t value;
3444         char buf[PATH_BUF_LEN];
3445         char tmpbuf[PATH_BUF_LEN];
3446
3447         env = getenv("ZPOOL_VDEV_NAME_PATH");
3448         if (env && (strtoul(env, NULL, 0) > 0 ||
3449             !strncasecmp(env, "YES", 3) || !strncasecmp(env, "ON", 2)))
3450                 name_flags |= VDEV_NAME_PATH;
3451
3452         env = getenv("ZPOOL_VDEV_NAME_GUID");
3453         if (env && (strtoul(env, NULL, 0) > 0 ||
3454             !strncasecmp(env, "YES", 3) || !strncasecmp(env, "ON", 2)))
3455                 name_flags |= VDEV_NAME_GUID;
3456
3457         env = getenv("ZPOOL_VDEV_NAME_FOLLOW_LINKS");
3458         if (env && (strtoul(env, NULL, 0) > 0 ||
3459             !strncasecmp(env, "YES", 3) || !strncasecmp(env, "ON", 2)))
3460                 name_flags |= VDEV_NAME_FOLLOW_LINKS;
3461
3462         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &value) == 0 ||
3463             name_flags & VDEV_NAME_GUID) {
3464                 nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value);
3465                 (void) snprintf(buf, sizeof (buf), "%llu", (u_longlong_t)value);
3466                 path = buf;
3467         } else if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0) {
3468 #if defined(__sun__) || defined(__sun)
3469                 /*
3470                  * Live VDEV path updates to a kernel VDEV during a
3471                  * zpool_vdev_name lookup are not supported on Linux.
3472                  */
3473                 char *devid;
3474                 vdev_stat_t *vs;
3475                 uint_t vsc;
3476
3477                 /*
3478                  * If the device is dead (faulted, offline, etc) then don't
3479                  * bother opening it.  Otherwise we may be forcing the user to
3480                  * open a misbehaving device, which can have undesirable
3481                  * effects.
3482                  */
3483                 if ((nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3484                     (uint64_t **)&vs, &vsc) != 0 ||
3485                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3486                     zhp != NULL &&
3487                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3488                         /*
3489                          * Determine if the current path is correct.
3490                          */
3491                         char *newdevid = path_to_devid(path);
3492
3493                         if (newdevid == NULL ||
3494                             strcmp(devid, newdevid) != 0) {
3495                                 char *newpath;
3496
3497                                 if ((newpath = devid_to_path(devid)) != NULL) {
3498                                         /*
3499                                          * Update the path appropriately.
3500                                          */
3501                                         set_path(zhp, nv, newpath);
3502                                         if (nvlist_add_string(nv,
3503                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3504                                                 verify(nvlist_lookup_string(nv,
3505                                                     ZPOOL_CONFIG_PATH,
3506                                                     &path) == 0);
3507                                         free(newpath);
3508                                 }
3509                         }
3510
3511                         if (newdevid)
3512                                 devid_str_free(newdevid);
3513                 }
3514 #endif /* sun */
3515
3516                 if (name_flags & VDEV_NAME_FOLLOW_LINKS) {
3517                         char *rp = realpath(path, NULL);
3518                         if (rp) {
3519                                 strlcpy(buf, rp, sizeof (buf));
3520                                 path = buf;
3521                                 free(rp);
3522                         }
3523                 }
3524
3525                 /*
3526                  * For a block device only use the name.
3527                  */
3528                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
3529                 if ((strcmp(type, VDEV_TYPE_DISK) == 0) &&
3530                     !(name_flags & VDEV_NAME_PATH)) {
3531                         path = strrchr(path, '/');
3532                         path++;
3533                 }
3534
3535                 /*
3536                  * Remove the partition from the path it this is a whole disk.
3537                  */
3538                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, &value)
3539                     == 0 && value && !(name_flags & VDEV_NAME_PATH)) {
3540                         return (strip_partition(hdl, path));
3541                 }
3542         } else {
3543                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3544
3545                 /*
3546                  * If it's a raidz device, we need to stick in the parity level.
3547                  */
3548                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3549                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3550                             &value) == 0);
3551                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3552                             (u_longlong_t)value);
3553                         path = buf;
3554                 }
3555
3556                 /*
3557                  * We identify each top-level vdev by using a <type-id>
3558                  * naming convention.
3559                  */
3560                 if (name_flags & VDEV_NAME_TYPE_ID) {
3561                         uint64_t id;
3562                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3563                             &id) == 0);
3564                         (void) snprintf(tmpbuf, sizeof (tmpbuf), "%s-%llu",
3565                             path, (u_longlong_t)id);
3566                         path = tmpbuf;
3567                 }
3568         }
3569
3570         return (zfs_strdup(hdl, path));
3571 }
3572
3573 static int
3574 zbookmark_mem_compare(const void *a, const void *b)
3575 {
3576         return (memcmp(a, b, sizeof (zbookmark_phys_t)));
3577 }
3578
3579 /*
3580  * Retrieve the persistent error log, uniquify the members, and return to the
3581  * caller.
3582  */
3583 int
3584 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3585 {
3586         zfs_cmd_t zc = {"\0"};
3587         uint64_t count;
3588         zbookmark_phys_t *zb = NULL;
3589         int i;
3590
3591         /*
3592          * Retrieve the raw error list from the kernel.  If the number of errors
3593          * has increased, allocate more space and continue until we get the
3594          * entire list.
3595          */
3596         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3597             &count) == 0);
3598         if (count == 0)
3599                 return (0);
3600         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3601             count * sizeof (zbookmark_phys_t))) == (uintptr_t)NULL)
3602                 return (-1);
3603         zc.zc_nvlist_dst_size = count;
3604         (void) strcpy(zc.zc_name, zhp->zpool_name);
3605         for (;;) {
3606                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3607                     &zc) != 0) {
3608                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3609                         if (errno == ENOMEM) {
3610                                 void *dst;
3611
3612                                 count = zc.zc_nvlist_dst_size;
3613                                 dst = zfs_alloc(zhp->zpool_hdl, count *
3614                                     sizeof (zbookmark_phys_t));
3615                                 if (dst == NULL)
3616                                         return (-1);
3617                                 zc.zc_nvlist_dst = (uintptr_t)dst;
3618                         } else {
3619                                 return (-1);
3620                         }
3621                 } else {
3622                         break;
3623                 }
3624         }
3625
3626         /*
3627          * Sort the resulting bookmarks.  This is a little confusing due to the
3628          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3629          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3630          * _not_ copied as part of the process.  So we point the start of our
3631          * array appropriate and decrement the total number of elements.
3632          */
3633         zb = ((zbookmark_phys_t *)(uintptr_t)zc.zc_nvlist_dst) +
3634             zc.zc_nvlist_dst_size;
3635         count -= zc.zc_nvlist_dst_size;
3636
3637         qsort(zb, count, sizeof (zbookmark_phys_t), zbookmark_mem_compare);
3638
3639         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3640
3641         /*
3642          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3643          */
3644         for (i = 0; i < count; i++) {
3645                 nvlist_t *nv;
3646
3647                 /* ignoring zb_blkid and zb_level for now */
3648                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3649                     zb[i-1].zb_object == zb[i].zb_object)
3650                         continue;
3651
3652                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3653                         goto nomem;
3654                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3655                     zb[i].zb_objset) != 0) {
3656                         nvlist_free(nv);
3657                         goto nomem;
3658                 }
3659                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3660                     zb[i].zb_object) != 0) {
3661                         nvlist_free(nv);
3662                         goto nomem;
3663                 }
3664                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3665                         nvlist_free(nv);
3666                         goto nomem;
3667                 }
3668                 nvlist_free(nv);
3669         }
3670
3671         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3672         return (0);
3673
3674 nomem:
3675         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3676         return (no_memory(zhp->zpool_hdl));
3677 }
3678
3679 /*
3680  * Upgrade a ZFS pool to the latest on-disk version.
3681  */
3682 int
3683 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3684 {
3685         zfs_cmd_t zc = {"\0"};
3686         libzfs_handle_t *hdl = zhp->zpool_hdl;
3687
3688         (void) strcpy(zc.zc_name, zhp->zpool_name);
3689         zc.zc_cookie = new_version;
3690
3691         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3692                 return (zpool_standard_error_fmt(hdl, errno,
3693                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3694                     zhp->zpool_name));
3695         return (0);
3696 }
3697
3698 void
3699 zfs_save_arguments(int argc, char **argv, char *string, int len)
3700 {
3701         int i;
3702
3703         (void) strlcpy(string, basename(argv[0]), len);
3704         for (i = 1; i < argc; i++) {
3705                 (void) strlcat(string, " ", len);
3706                 (void) strlcat(string, argv[i], len);
3707         }
3708 }
3709
3710 int
3711 zpool_log_history(libzfs_handle_t *hdl, const char *message)
3712 {
3713         zfs_cmd_t zc = {"\0"};
3714         nvlist_t *args;
3715         int err;
3716
3717         args = fnvlist_alloc();
3718         fnvlist_add_string(args, "message", message);
3719         err = zcmd_write_src_nvlist(hdl, &zc, args);
3720         if (err == 0)
3721                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_LOG_HISTORY, &zc);
3722         nvlist_free(args);
3723         zcmd_free_nvlists(&zc);
3724         return (err);
3725 }
3726
3727 /*
3728  * Perform ioctl to get some command history of a pool.
3729  *
3730  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3731  * logical offset of the history buffer to start reading from.
3732  *
3733  * Upon return, 'off' is the next logical offset to read from and
3734  * 'len' is the actual amount of bytes read into 'buf'.
3735  */
3736 static int
3737 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3738 {
3739         zfs_cmd_t zc = {"\0"};
3740         libzfs_handle_t *hdl = zhp->zpool_hdl;
3741
3742         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3743
3744         zc.zc_history = (uint64_t)(uintptr_t)buf;
3745         zc.zc_history_len = *len;
3746         zc.zc_history_offset = *off;
3747
3748         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3749                 switch (errno) {
3750                 case EPERM:
3751                         return (zfs_error_fmt(hdl, EZFS_PERM,
3752                             dgettext(TEXT_DOMAIN,
3753                             "cannot show history for pool '%s'"),
3754                             zhp->zpool_name));
3755                 case ENOENT:
3756                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3757                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3758                             "'%s'"), zhp->zpool_name));
3759                 case ENOTSUP:
3760                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3761                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3762                             "'%s', pool must be upgraded"), zhp->zpool_name));
3763                 default:
3764                         return (zpool_standard_error_fmt(hdl, errno,
3765                             dgettext(TEXT_DOMAIN,
3766                             "cannot get history for '%s'"), zhp->zpool_name));
3767                 }
3768         }
3769
3770         *len = zc.zc_history_len;
3771         *off = zc.zc_history_offset;
3772
3773         return (0);
3774 }
3775
3776 /*
3777  * Process the buffer of nvlists, unpacking and storing each nvlist record
3778  * into 'records'.  'leftover' is set to the number of bytes that weren't
3779  * processed as there wasn't a complete record.
3780  */
3781 int
3782 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3783     nvlist_t ***records, uint_t *numrecords)
3784 {
3785         uint64_t reclen;
3786         nvlist_t *nv;
3787         int i;
3788
3789         while (bytes_read > sizeof (reclen)) {
3790
3791                 /* get length of packed record (stored as little endian) */
3792                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3793                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3794
3795                 if (bytes_read < sizeof (reclen) + reclen)
3796                         break;
3797
3798                 /* unpack record */
3799                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3800                         return (ENOMEM);
3801                 bytes_read -= sizeof (reclen) + reclen;
3802                 buf += sizeof (reclen) + reclen;
3803
3804                 /* add record to nvlist array */
3805                 (*numrecords)++;
3806                 if (ISP2(*numrecords + 1)) {
3807                         *records = realloc(*records,
3808                             *numrecords * 2 * sizeof (nvlist_t *));
3809                 }
3810                 (*records)[*numrecords - 1] = nv;
3811         }
3812
3813         *leftover = bytes_read;
3814         return (0);
3815 }
3816
3817 /*
3818  * Retrieve the command history of a pool.
3819  */
3820 int
3821 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3822 {
3823         char *buf;
3824         int buflen = 128 * 1024;
3825         uint64_t off = 0;
3826         nvlist_t **records = NULL;
3827         uint_t numrecords = 0;
3828         int err, i;
3829
3830         buf = malloc(buflen);
3831         if (buf == NULL)
3832                 return (ENOMEM);
3833         do {
3834                 uint64_t bytes_read = buflen;
3835                 uint64_t leftover;
3836
3837                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3838                         break;
3839
3840                 /* if nothing else was read in, we're at EOF, just return */
3841                 if (!bytes_read)
3842                         break;
3843
3844                 if ((err = zpool_history_unpack(buf, bytes_read,
3845                     &leftover, &records, &numrecords)) != 0)
3846                         break;
3847                 off -= leftover;
3848                 if (leftover == bytes_read) {
3849                         /*
3850                          * no progress made, because buffer is not big enough
3851                          * to hold this record; resize and retry.
3852                          */
3853                         buflen *= 2;
3854                         free(buf);
3855                         buf = malloc(buflen);
3856                         if (buf == NULL)
3857                                 return (ENOMEM);
3858                 }
3859
3860                 /* CONSTCOND */
3861         } while (1);
3862
3863         free(buf);
3864
3865         if (!err) {
3866                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3867                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3868                     records, numrecords) == 0);
3869         }
3870         for (i = 0; i < numrecords; i++)
3871                 nvlist_free(records[i]);
3872         free(records);
3873
3874         return (err);
3875 }
3876
3877 /*
3878  * Retrieve the next event given the passed 'zevent_fd' file descriptor.
3879  * If there is a new event available 'nvp' will contain a newly allocated
3880  * nvlist and 'dropped' will be set to the number of missed events since
3881  * the last call to this function.  When 'nvp' is set to NULL it indicates
3882  * no new events are available.  In either case the function returns 0 and
3883  * it is up to the caller to free 'nvp'.  In the case of a fatal error the
3884  * function will return a non-zero value.  When the function is called in
3885  * blocking mode (the default, unless the ZEVENT_NONBLOCK flag is passed),
3886  * it will not return until a new event is available.
3887  */
3888 int
3889 zpool_events_next(libzfs_handle_t *hdl, nvlist_t **nvp,
3890     int *dropped, unsigned flags, int zevent_fd)
3891 {
3892         zfs_cmd_t zc = {"\0"};
3893         int error = 0;
3894
3895         *nvp = NULL;
3896         *dropped = 0;
3897         zc.zc_cleanup_fd = zevent_fd;
3898
3899         if (flags & ZEVENT_NONBLOCK)
3900                 zc.zc_guid = ZEVENT_NONBLOCK;
3901
3902         if (zcmd_alloc_dst_nvlist(hdl, &zc, ZEVENT_SIZE) != 0)
3903                 return (-1);
3904
3905 retry:
3906         if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_NEXT, &zc) != 0) {
3907                 switch (errno) {
3908                 case ESHUTDOWN:
3909                         error = zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
3910                             dgettext(TEXT_DOMAIN, "zfs shutdown"));
3911                         goto out;
3912                 case ENOENT:
3913                         /* Blocking error case should not occur */
3914                         if (!(flags & ZEVENT_NONBLOCK))
3915                                 error = zpool_standard_error_fmt(hdl, errno,
3916                                     dgettext(TEXT_DOMAIN, "cannot get event"));
3917
3918                         goto out;
3919                 case ENOMEM:
3920                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3921                                 error = zfs_error_fmt(hdl, EZFS_NOMEM,
3922                                     dgettext(TEXT_DOMAIN, "cannot get event"));
3923                                 goto out;
3924                         } else {
3925                                 goto retry;
3926                         }
3927                 default:
3928                         error = zpool_standard_error_fmt(hdl, errno,
3929                             dgettext(TEXT_DOMAIN, "cannot get event"));
3930                         goto out;
3931                 }
3932         }
3933
3934         error = zcmd_read_dst_nvlist(hdl, &zc, nvp);
3935         if (error != 0)
3936                 goto out;
3937
3938         *dropped = (int)zc.zc_cookie;
3939 out:
3940         zcmd_free_nvlists(&zc);
3941
3942         return (error);
3943 }
3944
3945 /*
3946  * Clear all events.
3947  */
3948 int
3949 zpool_events_clear(libzfs_handle_t *hdl, int *count)
3950 {
3951         zfs_cmd_t zc = {"\0"};
3952         char msg[1024];
3953
3954         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
3955             "cannot clear events"));
3956
3957         if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_CLEAR, &zc) != 0)
3958                 return (zpool_standard_error_fmt(hdl, errno, msg));
3959
3960         if (count != NULL)
3961                 *count = (int)zc.zc_cookie; /* # of events cleared */
3962
3963         return (0);
3964 }
3965
3966 /*
3967  * Seek to a specific EID, ZEVENT_SEEK_START, or ZEVENT_SEEK_END for
3968  * the passed zevent_fd file handle.  On success zero is returned,
3969  * otherwise -1 is returned and hdl->libzfs_error is set to the errno.
3970  */
3971 int
3972 zpool_events_seek(libzfs_handle_t *hdl, uint64_t eid, int zevent_fd)
3973 {
3974         zfs_cmd_t zc = {"\0"};
3975         int error = 0;
3976
3977         zc.zc_guid = eid;
3978         zc.zc_cleanup_fd = zevent_fd;
3979
3980         if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_SEEK, &zc) != 0) {
3981                 switch (errno) {
3982                 case ENOENT:
3983                         error = zfs_error_fmt(hdl, EZFS_NOENT,
3984                             dgettext(TEXT_DOMAIN, "cannot get event"));
3985                         break;
3986
3987                 case ENOMEM:
3988                         error = zfs_error_fmt(hdl, EZFS_NOMEM,
3989                             dgettext(TEXT_DOMAIN, "cannot get event"));
3990                         break;
3991
3992                 default:
3993                         error = zpool_standard_error_fmt(hdl, errno,
3994                             dgettext(TEXT_DOMAIN, "cannot get event"));
3995                         break;
3996                 }
3997         }
3998
3999         return (error);
4000 }
4001
4002 void
4003 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
4004     char *pathname, size_t len)
4005 {
4006         zfs_cmd_t zc = {"\0"};
4007         boolean_t mounted = B_FALSE;
4008         char *mntpnt = NULL;
4009         char dsname[MAXNAMELEN];
4010
4011         if (dsobj == 0) {
4012                 /* special case for the MOS */
4013                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>",
4014                     (longlong_t)obj);
4015                 return;
4016         }
4017
4018         /* get the dataset's name */
4019         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
4020         zc.zc_obj = dsobj;
4021         if (ioctl(zhp->zpool_hdl->libzfs_fd,
4022             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
4023                 /* just write out a path of two object numbers */
4024                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
4025                     (longlong_t)dsobj, (longlong_t)obj);
4026                 return;
4027         }
4028         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
4029
4030         /* find out if the dataset is mounted */
4031         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
4032
4033         /* get the corrupted object's path */
4034         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
4035         zc.zc_obj = obj;
4036         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
4037             &zc) == 0) {
4038                 if (mounted) {
4039                         (void) snprintf(pathname, len, "%s%s", mntpnt,
4040                             zc.zc_value);
4041                 } else {
4042                         (void) snprintf(pathname, len, "%s:%s",
4043                             dsname, zc.zc_value);
4044                 }
4045         } else {
4046                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname,
4047                     (longlong_t)obj);
4048         }
4049         free(mntpnt);
4050 }
4051
4052 /*
4053  * Read the EFI label from the config, if a label does not exist then
4054  * pass back the error to the caller. If the caller has passed a non-NULL
4055  * diskaddr argument then we set it to the starting address of the EFI
4056  * partition.
4057  */
4058 static int
4059 read_efi_label(nvlist_t *config, diskaddr_t *sb)
4060 {
4061         char *path;
4062         int fd;
4063         char diskname[MAXPATHLEN];
4064         int err = -1;
4065
4066         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
4067                 return (err);
4068
4069         (void) snprintf(diskname, sizeof (diskname), "%s%s", DISK_ROOT,
4070             strrchr(path, '/'));
4071         if ((fd = open(diskname, O_RDWR|O_DIRECT)) >= 0) {
4072                 struct dk_gpt *vtoc;
4073
4074                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
4075                         if (sb != NULL)
4076                                 *sb = vtoc->efi_parts[0].p_start;
4077                         efi_free(vtoc);
4078                 }
4079                 (void) close(fd);
4080         }
4081         return (err);
4082 }
4083
4084 /*
4085  * determine where a partition starts on a disk in the current
4086  * configuration
4087  */
4088 static diskaddr_t
4089 find_start_block(nvlist_t *config)
4090 {
4091         nvlist_t **child;
4092         uint_t c, children;
4093         diskaddr_t sb = MAXOFFSET_T;
4094         uint64_t wholedisk;
4095
4096         if (nvlist_lookup_nvlist_array(config,
4097             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
4098                 if (nvlist_lookup_uint64(config,
4099                     ZPOOL_CONFIG_WHOLE_DISK,
4100                     &wholedisk) != 0 || !wholedisk) {
4101                         return (MAXOFFSET_T);
4102                 }
4103                 if (read_efi_label(config, &sb) < 0)
4104                         sb = MAXOFFSET_T;
4105                 return (sb);
4106         }
4107
4108         for (c = 0; c < children; c++) {
4109                 sb = find_start_block(child[c]);
4110                 if (sb != MAXOFFSET_T) {
4111                         return (sb);
4112                 }
4113         }
4114         return (MAXOFFSET_T);
4115 }
4116
4117 static int
4118 zpool_label_disk_check(char *path)
4119 {
4120         struct dk_gpt *vtoc;
4121         int fd, err;
4122
4123         if ((fd = open(path, O_RDWR|O_DIRECT)) < 0)
4124                 return (errno);
4125
4126         if ((err = efi_alloc_and_read(fd, &vtoc)) != 0) {
4127                 (void) close(fd);
4128                 return (err);
4129         }
4130
4131         if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
4132                 efi_free(vtoc);
4133                 (void) close(fd);
4134                 return (EIDRM);
4135         }
4136
4137         efi_free(vtoc);
4138         (void) close(fd);
4139         return (0);
4140 }
4141
4142 /*
4143  * Generate a unique partition name for the ZFS member.  Partitions must
4144  * have unique names to ensure udev will be able to create symlinks under
4145  * /dev/disk/by-partlabel/ for all pool members.  The partition names are
4146  * of the form <pool>-<unique-id>.
4147  */
4148 static void
4149 zpool_label_name(char *label_name, int label_size)
4150 {
4151         uint64_t id = 0;
4152         int fd;
4153
4154         fd = open("/dev/urandom", O_RDONLY);
4155         if (fd > 0) {
4156                 if (read(fd, &id, sizeof (id)) != sizeof (id))
4157                         id = 0;
4158
4159                 close(fd);
4160         }
4161
4162         if (id == 0)
4163                 id = (((uint64_t)rand()) << 32) | (uint64_t)rand();
4164
4165         snprintf(label_name, label_size, "zfs-%016llx", (u_longlong_t) id);
4166 }
4167
4168 /*
4169  * Label an individual disk.  The name provided is the short name,
4170  * stripped of any leading /dev path.
4171  */
4172 int
4173 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, char *name)
4174 {
4175         char path[MAXPATHLEN];
4176         struct dk_gpt *vtoc;
4177         int rval, fd;
4178         size_t resv = EFI_MIN_RESV_SIZE;
4179         uint64_t slice_size;
4180         diskaddr_t start_block;
4181         char errbuf[1024];
4182
4183         /* prepare an error message just in case */
4184         (void) snprintf(errbuf, sizeof (errbuf),
4185             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
4186
4187         if (zhp) {
4188                 nvlist_t *nvroot;
4189
4190 #if defined(__sun__) || defined(__sun)
4191                 if (zpool_is_bootable(zhp)) {
4192                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4193                             "EFI labeled devices are not supported on root "
4194                             "pools."));
4195                         return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
4196                 }
4197 #endif
4198
4199                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
4200                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4201
4202                 if (zhp->zpool_start_block == 0)
4203                         start_block = find_start_block(nvroot);
4204                 else
4205                         start_block = zhp->zpool_start_block;
4206                 zhp->zpool_start_block = start_block;
4207         } else {
4208                 /* new pool */
4209                 start_block = NEW_START_BLOCK;
4210         }
4211
4212         (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT, name);
4213
4214         if ((fd = open(path, O_RDWR|O_DIRECT)) < 0) {
4215                 /*
4216                  * This shouldn't happen.  We've long since verified that this
4217                  * is a valid device.
4218                  */
4219                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
4220                     "label '%s': unable to open device: %d"), path, errno);
4221                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
4222         }
4223
4224         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
4225                 /*
4226                  * The only way this can fail is if we run out of memory, or we
4227                  * were unable to read the disk's capacity
4228                  */
4229                 if (errno == ENOMEM)
4230                         (void) no_memory(hdl);
4231
4232                 (void) close(fd);
4233                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
4234                     "label '%s': unable to read disk capacity"), path);
4235
4236                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
4237         }
4238
4239         slice_size = vtoc->efi_last_u_lba + 1;
4240         slice_size -= EFI_MIN_RESV_SIZE;
4241         if (start_block == MAXOFFSET_T)
4242                 start_block = NEW_START_BLOCK;
4243         slice_size -= start_block;
4244         slice_size = P2ALIGN(slice_size, PARTITION_END_ALIGNMENT);
4245
4246         vtoc->efi_parts[0].p_start = start_block;
4247         vtoc->efi_parts[0].p_size = slice_size;
4248
4249         /*
4250          * Why we use V_USR: V_BACKUP confuses users, and is considered
4251          * disposable by some EFI utilities (since EFI doesn't have a backup
4252          * slice).  V_UNASSIGNED is supposed to be used only for zero size
4253          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
4254          * etc. were all pretty specific.  V_USR is as close to reality as we
4255          * can get, in the absence of V_OTHER.
4256          */
4257         vtoc->efi_parts[0].p_tag = V_USR;
4258         zpool_label_name(vtoc->efi_parts[0].p_name, EFI_PART_NAME_LEN);
4259
4260         vtoc->efi_parts[8].p_start = slice_size + start_block;
4261         vtoc->efi_parts[8].p_size = resv;
4262         vtoc->efi_parts[8].p_tag = V_RESERVED;
4263
4264         if ((rval = efi_write(fd, vtoc)) != 0 || (rval = efi_rescan(fd)) != 0) {
4265                 /*
4266                  * Some block drivers (like pcata) may not support EFI
4267                  * GPT labels.  Print out a helpful error message dir-
4268                  * ecting the user to manually label the disk and give
4269                  * a specific slice.
4270                  */
4271                 (void) close(fd);
4272                 efi_free(vtoc);
4273
4274                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "try using "
4275                     "parted(8) and then provide a specific slice: %d"), rval);
4276                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4277         }
4278
4279         (void) close(fd);
4280         efi_free(vtoc);
4281
4282         (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT, name);
4283         (void) zfs_append_partition(path, MAXPATHLEN);
4284
4285         /* Wait to udev to signal use the device has settled. */
4286         rval = zpool_label_disk_wait(path, DISK_LABEL_WAIT);
4287         if (rval) {
4288                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "failed to "
4289                     "detect device partitions on '%s': %d"), path, rval);
4290                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4291         }
4292
4293         /* We can't be to paranoid.  Read the label back and verify it. */
4294         (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT, name);
4295         rval = zpool_label_disk_check(path);
4296         if (rval) {
4297                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "freshly written "
4298                     "EFI label on '%s' is damaged.  Ensure\nthis device "
4299                     "is not in in use, and is functioning properly: %d"),
4300                     path, rval);
4301                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4302         }
4303
4304         return (0);
4305 }