lib/libzfs/libzfs_pool.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  */
  25
  26 #include <ctype.h>
  27 #include <errno.h>
  28 #include <devid.h>
  29 #include <fcntl.h>
  30 #include <libintl.h>
  31 #include <stdio.h>
  32 #include <stdlib.h>
  33 #include <strings.h>
  34 #include <unistd.h>
  35 #include <sys/efi_partition.h>
  36 #include <sys/vtoc.h>
  37 #include <sys/zfs_ioctl.h>
  38 #include <dlfcn.h>
  39
  40 #include "zfs_namecheck.h"
  41 #include "zfs_prop.h"
  42 #include "libzfs_impl.h"
  43 #include "zfs_comutil.h"
  44
  45 static int read_efi_label(nvlist_t *config, diskaddr_t *sb);
  46
  47 #define DISK_ROOT       "/dev/dsk"
  48 #define RDISK_ROOT      "/dev/rdsk"
  49 #define BACKUP_SLICE    "s2"
  50
  51 typedef struct prop_flags {
  52         int create:1;   /* Validate property on creation */
  53         int import:1;   /* Validate property on import */
  54 } prop_flags_t;
  55
  56 /*
  57  * ====================================================================
  58  *   zpool property functions
  59  * ====================================================================
  60  */
  61
  62 static int
  63 zpool_get_all_props(zpool_handle_t *zhp)
  64 {
  65         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
  66         libzfs_handle_t *hdl = zhp->zpool_hdl;
  67
  68         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
  69
  70         if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
  71                 return (-1);
  72
  73         while (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
  74                 if (errno == ENOMEM) {
  75                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
  76                                 zcmd_free_nvlists(&zc);
  77                                 return (-1);
  78                         }
  79                 } else {
  80                         zcmd_free_nvlists(&zc);
  81                         return (-1);
  82                 }
  83         }
  84
  85         if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
  86                 zcmd_free_nvlists(&zc);
  87                 return (-1);
  88         }
  89
  90         zcmd_free_nvlists(&zc);
  91
  92         return (0);
  93 }
  94
  95 static int
  96 zpool_props_refresh(zpool_handle_t *zhp)
  97 {
  98         nvlist_t *old_props;
  99
 100         old_props = zhp->zpool_props;
 101
 102         if (zpool_get_all_props(zhp) != 0)
 103                 return (-1);
 104
 105         nvlist_free(old_props);
 106         return (0);
 107 }
 108
 109 static char *
 110 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
 111     zprop_source_t *src)
 112 {
 113         nvlist_t *nv, *nvl;
 114         uint64_t ival;
 115         char *value;
 116         zprop_source_t source;
 117
 118         nvl = zhp->zpool_props;
 119         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 120                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &ival) == 0);
 121                 source = ival;
 122                 verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
 123         } else {
 124                 source = ZPROP_SRC_DEFAULT;
 125                 if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
 126                         value = "-";
 127         }
 128
 129         if (src)
 130                 *src = source;
 131
 132         return (value);
 133 }
 134
 135 uint64_t
 136 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
 137 {
 138         nvlist_t *nv, *nvl;
 139         uint64_t value;
 140         zprop_source_t source;
 141
 142         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
 143                 /*
 144                  * zpool_get_all_props() has most likely failed because
 145                  * the pool is faulted, but if all we need is the top level
 146                  * vdev's guid then get it from the zhp config nvlist.
 147                  */
 148                 if ((prop == ZPOOL_PROP_GUID) &&
 149                     (nvlist_lookup_nvlist(zhp->zpool_config,
 150                     ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
 151                     (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
 152                     == 0)) {
 153                         return (value);
 154                 }
 155                 return (zpool_prop_default_numeric(prop));
 156         }
 157
 158         nvl = zhp->zpool_props;
 159         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 160                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &value) == 0);
 161                 source = value;
 162                 verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
 163         } else {
 164                 source = ZPROP_SRC_DEFAULT;
 165                 value = zpool_prop_default_numeric(prop);
 166         }
 167
 168         if (src)
 169                 *src = source;
 170
 171         return (value);
 172 }
 173
 174 /*
 175  * Map VDEV STATE to printed strings.
 176  */
 177 char *
 178 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
 179 {
 180         switch (state) {
 181         default:
 182                 break;
 183         case VDEV_STATE_CLOSED:
 184         case VDEV_STATE_OFFLINE:
 185                 return (gettext("OFFLINE"));
 186         case VDEV_STATE_REMOVED:
 187                 return (gettext("REMOVED"));
 188         case VDEV_STATE_CANT_OPEN:
 189                 if (aux == VDEV_AUX_CORRUPT_DATA || aux == VDEV_AUX_BAD_LOG)
 190                         return (gettext("FAULTED"));
 191                 else if (aux == VDEV_AUX_SPLIT_POOL)
 192                         return (gettext("SPLIT"));
 193                 else
 194                         return (gettext("UNAVAIL"));
 195         case VDEV_STATE_FAULTED:
 196                 return (gettext("FAULTED"));
 197         case VDEV_STATE_DEGRADED:
 198                 return (gettext("DEGRADED"));
 199         case VDEV_STATE_HEALTHY:
 200                 return (gettext("ONLINE"));
 201         }
 202
 203         return (gettext("UNKNOWN"));
 204 }
 205
 206 /*
 207  * Get a zpool property value for 'prop' and return the value in
 208  * a pre-allocated buffer.
 209  */
 210 int
 211 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf, size_t len,
 212     zprop_source_t *srctype)
 213 {
 214         uint64_t intval;
 215         const char *strval;
 216         zprop_source_t src = ZPROP_SRC_NONE;
 217         nvlist_t *nvroot;
 218         vdev_stat_t *vs;
 219         uint_t vsc;
 220
 221         if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
 222                 switch (prop) {
 223                 case ZPOOL_PROP_NAME:
 224                         (void) strlcpy(buf, zpool_get_name(zhp), len);
 225                         break;
 226
 227                 case ZPOOL_PROP_HEALTH:
 228                         (void) strlcpy(buf, "FAULTED", len);
 229                         break;
 230
 231                 case ZPOOL_PROP_GUID:
 232                         intval = zpool_get_prop_int(zhp, prop, &src);
 233                         (void) snprintf(buf, len, "%llu", (u_longlong_t)intval);
 234                         break;
 235
 236                 case ZPOOL_PROP_ALTROOT:
 237                 case ZPOOL_PROP_CACHEFILE:
 238                         if (zhp->zpool_props != NULL ||
 239                             zpool_get_all_props(zhp) == 0) {
 240                                 (void) strlcpy(buf,
 241                                     zpool_get_prop_string(zhp, prop, &src),
 242                                     len);
 243                                 if (srctype != NULL)
 244                                         *srctype = src;
 245                                 return (0);
 246                         }
 247                         /* FALLTHROUGH */
 248                 default:
 249                         (void) strlcpy(buf, "-", len);
 250                         break;
 251                 }
 252
 253                 if (srctype != NULL)
 254                         *srctype = src;
 255                 return (0);
 256         }
 257
 258         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
 259             prop != ZPOOL_PROP_NAME)
 260                 return (-1);
 261
 262         switch (zpool_prop_get_type(prop)) {
 263         case PROP_TYPE_STRING:
 264                 (void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
 265                     len);
 266                 break;
 267
 268         case PROP_TYPE_NUMBER:
 269                 intval = zpool_get_prop_int(zhp, prop, &src);
 270
 271                 switch (prop) {
 272                 case ZPOOL_PROP_SIZE:
 273                 case ZPOOL_PROP_ALLOCATED:
 274                 case ZPOOL_PROP_FREE:
 275                         (void) zfs_nicenum(intval, buf, len);
 276                         break;
 277
 278                 case ZPOOL_PROP_CAPACITY:
 279                         (void) snprintf(buf, len, "%llu%%",
 280                             (u_longlong_t)intval);
 281                         break;
 282
 283                 case ZPOOL_PROP_DEDUPRATIO:
 284                         (void) snprintf(buf, len, "%llu.%02llux",
 285                             (u_longlong_t)(intval / 100),
 286                             (u_longlong_t)(intval % 100));
 287                         break;
 288
 289                 case ZPOOL_PROP_HEALTH:
 290                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 291                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 292                         verify(nvlist_lookup_uint64_array(nvroot,
 293                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
 294                             == 0);
 295
 296                         (void) strlcpy(buf, zpool_state_to_name(intval,
 297                             vs->vs_aux), len);
 298                         break;
 299                 default:
 300                         (void) snprintf(buf, len, "%llu", (u_longlong_t)intval);
 301                 }
 302                 break;
 303
 304         case PROP_TYPE_INDEX:
 305                 intval = zpool_get_prop_int(zhp, prop, &src);
 306                 if (zpool_prop_index_to_string(prop, intval, &strval)
 307                     != 0)
 308                         return (-1);
 309                 (void) strlcpy(buf, strval, len);
 310                 break;
 311
 312         default:
 313                 abort();
 314         }
 315
 316         if (srctype)
 317                 *srctype = src;
 318
 319         return (0);
 320 }
 321
 322 /*
 323  * Check if the bootfs name has the same pool name as it is set to.
 324  * Assuming bootfs is a valid dataset name.
 325  */
 326 static boolean_t
 327 bootfs_name_valid(const char *pool, char *bootfs)
 328 {
 329         int len = strlen(pool);
 330
 331         if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
 332                 return (B_FALSE);
 333
 334         if (strncmp(pool, bootfs, len) == 0 &&
 335             (bootfs[len] == '/' || bootfs[len] == '\0'))
 336                 return (B_TRUE);
 337
 338         return (B_FALSE);
 339 }
 340
 341 /*
 342  * Inspect the configuration to determine if any of the devices contain
 343  * an EFI label.
 344  */
 345 static boolean_t
 346 pool_uses_efi(nvlist_t *config)
 347 {
 348         nvlist_t **child;
 349         uint_t c, children;
 350
 351         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
 352             &child, &children) != 0)
 353                 return (read_efi_label(config, NULL) >= 0);
 354
 355         for (c = 0; c < children; c++) {
 356                 if (pool_uses_efi(child[c]))
 357                         return (B_TRUE);
 358         }
 359         return (B_FALSE);
 360 }
 361
 362 static boolean_t
 363 pool_is_bootable(zpool_handle_t *zhp)
 364 {
 365         char bootfs[ZPOOL_MAXNAMELEN];
 366
 367         return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
 368             sizeof (bootfs), NULL) == 0 && strncmp(bootfs, "-",
 369             sizeof (bootfs)) != 0);
 370 }
 371
 372
 373 /*
 374  * Given an nvlist of zpool properties to be set, validate that they are
 375  * correct, and parse any numeric properties (index, boolean, etc) if they are
 376  * specified as strings.
 377  */
 378 static nvlist_t *
 379 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 380     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
 381 {
 382         nvpair_t *elem;
 383         nvlist_t *retprops;
 384         zpool_prop_t prop;
 385         char *strval;
 386         uint64_t intval;
 387         char *slash;
 388         struct stat64 statbuf;
 389         zpool_handle_t *zhp;
 390         nvlist_t *nvroot;
 391
 392         if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
 393                 (void) no_memory(hdl);
 394                 return (NULL);
 395         }
 396
 397         elem = NULL;
 398         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 399                 const char *propname = nvpair_name(elem);
 400
 401                 /*
 402                  * Make sure this property is valid and applies to this type.
 403                  */
 404                 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL) {
 405                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 406                             "invalid property '%s'"), propname);
 407                         (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 408                         goto error;
 409                 }
 410
 411                 if (zpool_prop_readonly(prop)) {
 412                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 413                             "is readonly"), propname);
 414                         (void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
 415                         goto error;
 416                 }
 417
 418                 if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
 419                     &strval, &intval, errbuf) != 0)
 420                         goto error;
 421
 422                 /*
 423                  * Perform additional checking for specific properties.
 424                  */
 425                 switch (prop) {
 426                 default:
 427                         break;
 428                 case ZPOOL_PROP_VERSION:
 429                         if (intval < version || intval > SPA_VERSION) {
 430                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 431                                     "property '%s' number %d is invalid."),
 432                                     propname, intval);
 433                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 434                                 goto error;
 435                         }
 436                         break;
 437
 438                 case ZPOOL_PROP_BOOTFS:
 439                         if (flags.create || flags.import) {
 440                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 441                                     "property '%s' cannot be set at creation "
 442                                     "or import time"), propname);
 443                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 444                                 goto error;
 445                         }
 446
 447                         if (version < SPA_VERSION_BOOTFS) {
 448                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 449                                     "pool must be upgraded to support "
 450                                     "'%s' property"), propname);
 451                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 452                                 goto error;
 453                         }
 454
 455                         /*
 456                          * bootfs property value has to be a dataset name and
 457                          * the dataset has to be in the same pool as it sets to.
 458                          */
 459                         if (strval[0] != '\0' && !bootfs_name_valid(poolname,
 460                             strval)) {
 461                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 462                                     "is an invalid name"), strval);
 463                                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
 464                                 goto error;
 465                         }
 466
 467                         if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
 468                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 469                                     "could not open pool '%s'"), poolname);
 470                                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
 471                                 goto error;
 472                         }
 473                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 474                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 475
 476                         /*
 477                          * bootfs property cannot be set on a disk which has
 478                          * been EFI labeled.
 479                          */
 480                         if (pool_uses_efi(nvroot)) {
 481                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 482                                     "property '%s' not supported on "
 483                                     "EFI labeled devices"), propname);
 484                                 (void) zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf);
 485                                 zpool_close(zhp);
 486                                 goto error;
 487                         }
 488                         zpool_close(zhp);
 489                         break;
 490
 491                 case ZPOOL_PROP_ALTROOT:
 492                         if (!flags.create && !flags.import) {
 493                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 494                                     "property '%s' can only be set during pool "
 495                                     "creation or import"), propname);
 496                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 497                                 goto error;
 498                         }
 499
 500                         if (strval[0] != '/') {
 501                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 502                                     "bad alternate root '%s'"), strval);
 503                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 504                                 goto error;
 505                         }
 506                         break;
 507
 508                 case ZPOOL_PROP_CACHEFILE:
 509                         if (strval[0] == '\0')
 510                                 break;
 511
 512                         if (strcmp(strval, "none") == 0)
 513                                 break;
 514
 515                         if (strval[0] != '/') {
 516                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 517                                     "property '%s' must be empty, an "
 518                                     "absolute path, or 'none'"), propname);
 519                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 520                                 goto error;
 521                         }
 522
 523                         slash = strrchr(strval, '/');
 524
 525                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 526                             strcmp(slash, "/..") == 0) {
 527                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 528                                     "'%s' is not a valid file"), strval);
 529                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 530                                 goto error;
 531                         }
 532
 533                         *slash = '\0';
 534
 535                         if (strval[0] != '\0' &&
 536                             (stat64(strval, &statbuf) != 0 ||
 537                             !S_ISDIR(statbuf.st_mode))) {
 538                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 539                                     "'%s' is not a valid directory"),
 540                                     strval);
 541                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 542                                 goto error;
 543                         }
 544
 545                         *slash = '/';
 546                         break;
 547
 548                 case ZPOOL_PROP_READONLY:
 549                         if (!flags.import) {
 550                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 551                                     "property '%s' can only be set at "
 552                                     "import time"), propname);
 553                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 554                                 goto error;
 555                         }
 556                         break;
 557                 }
 558         }
 559
 560         return (retprops);
 561 error:
 562         nvlist_free(retprops);
 563         return (NULL);
 564 }
 565
 566 /*
 567  * Set zpool property : propname=propval.
 568  */
 569 int
 570 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
 571 {
 572         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
 573         int ret = -1;
 574         char errbuf[1024];
 575         nvlist_t *nvl = NULL;
 576         nvlist_t *realprops;
 577         uint64_t version;
 578         prop_flags_t flags = { 0 };
 579
 580         (void) snprintf(errbuf, sizeof (errbuf),
 581             dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
 582             zhp->zpool_name);
 583
 584         if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
 585                 return (no_memory(zhp->zpool_hdl));
 586
 587         if (nvlist_add_string(nvl, propname, propval) != 0) {
 588                 nvlist_free(nvl);
 589                 return (no_memory(zhp->zpool_hdl));
 590         }
 591
 592         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
 593         if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
 594             zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
 595                 nvlist_free(nvl);
 596                 return (-1);
 597         }
 598
 599         nvlist_free(nvl);
 600         nvl = realprops;
 601
 602         /*
 603          * Execute the corresponding ioctl() to set this property.
 604          */
 605         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 606
 607         if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
 608                 nvlist_free(nvl);
 609                 return (-1);
 610         }
 611
 612         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
 613
 614         zcmd_free_nvlists(&zc);
 615         nvlist_free(nvl);
 616
 617         if (ret)
 618                 (void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
 619         else
 620                 (void) zpool_props_refresh(zhp);
 621
 622         return (ret);
 623 }
 624
 625 int
 626 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
 627 {
 628         libzfs_handle_t *hdl = zhp->zpool_hdl;
 629         zprop_list_t *entry;
 630         char buf[ZFS_MAXPROPLEN];
 631
 632         if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
 633                 return (-1);
 634
 635         for (entry = *plp; entry != NULL; entry = entry->pl_next) {
 636
 637                 if (entry->pl_fixed)
 638                         continue;
 639
 640                 if (entry->pl_prop != ZPROP_INVAL &&
 641                     zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
 642                     NULL) == 0) {
 643                         if (strlen(buf) > entry->pl_width)
 644                                 entry->pl_width = strlen(buf);
 645                 }
 646         }
 647
 648         return (0);
 649 }
 650
 651
 652 /*
 653  * Don't start the slice at the default block of 34; many storage
 654  * devices will use a stripe width of 128k, so start there instead.
 655  */
 656 #define NEW_START_BLOCK 256
 657
 658 /*
 659  * Validate the given pool name, optionally putting an extended error message in
 660  * 'buf'.
 661  */
 662 boolean_t
 663 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
 664 {
 665         namecheck_err_t why;
 666         char what;
 667         int ret;
 668
 669         ret = pool_namecheck(pool, &why, &what);
 670
 671         /*
 672          * The rules for reserved pool names were extended at a later point.
 673          * But we need to support users with existing pools that may now be
 674          * invalid.  So we only check for this expanded set of names during a
 675          * create (or import), and only in userland.
 676          */
 677         if (ret == 0 && !isopen &&
 678             (strncmp(pool, "mirror", 6) == 0 ||
 679             strncmp(pool, "raidz", 5) == 0 ||
 680             strncmp(pool, "spare", 5) == 0 ||
 681             strcmp(pool, "log") == 0)) {
 682                 if (hdl != NULL)
 683                         zfs_error_aux(hdl,
 684                             dgettext(TEXT_DOMAIN, "name is reserved"));
 685                 return (B_FALSE);
 686         }
 687
 688
 689         if (ret != 0) {
 690                 if (hdl != NULL) {
 691                         switch (why) {
 692                         case NAME_ERR_TOOLONG:
 693                                 zfs_error_aux(hdl,
 694                                     dgettext(TEXT_DOMAIN, "name is too long"));
 695                                 break;
 696
 697                         case NAME_ERR_INVALCHAR:
 698                                 zfs_error_aux(hdl,
 699                                     dgettext(TEXT_DOMAIN, "invalid character "
 700                                     "'%c' in pool name"), what);
 701                                 break;
 702
 703                         case NAME_ERR_NOLETTER:
 704                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 705                                     "name must begin with a letter"));
 706                                 break;
 707
 708                         case NAME_ERR_RESERVED:
 709                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 710                                     "name is reserved"));
 711                                 break;
 712
 713                         case NAME_ERR_DISKLIKE:
 714                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 715                                     "pool name is reserved"));
 716                                 break;
 717
 718                         case NAME_ERR_LEADING_SLASH:
 719                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 720                                     "leading slash in name"));
 721                                 break;
 722
 723                         case NAME_ERR_EMPTY_COMPONENT:
 724                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 725                                     "empty component in name"));
 726                                 break;
 727
 728                         case NAME_ERR_TRAILING_SLASH:
 729                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 730                                     "trailing slash in name"));
 731                                 break;
 732
 733                         case NAME_ERR_MULTIPLE_AT:
 734                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 735                                     "multiple '@' delimiters in name"));
 736                                 break;
 737                         case NAME_ERR_NO_AT:
 738                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 739                                     "permission set is missing '@'"));
 740                                 break;
 741                         }
 742                 }
 743                 return (B_FALSE);
 744         }
 745
 746         return (B_TRUE);
 747 }
 748
 749 /*
 750  * Open a handle to the given pool, even if the pool is currently in the FAULTED
 751  * state.
 752  */
 753 zpool_handle_t *
 754 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
 755 {
 756         zpool_handle_t *zhp;
 757         boolean_t missing;
 758
 759         /*
 760          * Make sure the pool name is valid.
 761          */
 762         if (!zpool_name_valid(hdl, B_TRUE, pool)) {
 763                 (void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
 764                     dgettext(TEXT_DOMAIN, "cannot open '%s'"),
 765                     pool);
 766                 return (NULL);
 767         }
 768
 769         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
 770                 return (NULL);
 771
 772         zhp->zpool_hdl = hdl;
 773         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
 774
 775         if (zpool_refresh_stats(zhp, &missing) != 0) {
 776                 zpool_close(zhp);
 777                 return (NULL);
 778         }
 779
 780         if (missing) {
 781                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
 782                 (void) zfs_error_fmt(hdl, EZFS_NOENT,
 783                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
 784                 zpool_close(zhp);
 785                 return (NULL);
 786         }
 787
 788         return (zhp);
 789 }
 790
 791 /*
 792  * Like the above, but silent on error.  Used when iterating over pools (because
 793  * the configuration cache may be out of date).
 794  */
 795 int
 796 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
 797 {
 798         zpool_handle_t *zhp;
 799         boolean_t missing;
 800
 801         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
 802                 return (-1);
 803
 804         zhp->zpool_hdl = hdl;
 805         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
 806
 807         if (zpool_refresh_stats(zhp, &missing) != 0) {
 808                 zpool_close(zhp);
 809                 return (-1);
 810         }
 811
 812         if (missing) {
 813                 zpool_close(zhp);
 814                 *ret = NULL;
 815                 return (0);
 816         }
 817
 818         *ret = zhp;
 819         return (0);
 820 }
 821
 822 /*
 823  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
 824  * state.
 825  */
 826 zpool_handle_t *
 827 zpool_open(libzfs_handle_t *hdl, const char *pool)
 828 {
 829         zpool_handle_t *zhp;
 830
 831         if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
 832                 return (NULL);
 833
 834         if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
 835                 (void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
 836                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
 837                 zpool_close(zhp);
 838                 return (NULL);
 839         }
 840
 841         return (zhp);
 842 }
 843
 844 /*
 845  * Close the handle.  Simply frees the memory associated with the handle.
 846  */
 847 void
 848 zpool_close(zpool_handle_t *zhp)
 849 {
 850         if (zhp->zpool_config)
 851                 nvlist_free(zhp->zpool_config);
 852         if (zhp->zpool_old_config)
 853                 nvlist_free(zhp->zpool_old_config);
 854         if (zhp->zpool_props)
 855                 nvlist_free(zhp->zpool_props);
 856         free(zhp);
 857 }
 858
 859 /*
 860  * Return the name of the pool.
 861  */
 862 const char *
 863 zpool_get_name(zpool_handle_t *zhp)
 864 {
 865         return (zhp->zpool_name);
 866 }
 867
 868
 869 /*
 870  * Return the state of the pool (ACTIVE or UNAVAILABLE)
 871  */
 872 int
 873 zpool_get_state(zpool_handle_t *zhp)
 874 {
 875         return (zhp->zpool_state);
 876 }
 877
 878 /*
 879  * Create the named pool, using the provided vdev list.  It is assumed
 880  * that the consumer has already validated the contents of the nvlist, so we
 881  * don't have to worry about error semantics.
 882  */
 883 int
 884 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
 885     nvlist_t *props, nvlist_t *fsprops)
 886 {
 887         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
 888         nvlist_t *zc_fsprops = NULL;
 889         nvlist_t *zc_props = NULL;
 890         char msg[1024];
 891         char *altroot;
 892         int ret = -1;
 893
 894         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
 895             "cannot create '%s'"), pool);
 896
 897         if (!zpool_name_valid(hdl, B_FALSE, pool))
 898                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
 899
 900         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
 901                 return (-1);
 902
 903         if (props) {
 904                 prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
 905
 906                 if ((zc_props = zpool_valid_proplist(hdl, pool, props,
 907                     SPA_VERSION_1, flags, msg)) == NULL) {
 908                         goto create_failed;
 909                 }
 910         }
 911
 912         if (fsprops) {
 913                 uint64_t zoned;
 914                 char *zonestr;
 915
 916                 zoned = ((nvlist_lookup_string(fsprops,
 917                     zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
 918                     strcmp(zonestr, "on") == 0);
 919
 920                 if ((zc_fsprops = zfs_valid_proplist(hdl,
 921                     ZFS_TYPE_FILESYSTEM, fsprops, zoned, NULL, msg)) == NULL) {
 922                         goto create_failed;
 923                 }
 924                 if (!zc_props &&
 925                     (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
 926                         goto create_failed;
 927                 }
 928                 if (nvlist_add_nvlist(zc_props,
 929                     ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
 930                         goto create_failed;
 931                 }
 932         }
 933
 934         if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
 935                 goto create_failed;
 936
 937         (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
 938
 939         if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
 940
 941                 zcmd_free_nvlists(&zc);
 942                 nvlist_free(zc_props);
 943                 nvlist_free(zc_fsprops);
 944
 945                 switch (errno) {
 946                 case EBUSY:
 947                         /*
 948                          * This can happen if the user has specified the same
 949                          * device multiple times.  We can't reliably detect this
 950                          * until we try to add it and see we already have a
 951                          * label.
 952                          */
 953                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 954                             "one or more vdevs refer to the same device"));
 955                         return (zfs_error(hdl, EZFS_BADDEV, msg));
 956
 957                 case EOVERFLOW:
 958                         /*
 959                          * This occurs when one of the devices is below
 960                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
 961                          * device was the problem device since there's no
 962                          * reliable way to determine device size from userland.
 963                          */
 964                         {
 965                                 char buf[64];
 966
 967                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
 968
 969                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 970                                     "one or more devices is less than the "
 971                                     "minimum size (%s)"), buf);
 972                         }
 973                         return (zfs_error(hdl, EZFS_BADDEV, msg));
 974
 975                 case ENOSPC:
 976                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 977                             "one or more devices is out of space"));
 978                         return (zfs_error(hdl, EZFS_BADDEV, msg));
 979
 980                 case ENOTBLK:
 981                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 982                             "cache device must be a disk or disk slice"));
 983                         return (zfs_error(hdl, EZFS_BADDEV, msg));
 984
 985                 default:
 986                         return (zpool_standard_error(hdl, errno, msg));
 987                 }
 988         }
 989
 990         /*
 991          * If this is an alternate root pool, then we automatically set the
 992          * mountpoint of the root dataset to be '/'.
 993          */
 994         if (nvlist_lookup_string(props, zpool_prop_to_name(ZPOOL_PROP_ALTROOT),
 995             &altroot) == 0) {
 996                 zfs_handle_t *zhp;
 997
 998                 verify((zhp = zfs_open(hdl, pool, ZFS_TYPE_DATASET)) != NULL);
 999                 verify(zfs_prop_set(zhp, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT),
1000                     "/") == 0);
1001
1002                 zfs_close(zhp);
1003         }
1004
1005 create_failed:
1006         zcmd_free_nvlists(&zc);
1007         nvlist_free(zc_props);
1008         nvlist_free(zc_fsprops);
1009         return (ret);
1010 }
1011
1012 /*
1013  * Destroy the given pool.  It is up to the caller to ensure that there are no
1014  * datasets left in the pool.
1015  */
1016 int
1017 zpool_destroy(zpool_handle_t *zhp)
1018 {
1019         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1020         zfs_handle_t *zfp = NULL;
1021         libzfs_handle_t *hdl = zhp->zpool_hdl;
1022         char msg[1024];
1023
1024         if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1025             (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1026                 return (-1);
1027
1028         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1029
1030         if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1031                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1032                     "cannot destroy '%s'"), zhp->zpool_name);
1033
1034                 if (errno == EROFS) {
1035                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1036                             "one or more devices is read only"));
1037                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1038                 } else {
1039                         (void) zpool_standard_error(hdl, errno, msg);
1040                 }
1041
1042                 if (zfp)
1043                         zfs_close(zfp);
1044                 return (-1);
1045         }
1046
1047         if (zfp) {
1048                 remove_mountpoint(zfp);
1049                 zfs_close(zfp);
1050         }
1051
1052         return (0);
1053 }
1054
1055 /*
1056  * Add the given vdevs to the pool.  The caller must have already performed the
1057  * necessary verification to ensure that the vdev specification is well-formed.
1058  */
1059 int
1060 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1061 {
1062         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1063         int ret;
1064         libzfs_handle_t *hdl = zhp->zpool_hdl;
1065         char msg[1024];
1066         nvlist_t **spares, **l2cache;
1067         uint_t nspares, nl2cache;
1068
1069         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1070             "cannot add to '%s'"), zhp->zpool_name);
1071
1072         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1073             SPA_VERSION_SPARES &&
1074             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1075             &spares, &nspares) == 0) {
1076                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1077                     "upgraded to add hot spares"));
1078                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1079         }
1080
1081         if (pool_is_bootable(zhp) && nvlist_lookup_nvlist_array(nvroot,
1082             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) {
1083                 uint64_t s;
1084
1085                 for (s = 0; s < nspares; s++) {
1086                         char *path;
1087
1088                         if (nvlist_lookup_string(spares[s], ZPOOL_CONFIG_PATH,
1089                             &path) == 0 && pool_uses_efi(spares[s])) {
1090                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1091                                     "device '%s' contains an EFI label and "
1092                                     "cannot be used on root pools."),
1093                                     zpool_vdev_name(hdl, NULL, spares[s],
1094                                     B_FALSE));
1095                                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
1096                         }
1097                 }
1098         }
1099
1100         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1101             SPA_VERSION_L2CACHE &&
1102             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1103             &l2cache, &nl2cache) == 0) {
1104                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1105                     "upgraded to add cache devices"));
1106                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1107         }
1108
1109         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1110                 return (-1);
1111         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1112
1113         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1114                 switch (errno) {
1115                 case EBUSY:
1116                         /*
1117                          * This can happen if the user has specified the same
1118                          * device multiple times.  We can't reliably detect this
1119                          * until we try to add it and see we already have a
1120                          * label.
1121                          */
1122                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1123                             "one or more vdevs refer to the same device"));
1124                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1125                         break;
1126
1127                 case EOVERFLOW:
1128                         /*
1129                          * This occurrs when one of the devices is below
1130                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1131                          * device was the problem device since there's no
1132                          * reliable way to determine device size from userland.
1133                          */
1134                         {
1135                                 char buf[64];
1136
1137                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1138
1139                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1140                                     "device is less than the minimum "
1141                                     "size (%s)"), buf);
1142                         }
1143                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1144                         break;
1145
1146                 case ENOTSUP:
1147                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1148                             "pool must be upgraded to add these vdevs"));
1149                         (void) zfs_error(hdl, EZFS_BADVERSION, msg);
1150                         break;
1151
1152                 case EDOM:
1153                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1154                             "root pool can not have multiple vdevs"
1155                             " or separate logs"));
1156                         (void) zfs_error(hdl, EZFS_POOL_NOTSUP, msg);
1157                         break;
1158
1159                 case ENOTBLK:
1160                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1161                             "cache device must be a disk or disk slice"));
1162                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1163                         break;
1164
1165                 default:
1166                         (void) zpool_standard_error(hdl, errno, msg);
1167                 }
1168
1169                 ret = -1;
1170         } else {
1171                 ret = 0;
1172         }
1173
1174         zcmd_free_nvlists(&zc);
1175
1176         return (ret);
1177 }
1178
1179 /*
1180  * Exports the pool from the system.  The caller must ensure that there are no
1181  * mounted datasets in the pool.
1182  */
1183 int
1184 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce)
1185 {
1186         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1187         char msg[1024];
1188
1189         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1190             "cannot export '%s'"), zhp->zpool_name);
1191
1192         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1193         zc.zc_cookie = force;
1194         zc.zc_guid = hardforce;
1195
1196         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1197                 switch (errno) {
1198                 case EXDEV:
1199                         zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1200                             "use '-f' to override the following errors:\n"
1201                             "'%s' has an active shared spare which could be"
1202                             " used by other pools once '%s' is exported."),
1203                             zhp->zpool_name, zhp->zpool_name);
1204                         return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1205                             msg));
1206                 default:
1207                         return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1208                             msg));
1209                 }
1210         }
1211
1212         return (0);
1213 }
1214
1215 int
1216 zpool_export(zpool_handle_t *zhp, boolean_t force)
1217 {
1218         return (zpool_export_common(zhp, force, B_FALSE));
1219 }
1220
1221 int
1222 zpool_export_force(zpool_handle_t *zhp)
1223 {
1224         return (zpool_export_common(zhp, B_TRUE, B_TRUE));
1225 }
1226
1227 static void
1228 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1229     nvlist_t *config)
1230 {
1231         nvlist_t *nv = NULL;
1232         uint64_t rewindto;
1233         int64_t loss = -1;
1234         struct tm t;
1235         char timestr[128];
1236
1237         if (!hdl->libzfs_printerr || config == NULL)
1238                 return;
1239
1240         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0)
1241                 return;
1242
1243         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1244                 return;
1245         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1246
1247         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1248             strftime(timestr, 128, "%c", &t) != 0) {
1249                 if (dryrun) {
1250                         (void) printf(dgettext(TEXT_DOMAIN,
1251                             "Would be able to return %s "
1252                             "to its state as of %s.\n"),
1253                             name, timestr);
1254                 } else {
1255                         (void) printf(dgettext(TEXT_DOMAIN,
1256                             "Pool %s returned to its state as of %s.\n"),
1257                             name, timestr);
1258                 }
1259                 if (loss > 120) {
1260                         (void) printf(dgettext(TEXT_DOMAIN,
1261                             "%s approximately %lld "),
1262                             dryrun ? "Would discard" : "Discarded",
1263                             ((longlong_t)loss + 30) / 60);
1264                         (void) printf(dgettext(TEXT_DOMAIN,
1265                             "minutes of transactions.\n"));
1266                 } else if (loss > 0) {
1267                         (void) printf(dgettext(TEXT_DOMAIN,
1268                             "%s approximately %lld "),
1269                             dryrun ? "Would discard" : "Discarded",
1270                             (longlong_t)loss);
1271                         (void) printf(dgettext(TEXT_DOMAIN,
1272                             "seconds of transactions.\n"));
1273                 }
1274         }
1275 }
1276
1277 void
1278 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1279     nvlist_t *config)
1280 {
1281         nvlist_t *nv = NULL;
1282         int64_t loss = -1;
1283         uint64_t edata = UINT64_MAX;
1284         uint64_t rewindto;
1285         struct tm t;
1286         char timestr[128];
1287
1288         if (!hdl->libzfs_printerr)
1289                 return;
1290
1291         if (reason >= 0)
1292                 (void) printf(dgettext(TEXT_DOMAIN, "action: "));
1293         else
1294                 (void) printf(dgettext(TEXT_DOMAIN, "\t"));
1295
1296         /* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1297         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1298             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1299                 goto no_info;
1300
1301         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1302         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1303             &edata);
1304
1305         (void) printf(dgettext(TEXT_DOMAIN,
1306             "Recovery is possible, but will result in some data loss.\n"));
1307
1308         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1309             strftime(timestr, 128, "%c", &t) != 0) {
1310                 (void) printf(dgettext(TEXT_DOMAIN,
1311                     "\tReturning the pool to its state as of %s\n"
1312                     "\tshould correct the problem.  "),
1313                     timestr);
1314         } else {
1315                 (void) printf(dgettext(TEXT_DOMAIN,
1316                     "\tReverting the pool to an earlier state "
1317                     "should correct the problem.\n\t"));
1318         }
1319
1320         if (loss > 120) {
1321                 (void) printf(dgettext(TEXT_DOMAIN,
1322                     "Approximately %lld minutes of data\n"
1323                     "\tmust be discarded, irreversibly.  "),
1324                     ((longlong_t)loss + 30) / 60);
1325         } else if (loss > 0) {
1326                 (void) printf(dgettext(TEXT_DOMAIN,
1327                     "Approximately %lld seconds of data\n"
1328                     "\tmust be discarded, irreversibly.  "),
1329                     (longlong_t)loss);
1330         }
1331         if (edata != 0 && edata != UINT64_MAX) {
1332                 if (edata == 1) {
1333                         (void) printf(dgettext(TEXT_DOMAIN,
1334                             "After rewind, at least\n"
1335                             "\tone persistent user-data error will remain.  "));
1336                 } else {
1337                         (void) printf(dgettext(TEXT_DOMAIN,
1338                             "After rewind, several\n"
1339                             "\tpersistent user-data errors will remain.  "));
1340                 }
1341         }
1342         (void) printf(dgettext(TEXT_DOMAIN,
1343             "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1344             reason >= 0 ? "clear" : "import", name);
1345
1346         (void) printf(dgettext(TEXT_DOMAIN,
1347             "A scrub of the pool\n"
1348             "\tis strongly recommended after recovery.\n"));
1349         return;
1350
1351 no_info:
1352         (void) printf(dgettext(TEXT_DOMAIN,
1353             "Destroy and re-create the pool from\n\ta backup source.\n"));
1354 }
1355
1356 /*
1357  * zpool_import() is a contracted interface. Should be kept the same
1358  * if possible.
1359  *
1360  * Applications should use zpool_import_props() to import a pool with
1361  * new properties value to be set.
1362  */
1363 int
1364 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1365     char *altroot)
1366 {
1367         nvlist_t *props = NULL;
1368         int ret;
1369
1370         if (altroot != NULL) {
1371                 if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1372                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1373                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1374                             newname));
1375                 }
1376
1377                 if (nvlist_add_string(props,
1378                     zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1379                     nvlist_add_string(props,
1380                     zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1381                         nvlist_free(props);
1382                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1383                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1384                             newname));
1385                 }
1386         }
1387
1388         ret = zpool_import_props(hdl, config, newname, props,
1389             ZFS_IMPORT_NORMAL);
1390         if (props)
1391                 nvlist_free(props);
1392         return (ret);
1393 }
1394
1395 static void
1396 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1397     int indent)
1398 {
1399         nvlist_t **child;
1400         uint_t c, children;
1401         char *vname;
1402         uint64_t is_log = 0;
1403
1404         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1405             &is_log);
1406
1407         if (name != NULL)
1408                 (void) printf("\t%*s%s%s\n", indent, "", name,
1409                     is_log ? " [log]" : "");
1410
1411         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1412             &child, &children) != 0)
1413                 return;
1414
1415         for (c = 0; c < children; c++) {
1416                 vname = zpool_vdev_name(hdl, NULL, child[c], B_TRUE);
1417                 print_vdev_tree(hdl, vname, child[c], indent + 2);
1418                 free(vname);
1419         }
1420 }
1421
1422 /*
1423  * Import the given pool using the known configuration and a list of
1424  * properties to be set. The configuration should have come from
1425  * zpool_find_import(). The 'newname' parameters control whether the pool
1426  * is imported with a different name.
1427  */
1428 int
1429 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1430     nvlist_t *props, int flags)
1431 {
1432         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1433         zpool_rewind_policy_t policy;
1434         nvlist_t *nv = NULL;
1435         nvlist_t *nvinfo = NULL;
1436         nvlist_t *missing = NULL;
1437         char *thename;
1438         char *origname;
1439         int ret;
1440         int error = 0;
1441         char errbuf[1024];
1442
1443         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1444             &origname) == 0);
1445
1446         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1447             "cannot import pool '%s'"), origname);
1448
1449         if (newname != NULL) {
1450                 if (!zpool_name_valid(hdl, B_FALSE, newname))
1451                         return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1452                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1453                             newname));
1454                 thename = (char *)newname;
1455         } else {
1456                 thename = origname;
1457         }
1458
1459         if (props) {
1460                 uint64_t version;
1461                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1462
1463                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1464                     &version) == 0);
1465
1466                 if ((props = zpool_valid_proplist(hdl, origname,
1467                     props, version, flags, errbuf)) == NULL) {
1468                         return (-1);
1469                 } else if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1470                         nvlist_free(props);
1471                         return (-1);
1472                 }
1473         }
1474
1475         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1476
1477         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1478             &zc.zc_guid) == 0);
1479
1480         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1481                 nvlist_free(props);
1482                 return (-1);
1483         }
1484         if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1485                 nvlist_free(props);
1486                 return (-1);
1487         }
1488
1489         zc.zc_cookie = flags;
1490         while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1491             errno == ENOMEM) {
1492                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1493                         zcmd_free_nvlists(&zc);
1494                         return (-1);
1495                 }
1496         }
1497         if (ret != 0)
1498                 error = errno;
1499
1500         (void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1501         zpool_get_rewind_policy(config, &policy);
1502
1503         if (error) {
1504                 char desc[1024];
1505
1506                 /*
1507                  * Dry-run failed, but we print out what success
1508                  * looks like if we found a best txg
1509                  */
1510                 if (policy.zrp_request & ZPOOL_TRY_REWIND) {
1511                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1512                             B_TRUE, nv);
1513                         nvlist_free(nv);
1514                         return (-1);
1515                 }
1516
1517                 if (newname == NULL)
1518                         (void) snprintf(desc, sizeof (desc),
1519                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1520                             thename);
1521                 else
1522                         (void) snprintf(desc, sizeof (desc),
1523                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1524                             origname, thename);
1525
1526                 switch (error) {
1527                 case ENOTSUP:
1528                         /*
1529                          * Unsupported version.
1530                          */
1531                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1532                         break;
1533
1534                 case EINVAL:
1535                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1536                         break;
1537
1538                 case EROFS:
1539                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1540                             "one or more devices is read only"));
1541                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1542                         break;
1543
1544                 case ENXIO:
1545                         if (nv && nvlist_lookup_nvlist(nv,
1546                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1547                             nvlist_lookup_nvlist(nvinfo,
1548                             ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1549                                 (void) printf(dgettext(TEXT_DOMAIN,
1550                                     "The devices below are missing, use "
1551                                     "'-m' to import the pool anyway:\n"));
1552                                 print_vdev_tree(hdl, NULL, missing, 2);
1553                                 (void) printf("\n");
1554                         }
1555                         (void) zpool_standard_error(hdl, error, desc);
1556                         break;
1557
1558                 case EEXIST:
1559                         (void) zpool_standard_error(hdl, error, desc);
1560                         break;
1561
1562                 default:
1563                         (void) zpool_standard_error(hdl, error, desc);
1564                         zpool_explain_recover(hdl,
1565                             newname ? origname : thename, -error, nv);
1566                         break;
1567                 }
1568
1569                 nvlist_free(nv);
1570                 ret = -1;
1571         } else {
1572                 zpool_handle_t *zhp;
1573
1574                 /*
1575                  * This should never fail, but play it safe anyway.
1576                  */
1577                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1578                         ret = -1;
1579                 else if (zhp != NULL)
1580                         zpool_close(zhp);
1581                 if (policy.zrp_request &
1582                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1583                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1584                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1585                 }
1586                 nvlist_free(nv);
1587                 return (0);
1588         }
1589
1590         zcmd_free_nvlists(&zc);
1591         nvlist_free(props);
1592
1593         return (ret);
1594 }
1595
1596 /*
1597  * Scan the pool.
1598  */
1599 int
1600 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1601 {
1602         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1603         char msg[1024];
1604         libzfs_handle_t *hdl = zhp->zpool_hdl;
1605
1606         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1607         zc.zc_cookie = func;
1608
1609         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1610             (errno == ENOENT && func != POOL_SCAN_NONE))
1611                 return (0);
1612
1613         if (func == POOL_SCAN_SCRUB) {
1614                 (void) snprintf(msg, sizeof (msg),
1615                     dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1616         } else if (func == POOL_SCAN_NONE) {
1617                 (void) snprintf(msg, sizeof (msg),
1618                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1619                     zc.zc_name);
1620         } else {
1621                 assert(!"unexpected result");
1622         }
1623
1624         if (errno == EBUSY) {
1625                 nvlist_t *nvroot;
1626                 pool_scan_stat_t *ps = NULL;
1627                 uint_t psc;
1628
1629                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
1630                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1631                 (void) nvlist_lookup_uint64_array(nvroot,
1632                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1633                 if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1634                         return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1635                 else
1636                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
1637         } else if (errno == ENOENT) {
1638                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1639         } else {
1640                 return (zpool_standard_error(hdl, errno, msg));
1641         }
1642 }
1643
1644 /*
1645  * This provides a very minimal check whether a given string is likely a
1646  * c#t#d# style string.  Users of this are expected to do their own
1647  * verification of the s# part.
1648  */
1649 #define CTD_CHECK(str)  (str && str[0] == 'c' && isdigit(str[1]))
1650
1651 /*
1652  * More elaborate version for ones which may start with "/dev/dsk/"
1653  * and the like.
1654  */
1655 static int
1656 ctd_check_path(char *str) {
1657         /*
1658          * If it starts with a slash, check the last component.
1659          */
1660         if (str && str[0] == '/') {
1661                 char *tmp = strrchr(str, '/');
1662
1663                 /*
1664                  * If it ends in "/old", check the second-to-last
1665                  * component of the string instead.
1666                  */
1667                 if (tmp != str && strcmp(tmp, "/old") == 0) {
1668                         for (tmp--; *tmp != '/'; tmp--)
1669                                 ;
1670                 }
1671                 str = tmp + 1;
1672         }
1673         return (CTD_CHECK(str));
1674 }
1675
1676 /*
1677  * Find a vdev that matches the search criteria specified. We use the
1678  * the nvpair name to determine how we should look for the device.
1679  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1680  * spare; but FALSE if its an INUSE spare.
1681  */
1682 static nvlist_t *
1683 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1684     boolean_t *l2cache, boolean_t *log)
1685 {
1686         uint_t c, children;
1687         nvlist_t **child;
1688         nvlist_t *ret;
1689         uint64_t is_log;
1690         char *srchkey;
1691         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
1692
1693         /* Nothing to look for */
1694         if (search == NULL || pair == NULL)
1695                 return (NULL);
1696
1697         /* Obtain the key we will use to search */
1698         srchkey = nvpair_name(pair);
1699
1700         switch (nvpair_type(pair)) {
1701         case DATA_TYPE_UINT64:
1702                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
1703                         uint64_t srchval, theguid;
1704
1705                         verify(nvpair_value_uint64(pair, &srchval) == 0);
1706                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
1707                             &theguid) == 0);
1708                         if (theguid == srchval)
1709                                 return (nv);
1710                 }
1711                 break;
1712
1713         case DATA_TYPE_STRING: {
1714                 char *srchval, *val;
1715
1716                 verify(nvpair_value_string(pair, &srchval) == 0);
1717                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
1718                         break;
1719
1720                 /*
1721                  * Search for the requested value. Special cases:
1722                  *
1723                  * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
1724                  *   "s0" or "s0/old".  The "s0" part is hidden from the user,
1725                  *   but included in the string, so this matches around it.
1726                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
1727                  *
1728                  * Otherwise, all other searches are simple string compares.
1729                  */
1730                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0 &&
1731                     ctd_check_path(val)) {
1732                         uint64_t wholedisk = 0;
1733
1734                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1735                             &wholedisk);
1736                         if (wholedisk) {
1737                                 int slen = strlen(srchval);
1738                                 int vlen = strlen(val);
1739
1740                                 if (slen != vlen - 2)
1741                                         break;
1742
1743                                 /*
1744                                  * make_leaf_vdev() should only set
1745                                  * wholedisk for ZPOOL_CONFIG_PATHs which
1746                                  * will include "/dev/dsk/", giving plenty of
1747                                  * room for the indices used next.
1748                                  */
1749                                 ASSERT(vlen >= 6);
1750
1751                                 /*
1752                                  * strings identical except trailing "s0"
1753                                  */
1754                                 if (strcmp(&val[vlen - 2], "s0") == 0 &&
1755                                     strncmp(srchval, val, slen) == 0)
1756                                         return (nv);
1757
1758                                 /*
1759                                  * strings identical except trailing "s0/old"
1760                                  */
1761                                 if (strcmp(&val[vlen - 6], "s0/old") == 0 &&
1762                                     strcmp(&srchval[slen - 4], "/old") == 0 &&
1763                                     strncmp(srchval, val, slen - 4) == 0)
1764                                         return (nv);
1765
1766                                 break;
1767                         }
1768                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
1769                         char *type, *idx, *end, *p;
1770                         uint64_t id, vdev_id;
1771
1772                         /*
1773                          * Determine our vdev type, keeping in mind
1774                          * that the srchval is composed of a type and
1775                          * vdev id pair (i.e. mirror-4).
1776                          */
1777                         if ((type = strdup(srchval)) == NULL)
1778                                 return (NULL);
1779
1780                         if ((p = strrchr(type, '-')) == NULL) {
1781                                 free(type);
1782                                 break;
1783                         }
1784                         idx = p + 1;
1785                         *p = '\0';
1786
1787                         /*
1788                          * If the types don't match then keep looking.
1789                          */
1790                         if (strncmp(val, type, strlen(val)) != 0) {
1791                                 free(type);
1792                                 break;
1793                         }
1794
1795                         verify(strncmp(type, VDEV_TYPE_RAIDZ,
1796                             strlen(VDEV_TYPE_RAIDZ)) == 0 ||
1797                             strncmp(type, VDEV_TYPE_MIRROR,
1798                             strlen(VDEV_TYPE_MIRROR)) == 0);
1799                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
1800                             &id) == 0);
1801
1802                         errno = 0;
1803                         vdev_id = strtoull(idx, &end, 10);
1804
1805                         free(type);
1806                         if (errno != 0)
1807                                 return (NULL);
1808
1809                         /*
1810                          * Now verify that we have the correct vdev id.
1811                          */
1812                         if (vdev_id == id)
1813                                 return (nv);
1814                 }
1815
1816                 /*
1817                  * Common case
1818                  */
1819                 if (strcmp(srchval, val) == 0)
1820                         return (nv);
1821                 break;
1822         }
1823
1824         default:
1825                 break;
1826         }
1827
1828         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1829             &child, &children) != 0)
1830                 return (NULL);
1831
1832         for (c = 0; c < children; c++) {
1833                 if ((ret = vdev_to_nvlist_iter(child[c], search,
1834                     avail_spare, l2cache, NULL)) != NULL) {
1835                         /*
1836                          * The 'is_log' value is only set for the toplevel
1837                          * vdev, not the leaf vdevs.  So we always lookup the
1838                          * log device from the root of the vdev tree (where
1839                          * 'log' is non-NULL).
1840                          */
1841                         if (log != NULL &&
1842                             nvlist_lookup_uint64(child[c],
1843                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
1844                             is_log) {
1845                                 *log = B_TRUE;
1846                         }
1847                         return (ret);
1848                 }
1849         }
1850
1851         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1852             &child, &children) == 0) {
1853                 for (c = 0; c < children; c++) {
1854                         if ((ret = vdev_to_nvlist_iter(child[c], search,
1855                             avail_spare, l2cache, NULL)) != NULL) {
1856                                 *avail_spare = B_TRUE;
1857                                 return (ret);
1858                         }
1859                 }
1860         }
1861
1862         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1863             &child, &children) == 0) {
1864                 for (c = 0; c < children; c++) {
1865                         if ((ret = vdev_to_nvlist_iter(child[c], search,
1866                             avail_spare, l2cache, NULL)) != NULL) {
1867                                 *l2cache = B_TRUE;
1868                                 return (ret);
1869                         }
1870                 }
1871         }
1872
1873         return (NULL);
1874 }
1875
1876 /*
1877  * Given a physical path (minus the "/devices" prefix), find the
1878  * associated vdev.
1879  */
1880 nvlist_t *
1881 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
1882     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
1883 {
1884         nvlist_t *search, *nvroot, *ret;
1885
1886         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1887         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
1888
1889         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1890             &nvroot) == 0);
1891
1892         *avail_spare = B_FALSE;
1893         *l2cache = B_FALSE;
1894         if (log != NULL)
1895                 *log = B_FALSE;
1896         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
1897         nvlist_free(search);
1898
1899         return (ret);
1900 }
1901
1902 /*
1903  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
1904  */
1905 boolean_t
1906 zpool_vdev_is_interior(const char *name)
1907 {
1908         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
1909             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
1910                 return (B_TRUE);
1911         return (B_FALSE);
1912 }
1913
1914 nvlist_t *
1915 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
1916     boolean_t *l2cache, boolean_t *log)
1917 {
1918         char buf[MAXPATHLEN];
1919         char *end;
1920         nvlist_t *nvroot, *search, *ret;
1921         uint64_t guid;
1922
1923         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1924
1925         guid = strtoull(path, &end, 10);
1926         if (guid != 0 && *end == '\0') {
1927                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
1928         } else if (zpool_vdev_is_interior(path)) {
1929                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
1930         } else if (path[0] != '/') {
1931                 (void) snprintf(buf, sizeof (buf), "%s%s", "/dev/dsk/", path);
1932                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, buf) == 0);
1933         } else {
1934                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
1935         }
1936
1937         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1938             &nvroot) == 0);
1939
1940         *avail_spare = B_FALSE;
1941         *l2cache = B_FALSE;
1942         if (log != NULL)
1943                 *log = B_FALSE;
1944         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
1945         nvlist_free(search);
1946
1947         return (ret);
1948 }
1949
1950 static int
1951 vdev_online(nvlist_t *nv)
1952 {
1953         uint64_t ival;
1954
1955         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
1956             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
1957             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
1958                 return (0);
1959
1960         return (1);
1961 }
1962
1963 /*
1964  * Helper function for zpool_get_physpaths().
1965  */
1966 static int
1967 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
1968     size_t *bytes_written)
1969 {
1970         size_t bytes_left, pos, rsz;
1971         char *tmppath;
1972         const char *format;
1973
1974         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
1975             &tmppath) != 0)
1976                 return (EZFS_NODEVICE);
1977
1978         pos = *bytes_written;
1979         bytes_left = physpath_size - pos;
1980         format = (pos == 0) ? "%s" : " %s";
1981
1982         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
1983         *bytes_written += rsz;
1984
1985         if (rsz >= bytes_left) {
1986                 /* if physpath was not copied properly, clear it */
1987                 if (bytes_left != 0) {
1988                         physpath[pos] = 0;
1989                 }
1990                 return (EZFS_NOSPC);
1991         }
1992         return (0);
1993 }
1994
1995 static int
1996 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
1997     size_t *rsz, boolean_t is_spare)
1998 {
1999         char *type;
2000         int ret;
2001
2002         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2003                 return (EZFS_INVALCONFIG);
2004
2005         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2006                 /*
2007                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2008                  * For a spare vdev, we only want to boot from the active
2009                  * spare device.
2010                  */
2011                 if (is_spare) {
2012                         uint64_t spare = 0;
2013                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2014                             &spare);
2015                         if (!spare)
2016                                 return (EZFS_INVALCONFIG);
2017                 }
2018
2019                 if (vdev_online(nv)) {
2020                         if ((ret = vdev_get_one_physpath(nv, physpath,
2021                             phypath_size, rsz)) != 0)
2022                                 return (ret);
2023                 }
2024         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2025             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2026             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2027                 nvlist_t **child;
2028                 uint_t count;
2029                 int i, ret;
2030
2031                 if (nvlist_lookup_nvlist_array(nv,
2032                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2033                         return (EZFS_INVALCONFIG);
2034
2035                 for (i = 0; i < count; i++) {
2036                         ret = vdev_get_physpaths(child[i], physpath,
2037                             phypath_size, rsz, is_spare);
2038                         if (ret == EZFS_NOSPC)
2039                                 return (ret);
2040                 }
2041         }
2042
2043         return (EZFS_POOL_INVALARG);
2044 }
2045
2046 /*
2047  * Get phys_path for a root pool config.
2048  * Return 0 on success; non-zero on failure.
2049  */
2050 static int
2051 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2052 {
2053         size_t rsz;
2054         nvlist_t *vdev_root;
2055         nvlist_t **child;
2056         uint_t count;
2057         char *type;
2058
2059         rsz = 0;
2060
2061         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2062             &vdev_root) != 0)
2063                 return (EZFS_INVALCONFIG);
2064
2065         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2066             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2067             &child, &count) != 0)
2068                 return (EZFS_INVALCONFIG);
2069
2070         /*
2071          * root pool can not have EFI labeled disks and can only have
2072          * a single top-level vdev.
2073          */
2074         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1 ||
2075             pool_uses_efi(vdev_root))
2076                 return (EZFS_POOL_INVALARG);
2077
2078         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2079             B_FALSE);
2080
2081         /* No online devices */
2082         if (rsz == 0)
2083                 return (EZFS_NODEVICE);
2084
2085         return (0);
2086 }
2087
2088 /*
2089  * Get phys_path for a root pool
2090  * Return 0 on success; non-zero on failure.
2091  */
2092 int
2093 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2094 {
2095         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2096             phypath_size));
2097 }
2098
2099 /*
2100  * If the device has being dynamically expanded then we need to relabel
2101  * the disk to use the new unallocated space.
2102  */
2103 static int
2104 zpool_relabel_disk(libzfs_handle_t *hdl, const char *name)
2105 {
2106         char path[MAXPATHLEN];
2107         char errbuf[1024];
2108         int fd, error;
2109         int (*_efi_use_whole_disk)(int);
2110
2111         if ((_efi_use_whole_disk = (int (*)(int))dlsym(RTLD_DEFAULT,
2112             "efi_use_whole_disk")) == NULL)
2113                 return (-1);
2114
2115         (void) snprintf(path, sizeof (path), "%s/%s", RDISK_ROOT, name);
2116
2117         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
2118                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2119                     "relabel '%s': unable to open device"), name);
2120                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
2121         }
2122
2123         /*
2124          * It's possible that we might encounter an error if the device
2125          * does not have any unallocated space left. If so, we simply
2126          * ignore that error and continue on.
2127          */
2128         error = _efi_use_whole_disk(fd);
2129         (void) close(fd);
2130         if (error && error != VT_ENOSPC) {
2131                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2132                     "relabel '%s': unable to read disk capacity"), name);
2133                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
2134         }
2135         return (0);
2136 }
2137
2138 /*
2139  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2140  * ZFS_ONLINE_* flags.
2141  */
2142 int
2143 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2144     vdev_state_t *newstate)
2145 {
2146         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2147         char msg[1024];
2148         nvlist_t *tgt;
2149         boolean_t avail_spare, l2cache, islog;
2150         libzfs_handle_t *hdl = zhp->zpool_hdl;
2151
2152         if (flags & ZFS_ONLINE_EXPAND) {
2153                 (void) snprintf(msg, sizeof (msg),
2154                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2155         } else {
2156                 (void) snprintf(msg, sizeof (msg),
2157                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2158         }
2159
2160         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2161         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2162             &islog)) == NULL)
2163                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2164
2165         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2166
2167         if (avail_spare)
2168                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2169
2170         if (flags & ZFS_ONLINE_EXPAND ||
2171             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2172                 char *pathname = NULL;
2173                 uint64_t wholedisk = 0;
2174
2175                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2176                     &wholedisk);
2177                 verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
2178                     &pathname) == 0);
2179
2180                 /*
2181                  * XXX - L2ARC 1.0 devices can't support expansion.
2182                  */
2183                 if (l2cache) {
2184                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2185                             "cannot expand cache devices"));
2186                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2187                 }
2188
2189                 if (wholedisk) {
2190                         pathname += strlen(DISK_ROOT) + 1;
2191                         (void) zpool_relabel_disk(hdl, pathname);
2192                 }
2193         }
2194
2195         zc.zc_cookie = VDEV_STATE_ONLINE;
2196         zc.zc_obj = flags;
2197
2198         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2199                 if (errno == EINVAL) {
2200                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2201                             "from this pool into a new one.  Use '%s' "
2202                             "instead"), "zpool detach");
2203                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2204                 }
2205                 return (zpool_standard_error(hdl, errno, msg));
2206         }
2207
2208         *newstate = zc.zc_cookie;
2209         return (0);
2210 }
2211
2212 /*
2213  * Take the specified vdev offline
2214  */
2215 int
2216 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2217 {
2218         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2219         char msg[1024];
2220         nvlist_t *tgt;
2221         boolean_t avail_spare, l2cache;
2222         libzfs_handle_t *hdl = zhp->zpool_hdl;
2223
2224         (void) snprintf(msg, sizeof (msg),
2225             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2226
2227         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2228         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2229             NULL)) == NULL)
2230                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2231
2232         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2233
2234         if (avail_spare)
2235                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2236
2237         zc.zc_cookie = VDEV_STATE_OFFLINE;
2238         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2239
2240         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2241                 return (0);
2242
2243         switch (errno) {
2244         case EBUSY:
2245
2246                 /*
2247                  * There are no other replicas of this device.
2248                  */
2249                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2250
2251         case EEXIST:
2252                 /*
2253                  * The log device has unplayed logs
2254                  */
2255                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2256
2257         default:
2258                 return (zpool_standard_error(hdl, errno, msg));
2259         }
2260 }
2261
2262 /*
2263  * Mark the given vdev faulted.
2264  */
2265 int
2266 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2267 {
2268         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2269         char msg[1024];
2270         libzfs_handle_t *hdl = zhp->zpool_hdl;
2271
2272         (void) snprintf(msg, sizeof (msg),
2273            dgettext(TEXT_DOMAIN, "cannot fault %llu"), (u_longlong_t)guid);
2274
2275         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2276         zc.zc_guid = guid;
2277         zc.zc_cookie = VDEV_STATE_FAULTED;
2278         zc.zc_obj = aux;
2279
2280         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2281                 return (0);
2282
2283         switch (errno) {
2284         case EBUSY:
2285
2286                 /*
2287                  * There are no other replicas of this device.
2288                  */
2289                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2290
2291         default:
2292                 return (zpool_standard_error(hdl, errno, msg));
2293         }
2294
2295 }
2296
2297 /*
2298  * Mark the given vdev degraded.
2299  */
2300 int
2301 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2302 {
2303         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2304         char msg[1024];
2305         libzfs_handle_t *hdl = zhp->zpool_hdl;
2306
2307         (void) snprintf(msg, sizeof (msg),
2308            dgettext(TEXT_DOMAIN, "cannot degrade %llu"), (u_longlong_t)guid);
2309
2310         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2311         zc.zc_guid = guid;
2312         zc.zc_cookie = VDEV_STATE_DEGRADED;
2313         zc.zc_obj = aux;
2314
2315         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2316                 return (0);
2317
2318         return (zpool_standard_error(hdl, errno, msg));
2319 }
2320
2321 /*
2322  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2323  * a hot spare.
2324  */
2325 static boolean_t
2326 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2327 {
2328         nvlist_t **child;
2329         uint_t c, children;
2330         char *type;
2331
2332         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2333             &children) == 0) {
2334                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2335                     &type) == 0);
2336
2337                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2338                     children == 2 && child[which] == tgt)
2339                         return (B_TRUE);
2340
2341                 for (c = 0; c < children; c++)
2342                         if (is_replacing_spare(child[c], tgt, which))
2343                                 return (B_TRUE);
2344         }
2345
2346         return (B_FALSE);
2347 }
2348
2349 /*
2350  * Attach new_disk (fully described by nvroot) to old_disk.
2351  * If 'replacing' is specified, the new disk will replace the old one.
2352  */
2353 int
2354 zpool_vdev_attach(zpool_handle_t *zhp,
2355     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2356 {
2357         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2358         char msg[1024];
2359         int ret;
2360         nvlist_t *tgt;
2361         boolean_t avail_spare, l2cache, islog;
2362         uint64_t val;
2363         char *newname;
2364         nvlist_t **child;
2365         uint_t children;
2366         nvlist_t *config_root;
2367         libzfs_handle_t *hdl = zhp->zpool_hdl;
2368         boolean_t rootpool = pool_is_bootable(zhp);
2369
2370         if (replacing)
2371                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2372                     "cannot replace %s with %s"), old_disk, new_disk);
2373         else
2374                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2375                     "cannot attach %s to %s"), new_disk, old_disk);
2376
2377         /*
2378          * If this is a root pool, make sure that we're not attaching an
2379          * EFI labeled device.
2380          */
2381         if (rootpool && pool_uses_efi(nvroot)) {
2382                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2383                     "EFI labeled devices are not supported on root pools."));
2384                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
2385         }
2386
2387         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2388         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2389             &islog)) == 0)
2390                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2391
2392         if (avail_spare)
2393                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2394
2395         if (l2cache)
2396                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2397
2398         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2399         zc.zc_cookie = replacing;
2400
2401         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2402             &child, &children) != 0 || children != 1) {
2403                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2404                     "new device must be a single disk"));
2405                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2406         }
2407
2408         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2409             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2410
2411         if ((newname = zpool_vdev_name(NULL, NULL, child[0], B_FALSE)) == NULL)
2412                 return (-1);
2413
2414         /*
2415          * If the target is a hot spare that has been swapped in, we can only
2416          * replace it with another hot spare.
2417          */
2418         if (replacing &&
2419             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2420             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2421             NULL) == NULL || !avail_spare) &&
2422             is_replacing_spare(config_root, tgt, 1)) {
2423                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2424                     "can only be replaced by another hot spare"));
2425                 free(newname);
2426                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2427         }
2428
2429         free(newname);
2430
2431         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2432                 return (-1);
2433
2434         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2435
2436         zcmd_free_nvlists(&zc);
2437
2438         if (ret == 0) {
2439                 if (rootpool) {
2440                         /*
2441                          * XXX need a better way to prevent user from
2442                          * booting up a half-baked vdev.
2443                          */
2444                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2445                             "sure to wait until resilver is done "
2446                             "before rebooting.\n"));
2447                 }
2448                 return (0);
2449         }
2450
2451         switch (errno) {
2452         case ENOTSUP:
2453                 /*
2454                  * Can't attach to or replace this type of vdev.
2455                  */
2456                 if (replacing) {
2457                         uint64_t version = zpool_get_prop_int(zhp,
2458                             ZPOOL_PROP_VERSION, NULL);
2459
2460                         if (islog)
2461                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2462                                     "cannot replace a log with a spare"));
2463                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2464                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2465                                     "already in replacing/spare config; wait "
2466                                     "for completion or use 'zpool detach'"));
2467                         else
2468                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2469                                     "cannot replace a replacing device"));
2470                 } else {
2471                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2472                             "can only attach to mirrors and top-level "
2473                             "disks"));
2474                 }
2475                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2476                 break;
2477
2478         case EINVAL:
2479                 /*
2480                  * The new device must be a single disk.
2481                  */
2482                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2483                     "new device must be a single disk"));
2484                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2485                 break;
2486
2487         case EBUSY:
2488                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2489                     new_disk);
2490                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2491                 break;
2492
2493         case EOVERFLOW:
2494                 /*
2495                  * The new device is too small.
2496                  */
2497                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2498                     "device is too small"));
2499                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2500                 break;
2501
2502         case EDOM:
2503                 /*
2504                  * The new device has a different alignment requirement.
2505                  */
2506                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2507                     "devices have different sector alignment"));
2508                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2509                 break;
2510
2511         case ENAMETOOLONG:
2512                 /*
2513                  * The resulting top-level vdev spec won't fit in the label.
2514                  */
2515                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2516                 break;
2517
2518         default:
2519                 (void) zpool_standard_error(hdl, errno, msg);
2520         }
2521
2522         return (-1);
2523 }
2524
2525 /*
2526  * Detach the specified device.
2527  */
2528 int
2529 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2530 {
2531         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2532         char msg[1024];
2533         nvlist_t *tgt;
2534         boolean_t avail_spare, l2cache;
2535         libzfs_handle_t *hdl = zhp->zpool_hdl;
2536
2537         (void) snprintf(msg, sizeof (msg),
2538             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2539
2540         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2541         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2542             NULL)) == 0)
2543                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2544
2545         if (avail_spare)
2546                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2547
2548         if (l2cache)
2549                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2550
2551         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2552
2553         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2554                 return (0);
2555
2556         switch (errno) {
2557
2558         case ENOTSUP:
2559                 /*
2560                  * Can't detach from this type of vdev.
2561                  */
2562                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2563                     "applicable to mirror and replacing vdevs"));
2564                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2565                 break;
2566
2567         case EBUSY:
2568                 /*
2569                  * There are no other replicas of this device.
2570                  */
2571                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2572                 break;
2573
2574         default:
2575                 (void) zpool_standard_error(hdl, errno, msg);
2576         }
2577
2578         return (-1);
2579 }
2580
2581 /*
2582  * Find a mirror vdev in the source nvlist.
2583  *
2584  * The mchild array contains a list of disks in one of the top-level mirrors
2585  * of the source pool.  The schild array contains a list of disks that the
2586  * user specified on the command line.  We loop over the mchild array to
2587  * see if any entry in the schild array matches.
2588  *
2589  * If a disk in the mchild array is found in the schild array, we return
2590  * the index of that entry.  Otherwise we return -1.
2591  */
2592 static int
2593 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2594     nvlist_t **schild, uint_t schildren)
2595 {
2596         uint_t mc;
2597
2598         for (mc = 0; mc < mchildren; mc++) {
2599                 uint_t sc;
2600                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2601                     mchild[mc], B_FALSE);
2602
2603                 for (sc = 0; sc < schildren; sc++) {
2604                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2605                             schild[sc], B_FALSE);
2606                         boolean_t result = (strcmp(mpath, spath) == 0);
2607
2608                         free(spath);
2609                         if (result) {
2610                                 free(mpath);
2611                                 return (mc);
2612                         }
2613                 }
2614
2615                 free(mpath);
2616         }
2617
2618         return (-1);
2619 }
2620
2621 /*
2622  * Split a mirror pool.  If newroot points to null, then a new nvlist
2623  * is generated and it is the responsibility of the caller to free it.
2624  */
2625 int
2626 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2627     nvlist_t *props, splitflags_t flags)
2628 {
2629         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2630         char msg[1024];
2631         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2632         nvlist_t **varray = NULL, *zc_props = NULL;
2633         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2634         libzfs_handle_t *hdl = zhp->zpool_hdl;
2635         uint64_t vers;
2636         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2637         int retval = 0;
2638
2639         (void) snprintf(msg, sizeof (msg),
2640             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2641
2642         if (!zpool_name_valid(hdl, B_FALSE, newname))
2643                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2644
2645         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2646                 (void) fprintf(stderr, gettext("Internal error: unable to "
2647                     "retrieve pool configuration\n"));
2648                 return (-1);
2649         }
2650
2651         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2652             == 0);
2653         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2654
2655         if (props) {
2656                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2657                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2658                     props, vers, flags, msg)) == NULL)
2659                         return (-1);
2660         }
2661
2662         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2663             &children) != 0) {
2664                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2665                     "Source pool is missing vdev tree"));
2666                 if (zc_props)
2667                         nvlist_free(zc_props);
2668                 return (-1);
2669         }
2670
2671         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2672         vcount = 0;
2673
2674         if (*newroot == NULL ||
2675             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2676             &newchild, &newchildren) != 0)
2677                 newchildren = 0;
2678
2679         for (c = 0; c < children; c++) {
2680                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2681                 char *type;
2682                 nvlist_t **mchild, *vdev;
2683                 uint_t mchildren;
2684                 int entry;
2685
2686                 /*
2687                  * Unlike cache & spares, slogs are stored in the
2688                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2689                  */
2690                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2691                     &is_log);
2692                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2693                     &is_hole);
2694                 if (is_log || is_hole) {
2695                         /*
2696                          * Create a hole vdev and put it in the config.
2697                          */
2698                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2699                                 goto out;
2700                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2701                             VDEV_TYPE_HOLE) != 0)
2702                                 goto out;
2703                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2704                             1) != 0)
2705                                 goto out;
2706                         if (lastlog == 0)
2707                                 lastlog = vcount;
2708                         varray[vcount++] = vdev;
2709                         continue;
2710                 }
2711                 lastlog = 0;
2712                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
2713                     == 0);
2714                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
2715                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2716                             "Source pool must be composed only of mirrors\n"));
2717                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2718                         goto out;
2719                 }
2720
2721                 verify(nvlist_lookup_nvlist_array(child[c],
2722                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2723
2724                 /* find or add an entry for this top-level vdev */
2725                 if (newchildren > 0 &&
2726                     (entry = find_vdev_entry(zhp, mchild, mchildren,
2727                     newchild, newchildren)) >= 0) {
2728                         /* We found a disk that the user specified. */
2729                         vdev = mchild[entry];
2730                         ++found;
2731                 } else {
2732                         /* User didn't specify a disk for this vdev. */
2733                         vdev = mchild[mchildren - 1];
2734                 }
2735
2736                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
2737                         goto out;
2738         }
2739
2740         /* did we find every disk the user specified? */
2741         if (found != newchildren) {
2742                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
2743                     "include at most one disk from each mirror"));
2744                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2745                 goto out;
2746         }
2747
2748         /* Prepare the nvlist for populating. */
2749         if (*newroot == NULL) {
2750                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
2751                         goto out;
2752                 freelist = B_TRUE;
2753                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
2754                     VDEV_TYPE_ROOT) != 0)
2755                         goto out;
2756         } else {
2757                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
2758         }
2759
2760         /* Add all the children we found */
2761         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
2762             lastlog == 0 ? vcount : lastlog) != 0)
2763                 goto out;
2764
2765         /*
2766          * If we're just doing a dry run, exit now with success.
2767          */
2768         if (flags.dryrun) {
2769                 memory_err = B_FALSE;
2770                 freelist = B_FALSE;
2771                 goto out;
2772         }
2773
2774         /* now build up the config list & call the ioctl */
2775         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
2776                 goto out;
2777
2778         if (nvlist_add_nvlist(newconfig,
2779             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
2780             nvlist_add_string(newconfig,
2781             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
2782             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
2783                 goto out;
2784
2785         /*
2786          * The new pool is automatically part of the namespace unless we
2787          * explicitly export it.
2788          */
2789         if (!flags.import)
2790                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
2791         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2792         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
2793         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
2794                 goto out;
2795         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
2796                 goto out;
2797
2798         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
2799                 retval = zpool_standard_error(hdl, errno, msg);
2800                 goto out;
2801         }
2802
2803         freelist = B_FALSE;
2804         memory_err = B_FALSE;
2805
2806 out:
2807         if (varray != NULL) {
2808                 int v;
2809
2810                 for (v = 0; v < vcount; v++)
2811                         nvlist_free(varray[v]);
2812                 free(varray);
2813         }
2814         zcmd_free_nvlists(&zc);
2815         if (zc_props)
2816                 nvlist_free(zc_props);
2817         if (newconfig)
2818                 nvlist_free(newconfig);
2819         if (freelist) {
2820                 nvlist_free(*newroot);
2821                 *newroot = NULL;
2822         }
2823
2824         if (retval != 0)
2825                 return (retval);
2826
2827         if (memory_err)
2828                 return (no_memory(hdl));
2829
2830         return (0);
2831 }
2832
2833 /*
2834  * Remove the given device.  Currently, this is supported only for hot spares
2835  * and level 2 cache devices.
2836  */
2837 int
2838 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
2839 {
2840         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2841         char msg[1024];
2842         nvlist_t *tgt;
2843         boolean_t avail_spare, l2cache, islog;
2844         libzfs_handle_t *hdl = zhp->zpool_hdl;
2845         uint64_t version;
2846
2847         (void) snprintf(msg, sizeof (msg),
2848             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
2849
2850         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2851         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2852             &islog)) == 0)
2853                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2854         /*
2855          * XXX - this should just go away.
2856          */
2857         if (!avail_spare && !l2cache && !islog) {
2858                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2859                     "only inactive hot spares, cache, top-level, "
2860                     "or log devices can be removed"));
2861                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2862         }
2863
2864         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
2865         if (islog && version < SPA_VERSION_HOLES) {
2866                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2867                     "pool must be upgrade to support log removal"));
2868                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
2869         }
2870
2871         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2872
2873         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
2874                 return (0);
2875
2876         return (zpool_standard_error(hdl, errno, msg));
2877 }
2878
2879 /*
2880  * Clear the errors for the pool, or the particular device if specified.
2881  */
2882 int
2883 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
2884 {
2885         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2886         char msg[1024];
2887         nvlist_t *tgt;
2888         zpool_rewind_policy_t policy;
2889         boolean_t avail_spare, l2cache;
2890         libzfs_handle_t *hdl = zhp->zpool_hdl;
2891         nvlist_t *nvi = NULL;
2892         int error;
2893
2894         if (path)
2895                 (void) snprintf(msg, sizeof (msg),
2896                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
2897                     path);
2898         else
2899                 (void) snprintf(msg, sizeof (msg),
2900                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
2901                     zhp->zpool_name);
2902
2903         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2904         if (path) {
2905                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
2906                     &l2cache, NULL)) == 0)
2907                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
2908
2909                 /*
2910                  * Don't allow error clearing for hot spares.  Do allow
2911                  * error clearing for l2cache devices.
2912                  */
2913                 if (avail_spare)
2914                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
2915
2916                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
2917                     &zc.zc_guid) == 0);
2918         }
2919
2920         zpool_get_rewind_policy(rewindnvl, &policy);
2921         zc.zc_cookie = policy.zrp_request;
2922
2923         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
2924                 return (-1);
2925
2926         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
2927                 return (-1);
2928
2929         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
2930             errno == ENOMEM) {
2931                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
2932                         zcmd_free_nvlists(&zc);
2933                         return (-1);
2934                 }
2935         }
2936
2937         if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
2938             errno != EPERM && errno != EACCES)) {
2939                 if (policy.zrp_request &
2940                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
2941                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
2942                         zpool_rewind_exclaim(hdl, zc.zc_name,
2943                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
2944                             nvi);
2945                         nvlist_free(nvi);
2946                 }
2947                 zcmd_free_nvlists(&zc);
2948                 return (0);
2949         }
2950
2951         zcmd_free_nvlists(&zc);
2952         return (zpool_standard_error(hdl, errno, msg));
2953 }
2954
2955 /*
2956  * Similar to zpool_clear(), but takes a GUID (used by fmd).
2957  */
2958 int
2959 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
2960 {
2961         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2962         char msg[1024];
2963         libzfs_handle_t *hdl = zhp->zpool_hdl;
2964
2965         (void) snprintf(msg, sizeof (msg),
2966             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
2967            (u_longlong_t)guid);
2968
2969         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2970         zc.zc_guid = guid;
2971         zc.zc_cookie = ZPOOL_NO_REWIND;
2972
2973         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
2974                 return (0);
2975
2976         return (zpool_standard_error(hdl, errno, msg));
2977 }
2978
2979 /*
2980  * Convert from a devid string to a path.
2981  */
2982 static char *
2983 devid_to_path(char *devid_str)
2984 {
2985         ddi_devid_t devid;
2986         char *minor;
2987         char *path;
2988         devid_nmlist_t *list = NULL;
2989         int ret;
2990
2991         if (devid_str_decode(devid_str, &devid, &minor) != 0)
2992                 return (NULL);
2993
2994         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
2995
2996         devid_str_free(minor);
2997         devid_free(devid);
2998
2999         if (ret != 0)
3000                 return (NULL);
3001
3002         if ((path = strdup(list[0].devname)) == NULL)
3003                 return (NULL);
3004
3005         devid_free_nmlist(list);
3006
3007         return (path);
3008 }
3009
3010 /*
3011  * Convert from a path to a devid string.
3012  */
3013 static char *
3014 path_to_devid(const char *path)
3015 {
3016         int fd;
3017         ddi_devid_t devid;
3018         char *minor, *ret;
3019
3020         if ((fd = open(path, O_RDONLY)) < 0)
3021                 return (NULL);
3022
3023         minor = NULL;
3024         ret = NULL;
3025         if (devid_get(fd, &devid) == 0) {
3026                 if (devid_get_minor_name(fd, &minor) == 0)
3027                         ret = devid_str_encode(devid, minor);
3028                 if (minor != NULL)
3029                         devid_str_free(minor);
3030                 devid_free(devid);
3031         }
3032         (void) close(fd);
3033
3034         return (ret);
3035 }
3036
3037 /*
3038  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3039  * ignore any failure here, since a common case is for an unprivileged user to
3040  * type 'zpool status', and we'll display the correct information anyway.
3041  */
3042 static void
3043 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3044 {
3045         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3046
3047         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3048         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3049         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3050             &zc.zc_guid) == 0);
3051
3052         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3053 }
3054
3055 /*
3056  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3057  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3058  * We also check if this is a whole disk, in which case we strip off the
3059  * trailing 's0' slice name.
3060  *
3061  * This routine is also responsible for identifying when disks have been
3062  * reconfigured in a new location.  The kernel will have opened the device by
3063  * devid, but the path will still refer to the old location.  To catch this, we
3064  * first do a path -> devid translation (which is fast for the common case).  If
3065  * the devid matches, we're done.  If not, we do a reverse devid -> path
3066  * translation and issue the appropriate ioctl() to update the path of the vdev.
3067  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3068  * of these checks.
3069  */
3070 char *
3071 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3072     boolean_t verbose)
3073 {
3074         char *path, *devid;
3075         uint64_t value;
3076         char buf[64];
3077         vdev_stat_t *vs;
3078         uint_t vsc;
3079
3080         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
3081             &value) == 0) {
3082                 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3083                     &value) == 0);
3084                 (void) snprintf(buf, sizeof (buf), "%llu",
3085                     (u_longlong_t)value);
3086                 path = buf;
3087         } else if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0) {
3088
3089                 /*
3090                  * If the device is dead (faulted, offline, etc) then don't
3091                  * bother opening it.  Otherwise we may be forcing the user to
3092                  * open a misbehaving device, which can have undesirable
3093                  * effects.
3094                  */
3095                 if ((nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3096                     (uint64_t **)&vs, &vsc) != 0 ||
3097                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3098                     zhp != NULL &&
3099                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3100                         /*
3101                          * Determine if the current path is correct.
3102                          */
3103                         char *newdevid = path_to_devid(path);
3104
3105                         if (newdevid == NULL ||
3106                             strcmp(devid, newdevid) != 0) {
3107                                 char *newpath;
3108
3109                                 if ((newpath = devid_to_path(devid)) != NULL) {
3110                                         /*
3111                                          * Update the path appropriately.
3112                                          */
3113                                         set_path(zhp, nv, newpath);
3114                                         if (nvlist_add_string(nv,
3115                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3116                                                 verify(nvlist_lookup_string(nv,
3117                                                     ZPOOL_CONFIG_PATH,
3118                                                     &path) == 0);
3119                                         free(newpath);
3120                                 }
3121                         }
3122
3123                         if (newdevid)
3124                                 devid_str_free(newdevid);
3125                 }
3126
3127                 if (strncmp(path, "/dev/dsk/", 9) == 0)
3128                         path += 9;
3129
3130                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
3131                     &value) == 0 && value) {
3132                         int pathlen = strlen(path);
3133                         char *tmp = zfs_strdup(hdl, path);
3134
3135                         /*
3136                          * If it starts with c#, and ends with "s0", chop
3137                          * the "s0" off, or if it ends with "s0/old", remove
3138                          * the "s0" from the middle.
3139                          */
3140                         if (CTD_CHECK(tmp)) {
3141                                 if (strcmp(&tmp[pathlen - 2], "s0") == 0) {
3142                                         tmp[pathlen - 2] = '\0';
3143                                 } else if (pathlen > 6 &&
3144                                     strcmp(&tmp[pathlen - 6], "s0/old") == 0) {
3145                                         (void) strcpy(&tmp[pathlen - 6],
3146                                             "/old");
3147                                 }
3148                         }
3149                         return (tmp);
3150                 }
3151         } else {
3152                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3153
3154                 /*
3155                  * If it's a raidz device, we need to stick in the parity level.
3156                  */
3157                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3158                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3159                             &value) == 0);
3160                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3161                             (u_longlong_t)value);
3162                         path = buf;
3163                 }
3164
3165                 /*
3166                  * We identify each top-level vdev by using a <type-id>
3167                  * naming convention.
3168                  */
3169                 if (verbose) {
3170                         uint64_t id;
3171
3172                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3173                             &id) == 0);
3174                         (void) snprintf(buf, sizeof (buf), "%s-%llu", path,
3175                             (u_longlong_t)id);
3176                         path = buf;
3177                 }
3178         }
3179
3180         return (zfs_strdup(hdl, path));
3181 }
3182
3183 static int
3184 zbookmark_compare(const void *a, const void *b)
3185 {
3186         return (memcmp(a, b, sizeof (zbookmark_t)));
3187 }
3188
3189 /*
3190  * Retrieve the persistent error log, uniquify the members, and return to the
3191  * caller.
3192  */
3193 int
3194 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3195 {
3196         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3197         uint64_t count;
3198         zbookmark_t *zb = NULL;
3199         int i;
3200
3201         /*
3202          * Retrieve the raw error list from the kernel.  If the number of errors
3203          * has increased, allocate more space and continue until we get the
3204          * entire list.
3205          */
3206         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3207             &count) == 0);
3208         if (count == 0)
3209                 return (0);
3210         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3211             count * sizeof (zbookmark_t))) == (uintptr_t)NULL)
3212                 return (-1);
3213         zc.zc_nvlist_dst_size = count;
3214         (void) strcpy(zc.zc_name, zhp->zpool_name);
3215         for (;;) {
3216                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3217                     &zc) != 0) {
3218                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3219                         if (errno == ENOMEM) {
3220                                 count = zc.zc_nvlist_dst_size;
3221                                 if ((zc.zc_nvlist_dst = (uintptr_t)
3222                                     zfs_alloc(zhp->zpool_hdl, count *
3223                                     sizeof (zbookmark_t))) == (uintptr_t)NULL)
3224                                         return (-1);
3225                         } else {
3226                                 return (-1);
3227                         }
3228                 } else {
3229                         break;
3230                 }
3231         }
3232
3233         /*
3234          * Sort the resulting bookmarks.  This is a little confusing due to the
3235          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3236          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3237          * _not_ copied as part of the process.  So we point the start of our
3238          * array appropriate and decrement the total number of elements.
3239          */
3240         zb = ((zbookmark_t *)(uintptr_t)zc.zc_nvlist_dst) +
3241             zc.zc_nvlist_dst_size;
3242         count -= zc.zc_nvlist_dst_size;
3243
3244         qsort(zb, count, sizeof (zbookmark_t), zbookmark_compare);
3245
3246         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3247
3248         /*
3249          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3250          */
3251         for (i = 0; i < count; i++) {
3252                 nvlist_t *nv;
3253
3254                 /* ignoring zb_blkid and zb_level for now */
3255                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3256                     zb[i-1].zb_object == zb[i].zb_object)
3257                         continue;
3258
3259                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3260                         goto nomem;
3261                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3262                     zb[i].zb_objset) != 0) {
3263                         nvlist_free(nv);
3264                         goto nomem;
3265                 }
3266                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3267                     zb[i].zb_object) != 0) {
3268                         nvlist_free(nv);
3269                         goto nomem;
3270                 }
3271                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3272                         nvlist_free(nv);
3273                         goto nomem;
3274                 }
3275                 nvlist_free(nv);
3276         }
3277
3278         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3279         return (0);
3280
3281 nomem:
3282         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3283         return (no_memory(zhp->zpool_hdl));
3284 }
3285
3286 /*
3287  * Upgrade a ZFS pool to the latest on-disk version.
3288  */
3289 int
3290 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3291 {
3292         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3293         libzfs_handle_t *hdl = zhp->zpool_hdl;
3294
3295         (void) strcpy(zc.zc_name, zhp->zpool_name);
3296         zc.zc_cookie = new_version;
3297
3298         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3299                 return (zpool_standard_error_fmt(hdl, errno,
3300                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3301                     zhp->zpool_name));
3302         return (0);
3303 }
3304
3305 void
3306 zpool_set_history_str(const char *subcommand, int argc, char **argv,
3307     char *history_str)
3308 {
3309         int i;
3310
3311         (void) strlcpy(history_str, subcommand, HIS_MAX_RECORD_LEN);
3312         for (i = 1; i < argc; i++) {
3313                 if (strlen(history_str) + 1 + strlen(argv[i]) >
3314                     HIS_MAX_RECORD_LEN)
3315                         break;
3316                 (void) strlcat(history_str, " ", HIS_MAX_RECORD_LEN);
3317                 (void) strlcat(history_str, argv[i], HIS_MAX_RECORD_LEN);
3318         }
3319 }
3320
3321 /*
3322  * Stage command history for logging.
3323  */
3324 int
3325 zpool_stage_history(libzfs_handle_t *hdl, const char *history_str)
3326 {
3327         if (history_str == NULL)
3328                 return (EINVAL);
3329
3330         if (strlen(history_str) > HIS_MAX_RECORD_LEN)
3331                 return (EINVAL);
3332
3333         if (hdl->libzfs_log_str != NULL)
3334                 free(hdl->libzfs_log_str);
3335
3336         if ((hdl->libzfs_log_str = strdup(history_str)) == NULL)
3337                 return (no_memory(hdl));
3338
3339         return (0);
3340 }
3341
3342 /*
3343  * Perform ioctl to get some command history of a pool.
3344  *
3345  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3346  * logical offset of the history buffer to start reading from.
3347  *
3348  * Upon return, 'off' is the next logical offset to read from and
3349  * 'len' is the actual amount of bytes read into 'buf'.
3350  */
3351 static int
3352 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3353 {
3354         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3355         libzfs_handle_t *hdl = zhp->zpool_hdl;
3356
3357         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3358
3359         zc.zc_history = (uint64_t)(uintptr_t)buf;
3360         zc.zc_history_len = *len;
3361         zc.zc_history_offset = *off;
3362
3363         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3364                 switch (errno) {
3365                 case EPERM:
3366                         return (zfs_error_fmt(hdl, EZFS_PERM,
3367                             dgettext(TEXT_DOMAIN,
3368                             "cannot show history for pool '%s'"),
3369                             zhp->zpool_name));
3370                 case ENOENT:
3371                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3372                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3373                             "'%s'"), zhp->zpool_name));
3374                 case ENOTSUP:
3375                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3376                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3377                             "'%s', pool must be upgraded"), zhp->zpool_name));
3378                 default:
3379                         return (zpool_standard_error_fmt(hdl, errno,
3380                             dgettext(TEXT_DOMAIN,
3381                             "cannot get history for '%s'"), zhp->zpool_name));
3382                 }
3383         }
3384
3385         *len = zc.zc_history_len;
3386         *off = zc.zc_history_offset;
3387
3388         return (0);
3389 }
3390
3391 /*
3392  * Process the buffer of nvlists, unpacking and storing each nvlist record
3393  * into 'records'.  'leftover' is set to the number of bytes that weren't
3394  * processed as there wasn't a complete record.
3395  */
3396 int
3397 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3398     nvlist_t ***records, uint_t *numrecords)
3399 {
3400         uint64_t reclen;
3401         nvlist_t *nv;
3402         int i;
3403
3404         while (bytes_read > sizeof (reclen)) {
3405
3406                 /* get length of packed record (stored as little endian) */
3407                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3408                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3409
3410                 if (bytes_read < sizeof (reclen) + reclen)
3411                         break;
3412
3413                 /* unpack record */
3414                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3415                         return (ENOMEM);
3416                 bytes_read -= sizeof (reclen) + reclen;
3417                 buf += sizeof (reclen) + reclen;
3418
3419                 /* add record to nvlist array */
3420                 (*numrecords)++;
3421                 if (ISP2(*numrecords + 1)) {
3422                         *records = realloc(*records,
3423                             *numrecords * 2 * sizeof (nvlist_t *));
3424                 }
3425                 (*records)[*numrecords - 1] = nv;
3426         }
3427
3428         *leftover = bytes_read;
3429         return (0);
3430 }
3431
3432 #define HIS_BUF_LEN     (128*1024)
3433
3434 /*
3435  * Retrieve the command history of a pool.
3436  */
3437 int
3438 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3439 {
3440         char buf[HIS_BUF_LEN];
3441         uint64_t off = 0;
3442         nvlist_t **records = NULL;
3443         uint_t numrecords = 0;
3444         int err, i;
3445
3446         do {
3447                 uint64_t bytes_read = sizeof (buf);
3448                 uint64_t leftover;
3449
3450                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3451                         break;
3452
3453                 /* if nothing else was read in, we're at EOF, just return */
3454                 if (!bytes_read)
3455                         break;
3456
3457                 if ((err = zpool_history_unpack(buf, bytes_read,
3458                     &leftover, &records, &numrecords)) != 0)
3459                         break;
3460                 off -= leftover;
3461
3462                 /* CONSTCOND */
3463         } while (1);
3464
3465         if (!err) {
3466                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3467                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3468                     records, numrecords) == 0);
3469         }
3470         for (i = 0; i < numrecords; i++)
3471                 nvlist_free(records[i]);
3472         free(records);
3473
3474         return (err);
3475 }
3476
3477 void
3478 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
3479     char *pathname, size_t len)
3480 {
3481         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3482         boolean_t mounted = B_FALSE;
3483         char *mntpnt = NULL;
3484         char dsname[MAXNAMELEN];
3485
3486         if (dsobj == 0) {
3487                 /* special case for the MOS */
3488                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>", (longlong_t)obj);
3489                 return;
3490         }
3491
3492         /* get the dataset's name */
3493         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3494         zc.zc_obj = dsobj;
3495         if (ioctl(zhp->zpool_hdl->libzfs_fd,
3496             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
3497                 /* just write out a path of two object numbers */
3498                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
3499                     (longlong_t)dsobj, (longlong_t)obj);
3500                 return;
3501         }
3502         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
3503
3504         /* find out if the dataset is mounted */
3505         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
3506
3507         /* get the corrupted object's path */
3508         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
3509         zc.zc_obj = obj;
3510         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
3511             &zc) == 0) {
3512                 if (mounted) {
3513                         (void) snprintf(pathname, len, "%s%s", mntpnt,
3514                             zc.zc_value);
3515                 } else {
3516                         (void) snprintf(pathname, len, "%s:%s",
3517                             dsname, zc.zc_value);
3518                 }
3519         } else {
3520                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname, (longlong_t)obj);
3521         }
3522         free(mntpnt);
3523 }
3524
3525 /*
3526  * Read the EFI label from the config, if a label does not exist then
3527  * pass back the error to the caller. If the caller has passed a non-NULL
3528  * diskaddr argument then we set it to the starting address of the EFI
3529  * partition.
3530  */
3531 static int
3532 read_efi_label(nvlist_t *config, diskaddr_t *sb)
3533 {
3534         char *path;
3535         int fd;
3536         char diskname[MAXPATHLEN];
3537         int err = -1;
3538
3539         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
3540                 return (err);
3541
3542         (void) snprintf(diskname, sizeof (diskname), "%s%s", RDISK_ROOT,
3543             strrchr(path, '/'));
3544         if ((fd = open(diskname, O_RDONLY|O_NDELAY)) >= 0) {
3545                 struct dk_gpt *vtoc;
3546
3547                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
3548                         if (sb != NULL)
3549                                 *sb = vtoc->efi_parts[0].p_start;
3550                         efi_free(vtoc);
3551                 }
3552                 (void) close(fd);
3553         }
3554         return (err);
3555 }
3556
3557 /*
3558  * determine where a partition starts on a disk in the current
3559  * configuration
3560  */
3561 static diskaddr_t
3562 find_start_block(nvlist_t *config)
3563 {
3564         nvlist_t **child;
3565         uint_t c, children;
3566         diskaddr_t sb = MAXOFFSET_T;
3567         uint64_t wholedisk;
3568
3569         if (nvlist_lookup_nvlist_array(config,
3570             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
3571                 if (nvlist_lookup_uint64(config,
3572                     ZPOOL_CONFIG_WHOLE_DISK,
3573                     &wholedisk) != 0 || !wholedisk) {
3574                         return (MAXOFFSET_T);
3575                 }
3576                 if (read_efi_label(config, &sb) < 0)
3577                         sb = MAXOFFSET_T;
3578                 return (sb);
3579         }
3580
3581         for (c = 0; c < children; c++) {
3582                 sb = find_start_block(child[c]);
3583                 if (sb != MAXOFFSET_T) {
3584                         return (sb);
3585                 }
3586         }
3587         return (MAXOFFSET_T);
3588 }
3589
3590 /*
3591  * Label an individual disk.  The name provided is the short name,
3592  * stripped of any leading /dev path.
3593  */
3594 int
3595 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, char *name)
3596 {
3597         char path[MAXPATHLEN];
3598         struct dk_gpt *vtoc;
3599         int fd;
3600         size_t resv = EFI_MIN_RESV_SIZE;
3601         uint64_t slice_size;
3602         diskaddr_t start_block;
3603         char errbuf[1024];
3604
3605         /* prepare an error message just in case */
3606         (void) snprintf(errbuf, sizeof (errbuf),
3607             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
3608
3609         if (zhp) {
3610                 nvlist_t *nvroot;
3611
3612                 if (pool_is_bootable(zhp)) {
3613                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3614                             "EFI labeled devices are not supported on root "
3615                             "pools."));
3616                         return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
3617                 }
3618
3619                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
3620                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3621
3622                 if (zhp->zpool_start_block == 0)
3623                         start_block = find_start_block(nvroot);
3624                 else
3625                         start_block = zhp->zpool_start_block;
3626                 zhp->zpool_start_block = start_block;
3627         } else {
3628                 /* new pool */
3629                 start_block = NEW_START_BLOCK;
3630         }
3631
3632         (void) snprintf(path, sizeof (path), "%s/%s%s", RDISK_ROOT, name,
3633             BACKUP_SLICE);
3634
3635         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
3636                 /*
3637                  * This shouldn't happen.  We've long since verified that this
3638                  * is a valid device.
3639                  */
3640                 zfs_error_aux(hdl,
3641                     dgettext(TEXT_DOMAIN, "unable to open device"));
3642                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
3643         }
3644
3645         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
3646                 /*
3647                  * The only way this can fail is if we run out of memory, or we
3648                  * were unable to read the disk's capacity
3649                  */
3650                 if (errno == ENOMEM)
3651                         (void) no_memory(hdl);
3652
3653                 (void) close(fd);
3654                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3655                     "unable to read disk capacity"), name);
3656
3657                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
3658         }
3659
3660         slice_size = vtoc->efi_last_u_lba + 1;
3661         slice_size -= EFI_MIN_RESV_SIZE;
3662         if (start_block == MAXOFFSET_T)
3663                 start_block = NEW_START_BLOCK;
3664         slice_size -= start_block;
3665
3666         vtoc->efi_parts[0].p_start = start_block;
3667         vtoc->efi_parts[0].p_size = slice_size;
3668
3669         /*
3670          * Why we use V_USR: V_BACKUP confuses users, and is considered
3671          * disposable by some EFI utilities (since EFI doesn't have a backup
3672          * slice).  V_UNASSIGNED is supposed to be used only for zero size
3673          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
3674          * etc. were all pretty specific.  V_USR is as close to reality as we
3675          * can get, in the absence of V_OTHER.
3676          */
3677         vtoc->efi_parts[0].p_tag = V_USR;
3678         (void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
3679
3680         vtoc->efi_parts[8].p_start = slice_size + start_block;
3681         vtoc->efi_parts[8].p_size = resv;
3682         vtoc->efi_parts[8].p_tag = V_RESERVED;
3683
3684         if (efi_write(fd, vtoc) != 0) {
3685                 /*
3686                  * Some block drivers (like pcata) may not support EFI
3687                  * GPT labels.  Print out a helpful error message dir-
3688                  * ecting the user to manually label the disk and give
3689                  * a specific slice.
3690                  */
3691                 (void) close(fd);
3692                 efi_free(vtoc);
3693
3694                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3695                     "try using fdisk(1M) and then provide a specific slice"));
3696                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
3697         }
3698
3699         (void) close(fd);
3700         efi_free(vtoc);
3701         return (0);
3702 }
3703
3704 static boolean_t
3705 supported_dump_vdev_type(libzfs_handle_t *hdl, nvlist_t *config, char *errbuf)
3706 {
3707         char *type;
3708         nvlist_t **child;
3709         uint_t children, c;
3710
3711         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_TYPE, &type) == 0);
3712         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0 ||
3713             strcmp(type, VDEV_TYPE_FILE) == 0 ||
3714             strcmp(type, VDEV_TYPE_LOG) == 0 ||
3715             strcmp(type, VDEV_TYPE_HOLE) == 0 ||
3716             strcmp(type, VDEV_TYPE_MISSING) == 0) {
3717                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3718                     "vdev type '%s' is not supported"), type);
3719                 (void) zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf);
3720                 return (B_FALSE);
3721         }
3722         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
3723             &child, &children) == 0) {
3724                 for (c = 0; c < children; c++) {
3725                         if (!supported_dump_vdev_type(hdl, child[c], errbuf))
3726                                 return (B_FALSE);
3727                 }
3728         }
3729         return (B_TRUE);
3730 }
3731
3732 /*
3733  * check if this zvol is allowable for use as a dump device; zero if
3734  * it is, > 0 if it isn't, < 0 if it isn't a zvol
3735  */
3736 int
3737 zvol_check_dump_config(char *arg)
3738 {
3739         zpool_handle_t *zhp = NULL;
3740         nvlist_t *config, *nvroot;
3741         char *p, *volname;
3742         nvlist_t **top;
3743         uint_t toplevels;
3744         libzfs_handle_t *hdl;
3745         char errbuf[1024];
3746         char poolname[ZPOOL_MAXNAMELEN];
3747         int pathlen = strlen(ZVOL_FULL_DEV_DIR);
3748         int ret = 1;
3749
3750         if (strncmp(arg, ZVOL_FULL_DEV_DIR, pathlen)) {
3751                 return (-1);
3752         }
3753
3754         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3755             "dump is not supported on device '%s'"), arg);
3756
3757         if ((hdl = libzfs_init()) == NULL)
3758                 return (1);
3759         libzfs_print_on_error(hdl, B_TRUE);
3760
3761         volname = arg + pathlen;
3762
3763         /* check the configuration of the pool */
3764         if ((p = strchr(volname, '/')) == NULL) {
3765                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3766                     "malformed dataset name"));
3767                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
3768                 return (1);
3769         } else if (p - volname >= ZFS_MAXNAMELEN) {
3770                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3771                     "dataset name is too long"));
3772                 (void) zfs_error(hdl, EZFS_NAMETOOLONG, errbuf);
3773                 return (1);
3774         } else {
3775                 (void) strncpy(poolname, volname, p - volname);
3776                 poolname[p - volname] = '\0';
3777         }
3778
3779         if ((zhp = zpool_open(hdl, poolname)) == NULL) {
3780                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3781                     "could not open pool '%s'"), poolname);
3782                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
3783                 goto out;
3784         }
3785         config = zpool_get_config(zhp, NULL);
3786         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3787             &nvroot) != 0) {
3788                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3789                     "could not obtain vdev configuration for  '%s'"), poolname);
3790                 (void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
3791                 goto out;
3792         }
3793
3794         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3795             &top, &toplevels) == 0);
3796         if (toplevels != 1) {
3797                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3798                     "'%s' has multiple top level vdevs"), poolname);
3799                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, errbuf);
3800                 goto out;
3801         }
3802
3803         if (!supported_dump_vdev_type(hdl, top[0], errbuf)) {
3804                 goto out;
3805         }
3806         ret = 0;
3807
3808 out:
3809         if (zhp)
3810                 zpool_close(zhp);
3811         libzfs_fini(hdl);
3812         return (ret);
3813 }