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