cddl/contrib/opensolaris/lib/libzfs/common/libzfs_pool.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  25  * Copyright (c) 2012 by Delphix. All rights reserved.
  26  */
  27
  28 #include <sys/types.h>
  29 #include <sys/stat.h>
  30 #include <ctype.h>
  31 #include <errno.h>
  32 #include <devid.h>
  33 #include <fcntl.h>
  34 #include <libintl.h>
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <strings.h>
  38 #include <unistd.h>
  39 #include <libgen.h>
  40 #include <sys/zfs_ioctl.h>
  41 #include <dlfcn.h>
  42
  43 #include "zfs_namecheck.h"
  44 #include "zfs_prop.h"
  45 #include "libzfs_impl.h"
  46 #include "zfs_comutil.h"
  47 #include "zfeature_common.h"
  48
  49 static int read_efi_label(nvlist_t *config, diskaddr_t *sb);
  50
  51 #define DISK_ROOT       "/dev/dsk"
  52 #define RDISK_ROOT      "/dev/rdsk"
  53 #define BACKUP_SLICE    "s2"
  54
  55 typedef struct prop_flags {
  56         int create:1;   /* Validate property on creation */
  57         int import:1;   /* Validate property on import */
  58 } prop_flags_t;
  59
  60 /*
  61  * ====================================================================
  62  *   zpool property functions
  63  * ====================================================================
  64  */
  65
  66 static int
  67 zpool_get_all_props(zpool_handle_t *zhp)
  68 {
  69         zfs_cmd_t zc = { 0 };
  70         libzfs_handle_t *hdl = zhp->zpool_hdl;
  71
  72         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
  73
  74         if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
  75                 return (-1);
  76
  77         while (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
  78                 if (errno == ENOMEM) {
  79                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
  80                                 zcmd_free_nvlists(&zc);
  81                                 return (-1);
  82                         }
  83                 } else {
  84                         zcmd_free_nvlists(&zc);
  85                         return (-1);
  86                 }
  87         }
  88
  89         if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
  90                 zcmd_free_nvlists(&zc);
  91                 return (-1);
  92         }
  93
  94         zcmd_free_nvlists(&zc);
  95
  96         return (0);
  97 }
  98
  99 static int
 100 zpool_props_refresh(zpool_handle_t *zhp)
 101 {
 102         nvlist_t *old_props;
 103
 104         old_props = zhp->zpool_props;
 105
 106         if (zpool_get_all_props(zhp) != 0)
 107                 return (-1);
 108
 109         nvlist_free(old_props);
 110         return (0);
 111 }
 112
 113 static char *
 114 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
 115     zprop_source_t *src)
 116 {
 117         nvlist_t *nv, *nvl;
 118         uint64_t ival;
 119         char *value;
 120         zprop_source_t source;
 121
 122         nvl = zhp->zpool_props;
 123         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 124                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &ival) == 0);
 125                 source = ival;
 126                 verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
 127         } else {
 128                 source = ZPROP_SRC_DEFAULT;
 129                 if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
 130                         value = "-";
 131         }
 132
 133         if (src)
 134                 *src = source;
 135
 136         return (value);
 137 }
 138
 139 uint64_t
 140 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
 141 {
 142         nvlist_t *nv, *nvl;
 143         uint64_t value;
 144         zprop_source_t source;
 145
 146         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
 147                 /*
 148                  * zpool_get_all_props() has most likely failed because
 149                  * the pool is faulted, but if all we need is the top level
 150                  * vdev's guid then get it from the zhp config nvlist.
 151                  */
 152                 if ((prop == ZPOOL_PROP_GUID) &&
 153                     (nvlist_lookup_nvlist(zhp->zpool_config,
 154                     ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
 155                     (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
 156                     == 0)) {
 157                         return (value);
 158                 }
 159                 return (zpool_prop_default_numeric(prop));
 160         }
 161
 162         nvl = zhp->zpool_props;
 163         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 164                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &value) == 0);
 165                 source = value;
 166                 verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
 167         } else {
 168                 source = ZPROP_SRC_DEFAULT;
 169                 value = zpool_prop_default_numeric(prop);
 170         }
 171
 172         if (src)
 173                 *src = source;
 174
 175         return (value);
 176 }
 177
 178 /*
 179  * Map VDEV STATE to printed strings.
 180  */
 181 const char *
 182 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
 183 {
 184         switch (state) {
 185         case VDEV_STATE_CLOSED:
 186         case VDEV_STATE_OFFLINE:
 187                 return (gettext("OFFLINE"));
 188         case VDEV_STATE_REMOVED:
 189                 return (gettext("REMOVED"));
 190         case VDEV_STATE_CANT_OPEN:
 191                 if (aux == VDEV_AUX_CORRUPT_DATA || aux == VDEV_AUX_BAD_LOG)
 192                         return (gettext("FAULTED"));
 193                 else if (aux == VDEV_AUX_SPLIT_POOL)
 194                         return (gettext("SPLIT"));
 195                 else
 196                         return (gettext("UNAVAIL"));
 197         case VDEV_STATE_FAULTED:
 198                 return (gettext("FAULTED"));
 199         case VDEV_STATE_DEGRADED:
 200                 return (gettext("DEGRADED"));
 201         case VDEV_STATE_HEALTHY:
 202                 return (gettext("ONLINE"));
 203         }
 204
 205         return (gettext("UNKNOWN"));
 206 }
 207
 208 /*
 209  * Map POOL STATE to printed strings.
 210  */
 211 const char *
 212 zpool_pool_state_to_name(pool_state_t state)
 213 {
 214         switch (state) {
 215         case POOL_STATE_ACTIVE:
 216                 return (gettext("ACTIVE"));
 217         case POOL_STATE_EXPORTED:
 218                 return (gettext("EXPORTED"));
 219         case POOL_STATE_DESTROYED:
 220                 return (gettext("DESTROYED"));
 221         case POOL_STATE_SPARE:
 222                 return (gettext("SPARE"));
 223         case POOL_STATE_L2CACHE:
 224                 return (gettext("L2CACHE"));
 225         case POOL_STATE_UNINITIALIZED:
 226                 return (gettext("UNINITIALIZED"));
 227         case POOL_STATE_UNAVAIL:
 228                 return (gettext("UNAVAIL"));
 229         case POOL_STATE_POTENTIALLY_ACTIVE:
 230                 return (gettext("POTENTIALLY_ACTIVE"));
 231         }
 232
 233         return (gettext("UNKNOWN"));
 234 }
 235
 236 /*
 237  * Get a zpool property value for 'prop' and return the value in
 238  * a pre-allocated buffer.
 239  */
 240 int
 241 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf, size_t len,
 242     zprop_source_t *srctype)
 243 {
 244         uint64_t intval;
 245         const char *strval;
 246         zprop_source_t src = ZPROP_SRC_NONE;
 247         nvlist_t *nvroot;
 248         vdev_stat_t *vs;
 249         uint_t vsc;
 250
 251         if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
 252                 switch (prop) {
 253                 case ZPOOL_PROP_NAME:
 254                         (void) strlcpy(buf, zpool_get_name(zhp), len);
 255                         break;
 256
 257                 case ZPOOL_PROP_HEALTH:
 258                         (void) strlcpy(buf, "FAULTED", len);
 259                         break;
 260
 261                 case ZPOOL_PROP_GUID:
 262                         intval = zpool_get_prop_int(zhp, prop, &src);
 263                         (void) snprintf(buf, len, "%llu", intval);
 264                         break;
 265
 266                 case ZPOOL_PROP_ALTROOT:
 267                 case ZPOOL_PROP_CACHEFILE:
 268                 case ZPOOL_PROP_COMMENT:
 269                         if (zhp->zpool_props != NULL ||
 270                             zpool_get_all_props(zhp) == 0) {
 271                                 (void) strlcpy(buf,
 272                                     zpool_get_prop_string(zhp, prop, &src),
 273                                     len);
 274                                 if (srctype != NULL)
 275                                         *srctype = src;
 276                                 return (0);
 277                         }
 278                         /* FALLTHROUGH */
 279                 default:
 280                         (void) strlcpy(buf, "-", len);
 281                         break;
 282                 }
 283
 284                 if (srctype != NULL)
 285                         *srctype = src;
 286                 return (0);
 287         }
 288
 289         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
 290             prop != ZPOOL_PROP_NAME)
 291                 return (-1);
 292
 293         switch (zpool_prop_get_type(prop)) {
 294         case PROP_TYPE_STRING:
 295                 (void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
 296                     len);
 297                 break;
 298
 299         case PROP_TYPE_NUMBER:
 300                 intval = zpool_get_prop_int(zhp, prop, &src);
 301
 302                 switch (prop) {
 303                 case ZPOOL_PROP_SIZE:
 304                 case ZPOOL_PROP_ALLOCATED:
 305                 case ZPOOL_PROP_FREE:
 306                 case ZPOOL_PROP_FREEING:
 307                 case ZPOOL_PROP_EXPANDSZ:
 308                         (void) zfs_nicenum(intval, buf, len);
 309                         break;
 310
 311                 case ZPOOL_PROP_CAPACITY:
 312                         (void) snprintf(buf, len, "%llu%%",
 313                             (u_longlong_t)intval);
 314                         break;
 315
 316                 case ZPOOL_PROP_DEDUPRATIO:
 317                         (void) snprintf(buf, len, "%llu.%02llux",
 318                             (u_longlong_t)(intval / 100),
 319                             (u_longlong_t)(intval % 100));
 320                         break;
 321
 322                 case ZPOOL_PROP_HEALTH:
 323                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 324                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 325                         verify(nvlist_lookup_uint64_array(nvroot,
 326                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
 327                             == 0);
 328
 329                         (void) strlcpy(buf, zpool_state_to_name(intval,
 330                             vs->vs_aux), len);
 331                         break;
 332                 case ZPOOL_PROP_VERSION:
 333                         if (intval >= SPA_VERSION_FEATURES) {
 334                                 (void) snprintf(buf, len, "-");
 335                                 break;
 336                         }
 337                         /* FALLTHROUGH */
 338                 default:
 339                         (void) snprintf(buf, len, "%llu", intval);
 340                 }
 341                 break;
 342
 343         case PROP_TYPE_INDEX:
 344                 intval = zpool_get_prop_int(zhp, prop, &src);
 345                 if (zpool_prop_index_to_string(prop, intval, &strval)
 346                     != 0)
 347                         return (-1);
 348                 (void) strlcpy(buf, strval, len);
 349                 break;
 350
 351         default:
 352                 abort();
 353         }
 354
 355         if (srctype)
 356                 *srctype = src;
 357
 358         return (0);
 359 }
 360
 361 /*
 362  * Check if the bootfs name has the same pool name as it is set to.
 363  * Assuming bootfs is a valid dataset name.
 364  */
 365 static boolean_t
 366 bootfs_name_valid(const char *pool, char *bootfs)
 367 {
 368         int len = strlen(pool);
 369
 370         if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
 371                 return (B_FALSE);
 372
 373         if (strncmp(pool, bootfs, len) == 0 &&
 374             (bootfs[len] == '/' || bootfs[len] == '\0'))
 375                 return (B_TRUE);
 376
 377         return (B_FALSE);
 378 }
 379
 380 /*
 381  * Inspect the configuration to determine if any of the devices contain
 382  * an EFI label.
 383  */
 384 static boolean_t
 385 pool_uses_efi(nvlist_t *config)
 386 {
 387 #ifdef sun
 388         nvlist_t **child;
 389         uint_t c, children;
 390
 391         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
 392             &child, &children) != 0)
 393                 return (read_efi_label(config, NULL) >= 0);
 394
 395         for (c = 0; c < children; c++) {
 396                 if (pool_uses_efi(child[c]))
 397                         return (B_TRUE);
 398         }
 399 #endif  /* sun */
 400         return (B_FALSE);
 401 }
 402
 403 boolean_t
 404 zpool_is_bootable(zpool_handle_t *zhp)
 405 {
 406         char bootfs[ZPOOL_MAXNAMELEN];
 407
 408         return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
 409             sizeof (bootfs), NULL) == 0 && strncmp(bootfs, "-",
 410             sizeof (bootfs)) != 0);
 411 }
 412
 413
 414 /*
 415  * Given an nvlist of zpool properties to be set, validate that they are
 416  * correct, and parse any numeric properties (index, boolean, etc) if they are
 417  * specified as strings.
 418  */
 419 static nvlist_t *
 420 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 421     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
 422 {
 423         nvpair_t *elem;
 424         nvlist_t *retprops;
 425         zpool_prop_t prop;
 426         char *strval;
 427         uint64_t intval;
 428         char *slash, *check;
 429         struct stat64 statbuf;
 430         zpool_handle_t *zhp;
 431         nvlist_t *nvroot;
 432
 433         if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
 434                 (void) no_memory(hdl);
 435                 return (NULL);
 436         }
 437
 438         elem = NULL;
 439         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 440                 const char *propname = nvpair_name(elem);
 441
 442                 prop = zpool_name_to_prop(propname);
 443                 if (prop == ZPROP_INVAL && zpool_prop_feature(propname)) {
 444                         int err;
 445                         zfeature_info_t *feature;
 446                         char *fname = strchr(propname, '@') + 1;
 447
 448                         err = zfeature_lookup_name(fname, &feature);
 449                         if (err != 0) {
 450                                 ASSERT3U(err, ==, ENOENT);
 451                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 452                                     "invalid feature '%s'"), fname);
 453                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 454                                 goto error;
 455                         }
 456
 457                         if (nvpair_type(elem) != DATA_TYPE_STRING) {
 458                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 459                                     "'%s' must be a string"), propname);
 460                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 461                                 goto error;
 462                         }
 463
 464                         (void) nvpair_value_string(elem, &strval);
 465                         if (strcmp(strval, ZFS_FEATURE_ENABLED) != 0) {
 466                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 467                                     "property '%s' can only be set to "
 468                                     "'enabled'"), propname);
 469                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 470                                 goto error;
 471                         }
 472
 473                         if (nvlist_add_uint64(retprops, propname, 0) != 0) {
 474                                 (void) no_memory(hdl);
 475                                 goto error;
 476                         }
 477                         continue;
 478                 }
 479
 480                 /*
 481                  * Make sure this property is valid and applies to this type.
 482                  */
 483                 if (prop == ZPROP_INVAL) {
 484                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 485                             "invalid property '%s'"), propname);
 486                         (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 487                         goto error;
 488                 }
 489
 490                 if (zpool_prop_readonly(prop)) {
 491                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 492                             "is readonly"), propname);
 493                         (void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
 494                         goto error;
 495                 }
 496
 497                 if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
 498                     &strval, &intval, errbuf) != 0)
 499                         goto error;
 500
 501                 /*
 502                  * Perform additional checking for specific properties.
 503                  */
 504                 switch (prop) {
 505                 case ZPOOL_PROP_VERSION:
 506                         if (intval < version ||
 507                             !SPA_VERSION_IS_SUPPORTED(intval)) {
 508                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 509                                     "property '%s' number %d is invalid."),
 510                                     propname, intval);
 511                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 512                                 goto error;
 513                         }
 514                         break;
 515
 516                 case ZPOOL_PROP_BOOTFS:
 517                         if (flags.create || flags.import) {
 518                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 519                                     "property '%s' cannot be set at creation "
 520                                     "or import time"), propname);
 521                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 522                                 goto error;
 523                         }
 524
 525                         if (version < SPA_VERSION_BOOTFS) {
 526                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 527                                     "pool must be upgraded to support "
 528                                     "'%s' property"), propname);
 529                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 530                                 goto error;
 531                         }
 532
 533                         /*
 534                          * bootfs property value has to be a dataset name and
 535                          * the dataset has to be in the same pool as it sets to.
 536                          */
 537                         if (strval[0] != '\0' && !bootfs_name_valid(poolname,
 538                             strval)) {
 539                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 540                                     "is an invalid name"), strval);
 541                                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
 542                                 goto error;
 543                         }
 544
 545                         if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
 546                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 547                                     "could not open pool '%s'"), poolname);
 548                                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
 549                                 goto error;
 550                         }
 551                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 552                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 553
 554 #ifdef sun
 555                         /*
 556                          * bootfs property cannot be set on a disk which has
 557                          * been EFI labeled.
 558                          */
 559                         if (pool_uses_efi(nvroot)) {
 560                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 561                                     "property '%s' not supported on "
 562                                     "EFI labeled devices"), propname);
 563                                 (void) zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf);
 564                                 zpool_close(zhp);
 565                                 goto error;
 566                         }
 567 #endif  /* sun */
 568                         zpool_close(zhp);
 569                         break;
 570
 571                 case ZPOOL_PROP_ALTROOT:
 572                         if (!flags.create && !flags.import) {
 573                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 574                                     "property '%s' can only be set during pool "
 575                                     "creation or import"), propname);
 576                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 577                                 goto error;
 578                         }
 579
 580                         if (strval[0] != '/') {
 581                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 582                                     "bad alternate root '%s'"), strval);
 583                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 584                                 goto error;
 585                         }
 586                         break;
 587
 588                 case ZPOOL_PROP_CACHEFILE:
 589                         if (strval[0] == '\0')
 590                                 break;
 591
 592                         if (strcmp(strval, "none") == 0)
 593                                 break;
 594
 595                         if (strval[0] != '/') {
 596                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 597                                     "property '%s' must be empty, an "
 598                                     "absolute path, or 'none'"), propname);
 599                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 600                                 goto error;
 601                         }
 602
 603                         slash = strrchr(strval, '/');
 604
 605                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 606                             strcmp(slash, "/..") == 0) {
 607                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 608                                     "'%s' is not a valid file"), strval);
 609                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 610                                 goto error;
 611                         }
 612
 613                         *slash = '\0';
 614
 615                         if (strval[0] != '\0' &&
 616                             (stat64(strval, &statbuf) != 0 ||
 617                             !S_ISDIR(statbuf.st_mode))) {
 618                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 619                                     "'%s' is not a valid directory"),
 620                                     strval);
 621                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 622                                 goto error;
 623                         }
 624
 625                         *slash = '/';
 626                         break;
 627
 628                 case ZPOOL_PROP_COMMENT:
 629                         for (check = strval; *check != '\0'; check++) {
 630                                 if (!isprint(*check)) {
 631                                         zfs_error_aux(hdl,
 632                                             dgettext(TEXT_DOMAIN,
 633                                             "comment may only have printable "
 634                                             "characters"));
 635                                         (void) zfs_error(hdl, EZFS_BADPROP,
 636                                             errbuf);
 637                                         goto error;
 638                                 }
 639                         }
 640                         if (strlen(strval) > ZPROP_MAX_COMMENT) {
 641                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 642                                     "comment must not exceed %d characters"),
 643                                     ZPROP_MAX_COMMENT);
 644                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 645                                 goto error;
 646                         }
 647                         break;
 648                 case ZPOOL_PROP_READONLY:
 649                         if (!flags.import) {
 650                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 651                                     "property '%s' can only be set at "
 652                                     "import time"), propname);
 653                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 654                                 goto error;
 655                         }
 656                         break;
 657                 }
 658         }
 659
 660         return (retprops);
 661 error:
 662         nvlist_free(retprops);
 663         return (NULL);
 664 }
 665
 666 /*
 667  * Set zpool property : propname=propval.
 668  */
 669 int
 670 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
 671 {
 672         zfs_cmd_t zc = { 0 };
 673         int ret = -1;
 674         char errbuf[1024];
 675         nvlist_t *nvl = NULL;
 676         nvlist_t *realprops;
 677         uint64_t version;
 678         prop_flags_t flags = { 0 };
 679
 680         (void) snprintf(errbuf, sizeof (errbuf),
 681             dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
 682             zhp->zpool_name);
 683
 684         if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
 685                 return (no_memory(zhp->zpool_hdl));
 686
 687         if (nvlist_add_string(nvl, propname, propval) != 0) {
 688                 nvlist_free(nvl);
 689                 return (no_memory(zhp->zpool_hdl));
 690         }
 691
 692         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
 693         if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
 694             zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
 695                 nvlist_free(nvl);
 696                 return (-1);
 697         }
 698
 699         nvlist_free(nvl);
 700         nvl = realprops;
 701
 702         /*
 703          * Execute the corresponding ioctl() to set this property.
 704          */
 705         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 706
 707         if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
 708                 nvlist_free(nvl);
 709                 return (-1);
 710         }
 711
 712         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
 713
 714         zcmd_free_nvlists(&zc);
 715         nvlist_free(nvl);
 716
 717         if (ret)
 718                 (void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
 719         else
 720                 (void) zpool_props_refresh(zhp);
 721
 722         return (ret);
 723 }
 724
 725 int
 726 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
 727 {
 728         libzfs_handle_t *hdl = zhp->zpool_hdl;
 729         zprop_list_t *entry;
 730         char buf[ZFS_MAXPROPLEN];
 731         nvlist_t *features = NULL;
 732         zprop_list_t **last;
 733         boolean_t firstexpand = (NULL == *plp);
 734
 735         if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
 736                 return (-1);
 737
 738         last = plp;
 739         while (*last != NULL)
 740                 last = &(*last)->pl_next;
 741
 742         if ((*plp)->pl_all)
 743                 features = zpool_get_features(zhp);
 744
 745         if ((*plp)->pl_all && firstexpand) {
 746                 for (int i = 0; i < SPA_FEATURES; i++) {
 747                         zprop_list_t *entry = zfs_alloc(hdl,
 748                             sizeof (zprop_list_t));
 749                         entry->pl_prop = ZPROP_INVAL;
 750                         entry->pl_user_prop = zfs_asprintf(hdl, "feature@%s",
 751                             spa_feature_table[i].fi_uname);
 752                         entry->pl_width = strlen(entry->pl_user_prop);
 753                         entry->pl_all = B_TRUE;
 754
 755                         *last = entry;
 756                         last = &entry->pl_next;
 757                 }
 758         }
 759
 760         /* add any unsupported features */
 761         for (nvpair_t *nvp = nvlist_next_nvpair(features, NULL);
 762             nvp != NULL; nvp = nvlist_next_nvpair(features, nvp)) {
 763                 char *propname;
 764                 boolean_t found;
 765                 zprop_list_t *entry;
 766
 767                 if (zfeature_is_supported(nvpair_name(nvp)))
 768                         continue;
 769
 770                 propname = zfs_asprintf(hdl, "unsupported@%s",
 771                     nvpair_name(nvp));
 772
 773                 /*
 774                  * Before adding the property to the list make sure that no
 775                  * other pool already added the same property.
 776                  */
 777                 found = B_FALSE;
 778                 entry = *plp;
 779                 while (entry != NULL) {
 780                         if (entry->pl_user_prop != NULL &&
 781                             strcmp(propname, entry->pl_user_prop) == 0) {
 782                                 found = B_TRUE;
 783                                 break;
 784                         }
 785                         entry = entry->pl_next;
 786                 }
 787                 if (found) {
 788                         free(propname);
 789                         continue;
 790                 }
 791
 792                 entry = zfs_alloc(hdl, sizeof (zprop_list_t));
 793                 entry->pl_prop = ZPROP_INVAL;
 794                 entry->pl_user_prop = propname;
 795                 entry->pl_width = strlen(entry->pl_user_prop);
 796                 entry->pl_all = B_TRUE;
 797
 798                 *last = entry;
 799                 last = &entry->pl_next;
 800         }
 801
 802         for (entry = *plp; entry != NULL; entry = entry->pl_next) {
 803
 804                 if (entry->pl_fixed)
 805                         continue;
 806
 807                 if (entry->pl_prop != ZPROP_INVAL &&
 808                     zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
 809                     NULL) == 0) {
 810                         if (strlen(buf) > entry->pl_width)
 811                                 entry->pl_width = strlen(buf);
 812                 }
 813         }
 814
 815         return (0);
 816 }
 817
 818 /*
 819  * Get the state for the given feature on the given ZFS pool.
 820  */
 821 int
 822 zpool_prop_get_feature(zpool_handle_t *zhp, const char *propname, char *buf,
 823     size_t len)
 824 {
 825         uint64_t refcount;
 826         boolean_t found = B_FALSE;
 827         nvlist_t *features = zpool_get_features(zhp);
 828         boolean_t supported;
 829         const char *feature = strchr(propname, '@') + 1;
 830
 831         supported = zpool_prop_feature(propname);
 832         ASSERT(supported || zpool_prop_unsupported(propname));
 833
 834         /*
 835          * Convert from feature name to feature guid. This conversion is
 836          * unecessary for unsupported@... properties because they already
 837          * use guids.
 838          */
 839         if (supported) {
 840                 int ret;
 841                 zfeature_info_t *fi;
 842
 843                 ret = zfeature_lookup_name(feature, &fi);
 844                 if (ret != 0) {
 845                         (void) strlcpy(buf, "-", len);
 846                         return (ENOTSUP);
 847                 }
 848                 feature = fi->fi_guid;
 849         }
 850
 851         if (nvlist_lookup_uint64(features, feature, &refcount) == 0)
 852                 found = B_TRUE;
 853
 854         if (supported) {
 855                 if (!found) {
 856                         (void) strlcpy(buf, ZFS_FEATURE_DISABLED, len);
 857                 } else  {
 858                         if (refcount == 0)
 859                                 (void) strlcpy(buf, ZFS_FEATURE_ENABLED, len);
 860                         else
 861                                 (void) strlcpy(buf, ZFS_FEATURE_ACTIVE, len);
 862                 }
 863         } else {
 864                 if (found) {
 865                         if (refcount == 0) {
 866                                 (void) strcpy(buf, ZFS_UNSUPPORTED_INACTIVE);
 867                         } else {
 868                                 (void) strcpy(buf, ZFS_UNSUPPORTED_READONLY);
 869                         }
 870                 } else {
 871                         (void) strlcpy(buf, "-", len);
 872                         return (ENOTSUP);
 873                 }
 874         }
 875
 876         return (0);
 877 }
 878
 879 /*
 880  * Don't start the slice at the default block of 34; many storage
 881  * devices will use a stripe width of 128k, so start there instead.
 882  */
 883 #define NEW_START_BLOCK 256
 884
 885 /*
 886  * Validate the given pool name, optionally putting an extended error message in
 887  * 'buf'.
 888  */
 889 boolean_t
 890 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
 891 {
 892         namecheck_err_t why;
 893         char what;
 894         int ret;
 895
 896         ret = pool_namecheck(pool, &why, &what);
 897
 898         /*
 899          * The rules for reserved pool names were extended at a later point.
 900          * But we need to support users with existing pools that may now be
 901          * invalid.  So we only check for this expanded set of names during a
 902          * create (or import), and only in userland.
 903          */
 904         if (ret == 0 && !isopen &&
 905             (strncmp(pool, "mirror", 6) == 0 ||
 906             strncmp(pool, "raidz", 5) == 0 ||
 907             strncmp(pool, "spare", 5) == 0 ||
 908             strcmp(pool, "log") == 0)) {
 909                 if (hdl != NULL)
 910                         zfs_error_aux(hdl,
 911                             dgettext(TEXT_DOMAIN, "name is reserved"));
 912                 return (B_FALSE);
 913         }
 914
 915
 916         if (ret != 0) {
 917                 if (hdl != NULL) {
 918                         switch (why) {
 919                         case NAME_ERR_TOOLONG:
 920                                 zfs_error_aux(hdl,
 921                                     dgettext(TEXT_DOMAIN, "name is too long"));
 922                                 break;
 923
 924                         case NAME_ERR_INVALCHAR:
 925                                 zfs_error_aux(hdl,
 926                                     dgettext(TEXT_DOMAIN, "invalid character "
 927                                     "'%c' in pool name"), what);
 928                                 break;
 929
 930                         case NAME_ERR_NOLETTER:
 931                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 932                                     "name must begin with a letter"));
 933                                 break;
 934
 935                         case NAME_ERR_RESERVED:
 936                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 937                                     "name is reserved"));
 938                                 break;
 939
 940                         case NAME_ERR_DISKLIKE:
 941                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 942                                     "pool name is reserved"));
 943                                 break;
 944
 945                         case NAME_ERR_LEADING_SLASH:
 946                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 947                                     "leading slash in name"));
 948                                 break;
 949
 950                         case NAME_ERR_EMPTY_COMPONENT:
 951                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 952                                     "empty component in name"));
 953                                 break;
 954
 955                         case NAME_ERR_TRAILING_SLASH:
 956                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 957                                     "trailing slash in name"));
 958                                 break;
 959
 960                         case NAME_ERR_MULTIPLE_AT:
 961                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 962                                     "multiple '@' delimiters in name"));
 963                                 break;
 964
 965                         }
 966                 }
 967                 return (B_FALSE);
 968         }
 969
 970         return (B_TRUE);
 971 }
 972
 973 /*
 974  * Open a handle to the given pool, even if the pool is currently in the FAULTED
 975  * state.
 976  */
 977 zpool_handle_t *
 978 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
 979 {
 980         zpool_handle_t *zhp;
 981         boolean_t missing;
 982
 983         /*
 984          * Make sure the pool name is valid.
 985          */
 986         if (!zpool_name_valid(hdl, B_TRUE, pool)) {
 987                 (void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
 988                     dgettext(TEXT_DOMAIN, "cannot open '%s'"),
 989                     pool);
 990                 return (NULL);
 991         }
 992
 993         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
 994                 return (NULL);
 995
 996         zhp->zpool_hdl = hdl;
 997         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
 998
 999         if (zpool_refresh_stats(zhp, &missing) != 0) {
1000                 zpool_close(zhp);
1001                 return (NULL);
1002         }
1003
1004         if (missing) {
1005                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
1006                 (void) zfs_error_fmt(hdl, EZFS_NOENT,
1007                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
1008                 zpool_close(zhp);
1009                 return (NULL);
1010         }
1011
1012         return (zhp);
1013 }
1014
1015 /*
1016  * Like the above, but silent on error.  Used when iterating over pools (because
1017  * the configuration cache may be out of date).
1018  */
1019 int
1020 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
1021 {
1022         zpool_handle_t *zhp;
1023         boolean_t missing;
1024
1025         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1026                 return (-1);
1027
1028         zhp->zpool_hdl = hdl;
1029         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1030
1031         if (zpool_refresh_stats(zhp, &missing) != 0) {
1032                 zpool_close(zhp);
1033                 return (-1);
1034         }
1035
1036         if (missing) {
1037                 zpool_close(zhp);
1038                 *ret = NULL;
1039                 return (0);
1040         }
1041
1042         *ret = zhp;
1043         return (0);
1044 }
1045
1046 /*
1047  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
1048  * state.
1049  */
1050 zpool_handle_t *
1051 zpool_open(libzfs_handle_t *hdl, const char *pool)
1052 {
1053         zpool_handle_t *zhp;
1054
1055         if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
1056                 return (NULL);
1057
1058         if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
1059                 (void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
1060                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
1061                 zpool_close(zhp);
1062                 return (NULL);
1063         }
1064
1065         return (zhp);
1066 }
1067
1068 /*
1069  * Close the handle.  Simply frees the memory associated with the handle.
1070  */
1071 void
1072 zpool_close(zpool_handle_t *zhp)
1073 {
1074         if (zhp->zpool_config)
1075                 nvlist_free(zhp->zpool_config);
1076         if (zhp->zpool_old_config)
1077                 nvlist_free(zhp->zpool_old_config);
1078         if (zhp->zpool_props)
1079                 nvlist_free(zhp->zpool_props);
1080         free(zhp);
1081 }
1082
1083 /*
1084  * Return the name of the pool.
1085  */
1086 const char *
1087 zpool_get_name(zpool_handle_t *zhp)
1088 {
1089         return (zhp->zpool_name);
1090 }
1091
1092
1093 /*
1094  * Return the state of the pool (ACTIVE or UNAVAILABLE)
1095  */
1096 int
1097 zpool_get_state(zpool_handle_t *zhp)
1098 {
1099         return (zhp->zpool_state);
1100 }
1101
1102 /*
1103  * Create the named pool, using the provided vdev list.  It is assumed
1104  * that the consumer has already validated the contents of the nvlist, so we
1105  * don't have to worry about error semantics.
1106  */
1107 int
1108 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
1109     nvlist_t *props, nvlist_t *fsprops)
1110 {
1111         zfs_cmd_t zc = { 0 };
1112         nvlist_t *zc_fsprops = NULL;
1113         nvlist_t *zc_props = NULL;
1114         char msg[1024];
1115         int ret = -1;
1116
1117         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1118             "cannot create '%s'"), pool);
1119
1120         if (!zpool_name_valid(hdl, B_FALSE, pool))
1121                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
1122
1123         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1124                 return (-1);
1125
1126         if (props) {
1127                 prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
1128
1129                 if ((zc_props = zpool_valid_proplist(hdl, pool, props,
1130                     SPA_VERSION_1, flags, msg)) == NULL) {
1131                         goto create_failed;
1132                 }
1133         }
1134
1135         if (fsprops) {
1136                 uint64_t zoned;
1137                 char *zonestr;
1138
1139                 zoned = ((nvlist_lookup_string(fsprops,
1140                     zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
1141                     strcmp(zonestr, "on") == 0);
1142
1143                 if ((zc_fsprops = zfs_valid_proplist(hdl,
1144                     ZFS_TYPE_FILESYSTEM, fsprops, zoned, NULL, msg)) == NULL) {
1145                         goto create_failed;
1146                 }
1147                 if (!zc_props &&
1148                     (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
1149                         goto create_failed;
1150                 }
1151                 if (nvlist_add_nvlist(zc_props,
1152                     ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
1153                         goto create_failed;
1154                 }
1155         }
1156
1157         if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
1158                 goto create_failed;
1159
1160         (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
1161
1162         if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
1163
1164                 zcmd_free_nvlists(&zc);
1165                 nvlist_free(zc_props);
1166                 nvlist_free(zc_fsprops);
1167
1168                 switch (errno) {
1169                 case EBUSY:
1170                         /*
1171                          * This can happen if the user has specified the same
1172                          * device multiple times.  We can't reliably detect this
1173                          * until we try to add it and see we already have a
1174                          * label.
1175                          */
1176                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1177                             "one or more vdevs refer to the same device"));
1178                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1179
1180                 case EOVERFLOW:
1181                         /*
1182                          * This occurs when one of the devices is below
1183                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1184                          * device was the problem device since there's no
1185                          * reliable way to determine device size from userland.
1186                          */
1187                         {
1188                                 char buf[64];
1189
1190                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1191
1192                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1193                                     "one or more devices is less than the "
1194                                     "minimum size (%s)"), buf);
1195                         }
1196                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1197
1198                 case ENOSPC:
1199                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1200                             "one or more devices is out of space"));
1201                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1202
1203                 case ENOTBLK:
1204                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1205                             "cache device must be a disk or disk slice"));
1206                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1207
1208                 default:
1209                         return (zpool_standard_error(hdl, errno, msg));
1210                 }
1211         }
1212
1213 create_failed:
1214         zcmd_free_nvlists(&zc);
1215         nvlist_free(zc_props);
1216         nvlist_free(zc_fsprops);
1217         return (ret);
1218 }
1219
1220 /*
1221  * Destroy the given pool.  It is up to the caller to ensure that there are no
1222  * datasets left in the pool.
1223  */
1224 int
1225 zpool_destroy(zpool_handle_t *zhp, const char *log_str)
1226 {
1227         zfs_cmd_t zc = { 0 };
1228         zfs_handle_t *zfp = NULL;
1229         libzfs_handle_t *hdl = zhp->zpool_hdl;
1230         char msg[1024];
1231
1232         if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1233             (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1234                 return (-1);
1235
1236         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1237         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1238
1239         if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1240                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1241                     "cannot destroy '%s'"), zhp->zpool_name);
1242
1243                 if (errno == EROFS) {
1244                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1245                             "one or more devices is read only"));
1246                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1247                 } else {
1248                         (void) zpool_standard_error(hdl, errno, msg);
1249                 }
1250
1251                 if (zfp)
1252                         zfs_close(zfp);
1253                 return (-1);
1254         }
1255
1256         if (zfp) {
1257                 remove_mountpoint(zfp);
1258                 zfs_close(zfp);
1259         }
1260
1261         return (0);
1262 }
1263
1264 /*
1265  * Add the given vdevs to the pool.  The caller must have already performed the
1266  * necessary verification to ensure that the vdev specification is well-formed.
1267  */
1268 int
1269 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1270 {
1271         zfs_cmd_t zc = { 0 };
1272         int ret;
1273         libzfs_handle_t *hdl = zhp->zpool_hdl;
1274         char msg[1024];
1275         nvlist_t **spares, **l2cache;
1276         uint_t nspares, nl2cache;
1277
1278         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1279             "cannot add to '%s'"), zhp->zpool_name);
1280
1281         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1282             SPA_VERSION_SPARES &&
1283             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1284             &spares, &nspares) == 0) {
1285                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1286                     "upgraded to add hot spares"));
1287                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1288         }
1289
1290         if (zpool_is_bootable(zhp) && nvlist_lookup_nvlist_array(nvroot,
1291             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) {
1292                 uint64_t s;
1293
1294                 for (s = 0; s < nspares; s++) {
1295                         char *path;
1296
1297                         if (nvlist_lookup_string(spares[s], ZPOOL_CONFIG_PATH,
1298                             &path) == 0 && pool_uses_efi(spares[s])) {
1299                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1300                                     "device '%s' contains an EFI label and "
1301                                     "cannot be used on root pools."),
1302                                     zpool_vdev_name(hdl, NULL, spares[s],
1303                                     B_FALSE));
1304                                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
1305                         }
1306                 }
1307         }
1308
1309         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1310             SPA_VERSION_L2CACHE &&
1311             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1312             &l2cache, &nl2cache) == 0) {
1313                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1314                     "upgraded to add cache devices"));
1315                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1316         }
1317
1318         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1319                 return (-1);
1320         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1321
1322         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1323                 switch (errno) {
1324                 case EBUSY:
1325                         /*
1326                          * This can happen if the user has specified the same
1327                          * device multiple times.  We can't reliably detect this
1328                          * until we try to add it and see we already have a
1329                          * label.
1330                          */
1331                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1332                             "one or more vdevs refer to the same device"));
1333                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1334                         break;
1335
1336                 case EOVERFLOW:
1337                         /*
1338                          * This occurrs when one of the devices is below
1339                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1340                          * device was the problem device since there's no
1341                          * reliable way to determine device size from userland.
1342                          */
1343                         {
1344                                 char buf[64];
1345
1346                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1347
1348                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1349                                     "device is less than the minimum "
1350                                     "size (%s)"), buf);
1351                         }
1352                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1353                         break;
1354
1355                 case ENOTSUP:
1356                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1357                             "pool must be upgraded to add these vdevs"));
1358                         (void) zfs_error(hdl, EZFS_BADVERSION, msg);
1359                         break;
1360
1361                 case EDOM:
1362                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1363                             "root pool can not have multiple vdevs"
1364                             " or separate logs"));
1365                         (void) zfs_error(hdl, EZFS_POOL_NOTSUP, msg);
1366                         break;
1367
1368                 case ENOTBLK:
1369                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1370                             "cache device must be a disk or disk slice"));
1371                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1372                         break;
1373
1374                 default:
1375                         (void) zpool_standard_error(hdl, errno, msg);
1376                 }
1377
1378                 ret = -1;
1379         } else {
1380                 ret = 0;
1381         }
1382
1383         zcmd_free_nvlists(&zc);
1384
1385         return (ret);
1386 }
1387
1388 /*
1389  * Exports the pool from the system.  The caller must ensure that there are no
1390  * mounted datasets in the pool.
1391  */
1392 static int
1393 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce,
1394     const char *log_str)
1395 {
1396         zfs_cmd_t zc = { 0 };
1397         char msg[1024];
1398
1399         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1400             "cannot export '%s'"), zhp->zpool_name);
1401
1402         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1403         zc.zc_cookie = force;
1404         zc.zc_guid = hardforce;
1405         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1406
1407         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1408                 switch (errno) {
1409                 case EXDEV:
1410                         zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1411                             "use '-f' to override the following errors:\n"
1412                             "'%s' has an active shared spare which could be"
1413                             " used by other pools once '%s' is exported."),
1414                             zhp->zpool_name, zhp->zpool_name);
1415                         return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1416                             msg));
1417                 default:
1418                         return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1419                             msg));
1420                 }
1421         }
1422
1423         return (0);
1424 }
1425
1426 int
1427 zpool_export(zpool_handle_t *zhp, boolean_t force, const char *log_str)
1428 {
1429         return (zpool_export_common(zhp, force, B_FALSE, log_str));
1430 }
1431
1432 int
1433 zpool_export_force(zpool_handle_t *zhp, const char *log_str)
1434 {
1435         return (zpool_export_common(zhp, B_TRUE, B_TRUE, log_str));
1436 }
1437
1438 static void
1439 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1440     nvlist_t *config)
1441 {
1442         nvlist_t *nv = NULL;
1443         uint64_t rewindto;
1444         int64_t loss = -1;
1445         struct tm t;
1446         char timestr[128];
1447
1448         if (!hdl->libzfs_printerr || config == NULL)
1449                 return;
1450
1451         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1452             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0) {
1453                 return;
1454         }
1455
1456         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1457                 return;
1458         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1459
1460         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1461             strftime(timestr, 128, 0, &t) != 0) {
1462                 if (dryrun) {
1463                         (void) printf(dgettext(TEXT_DOMAIN,
1464                             "Would be able to return %s "
1465                             "to its state as of %s.\n"),
1466                             name, timestr);
1467                 } else {
1468                         (void) printf(dgettext(TEXT_DOMAIN,
1469                             "Pool %s returned to its state as of %s.\n"),
1470                             name, timestr);
1471                 }
1472                 if (loss > 120) {
1473                         (void) printf(dgettext(TEXT_DOMAIN,
1474                             "%s approximately %lld "),
1475                             dryrun ? "Would discard" : "Discarded",
1476                             (loss + 30) / 60);
1477                         (void) printf(dgettext(TEXT_DOMAIN,
1478                             "minutes of transactions.\n"));
1479                 } else if (loss > 0) {
1480                         (void) printf(dgettext(TEXT_DOMAIN,
1481                             "%s approximately %lld "),
1482                             dryrun ? "Would discard" : "Discarded", loss);
1483                         (void) printf(dgettext(TEXT_DOMAIN,
1484                             "seconds of transactions.\n"));
1485                 }
1486         }
1487 }
1488
1489 void
1490 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1491     nvlist_t *config)
1492 {
1493         nvlist_t *nv = NULL;
1494         int64_t loss = -1;
1495         uint64_t edata = UINT64_MAX;
1496         uint64_t rewindto;
1497         struct tm t;
1498         char timestr[128];
1499
1500         if (!hdl->libzfs_printerr)
1501                 return;
1502
1503         if (reason >= 0)
1504                 (void) printf(dgettext(TEXT_DOMAIN, "action: "));
1505         else
1506                 (void) printf(dgettext(TEXT_DOMAIN, "\t"));
1507
1508         /* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1509         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1510             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0 ||
1511             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1512                 goto no_info;
1513
1514         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1515         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1516             &edata);
1517
1518         (void) printf(dgettext(TEXT_DOMAIN,
1519             "Recovery is possible, but will result in some data loss.\n"));
1520
1521         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1522             strftime(timestr, 128, 0, &t) != 0) {
1523                 (void) printf(dgettext(TEXT_DOMAIN,
1524                     "\tReturning the pool to its state as of %s\n"
1525                     "\tshould correct the problem.  "),
1526                     timestr);
1527         } else {
1528                 (void) printf(dgettext(TEXT_DOMAIN,
1529                     "\tReverting the pool to an earlier state "
1530                     "should correct the problem.\n\t"));
1531         }
1532
1533         if (loss > 120) {
1534                 (void) printf(dgettext(TEXT_DOMAIN,
1535                     "Approximately %lld minutes of data\n"
1536                     "\tmust be discarded, irreversibly.  "), (loss + 30) / 60);
1537         } else if (loss > 0) {
1538                 (void) printf(dgettext(TEXT_DOMAIN,
1539                     "Approximately %lld seconds of data\n"
1540                     "\tmust be discarded, irreversibly.  "), loss);
1541         }
1542         if (edata != 0 && edata != UINT64_MAX) {
1543                 if (edata == 1) {
1544                         (void) printf(dgettext(TEXT_DOMAIN,
1545                             "After rewind, at least\n"
1546                             "\tone persistent user-data error will remain.  "));
1547                 } else {
1548                         (void) printf(dgettext(TEXT_DOMAIN,
1549                             "After rewind, several\n"
1550                             "\tpersistent user-data errors will remain.  "));
1551                 }
1552         }
1553         (void) printf(dgettext(TEXT_DOMAIN,
1554             "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1555             reason >= 0 ? "clear" : "import", name);
1556
1557         (void) printf(dgettext(TEXT_DOMAIN,
1558             "A scrub of the pool\n"
1559             "\tis strongly recommended after recovery.\n"));
1560         return;
1561
1562 no_info:
1563         (void) printf(dgettext(TEXT_DOMAIN,
1564             "Destroy and re-create the pool from\n\ta backup source.\n"));
1565 }
1566
1567 /*
1568  * zpool_import() is a contracted interface. Should be kept the same
1569  * if possible.
1570  *
1571  * Applications should use zpool_import_props() to import a pool with
1572  * new properties value to be set.
1573  */
1574 int
1575 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1576     char *altroot)
1577 {
1578         nvlist_t *props = NULL;
1579         int ret;
1580
1581         if (altroot != NULL) {
1582                 if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1583                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1584                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1585                             newname));
1586                 }
1587
1588                 if (nvlist_add_string(props,
1589                     zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1590                     nvlist_add_string(props,
1591                     zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1592                         nvlist_free(props);
1593                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1594                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1595                             newname));
1596                 }
1597         }
1598
1599         ret = zpool_import_props(hdl, config, newname, props,
1600             ZFS_IMPORT_NORMAL);
1601         if (props)
1602                 nvlist_free(props);
1603         return (ret);
1604 }
1605
1606 static void
1607 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1608     int indent)
1609 {
1610         nvlist_t **child;
1611         uint_t c, children;
1612         char *vname;
1613         uint64_t is_log = 0;
1614
1615         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1616             &is_log);
1617
1618         if (name != NULL)
1619                 (void) printf("\t%*s%s%s\n", indent, "", name,
1620                     is_log ? " [log]" : "");
1621
1622         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1623             &child, &children) != 0)
1624                 return;
1625
1626         for (c = 0; c < children; c++) {
1627                 vname = zpool_vdev_name(hdl, NULL, child[c], B_TRUE);
1628                 print_vdev_tree(hdl, vname, child[c], indent + 2);
1629                 free(vname);
1630         }
1631 }
1632
1633 void
1634 zpool_print_unsup_feat(nvlist_t *config)
1635 {
1636         nvlist_t *nvinfo, *unsup_feat;
1637
1638         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nvinfo) ==
1639             0);
1640         verify(nvlist_lookup_nvlist(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT,
1641             &unsup_feat) == 0);
1642
1643         for (nvpair_t *nvp = nvlist_next_nvpair(unsup_feat, NULL); nvp != NULL;
1644             nvp = nvlist_next_nvpair(unsup_feat, nvp)) {
1645                 char *desc;
1646
1647                 verify(nvpair_type(nvp) == DATA_TYPE_STRING);
1648                 verify(nvpair_value_string(nvp, &desc) == 0);
1649
1650                 if (strlen(desc) > 0)
1651                         (void) printf("\t%s (%s)\n", nvpair_name(nvp), desc);
1652                 else
1653                         (void) printf("\t%s\n", nvpair_name(nvp));
1654         }
1655 }
1656
1657 /*
1658  * Import the given pool using the known configuration and a list of
1659  * properties to be set. The configuration should have come from
1660  * zpool_find_import(). The 'newname' parameters control whether the pool
1661  * is imported with a different name.
1662  */
1663 int
1664 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1665     nvlist_t *props, int flags)
1666 {
1667         zfs_cmd_t zc = { 0 };
1668         zpool_rewind_policy_t policy;
1669         nvlist_t *nv = NULL;
1670         nvlist_t *nvinfo = NULL;
1671         nvlist_t *missing = NULL;
1672         char *thename;
1673         char *origname;
1674         int ret;
1675         int error = 0;
1676         char errbuf[1024];
1677
1678         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1679             &origname) == 0);
1680
1681         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1682             "cannot import pool '%s'"), origname);
1683
1684         if (newname != NULL) {
1685                 if (!zpool_name_valid(hdl, B_FALSE, newname))
1686                         return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1687                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1688                             newname));
1689                 thename = (char *)newname;
1690         } else {
1691                 thename = origname;
1692         }
1693
1694         if (props) {
1695                 uint64_t version;
1696                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1697
1698                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1699                     &version) == 0);
1700
1701                 if ((props = zpool_valid_proplist(hdl, origname,
1702                     props, version, flags, errbuf)) == NULL) {
1703                         return (-1);
1704                 } else if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1705                         nvlist_free(props);
1706                         return (-1);
1707                 }
1708         }
1709
1710         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1711
1712         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1713             &zc.zc_guid) == 0);
1714
1715         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1716                 nvlist_free(props);
1717                 return (-1);
1718         }
1719         if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1720                 nvlist_free(props);
1721                 return (-1);
1722         }
1723
1724         zc.zc_cookie = flags;
1725         while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1726             errno == ENOMEM) {
1727                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1728                         zcmd_free_nvlists(&zc);
1729                         return (-1);
1730                 }
1731         }
1732         if (ret != 0)
1733                 error = errno;
1734
1735         (void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1736         zpool_get_rewind_policy(config, &policy);
1737
1738         if (error) {
1739                 char desc[1024];
1740
1741                 /*
1742                  * Dry-run failed, but we print out what success
1743                  * looks like if we found a best txg
1744                  */
1745                 if (policy.zrp_request & ZPOOL_TRY_REWIND) {
1746                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1747                             B_TRUE, nv);
1748                         nvlist_free(nv);
1749                         return (-1);
1750                 }
1751
1752                 if (newname == NULL)
1753                         (void) snprintf(desc, sizeof (desc),
1754                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1755                             thename);
1756                 else
1757                         (void) snprintf(desc, sizeof (desc),
1758                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1759                             origname, thename);
1760
1761                 switch (error) {
1762                 case ENOTSUP:
1763                         if (nv != NULL && nvlist_lookup_nvlist(nv,
1764                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1765                             nvlist_exists(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT)) {
1766                                 (void) printf(dgettext(TEXT_DOMAIN, "This "
1767                                     "pool uses the following feature(s) not "
1768                                     "supported by this system:\n"));
1769                                 zpool_print_unsup_feat(nv);
1770                                 if (nvlist_exists(nvinfo,
1771                                     ZPOOL_CONFIG_CAN_RDONLY)) {
1772                                         (void) printf(dgettext(TEXT_DOMAIN,
1773                                             "All unsupported features are only "
1774                                             "required for writing to the pool."
1775                                             "\nThe pool can be imported using "
1776                                             "'-o readonly=on'.\n"));
1777                                 }
1778                         }
1779                         /*
1780                          * Unsupported version.
1781                          */
1782                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1783                         break;
1784
1785                 case EINVAL:
1786                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1787                         break;
1788
1789                 case EROFS:
1790                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1791                             "one or more devices is read only"));
1792                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1793                         break;
1794
1795                 case ENXIO:
1796                         if (nv && nvlist_lookup_nvlist(nv,
1797                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1798                             nvlist_lookup_nvlist(nvinfo,
1799                             ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1800                                 (void) printf(dgettext(TEXT_DOMAIN,
1801                                     "The devices below are missing, use "
1802                                     "'-m' to import the pool anyway:\n"));
1803                                 print_vdev_tree(hdl, NULL, missing, 2);
1804                                 (void) printf("\n");
1805                         }
1806                         (void) zpool_standard_error(hdl, error, desc);
1807                         break;
1808
1809                 case EEXIST:
1810                         (void) zpool_standard_error(hdl, error, desc);
1811                         break;
1812
1813                 default:
1814                         (void) zpool_standard_error(hdl, error, desc);
1815                         zpool_explain_recover(hdl,
1816                             newname ? origname : thename, -error, nv);
1817                         break;
1818                 }
1819
1820                 nvlist_free(nv);
1821                 ret = -1;
1822         } else {
1823                 zpool_handle_t *zhp;
1824
1825                 /*
1826                  * This should never fail, but play it safe anyway.
1827                  */
1828                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1829                         ret = -1;
1830                 else if (zhp != NULL)
1831                         zpool_close(zhp);
1832                 if (policy.zrp_request &
1833                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1834                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1835                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1836                 }
1837                 nvlist_free(nv);
1838                 return (0);
1839         }
1840
1841         zcmd_free_nvlists(&zc);
1842         nvlist_free(props);
1843
1844         return (ret);
1845 }
1846
1847 /*
1848  * Scan the pool.
1849  */
1850 int
1851 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1852 {
1853         zfs_cmd_t zc = { 0 };
1854         char msg[1024];
1855         libzfs_handle_t *hdl = zhp->zpool_hdl;
1856
1857         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1858         zc.zc_cookie = func;
1859
1860         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1861             (errno == ENOENT && func != POOL_SCAN_NONE))
1862                 return (0);
1863
1864         if (func == POOL_SCAN_SCRUB) {
1865                 (void) snprintf(msg, sizeof (msg),
1866                     dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1867         } else if (func == POOL_SCAN_NONE) {
1868                 (void) snprintf(msg, sizeof (msg),
1869                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1870                     zc.zc_name);
1871         } else {
1872                 assert(!"unexpected result");
1873         }
1874
1875         if (errno == EBUSY) {
1876                 nvlist_t *nvroot;
1877                 pool_scan_stat_t *ps = NULL;
1878                 uint_t psc;
1879
1880                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
1881                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1882                 (void) nvlist_lookup_uint64_array(nvroot,
1883                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1884                 if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1885                         return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1886                 else
1887                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
1888         } else if (errno == ENOENT) {
1889                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1890         } else {
1891                 return (zpool_standard_error(hdl, errno, msg));
1892         }
1893 }
1894
1895 /*
1896  * This provides a very minimal check whether a given string is likely a
1897  * c#t#d# style string.  Users of this are expected to do their own
1898  * verification of the s# part.
1899  */
1900 #define CTD_CHECK(str)  (str && str[0] == 'c' && isdigit(str[1]))
1901
1902 /*
1903  * More elaborate version for ones which may start with "/dev/dsk/"
1904  * and the like.
1905  */
1906 static int
1907 ctd_check_path(char *str) {
1908         /*
1909          * If it starts with a slash, check the last component.
1910          */
1911         if (str && str[0] == '/') {
1912                 char *tmp = strrchr(str, '/');
1913
1914                 /*
1915                  * If it ends in "/old", check the second-to-last
1916                  * component of the string instead.
1917                  */
1918                 if (tmp != str && strcmp(tmp, "/old") == 0) {
1919                         for (tmp--; *tmp != '/'; tmp--)
1920                                 ;
1921                 }
1922                 str = tmp + 1;
1923         }
1924         return (CTD_CHECK(str));
1925 }
1926
1927 /*
1928  * Find a vdev that matches the search criteria specified. We use the
1929  * the nvpair name to determine how we should look for the device.
1930  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1931  * spare; but FALSE if its an INUSE spare.
1932  */
1933 static nvlist_t *
1934 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1935     boolean_t *l2cache, boolean_t *log)
1936 {
1937         uint_t c, children;
1938         nvlist_t **child;
1939         nvlist_t *ret;
1940         uint64_t is_log;
1941         char *srchkey;
1942         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
1943
1944         /* Nothing to look for */
1945         if (search == NULL || pair == NULL)
1946                 return (NULL);
1947
1948         /* Obtain the key we will use to search */
1949         srchkey = nvpair_name(pair);
1950
1951         switch (nvpair_type(pair)) {
1952         case DATA_TYPE_UINT64:
1953                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
1954                         uint64_t srchval, theguid;
1955
1956                         verify(nvpair_value_uint64(pair, &srchval) == 0);
1957                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
1958                             &theguid) == 0);
1959                         if (theguid == srchval)
1960                                 return (nv);
1961                 }
1962                 break;
1963
1964         case DATA_TYPE_STRING: {
1965                 char *srchval, *val;
1966
1967                 verify(nvpair_value_string(pair, &srchval) == 0);
1968                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
1969                         break;
1970
1971                 /*
1972                  * Search for the requested value. Special cases:
1973                  *
1974                  * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
1975                  *   "s0" or "s0/old".  The "s0" part is hidden from the user,
1976                  *   but included in the string, so this matches around it.
1977                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
1978                  *
1979                  * Otherwise, all other searches are simple string compares.
1980                  */
1981                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0 &&
1982                     ctd_check_path(val)) {
1983                         uint64_t wholedisk = 0;
1984
1985                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1986                             &wholedisk);
1987                         if (wholedisk) {
1988                                 int slen = strlen(srchval);
1989                                 int vlen = strlen(val);
1990
1991                                 if (slen != vlen - 2)
1992                                         break;
1993
1994                                 /*
1995                                  * make_leaf_vdev() should only set
1996                                  * wholedisk for ZPOOL_CONFIG_PATHs which
1997                                  * will include "/dev/dsk/", giving plenty of
1998                                  * room for the indices used next.
1999                                  */
2000                                 ASSERT(vlen >= 6);
2001
2002                                 /*
2003                                  * strings identical except trailing "s0"
2004                                  */
2005                                 if (strcmp(&val[vlen - 2], "s0") == 0 &&
2006                                     strncmp(srchval, val, slen) == 0)
2007                                         return (nv);
2008
2009                                 /*
2010                                  * strings identical except trailing "s0/old"
2011                                  */
2012                                 if (strcmp(&val[vlen - 6], "s0/old") == 0 &&
2013                                     strcmp(&srchval[slen - 4], "/old") == 0 &&
2014                                     strncmp(srchval, val, slen - 4) == 0)
2015                                         return (nv);
2016
2017                                 break;
2018                         }
2019                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2020                         char *type, *idx, *end, *p;
2021                         uint64_t id, vdev_id;
2022
2023                         /*
2024                          * Determine our vdev type, keeping in mind
2025                          * that the srchval is composed of a type and
2026                          * vdev id pair (i.e. mirror-4).
2027                          */
2028                         if ((type = strdup(srchval)) == NULL)
2029                                 return (NULL);
2030
2031                         if ((p = strrchr(type, '-')) == NULL) {
2032                                 free(type);
2033                                 break;
2034                         }
2035                         idx = p + 1;
2036                         *p = '\0';
2037
2038                         /*
2039                          * If the types don't match then keep looking.
2040                          */
2041                         if (strncmp(val, type, strlen(val)) != 0) {
2042                                 free(type);
2043                                 break;
2044                         }
2045
2046                         verify(strncmp(type, VDEV_TYPE_RAIDZ,
2047                             strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2048                             strncmp(type, VDEV_TYPE_MIRROR,
2049                             strlen(VDEV_TYPE_MIRROR)) == 0);
2050                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
2051                             &id) == 0);
2052
2053                         errno = 0;
2054                         vdev_id = strtoull(idx, &end, 10);
2055
2056                         free(type);
2057                         if (errno != 0)
2058                                 return (NULL);
2059
2060                         /*
2061                          * Now verify that we have the correct vdev id.
2062                          */
2063                         if (vdev_id == id)
2064                                 return (nv);
2065                 }
2066
2067                 /*
2068                  * Common case
2069                  */
2070                 if (strcmp(srchval, val) == 0)
2071                         return (nv);
2072                 break;
2073         }
2074
2075         default:
2076                 break;
2077         }
2078
2079         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2080             &child, &children) != 0)
2081                 return (NULL);
2082
2083         for (c = 0; c < children; c++) {
2084                 if ((ret = vdev_to_nvlist_iter(child[c], search,
2085                     avail_spare, l2cache, NULL)) != NULL) {
2086                         /*
2087                          * The 'is_log' value is only set for the toplevel
2088                          * vdev, not the leaf vdevs.  So we always lookup the
2089                          * log device from the root of the vdev tree (where
2090                          * 'log' is non-NULL).
2091                          */
2092                         if (log != NULL &&
2093                             nvlist_lookup_uint64(child[c],
2094                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2095                             is_log) {
2096                                 *log = B_TRUE;
2097                         }
2098                         return (ret);
2099                 }
2100         }
2101
2102         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2103             &child, &children) == 0) {
2104                 for (c = 0; c < children; c++) {
2105                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2106                             avail_spare, l2cache, NULL)) != NULL) {
2107                                 *avail_spare = B_TRUE;
2108                                 return (ret);
2109                         }
2110                 }
2111         }
2112
2113         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2114             &child, &children) == 0) {
2115                 for (c = 0; c < children; c++) {
2116                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2117                             avail_spare, l2cache, NULL)) != NULL) {
2118                                 *l2cache = B_TRUE;
2119                                 return (ret);
2120                         }
2121                 }
2122         }
2123
2124         return (NULL);
2125 }
2126
2127 /*
2128  * Given a physical path (minus the "/devices" prefix), find the
2129  * associated vdev.
2130  */
2131 nvlist_t *
2132 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2133     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2134 {
2135         nvlist_t *search, *nvroot, *ret;
2136
2137         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2138         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
2139
2140         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2141             &nvroot) == 0);
2142
2143         *avail_spare = B_FALSE;
2144         *l2cache = B_FALSE;
2145         if (log != NULL)
2146                 *log = B_FALSE;
2147         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2148         nvlist_free(search);
2149
2150         return (ret);
2151 }
2152
2153 /*
2154  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2155  */
2156 boolean_t
2157 zpool_vdev_is_interior(const char *name)
2158 {
2159         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2160             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2161                 return (B_TRUE);
2162         return (B_FALSE);
2163 }
2164
2165 nvlist_t *
2166 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2167     boolean_t *l2cache, boolean_t *log)
2168 {
2169         char buf[MAXPATHLEN];
2170         char *end;
2171         nvlist_t *nvroot, *search, *ret;
2172         uint64_t guid;
2173
2174         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2175
2176         guid = strtoull(path, &end, 10);
2177         if (guid != 0 && *end == '\0') {
2178                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
2179         } else if (zpool_vdev_is_interior(path)) {
2180                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
2181         } else if (path[0] != '/') {
2182                 (void) snprintf(buf, sizeof (buf), "%s%s", _PATH_DEV, path);
2183                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, buf) == 0);
2184         } else {
2185                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
2186         }
2187
2188         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2189             &nvroot) == 0);
2190
2191         *avail_spare = B_FALSE;
2192         *l2cache = B_FALSE;
2193         if (log != NULL)
2194                 *log = B_FALSE;
2195         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2196         nvlist_free(search);
2197
2198         return (ret);
2199 }
2200
2201 static int
2202 vdev_online(nvlist_t *nv)
2203 {
2204         uint64_t ival;
2205
2206         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2207             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2208             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2209                 return (0);
2210
2211         return (1);
2212 }
2213
2214 /*
2215  * Helper function for zpool_get_physpaths().
2216  */
2217 static int
2218 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2219     size_t *bytes_written)
2220 {
2221         size_t bytes_left, pos, rsz;
2222         char *tmppath;
2223         const char *format;
2224
2225         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2226             &tmppath) != 0)
2227                 return (EZFS_NODEVICE);
2228
2229         pos = *bytes_written;
2230         bytes_left = physpath_size - pos;
2231         format = (pos == 0) ? "%s" : " %s";
2232
2233         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2234         *bytes_written += rsz;
2235
2236         if (rsz >= bytes_left) {
2237                 /* if physpath was not copied properly, clear it */
2238                 if (bytes_left != 0) {
2239                         physpath[pos] = 0;
2240                 }
2241                 return (EZFS_NOSPC);
2242         }
2243         return (0);
2244 }
2245
2246 static int
2247 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2248     size_t *rsz, boolean_t is_spare)
2249 {
2250         char *type;
2251         int ret;
2252
2253         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2254                 return (EZFS_INVALCONFIG);
2255
2256         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2257                 /*
2258                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2259                  * For a spare vdev, we only want to boot from the active
2260                  * spare device.
2261                  */
2262                 if (is_spare) {
2263                         uint64_t spare = 0;
2264                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2265                             &spare);
2266                         if (!spare)
2267                                 return (EZFS_INVALCONFIG);
2268                 }
2269
2270                 if (vdev_online(nv)) {
2271                         if ((ret = vdev_get_one_physpath(nv, physpath,
2272                             phypath_size, rsz)) != 0)
2273                                 return (ret);
2274                 }
2275         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2276             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2277             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2278                 nvlist_t **child;
2279                 uint_t count;
2280                 int i, ret;
2281
2282                 if (nvlist_lookup_nvlist_array(nv,
2283                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2284                         return (EZFS_INVALCONFIG);
2285
2286                 for (i = 0; i < count; i++) {
2287                         ret = vdev_get_physpaths(child[i], physpath,
2288                             phypath_size, rsz, is_spare);
2289                         if (ret == EZFS_NOSPC)
2290                                 return (ret);
2291                 }
2292         }
2293
2294         return (EZFS_POOL_INVALARG);
2295 }
2296
2297 /*
2298  * Get phys_path for a root pool config.
2299  * Return 0 on success; non-zero on failure.
2300  */
2301 static int
2302 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2303 {
2304         size_t rsz;
2305         nvlist_t *vdev_root;
2306         nvlist_t **child;
2307         uint_t count;
2308         char *type;
2309
2310         rsz = 0;
2311
2312         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2313             &vdev_root) != 0)
2314                 return (EZFS_INVALCONFIG);
2315
2316         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2317             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2318             &child, &count) != 0)
2319                 return (EZFS_INVALCONFIG);
2320
2321         /*
2322          * root pool can not have EFI labeled disks and can only have
2323          * a single top-level vdev.
2324          */
2325         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1 ||
2326             pool_uses_efi(vdev_root))
2327                 return (EZFS_POOL_INVALARG);
2328
2329         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2330             B_FALSE);
2331
2332         /* No online devices */
2333         if (rsz == 0)
2334                 return (EZFS_NODEVICE);
2335
2336         return (0);
2337 }
2338
2339 /*
2340  * Get phys_path for a root pool
2341  * Return 0 on success; non-zero on failure.
2342  */
2343 int
2344 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2345 {
2346         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2347             phypath_size));
2348 }
2349
2350 /*
2351  * If the device has being dynamically expanded then we need to relabel
2352  * the disk to use the new unallocated space.
2353  */
2354 static int
2355 zpool_relabel_disk(libzfs_handle_t *hdl, const char *name)
2356 {
2357 #ifdef sun
2358         char path[MAXPATHLEN];
2359         char errbuf[1024];
2360         int fd, error;
2361         int (*_efi_use_whole_disk)(int);
2362
2363         if ((_efi_use_whole_disk = (int (*)(int))dlsym(RTLD_DEFAULT,
2364             "efi_use_whole_disk")) == NULL)
2365                 return (-1);
2366
2367         (void) snprintf(path, sizeof (path), "%s/%s", RDISK_ROOT, name);
2368
2369         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
2370                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2371                     "relabel '%s': unable to open device"), name);
2372                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
2373         }
2374
2375         /*
2376          * It's possible that we might encounter an error if the device
2377          * does not have any unallocated space left. If so, we simply
2378          * ignore that error and continue on.
2379          */
2380         error = _efi_use_whole_disk(fd);
2381         (void) close(fd);
2382         if (error && error != VT_ENOSPC) {
2383                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2384                     "relabel '%s': unable to read disk capacity"), name);
2385                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
2386         }
2387 #endif  /* sun */
2388         return (0);
2389 }
2390
2391 /*
2392  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2393  * ZFS_ONLINE_* flags.
2394  */
2395 int
2396 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2397     vdev_state_t *newstate)
2398 {
2399         zfs_cmd_t zc = { 0 };
2400         char msg[1024];
2401         nvlist_t *tgt;
2402         boolean_t avail_spare, l2cache, islog;
2403         libzfs_handle_t *hdl = zhp->zpool_hdl;
2404
2405         if (flags & ZFS_ONLINE_EXPAND) {
2406                 (void) snprintf(msg, sizeof (msg),
2407                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2408         } else {
2409                 (void) snprintf(msg, sizeof (msg),
2410                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2411         }
2412
2413         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2414         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2415             &islog)) == NULL)
2416                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2417
2418         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2419
2420         if (avail_spare)
2421                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2422
2423         if (flags & ZFS_ONLINE_EXPAND ||
2424             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2425                 char *pathname = NULL;
2426                 uint64_t wholedisk = 0;
2427
2428                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2429                     &wholedisk);
2430                 verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
2431                     &pathname) == 0);
2432
2433                 /*
2434                  * XXX - L2ARC 1.0 devices can't support expansion.
2435                  */
2436                 if (l2cache) {
2437                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2438                             "cannot expand cache devices"));
2439                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2440                 }
2441
2442                 if (wholedisk) {
2443                         pathname += strlen(DISK_ROOT) + 1;
2444                         (void) zpool_relabel_disk(hdl, pathname);
2445                 }
2446         }
2447
2448         zc.zc_cookie = VDEV_STATE_ONLINE;
2449         zc.zc_obj = flags;
2450
2451         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2452                 if (errno == EINVAL) {
2453                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2454                             "from this pool into a new one.  Use '%s' "
2455                             "instead"), "zpool detach");
2456                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2457                 }
2458                 return (zpool_standard_error(hdl, errno, msg));
2459         }
2460
2461         *newstate = zc.zc_cookie;
2462         return (0);
2463 }
2464
2465 /*
2466  * Take the specified vdev offline
2467  */
2468 int
2469 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2470 {
2471         zfs_cmd_t zc = { 0 };
2472         char msg[1024];
2473         nvlist_t *tgt;
2474         boolean_t avail_spare, l2cache;
2475         libzfs_handle_t *hdl = zhp->zpool_hdl;
2476
2477         (void) snprintf(msg, sizeof (msg),
2478             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2479
2480         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2481         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2482             NULL)) == NULL)
2483                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2484
2485         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2486
2487         if (avail_spare)
2488                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2489
2490         zc.zc_cookie = VDEV_STATE_OFFLINE;
2491         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2492
2493         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2494                 return (0);
2495
2496         switch (errno) {
2497         case EBUSY:
2498
2499                 /*
2500                  * There are no other replicas of this device.
2501                  */
2502                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2503
2504         case EEXIST:
2505                 /*
2506                  * The log device has unplayed logs
2507                  */
2508                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2509
2510         default:
2511                 return (zpool_standard_error(hdl, errno, msg));
2512         }
2513 }
2514
2515 /*
2516  * Mark the given vdev faulted.
2517  */
2518 int
2519 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2520 {
2521         zfs_cmd_t zc = { 0 };
2522         char msg[1024];
2523         libzfs_handle_t *hdl = zhp->zpool_hdl;
2524
2525         (void) snprintf(msg, sizeof (msg),
2526             dgettext(TEXT_DOMAIN, "cannot fault %llu"), guid);
2527
2528         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2529         zc.zc_guid = guid;
2530         zc.zc_cookie = VDEV_STATE_FAULTED;
2531         zc.zc_obj = aux;
2532
2533         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2534                 return (0);
2535
2536         switch (errno) {
2537         case EBUSY:
2538
2539                 /*
2540                  * There are no other replicas of this device.
2541                  */
2542                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2543
2544         default:
2545                 return (zpool_standard_error(hdl, errno, msg));
2546         }
2547
2548 }
2549
2550 /*
2551  * Mark the given vdev degraded.
2552  */
2553 int
2554 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2555 {
2556         zfs_cmd_t zc = { 0 };
2557         char msg[1024];
2558         libzfs_handle_t *hdl = zhp->zpool_hdl;
2559
2560         (void) snprintf(msg, sizeof (msg),
2561             dgettext(TEXT_DOMAIN, "cannot degrade %llu"), guid);
2562
2563         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2564         zc.zc_guid = guid;
2565         zc.zc_cookie = VDEV_STATE_DEGRADED;
2566         zc.zc_obj = aux;
2567
2568         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2569                 return (0);
2570
2571         return (zpool_standard_error(hdl, errno, msg));
2572 }
2573
2574 /*
2575  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2576  * a hot spare.
2577  */
2578 static boolean_t
2579 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2580 {
2581         nvlist_t **child;
2582         uint_t c, children;
2583         char *type;
2584
2585         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2586             &children) == 0) {
2587                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2588                     &type) == 0);
2589
2590                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2591                     children == 2 && child[which] == tgt)
2592                         return (B_TRUE);
2593
2594                 for (c = 0; c < children; c++)
2595                         if (is_replacing_spare(child[c], tgt, which))
2596                                 return (B_TRUE);
2597         }
2598
2599         return (B_FALSE);
2600 }
2601
2602 /*
2603  * Attach new_disk (fully described by nvroot) to old_disk.
2604  * If 'replacing' is specified, the new disk will replace the old one.
2605  */
2606 int
2607 zpool_vdev_attach(zpool_handle_t *zhp,
2608     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2609 {
2610         zfs_cmd_t zc = { 0 };
2611         char msg[1024];
2612         int ret;
2613         nvlist_t *tgt;
2614         boolean_t avail_spare, l2cache, islog;
2615         uint64_t val;
2616         char *newname;
2617         nvlist_t **child;
2618         uint_t children;
2619         nvlist_t *config_root;
2620         libzfs_handle_t *hdl = zhp->zpool_hdl;
2621         boolean_t rootpool = zpool_is_bootable(zhp);
2622
2623         if (replacing)
2624                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2625                     "cannot replace %s with %s"), old_disk, new_disk);
2626         else
2627                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2628                     "cannot attach %s to %s"), new_disk, old_disk);
2629
2630         /*
2631          * If this is a root pool, make sure that we're not attaching an
2632          * EFI labeled device.
2633          */
2634         if (rootpool && pool_uses_efi(nvroot)) {
2635                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2636                     "EFI labeled devices are not supported on root pools."));
2637                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
2638         }
2639
2640         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2641         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2642             &islog)) == 0)
2643                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2644
2645         if (avail_spare)
2646                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2647
2648         if (l2cache)
2649                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2650
2651         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2652         zc.zc_cookie = replacing;
2653
2654         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2655             &child, &children) != 0 || children != 1) {
2656                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2657                     "new device must be a single disk"));
2658                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2659         }
2660
2661         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2662             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2663
2664         if ((newname = zpool_vdev_name(NULL, NULL, child[0], B_FALSE)) == NULL)
2665                 return (-1);
2666
2667         /*
2668          * If the target is a hot spare that has been swapped in, we can only
2669          * replace it with another hot spare.
2670          */
2671         if (replacing &&
2672             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2673             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2674             NULL) == NULL || !avail_spare) &&
2675             is_replacing_spare(config_root, tgt, 1)) {
2676                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2677                     "can only be replaced by another hot spare"));
2678                 free(newname);
2679                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2680         }
2681
2682         free(newname);
2683
2684         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2685                 return (-1);
2686
2687         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2688
2689         zcmd_free_nvlists(&zc);
2690
2691         if (ret == 0) {
2692                 if (rootpool) {
2693                         /*
2694                          * XXX need a better way to prevent user from
2695                          * booting up a half-baked vdev.
2696                          */
2697                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2698                             "sure to wait until resilver is done "
2699                             "before rebooting.\n"));
2700                         (void) fprintf(stderr, "\n");
2701                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "If "
2702                             "you boot from pool '%s', you may need to update\n"
2703                             "boot code on newly attached disk '%s'.\n\n"
2704                             "Assuming you use GPT partitioning and 'da0' is "
2705                             "your new boot disk\n"
2706                             "you may use the following command:\n\n"
2707                             "\tgpart bootcode -b /boot/pmbr -p "
2708                             "/boot/gptzfsboot -i 1 da0\n\n"),
2709                             zhp->zpool_name, new_disk);
2710                 }
2711                 return (0);
2712         }
2713
2714         switch (errno) {
2715         case ENOTSUP:
2716                 /*
2717                  * Can't attach to or replace this type of vdev.
2718                  */
2719                 if (replacing) {
2720                         uint64_t version = zpool_get_prop_int(zhp,
2721                             ZPOOL_PROP_VERSION, NULL);
2722
2723                         if (islog)
2724                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2725                                     "cannot replace a log with a spare"));
2726                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2727                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2728                                     "already in replacing/spare config; wait "
2729                                     "for completion or use 'zpool detach'"));
2730                         else
2731                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2732                                     "cannot replace a replacing device"));
2733                 } else {
2734                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2735                             "can only attach to mirrors and top-level "
2736                             "disks"));
2737                 }
2738                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2739                 break;
2740
2741         case EINVAL:
2742                 /*
2743                  * The new device must be a single disk.
2744                  */
2745                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2746                     "new device must be a single disk"));
2747                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2748                 break;
2749
2750         case EBUSY:
2751                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2752                     new_disk);
2753                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2754                 break;
2755
2756         case EOVERFLOW:
2757                 /*
2758                  * The new device is too small.
2759                  */
2760                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2761                     "device is too small"));
2762                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2763                 break;
2764
2765         case EDOM:
2766                 /*
2767                  * The new device has a different alignment requirement.
2768                  */
2769                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2770                     "devices have different sector alignment"));
2771                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2772                 break;
2773
2774         case ENAMETOOLONG:
2775                 /*
2776                  * The resulting top-level vdev spec won't fit in the label.
2777                  */
2778                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2779                 break;
2780
2781         default:
2782                 (void) zpool_standard_error(hdl, errno, msg);
2783         }
2784
2785         return (-1);
2786 }
2787
2788 /*
2789  * Detach the specified device.
2790  */
2791 int
2792 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2793 {
2794         zfs_cmd_t zc = { 0 };
2795         char msg[1024];
2796         nvlist_t *tgt;
2797         boolean_t avail_spare, l2cache;
2798         libzfs_handle_t *hdl = zhp->zpool_hdl;
2799
2800         (void) snprintf(msg, sizeof (msg),
2801             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2802
2803         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2804         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2805             NULL)) == 0)
2806                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2807
2808         if (avail_spare)
2809                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2810
2811         if (l2cache)
2812                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2813
2814         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2815
2816         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2817                 return (0);
2818
2819         switch (errno) {
2820
2821         case ENOTSUP:
2822                 /*
2823                  * Can't detach from this type of vdev.
2824                  */
2825                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2826                     "applicable to mirror and replacing vdevs"));
2827                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2828                 break;
2829
2830         case EBUSY:
2831                 /*
2832                  * There are no other replicas of this device.
2833                  */
2834                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2835                 break;
2836
2837         default:
2838                 (void) zpool_standard_error(hdl, errno, msg);
2839         }
2840
2841         return (-1);
2842 }
2843
2844 /*
2845  * Find a mirror vdev in the source nvlist.
2846  *
2847  * The mchild array contains a list of disks in one of the top-level mirrors
2848  * of the source pool.  The schild array contains a list of disks that the
2849  * user specified on the command line.  We loop over the mchild array to
2850  * see if any entry in the schild array matches.
2851  *
2852  * If a disk in the mchild array is found in the schild array, we return
2853  * the index of that entry.  Otherwise we return -1.
2854  */
2855 static int
2856 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2857     nvlist_t **schild, uint_t schildren)
2858 {
2859         uint_t mc;
2860
2861         for (mc = 0; mc < mchildren; mc++) {
2862                 uint_t sc;
2863                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2864                     mchild[mc], B_FALSE);
2865
2866                 for (sc = 0; sc < schildren; sc++) {
2867                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2868                             schild[sc], B_FALSE);
2869                         boolean_t result = (strcmp(mpath, spath) == 0);
2870
2871                         free(spath);
2872                         if (result) {
2873                                 free(mpath);
2874                                 return (mc);
2875                         }
2876                 }
2877
2878                 free(mpath);
2879         }
2880
2881         return (-1);
2882 }
2883
2884 /*
2885  * Split a mirror pool.  If newroot points to null, then a new nvlist
2886  * is generated and it is the responsibility of the caller to free it.
2887  */
2888 int
2889 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2890     nvlist_t *props, splitflags_t flags)
2891 {
2892         zfs_cmd_t zc = { 0 };
2893         char msg[1024];
2894         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2895         nvlist_t **varray = NULL, *zc_props = NULL;
2896         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2897         libzfs_handle_t *hdl = zhp->zpool_hdl;
2898         uint64_t vers;
2899         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2900         int retval = 0;
2901
2902         (void) snprintf(msg, sizeof (msg),
2903             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2904
2905         if (!zpool_name_valid(hdl, B_FALSE, newname))
2906                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2907
2908         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2909                 (void) fprintf(stderr, gettext("Internal error: unable to "
2910                     "retrieve pool configuration\n"));
2911                 return (-1);
2912         }
2913
2914         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2915             == 0);
2916         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2917
2918         if (props) {
2919                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2920                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2921                     props, vers, flags, msg)) == NULL)
2922                         return (-1);
2923         }
2924
2925         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2926             &children) != 0) {
2927                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2928                     "Source pool is missing vdev tree"));
2929                 if (zc_props)
2930                         nvlist_free(zc_props);
2931                 return (-1);
2932         }
2933
2934         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2935         vcount = 0;
2936
2937         if (*newroot == NULL ||
2938             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2939             &newchild, &newchildren) != 0)
2940                 newchildren = 0;
2941
2942         for (c = 0; c < children; c++) {
2943                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2944                 char *type;
2945                 nvlist_t **mchild, *vdev;
2946                 uint_t mchildren;
2947                 int entry;
2948
2949                 /*
2950                  * Unlike cache & spares, slogs are stored in the
2951                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2952                  */
2953                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2954                     &is_log);
2955                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2956                     &is_hole);
2957                 if (is_log || is_hole) {
2958                         /*
2959                          * Create a hole vdev and put it in the config.
2960                          */
2961                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2962                                 goto out;
2963                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2964                             VDEV_TYPE_HOLE) != 0)
2965                                 goto out;
2966                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2967                             1) != 0)
2968                                 goto out;
2969                         if (lastlog == 0)
2970                                 lastlog = vcount;
2971                         varray[vcount++] = vdev;
2972                         continue;
2973                 }
2974                 lastlog = 0;
2975                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
2976                     == 0);
2977                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
2978                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2979                             "Source pool must be composed only of mirrors\n"));
2980                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2981                         goto out;
2982                 }
2983
2984                 verify(nvlist_lookup_nvlist_array(child[c],
2985                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2986
2987                 /* find or add an entry for this top-level vdev */
2988                 if (newchildren > 0 &&
2989                     (entry = find_vdev_entry(zhp, mchild, mchildren,
2990                     newchild, newchildren)) >= 0) {
2991                         /* We found a disk that the user specified. */
2992                         vdev = mchild[entry];
2993                         ++found;
2994                 } else {
2995                         /* User didn't specify a disk for this vdev. */
2996                         vdev = mchild[mchildren - 1];
2997                 }
2998
2999                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
3000                         goto out;
3001         }
3002
3003         /* did we find every disk the user specified? */
3004         if (found != newchildren) {
3005                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
3006                     "include at most one disk from each mirror"));
3007                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3008                 goto out;
3009         }
3010
3011         /* Prepare the nvlist for populating. */
3012         if (*newroot == NULL) {
3013                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
3014                         goto out;
3015                 freelist = B_TRUE;
3016                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
3017                     VDEV_TYPE_ROOT) != 0)
3018                         goto out;
3019         } else {
3020                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
3021         }
3022
3023         /* Add all the children we found */
3024         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
3025             lastlog == 0 ? vcount : lastlog) != 0)
3026                 goto out;
3027
3028         /*
3029          * If we're just doing a dry run, exit now with success.
3030          */
3031         if (flags.dryrun) {
3032                 memory_err = B_FALSE;
3033                 freelist = B_FALSE;
3034                 goto out;
3035         }
3036
3037         /* now build up the config list & call the ioctl */
3038         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3039                 goto out;
3040
3041         if (nvlist_add_nvlist(newconfig,
3042             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3043             nvlist_add_string(newconfig,
3044             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3045             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3046                 goto out;
3047
3048         /*
3049          * The new pool is automatically part of the namespace unless we
3050          * explicitly export it.
3051          */
3052         if (!flags.import)
3053                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3054         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3055         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3056         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
3057                 goto out;
3058         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
3059                 goto out;
3060
3061         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3062                 retval = zpool_standard_error(hdl, errno, msg);
3063                 goto out;
3064         }
3065
3066         freelist = B_FALSE;
3067         memory_err = B_FALSE;
3068
3069 out:
3070         if (varray != NULL) {
3071                 int v;
3072
3073                 for (v = 0; v < vcount; v++)
3074                         nvlist_free(varray[v]);
3075                 free(varray);
3076         }
3077         zcmd_free_nvlists(&zc);
3078         if (zc_props)
3079                 nvlist_free(zc_props);
3080         if (newconfig)
3081                 nvlist_free(newconfig);
3082         if (freelist) {
3083                 nvlist_free(*newroot);
3084                 *newroot = NULL;
3085         }
3086
3087         if (retval != 0)
3088                 return (retval);
3089
3090         if (memory_err)
3091                 return (no_memory(hdl));
3092
3093         return (0);
3094 }
3095
3096 /*
3097  * Remove the given device.  Currently, this is supported only for hot spares
3098  * and level 2 cache devices.
3099  */
3100 int
3101 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3102 {
3103         zfs_cmd_t zc = { 0 };
3104         char msg[1024];
3105         nvlist_t *tgt;
3106         boolean_t avail_spare, l2cache, islog;
3107         libzfs_handle_t *hdl = zhp->zpool_hdl;
3108         uint64_t version;
3109
3110         (void) snprintf(msg, sizeof (msg),
3111             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3112
3113         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3114         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3115             &islog)) == 0)
3116                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3117         /*
3118          * XXX - this should just go away.
3119          */
3120         if (!avail_spare && !l2cache && !islog) {
3121                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3122                     "only inactive hot spares, cache, top-level, "
3123                     "or log devices can be removed"));
3124                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3125         }
3126
3127         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3128         if (islog && version < SPA_VERSION_HOLES) {
3129                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3130                     "pool must be upgrade to support log removal"));
3131                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
3132         }
3133
3134         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
3135
3136         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3137                 return (0);
3138
3139         return (zpool_standard_error(hdl, errno, msg));
3140 }
3141
3142 /*
3143  * Clear the errors for the pool, or the particular device if specified.
3144  */
3145 int
3146 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3147 {
3148         zfs_cmd_t zc = { 0 };
3149         char msg[1024];
3150         nvlist_t *tgt;
3151         zpool_rewind_policy_t policy;
3152         boolean_t avail_spare, l2cache;
3153         libzfs_handle_t *hdl = zhp->zpool_hdl;
3154         nvlist_t *nvi = NULL;
3155         int error;
3156
3157         if (path)
3158                 (void) snprintf(msg, sizeof (msg),
3159                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3160                     path);
3161         else
3162                 (void) snprintf(msg, sizeof (msg),
3163                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3164                     zhp->zpool_name);
3165
3166         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3167         if (path) {
3168                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3169                     &l2cache, NULL)) == 0)
3170                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
3171
3172                 /*
3173                  * Don't allow error clearing for hot spares.  Do allow
3174                  * error clearing for l2cache devices.
3175                  */
3176                 if (avail_spare)
3177                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
3178
3179                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
3180                     &zc.zc_guid) == 0);
3181         }
3182
3183         zpool_get_rewind_policy(rewindnvl, &policy);
3184         zc.zc_cookie = policy.zrp_request;
3185
3186         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
3187                 return (-1);
3188
3189         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
3190                 return (-1);
3191
3192         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3193             errno == ENOMEM) {
3194                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3195                         zcmd_free_nvlists(&zc);
3196                         return (-1);
3197                 }
3198         }
3199
3200         if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
3201             errno != EPERM && errno != EACCES)) {
3202                 if (policy.zrp_request &
3203                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
3204                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3205                         zpool_rewind_exclaim(hdl, zc.zc_name,
3206                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
3207                             nvi);
3208                         nvlist_free(nvi);
3209                 }
3210                 zcmd_free_nvlists(&zc);
3211                 return (0);
3212         }
3213
3214         zcmd_free_nvlists(&zc);
3215         return (zpool_standard_error(hdl, errno, msg));
3216 }
3217
3218 /*
3219  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3220  */
3221 int
3222 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3223 {
3224         zfs_cmd_t zc = { 0 };
3225         char msg[1024];
3226         libzfs_handle_t *hdl = zhp->zpool_hdl;
3227
3228         (void) snprintf(msg, sizeof (msg),
3229             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3230             guid);
3231
3232         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3233         zc.zc_guid = guid;
3234         zc.zc_cookie = ZPOOL_NO_REWIND;
3235
3236         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
3237                 return (0);
3238
3239         return (zpool_standard_error(hdl, errno, msg));
3240 }
3241
3242 /*
3243  * Change the GUID for a pool.
3244  */
3245 int
3246 zpool_reguid(zpool_handle_t *zhp)
3247 {
3248         char msg[1024];
3249         libzfs_handle_t *hdl = zhp->zpool_hdl;
3250         zfs_cmd_t zc = { 0 };
3251
3252         (void) snprintf(msg, sizeof (msg),
3253             dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3254
3255         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3256         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3257                 return (0);
3258
3259         return (zpool_standard_error(hdl, errno, msg));
3260 }
3261
3262 /*
3263  * Reopen the pool.
3264  */
3265 int
3266 zpool_reopen(zpool_handle_t *zhp)
3267 {
3268         zfs_cmd_t zc = { 0 };
3269         char msg[1024];
3270         libzfs_handle_t *hdl = zhp->zpool_hdl;
3271
3272         (void) snprintf(msg, sizeof (msg),
3273             dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3274             zhp->zpool_name);
3275
3276         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3277         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3278                 return (0);
3279         return (zpool_standard_error(hdl, errno, msg));
3280 }
3281
3282 /*
3283  * Convert from a devid string to a path.
3284  */
3285 static char *
3286 devid_to_path(char *devid_str)
3287 {
3288         ddi_devid_t devid;
3289         char *minor;
3290         char *path;
3291         devid_nmlist_t *list = NULL;
3292         int ret;
3293
3294         if (devid_str_decode(devid_str, &devid, &minor) != 0)
3295                 return (NULL);
3296
3297         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3298
3299         devid_str_free(minor);
3300         devid_free(devid);
3301
3302         if (ret != 0)
3303                 return (NULL);
3304
3305         if ((path = strdup(list[0].devname)) == NULL)
3306                 return (NULL);
3307
3308         devid_free_nmlist(list);
3309
3310         return (path);
3311 }
3312
3313 /*
3314  * Convert from a path to a devid string.
3315  */
3316 static char *
3317 path_to_devid(const char *path)
3318 {
3319         int fd;
3320         ddi_devid_t devid;
3321         char *minor, *ret;
3322
3323         if ((fd = open(path, O_RDONLY)) < 0)
3324                 return (NULL);
3325
3326         minor = NULL;
3327         ret = NULL;
3328         if (devid_get(fd, &devid) == 0) {
3329                 if (devid_get_minor_name(fd, &minor) == 0)
3330                         ret = devid_str_encode(devid, minor);
3331                 if (minor != NULL)
3332                         devid_str_free(minor);
3333                 devid_free(devid);
3334         }
3335         (void) close(fd);
3336
3337         return (ret);
3338 }
3339
3340 /*
3341  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3342  * ignore any failure here, since a common case is for an unprivileged user to
3343  * type 'zpool status', and we'll display the correct information anyway.
3344  */
3345 static void
3346 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3347 {
3348         zfs_cmd_t zc = { 0 };
3349
3350         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3351         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3352         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3353             &zc.zc_guid) == 0);
3354
3355         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3356 }
3357
3358 /*
3359  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3360  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3361  * We also check if this is a whole disk, in which case we strip off the
3362  * trailing 's0' slice name.
3363  *
3364  * This routine is also responsible for identifying when disks have been
3365  * reconfigured in a new location.  The kernel will have opened the device by
3366  * devid, but the path will still refer to the old location.  To catch this, we
3367  * first do a path -> devid translation (which is fast for the common case).  If
3368  * the devid matches, we're done.  If not, we do a reverse devid -> path
3369  * translation and issue the appropriate ioctl() to update the path of the vdev.
3370  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3371  * of these checks.
3372  */
3373 char *
3374 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3375     boolean_t verbose)
3376 {
3377         char *path, *devid;
3378         uint64_t value;
3379         char buf[64];
3380         vdev_stat_t *vs;
3381         uint_t vsc;
3382         int have_stats;
3383         int have_path;
3384
3385         have_stats = nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3386             (uint64_t **)&vs, &vsc) == 0;
3387         have_path = nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0;
3388
3389         /*
3390          * If the device is not currently present, assume it will not
3391          * come back at the same device path.  Display the device by GUID.
3392          */
3393         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &value) == 0 ||
3394             have_path && have_stats && vs->vs_state <= VDEV_STATE_CANT_OPEN) {
3395                 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3396                     &value) == 0);
3397                 (void) snprintf(buf, sizeof (buf), "%llu",
3398                     (u_longlong_t)value);
3399                 path = buf;
3400         } else if (have_path) {
3401
3402                 /*
3403                  * If the device is dead (faulted, offline, etc) then don't
3404                  * bother opening it.  Otherwise we may be forcing the user to
3405                  * open a misbehaving device, which can have undesirable
3406                  * effects.
3407                  */
3408                 if ((have_stats == 0 ||
3409                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3410                     zhp != NULL &&
3411                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3412                         /*
3413                          * Determine if the current path is correct.
3414                          */
3415                         char *newdevid = path_to_devid(path);
3416
3417                         if (newdevid == NULL ||
3418                             strcmp(devid, newdevid) != 0) {
3419                                 char *newpath;
3420
3421                                 if ((newpath = devid_to_path(devid)) != NULL) {
3422                                         /*
3423                                          * Update the path appropriately.
3424                                          */
3425                                         set_path(zhp, nv, newpath);
3426                                         if (nvlist_add_string(nv,
3427                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3428                                                 verify(nvlist_lookup_string(nv,
3429                                                     ZPOOL_CONFIG_PATH,
3430                                                     &path) == 0);
3431                                         free(newpath);
3432                                 }
3433                         }
3434
3435                         if (newdevid)
3436                                 devid_str_free(newdevid);
3437                 }
3438
3439 #ifdef sun
3440                 if (strncmp(path, "/dev/dsk/", 9) == 0)
3441                         path += 9;
3442
3443                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
3444                     &value) == 0 && value) {
3445                         int pathlen = strlen(path);
3446                         char *tmp = zfs_strdup(hdl, path);
3447
3448                         /*
3449                          * If it starts with c#, and ends with "s0", chop
3450                          * the "s0" off, or if it ends with "s0/old", remove
3451                          * the "s0" from the middle.
3452                          */
3453                         if (CTD_CHECK(tmp)) {
3454                                 if (strcmp(&tmp[pathlen - 2], "s0") == 0) {
3455                                         tmp[pathlen - 2] = '\0';
3456                                 } else if (pathlen > 6 &&
3457                                     strcmp(&tmp[pathlen - 6], "s0/old") == 0) {
3458                                         (void) strcpy(&tmp[pathlen - 6],
3459                                             "/old");
3460                                 }
3461                         }
3462                         return (tmp);
3463                 }
3464 #else   /* !sun */
3465                 if (strncmp(path, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
3466                         path += sizeof(_PATH_DEV) - 1;
3467 #endif  /* !sun */
3468         } else {
3469                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3470
3471                 /*
3472                  * If it's a raidz device, we need to stick in the parity level.
3473                  */
3474                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3475                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3476                             &value) == 0);
3477                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3478                             (u_longlong_t)value);
3479                         path = buf;
3480                 }
3481
3482                 /*
3483                  * We identify each top-level vdev by using a <type-id>
3484                  * naming convention.
3485                  */
3486                 if (verbose) {
3487                         uint64_t id;
3488
3489                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3490                             &id) == 0);
3491                         (void) snprintf(buf, sizeof (buf), "%s-%llu", path,
3492                             (u_longlong_t)id);
3493                         path = buf;
3494                 }
3495         }
3496
3497         return (zfs_strdup(hdl, path));
3498 }
3499
3500 static int
3501 zbookmark_compare(const void *a, const void *b)
3502 {
3503         return (memcmp(a, b, sizeof (zbookmark_t)));
3504 }
3505
3506 /*
3507  * Retrieve the persistent error log, uniquify the members, and return to the
3508  * caller.
3509  */
3510 int
3511 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3512 {
3513         zfs_cmd_t zc = { 0 };
3514         uint64_t count;
3515         zbookmark_t *zb = NULL;
3516         int i;
3517
3518         /*
3519          * Retrieve the raw error list from the kernel.  If the number of errors
3520          * has increased, allocate more space and continue until we get the
3521          * entire list.
3522          */
3523         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3524             &count) == 0);
3525         if (count == 0)
3526                 return (0);
3527         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3528             count * sizeof (zbookmark_t))) == (uintptr_t)NULL)
3529                 return (-1);
3530         zc.zc_nvlist_dst_size = count;
3531         (void) strcpy(zc.zc_name, zhp->zpool_name);
3532         for (;;) {
3533                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3534                     &zc) != 0) {
3535                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3536                         if (errno == ENOMEM) {
3537                                 count = zc.zc_nvlist_dst_size;
3538                                 if ((zc.zc_nvlist_dst = (uintptr_t)
3539                                     zfs_alloc(zhp->zpool_hdl, count *
3540                                     sizeof (zbookmark_t))) == (uintptr_t)NULL)
3541                                         return (-1);
3542                         } else {
3543                                 return (-1);
3544                         }
3545                 } else {
3546                         break;
3547                 }
3548         }
3549
3550         /*
3551          * Sort the resulting bookmarks.  This is a little confusing due to the
3552          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3553          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3554          * _not_ copied as part of the process.  So we point the start of our
3555          * array appropriate and decrement the total number of elements.
3556          */
3557         zb = ((zbookmark_t *)(uintptr_t)zc.zc_nvlist_dst) +
3558             zc.zc_nvlist_dst_size;
3559         count -= zc.zc_nvlist_dst_size;
3560
3561         qsort(zb, count, sizeof (zbookmark_t), zbookmark_compare);
3562
3563         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3564
3565         /*
3566          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3567          */
3568         for (i = 0; i < count; i++) {
3569                 nvlist_t *nv;
3570
3571                 /* ignoring zb_blkid and zb_level for now */
3572                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3573                     zb[i-1].zb_object == zb[i].zb_object)
3574                         continue;
3575
3576                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3577                         goto nomem;
3578                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3579                     zb[i].zb_objset) != 0) {
3580                         nvlist_free(nv);
3581                         goto nomem;
3582                 }
3583                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3584                     zb[i].zb_object) != 0) {
3585                         nvlist_free(nv);
3586                         goto nomem;
3587                 }
3588                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3589                         nvlist_free(nv);
3590                         goto nomem;
3591                 }
3592                 nvlist_free(nv);
3593         }
3594
3595         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3596         return (0);
3597
3598 nomem:
3599         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3600         return (no_memory(zhp->zpool_hdl));
3601 }
3602
3603 /*
3604  * Upgrade a ZFS pool to the latest on-disk version.
3605  */
3606 int
3607 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3608 {
3609         zfs_cmd_t zc = { 0 };
3610         libzfs_handle_t *hdl = zhp->zpool_hdl;
3611
3612         (void) strcpy(zc.zc_name, zhp->zpool_name);
3613         zc.zc_cookie = new_version;
3614
3615         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3616                 return (zpool_standard_error_fmt(hdl, errno,
3617                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3618                     zhp->zpool_name));
3619         return (0);
3620 }
3621
3622 void
3623 zfs_save_arguments(int argc, char **argv, char *string, int len)
3624 {
3625         (void) strlcpy(string, basename(argv[0]), len);
3626         for (int i = 1; i < argc; i++) {
3627                 (void) strlcat(string, " ", len);
3628                 (void) strlcat(string, argv[i], len);
3629         }
3630 }
3631
3632 int
3633 zpool_log_history(libzfs_handle_t *hdl, const char *message)
3634 {
3635         zfs_cmd_t zc = { 0 };
3636         nvlist_t *args;
3637         int err;
3638
3639         args = fnvlist_alloc();
3640         fnvlist_add_string(args, "message", message);
3641         err = zcmd_write_src_nvlist(hdl, &zc, args);
3642         if (err == 0)
3643                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_LOG_HISTORY, &zc);
3644         nvlist_free(args);
3645         zcmd_free_nvlists(&zc);
3646         return (err);
3647 }
3648
3649 /*
3650  * Perform ioctl to get some command history of a pool.
3651  *
3652  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3653  * logical offset of the history buffer to start reading from.
3654  *
3655  * Upon return, 'off' is the next logical offset to read from and
3656  * 'len' is the actual amount of bytes read into 'buf'.
3657  */
3658 static int
3659 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3660 {
3661         zfs_cmd_t zc = { 0 };
3662         libzfs_handle_t *hdl = zhp->zpool_hdl;
3663
3664         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3665
3666         zc.zc_history = (uint64_t)(uintptr_t)buf;
3667         zc.zc_history_len = *len;
3668         zc.zc_history_offset = *off;
3669
3670         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3671                 switch (errno) {
3672                 case EPERM:
3673                         return (zfs_error_fmt(hdl, EZFS_PERM,
3674                             dgettext(TEXT_DOMAIN,
3675                             "cannot show history for pool '%s'"),
3676                             zhp->zpool_name));
3677                 case ENOENT:
3678                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3679                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3680                             "'%s'"), zhp->zpool_name));
3681                 case ENOTSUP:
3682                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3683                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3684                             "'%s', pool must be upgraded"), zhp->zpool_name));
3685                 default:
3686                         return (zpool_standard_error_fmt(hdl, errno,
3687                             dgettext(TEXT_DOMAIN,
3688                             "cannot get history for '%s'"), zhp->zpool_name));
3689                 }
3690         }
3691
3692         *len = zc.zc_history_len;
3693         *off = zc.zc_history_offset;
3694
3695         return (0);
3696 }
3697
3698 /*
3699  * Process the buffer of nvlists, unpacking and storing each nvlist record
3700  * into 'records'.  'leftover' is set to the number of bytes that weren't
3701  * processed as there wasn't a complete record.
3702  */
3703 int
3704 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3705     nvlist_t ***records, uint_t *numrecords)
3706 {
3707         uint64_t reclen;
3708         nvlist_t *nv;
3709         int i;
3710
3711         while (bytes_read > sizeof (reclen)) {
3712
3713                 /* get length of packed record (stored as little endian) */
3714                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3715                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3716
3717                 if (bytes_read < sizeof (reclen) + reclen)
3718                         break;
3719
3720                 /* unpack record */
3721                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3722                         return (ENOMEM);
3723                 bytes_read -= sizeof (reclen) + reclen;
3724                 buf += sizeof (reclen) + reclen;
3725
3726                 /* add record to nvlist array */
3727                 (*numrecords)++;
3728                 if (ISP2(*numrecords + 1)) {
3729                         *records = realloc(*records,
3730                             *numrecords * 2 * sizeof (nvlist_t *));
3731                 }
3732                 (*records)[*numrecords - 1] = nv;
3733         }
3734
3735         *leftover = bytes_read;
3736         return (0);
3737 }
3738
3739 #define HIS_BUF_LEN     (128*1024)
3740
3741 /*
3742  * Retrieve the command history of a pool.
3743  */
3744 int
3745 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3746 {
3747         char buf[HIS_BUF_LEN];
3748         uint64_t off = 0;
3749         nvlist_t **records = NULL;
3750         uint_t numrecords = 0;
3751         int err, i;
3752
3753         do {
3754                 uint64_t bytes_read = sizeof (buf);
3755                 uint64_t leftover;
3756
3757                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3758                         break;
3759
3760                 /* if nothing else was read in, we're at EOF, just return */
3761                 if (!bytes_read)
3762                         break;
3763
3764                 if ((err = zpool_history_unpack(buf, bytes_read,
3765                     &leftover, &records, &numrecords)) != 0)
3766                         break;
3767                 off -= leftover;
3768
3769                 /* CONSTCOND */
3770         } while (1);
3771
3772         if (!err) {
3773                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3774                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3775                     records, numrecords) == 0);
3776         }
3777         for (i = 0; i < numrecords; i++)
3778                 nvlist_free(records[i]);
3779         free(records);
3780
3781         return (err);
3782 }
3783
3784 void
3785 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
3786     char *pathname, size_t len)
3787 {
3788         zfs_cmd_t zc = { 0 };
3789         boolean_t mounted = B_FALSE;
3790         char *mntpnt = NULL;
3791         char dsname[MAXNAMELEN];
3792
3793         if (dsobj == 0) {
3794                 /* special case for the MOS */
3795                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>", obj);
3796                 return;
3797         }
3798
3799         /* get the dataset's name */
3800         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3801         zc.zc_obj = dsobj;
3802         if (ioctl(zhp->zpool_hdl->libzfs_fd,
3803             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
3804                 /* just write out a path of two object numbers */
3805                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
3806                     dsobj, obj);
3807                 return;
3808         }
3809         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
3810
3811         /* find out if the dataset is mounted */
3812         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
3813
3814         /* get the corrupted object's path */
3815         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
3816         zc.zc_obj = obj;
3817         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
3818             &zc) == 0) {
3819                 if (mounted) {
3820                         (void) snprintf(pathname, len, "%s%s", mntpnt,
3821                             zc.zc_value);
3822                 } else {
3823                         (void) snprintf(pathname, len, "%s:%s",
3824                             dsname, zc.zc_value);
3825                 }
3826         } else {
3827                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname, obj);
3828         }
3829         free(mntpnt);
3830 }
3831
3832 #ifdef sun
3833 /*
3834  * Read the EFI label from the config, if a label does not exist then
3835  * pass back the error to the caller. If the caller has passed a non-NULL
3836  * diskaddr argument then we set it to the starting address of the EFI
3837  * partition.
3838  */
3839 static int
3840 read_efi_label(nvlist_t *config, diskaddr_t *sb)
3841 {
3842         char *path;
3843         int fd;
3844         char diskname[MAXPATHLEN];
3845         int err = -1;
3846
3847         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
3848                 return (err);
3849
3850         (void) snprintf(diskname, sizeof (diskname), "%s%s", RDISK_ROOT,
3851             strrchr(path, '/'));
3852         if ((fd = open(diskname, O_RDONLY|O_NDELAY)) >= 0) {
3853                 struct dk_gpt *vtoc;
3854
3855                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
3856                         if (sb != NULL)
3857                                 *sb = vtoc->efi_parts[0].p_start;
3858                         efi_free(vtoc);
3859                 }
3860                 (void) close(fd);
3861         }
3862         return (err);
3863 }
3864
3865 /*
3866  * determine where a partition starts on a disk in the current
3867  * configuration
3868  */
3869 static diskaddr_t
3870 find_start_block(nvlist_t *config)
3871 {
3872         nvlist_t **child;
3873         uint_t c, children;
3874         diskaddr_t sb = MAXOFFSET_T;
3875         uint64_t wholedisk;
3876
3877         if (nvlist_lookup_nvlist_array(config,
3878             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
3879                 if (nvlist_lookup_uint64(config,
3880                     ZPOOL_CONFIG_WHOLE_DISK,
3881                     &wholedisk) != 0 || !wholedisk) {
3882                         return (MAXOFFSET_T);
3883                 }
3884                 if (read_efi_label(config, &sb) < 0)
3885                         sb = MAXOFFSET_T;
3886                 return (sb);
3887         }
3888
3889         for (c = 0; c < children; c++) {
3890                 sb = find_start_block(child[c]);
3891                 if (sb != MAXOFFSET_T) {
3892                         return (sb);
3893                 }
3894         }
3895         return (MAXOFFSET_T);
3896 }
3897 #endif /* sun */
3898
3899 /*
3900  * Label an individual disk.  The name provided is the short name,
3901  * stripped of any leading /dev path.
3902  */
3903 int
3904 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, const char *name)
3905 {
3906 #ifdef sun
3907         char path[MAXPATHLEN];
3908         struct dk_gpt *vtoc;
3909         int fd;
3910         size_t resv = EFI_MIN_RESV_SIZE;
3911         uint64_t slice_size;
3912         diskaddr_t start_block;
3913         char errbuf[1024];
3914
3915         /* prepare an error message just in case */
3916         (void) snprintf(errbuf, sizeof (errbuf),
3917             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
3918
3919         if (zhp) {
3920                 nvlist_t *nvroot;
3921
3922                 if (zpool_is_bootable(zhp)) {
3923                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3924                             "EFI labeled devices are not supported on root "
3925                             "pools."));
3926                         return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
3927                 }
3928
3929                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
3930                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3931
3932                 if (zhp->zpool_start_block == 0)
3933                         start_block = find_start_block(nvroot);
3934                 else
3935                         start_block = zhp->zpool_start_block;
3936                 zhp->zpool_start_block = start_block;
3937         } else {
3938                 /* new pool */
3939                 start_block = NEW_START_BLOCK;
3940         }
3941
3942         (void) snprintf(path, sizeof (path), "%s/%s%s", RDISK_ROOT, name,
3943             BACKUP_SLICE);
3944
3945         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
3946                 /*
3947                  * This shouldn't happen.  We've long since verified that this
3948                  * is a valid device.
3949                  */
3950                 zfs_error_aux(hdl,
3951                     dgettext(TEXT_DOMAIN, "unable to open device"));
3952                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
3953         }
3954
3955         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
3956                 /*
3957                  * The only way this can fail is if we run out of memory, or we
3958                  * were unable to read the disk's capacity
3959                  */
3960                 if (errno == ENOMEM)
3961                         (void) no_memory(hdl);
3962
3963                 (void) close(fd);
3964                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3965                     "unable to read disk capacity"), name);
3966
3967                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
3968         }
3969
3970         slice_size = vtoc->efi_last_u_lba + 1;
3971         slice_size -= EFI_MIN_RESV_SIZE;
3972         if (start_block == MAXOFFSET_T)
3973                 start_block = NEW_START_BLOCK;
3974         slice_size -= start_block;
3975
3976         vtoc->efi_parts[0].p_start = start_block;
3977         vtoc->efi_parts[0].p_size = slice_size;
3978
3979         /*
3980          * Why we use V_USR: V_BACKUP confuses users, and is considered
3981          * disposable by some EFI utilities (since EFI doesn't have a backup
3982          * slice).  V_UNASSIGNED is supposed to be used only for zero size
3983          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
3984          * etc. were all pretty specific.  V_USR is as close to reality as we
3985          * can get, in the absence of V_OTHER.
3986          */
3987         vtoc->efi_parts[0].p_tag = V_USR;
3988         (void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
3989
3990         vtoc->efi_parts[8].p_start = slice_size + start_block;
3991         vtoc->efi_parts[8].p_size = resv;
3992         vtoc->efi_parts[8].p_tag = V_RESERVED;
3993
3994         if (efi_write(fd, vtoc) != 0) {
3995                 /*
3996                  * Some block drivers (like pcata) may not support EFI
3997                  * GPT labels.  Print out a helpful error message dir-
3998                  * ecting the user to manually label the disk and give
3999                  * a specific slice.
4000                  */
4001                 (void) close(fd);
4002                 efi_free(vtoc);
4003
4004                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4005                     "try using fdisk(1M) and then provide a specific slice"));
4006                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4007         }
4008
4009         (void) close(fd);
4010         efi_free(vtoc);
4011 #endif /* sun */
4012         return (0);
4013 }
4014
4015 static boolean_t
4016 supported_dump_vdev_type(libzfs_handle_t *hdl, nvlist_t *config, char *errbuf)
4017 {
4018         char *type;
4019         nvlist_t **child;
4020         uint_t children, c;
4021
4022         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_TYPE, &type) == 0);
4023         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0 ||
4024             strcmp(type, VDEV_TYPE_FILE) == 0 ||
4025             strcmp(type, VDEV_TYPE_LOG) == 0 ||
4026             strcmp(type, VDEV_TYPE_HOLE) == 0 ||
4027             strcmp(type, VDEV_TYPE_MISSING) == 0) {
4028                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4029                     "vdev type '%s' is not supported"), type);
4030                 (void) zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf);
4031                 return (B_FALSE);
4032         }
4033         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
4034             &child, &children) == 0) {
4035                 for (c = 0; c < children; c++) {
4036                         if (!supported_dump_vdev_type(hdl, child[c], errbuf))
4037                                 return (B_FALSE);
4038                 }
4039         }
4040         return (B_TRUE);
4041 }
4042
4043 /*
4044  * check if this zvol is allowable for use as a dump device; zero if
4045  * it is, > 0 if it isn't, < 0 if it isn't a zvol
4046  */
4047 int
4048 zvol_check_dump_config(char *arg)
4049 {
4050         zpool_handle_t *zhp = NULL;
4051         nvlist_t *config, *nvroot;
4052         char *p, *volname;
4053         nvlist_t **top;
4054         uint_t toplevels;
4055         libzfs_handle_t *hdl;
4056         char errbuf[1024];
4057         char poolname[ZPOOL_MAXNAMELEN];
4058         int pathlen = strlen(ZVOL_FULL_DEV_DIR);
4059         int ret = 1;
4060
4061         if (strncmp(arg, ZVOL_FULL_DEV_DIR, pathlen)) {
4062                 return (-1);
4063         }
4064
4065         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
4066             "dump is not supported on device '%s'"), arg);
4067
4068         if ((hdl = libzfs_init()) == NULL)
4069                 return (1);
4070         libzfs_print_on_error(hdl, B_TRUE);
4071
4072         volname = arg + pathlen;
4073
4074         /* check the configuration of the pool */
4075         if ((p = strchr(volname, '/')) == NULL) {
4076                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4077                     "malformed dataset name"));
4078                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4079                 return (1);
4080         } else if (p - volname >= ZFS_MAXNAMELEN) {
4081                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4082                     "dataset name is too long"));
4083                 (void) zfs_error(hdl, EZFS_NAMETOOLONG, errbuf);
4084                 return (1);
4085         } else {
4086                 (void) strncpy(poolname, volname, p - volname);
4087                 poolname[p - volname] = '\0';
4088         }
4089
4090         if ((zhp = zpool_open(hdl, poolname)) == NULL) {
4091                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4092                     "could not open pool '%s'"), poolname);
4093                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
4094                 goto out;
4095         }
4096         config = zpool_get_config(zhp, NULL);
4097         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4098             &nvroot) != 0) {
4099                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4100                     "could not obtain vdev configuration for  '%s'"), poolname);
4101                 (void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
4102                 goto out;
4103         }
4104
4105         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4106             &top, &toplevels) == 0);
4107         if (toplevels != 1) {
4108                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4109                     "'%s' has multiple top level vdevs"), poolname);
4110                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, errbuf);
4111                 goto out;
4112         }
4113
4114         if (!supported_dump_vdev_type(hdl, top[0], errbuf)) {
4115                 goto out;
4116         }
4117         ret = 0;
4118
4119 out:
4120         if (zhp)
4121                 zpool_close(zhp);
4122         libzfs_fini(hdl);
4123         return (ret);
4124 }