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 (c) 2011, 2015 by Delphix. All rights reserved.
  25  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  26  * Copyright 2016 Nexenta Systems, Inc.
  27  * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>
  28  */
  29
  30 #include <sys/types.h>
  31 #include <sys/stat.h>
  32 #include <ctype.h>
  33 #include <errno.h>
  34 #include <devid.h>
  35 #include <fcntl.h>
  36 #include <libintl.h>
  37 #include <stdio.h>
  38 #include <stdlib.h>
  39 #include <strings.h>
  40 #include <unistd.h>
  41 #include <libgen.h>
  42 #include <sys/zfs_ioctl.h>
  43 #include <dlfcn.h>
  44
  45 #include "zfs_namecheck.h"
  46 #include "zfs_prop.h"
  47 #include "libzfs_impl.h"
  48 #include "zfs_comutil.h"
  49 #include "zfeature_common.h"
  50
  51 static int read_efi_label(nvlist_t *config, diskaddr_t *sb);
  52
  53 #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         default:
 205                 break;
 206         }
 207
 208         return (gettext("UNKNOWN"));
 209 }
 210
 211 /*
 212  * Map POOL STATE to printed strings.
 213  */
 214 const char *
 215 zpool_pool_state_to_name(pool_state_t state)
 216 {
 217         switch (state) {
 218         case POOL_STATE_ACTIVE:
 219                 return (gettext("ACTIVE"));
 220         case POOL_STATE_EXPORTED:
 221                 return (gettext("EXPORTED"));
 222         case POOL_STATE_DESTROYED:
 223                 return (gettext("DESTROYED"));
 224         case POOL_STATE_SPARE:
 225                 return (gettext("SPARE"));
 226         case POOL_STATE_L2CACHE:
 227                 return (gettext("L2CACHE"));
 228         case POOL_STATE_UNINITIALIZED:
 229                 return (gettext("UNINITIALIZED"));
 230         case POOL_STATE_UNAVAIL:
 231                 return (gettext("UNAVAIL"));
 232         case POOL_STATE_POTENTIALLY_ACTIVE:
 233                 return (gettext("POTENTIALLY_ACTIVE"));
 234         }
 235
 236         return (gettext("UNKNOWN"));
 237 }
 238
 239 /*
 240  * Get a zpool property value for 'prop' and return the value in
 241  * a pre-allocated buffer.
 242  */
 243 int
 244 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf, size_t len,
 245     zprop_source_t *srctype, boolean_t literal)
 246 {
 247         uint64_t intval;
 248         const char *strval;
 249         zprop_source_t src = ZPROP_SRC_NONE;
 250         nvlist_t *nvroot;
 251         vdev_stat_t *vs;
 252         uint_t vsc;
 253
 254         if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
 255                 switch (prop) {
 256                 case ZPOOL_PROP_NAME:
 257                         (void) strlcpy(buf, zpool_get_name(zhp), len);
 258                         break;
 259
 260                 case ZPOOL_PROP_HEALTH:
 261                         (void) strlcpy(buf,
 262                             zpool_pool_state_to_name(POOL_STATE_UNAVAIL), len);
 263                         break;
 264
 265                 case ZPOOL_PROP_GUID:
 266                         intval = zpool_get_prop_int(zhp, prop, &src);
 267                         (void) snprintf(buf, len, "%llu", intval);
 268                         break;
 269
 270                 case ZPOOL_PROP_ALTROOT:
 271                 case ZPOOL_PROP_CACHEFILE:
 272                 case ZPOOL_PROP_COMMENT:
 273                         if (zhp->zpool_props != NULL ||
 274                             zpool_get_all_props(zhp) == 0) {
 275                                 (void) strlcpy(buf,
 276                                     zpool_get_prop_string(zhp, prop, &src),
 277                                     len);
 278                                 break;
 279                         }
 280                         /* FALLTHROUGH */
 281                 default:
 282                         (void) strlcpy(buf, "-", len);
 283                         break;
 284                 }
 285
 286                 if (srctype != NULL)
 287                         *srctype = src;
 288                 return (0);
 289         }
 290
 291         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
 292             prop != ZPOOL_PROP_NAME)
 293                 return (-1);
 294
 295         switch (zpool_prop_get_type(prop)) {
 296         case PROP_TYPE_STRING:
 297                 (void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
 298                     len);
 299                 break;
 300
 301         case PROP_TYPE_NUMBER:
 302                 intval = zpool_get_prop_int(zhp, prop, &src);
 303
 304                 switch (prop) {
 305                 case ZPOOL_PROP_SIZE:
 306                 case ZPOOL_PROP_ALLOCATED:
 307                 case ZPOOL_PROP_FREE:
 308                 case ZPOOL_PROP_FREEING:
 309                 case ZPOOL_PROP_LEAKED:
 310                         if (literal) {
 311                                 (void) snprintf(buf, len, "%llu",
 312                                     (u_longlong_t)intval);
 313                         } else {
 314                                 (void) zfs_nicenum(intval, buf, len);
 315                         }
 316                         break;
 317                 case ZPOOL_PROP_EXPANDSZ:
 318                         if (intval == 0) {
 319                                 (void) strlcpy(buf, "-", len);
 320                         } else if (literal) {
 321                                 (void) snprintf(buf, len, "%llu",
 322                                     (u_longlong_t)intval);
 323                         } else {
 324                                 (void) zfs_nicenum(intval, buf, len);
 325                         }
 326                         break;
 327                 case ZPOOL_PROP_CAPACITY:
 328                         if (literal) {
 329                                 (void) snprintf(buf, len, "%llu",
 330                                     (u_longlong_t)intval);
 331                         } else {
 332                                 (void) snprintf(buf, len, "%llu%%",
 333                                     (u_longlong_t)intval);
 334                         }
 335                         break;
 336                 case ZPOOL_PROP_FRAGMENTATION:
 337                         if (intval == UINT64_MAX) {
 338                                 (void) strlcpy(buf, "-", len);
 339                         } else {
 340                                 (void) snprintf(buf, len, "%llu%%",
 341                                     (u_longlong_t)intval);
 342                         }
 343                         break;
 344                 case ZPOOL_PROP_DEDUPRATIO:
 345                         (void) snprintf(buf, len, "%llu.%02llux",
 346                             (u_longlong_t)(intval / 100),
 347                             (u_longlong_t)(intval % 100));
 348                         break;
 349                 case ZPOOL_PROP_HEALTH:
 350                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 351                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 352                         verify(nvlist_lookup_uint64_array(nvroot,
 353                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
 354                             == 0);
 355
 356                         (void) strlcpy(buf, zpool_state_to_name(intval,
 357                             vs->vs_aux), len);
 358                         break;
 359                 case ZPOOL_PROP_VERSION:
 360                         if (intval >= SPA_VERSION_FEATURES) {
 361                                 (void) snprintf(buf, len, "-");
 362                                 break;
 363                         }
 364                         /* FALLTHROUGH */
 365                 default:
 366                         (void) snprintf(buf, len, "%llu", intval);
 367                 }
 368                 break;
 369
 370         case PROP_TYPE_INDEX:
 371                 intval = zpool_get_prop_int(zhp, prop, &src);
 372                 if (zpool_prop_index_to_string(prop, intval, &strval)
 373                     != 0)
 374                         return (-1);
 375                 (void) strlcpy(buf, strval, len);
 376                 break;
 377
 378         default:
 379                 abort();
 380         }
 381
 382         if (srctype)
 383                 *srctype = src;
 384
 385         return (0);
 386 }
 387
 388 /*
 389  * Check if the bootfs name has the same pool name as it is set to.
 390  * Assuming bootfs is a valid dataset name.
 391  */
 392 static boolean_t
 393 bootfs_name_valid(const char *pool, char *bootfs)
 394 {
 395         int len = strlen(pool);
 396
 397         if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
 398                 return (B_FALSE);
 399
 400         if (strncmp(pool, bootfs, len) == 0 &&
 401             (bootfs[len] == '/' || bootfs[len] == '\0'))
 402                 return (B_TRUE);
 403
 404         return (B_FALSE);
 405 }
 406
 407 boolean_t
 408 zpool_is_bootable(zpool_handle_t *zhp)
 409 {
 410         char bootfs[ZFS_MAX_DATASET_NAME_LEN];
 411
 412         return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
 413             sizeof (bootfs), NULL, B_FALSE) == 0 && strncmp(bootfs, "-",
 414             sizeof (bootfs)) != 0);
 415 }
 416
 417
 418 /*
 419  * Given an nvlist of zpool properties to be set, validate that they are
 420  * correct, and parse any numeric properties (index, boolean, etc) if they are
 421  * specified as strings.
 422  */
 423 static nvlist_t *
 424 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 425     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
 426 {
 427         nvpair_t *elem;
 428         nvlist_t *retprops;
 429         zpool_prop_t prop;
 430         char *strval;
 431         uint64_t intval;
 432         char *slash, *check;
 433         struct stat64 statbuf;
 434         zpool_handle_t *zhp;
 435
 436         if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
 437                 (void) no_memory(hdl);
 438                 return (NULL);
 439         }
 440
 441         elem = NULL;
 442         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 443                 const char *propname = nvpair_name(elem);
 444
 445                 prop = zpool_name_to_prop(propname);
 446                 if (prop == ZPROP_INVAL && zpool_prop_feature(propname)) {
 447                         int err;
 448                         char *fname = strchr(propname, '@') + 1;
 449
 450                         err = zfeature_lookup_name(fname, NULL);
 451                         if (err != 0) {
 452                                 ASSERT3U(err, ==, ENOENT);
 453                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 454                                     "invalid feature '%s'"), fname);
 455                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 456                                 goto error;
 457                         }
 458
 459                         if (nvpair_type(elem) != DATA_TYPE_STRING) {
 460                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 461                                     "'%s' must be a string"), propname);
 462                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 463                                 goto error;
 464                         }
 465
 466                         (void) nvpair_value_string(elem, &strval);
 467                         if (strcmp(strval, ZFS_FEATURE_ENABLED) != 0) {
 468                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 469                                     "property '%s' can only be set to "
 470                                     "'enabled'"), propname);
 471                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 472                                 goto error;
 473                         }
 474
 475                         if (nvlist_add_uint64(retprops, propname, 0) != 0) {
 476                                 (void) no_memory(hdl);
 477                                 goto error;
 478                         }
 479                         continue;
 480                 }
 481
 482                 /*
 483                  * Make sure this property is valid and applies to this type.
 484                  */
 485                 if (prop == ZPROP_INVAL) {
 486                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 487                             "invalid property '%s'"), propname);
 488                         (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 489                         goto error;
 490                 }
 491
 492                 if (zpool_prop_readonly(prop)) {
 493                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 494                             "is readonly"), propname);
 495                         (void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
 496                         goto error;
 497                 }
 498
 499                 if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
 500                     &strval, &intval, errbuf) != 0)
 501                         goto error;
 502
 503                 /*
 504                  * Perform additional checking for specific properties.
 505                  */
 506                 switch (prop) {
 507                 case ZPOOL_PROP_VERSION:
 508                         if (intval < version ||
 509                             !SPA_VERSION_IS_SUPPORTED(intval)) {
 510                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 511                                     "property '%s' number %d is invalid."),
 512                                     propname, intval);
 513                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 514                                 goto error;
 515                         }
 516                         break;
 517
 518                 case ZPOOL_PROP_BOOTFS:
 519                         if (flags.create || flags.import) {
 520                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 521                                     "property '%s' cannot be set at creation "
 522                                     "or import time"), propname);
 523                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 524                                 goto error;
 525                         }
 526
 527                         if (version < SPA_VERSION_BOOTFS) {
 528                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 529                                     "pool must be upgraded to support "
 530                                     "'%s' property"), propname);
 531                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 532                                 goto error;
 533                         }
 534
 535                         /*
 536                          * bootfs property value has to be a dataset name and
 537                          * the dataset has to be in the same pool as it sets to.
 538                          */
 539                         if (strval[0] != '\0' && !bootfs_name_valid(poolname,
 540                             strval)) {
 541                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 542                                     "is an invalid name"), strval);
 543                                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
 544                                 goto error;
 545                         }
 546
 547                         if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
 548                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 549                                     "could not open pool '%s'"), poolname);
 550                                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
 551                                 goto error;
 552                         }
 553                         zpool_close(zhp);
 554                         break;
 555
 556                 case ZPOOL_PROP_ALTROOT:
 557                         if (!flags.create && !flags.import) {
 558                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 559                                     "property '%s' can only be set during pool "
 560                                     "creation or import"), propname);
 561                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 562                                 goto error;
 563                         }
 564
 565                         if (strval[0] != '/') {
 566                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 567                                     "bad alternate root '%s'"), strval);
 568                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 569                                 goto error;
 570                         }
 571                         break;
 572
 573                 case ZPOOL_PROP_CACHEFILE:
 574                         if (strval[0] == '\0')
 575                                 break;
 576
 577                         if (strcmp(strval, "none") == 0)
 578                                 break;
 579
 580                         if (strval[0] != '/') {
 581                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 582                                     "property '%s' must be empty, an "
 583                                     "absolute path, or 'none'"), propname);
 584                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 585                                 goto error;
 586                         }
 587
 588                         slash = strrchr(strval, '/');
 589
 590                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 591                             strcmp(slash, "/..") == 0) {
 592                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 593                                     "'%s' is not a valid file"), strval);
 594                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 595                                 goto error;
 596                         }
 597
 598                         *slash = '\0';
 599
 600                         if (strval[0] != '\0' &&
 601                             (stat64(strval, &statbuf) != 0 ||
 602                             !S_ISDIR(statbuf.st_mode))) {
 603                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 604                                     "'%s' is not a valid directory"),
 605                                     strval);
 606                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 607                                 goto error;
 608                         }
 609
 610                         *slash = '/';
 611                         break;
 612
 613                 case ZPOOL_PROP_COMMENT:
 614                         for (check = strval; *check != '\0'; check++) {
 615                                 if (!isprint(*check)) {
 616                                         zfs_error_aux(hdl,
 617                                             dgettext(TEXT_DOMAIN,
 618                                             "comment may only have printable "
 619                                             "characters"));
 620                                         (void) zfs_error(hdl, EZFS_BADPROP,
 621                                             errbuf);
 622                                         goto error;
 623                                 }
 624                         }
 625                         if (strlen(strval) > ZPROP_MAX_COMMENT) {
 626                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 627                                     "comment must not exceed %d characters"),
 628                                     ZPROP_MAX_COMMENT);
 629                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 630                                 goto error;
 631                         }
 632                         break;
 633
 634                 case ZPOOL_PROP_READONLY:
 635                         if (!flags.import) {
 636                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 637                                     "property '%s' can only be set at "
 638                                     "import time"), propname);
 639                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 640                                 goto error;
 641                         }
 642                         break;
 643
 644                 case ZPOOL_PROP_TNAME:
 645                         if (!flags.create) {
 646                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 647                                     "property '%s' can only be set at "
 648                                     "creation time"), propname);
 649                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 650                                 goto error;
 651                         }
 652                         break;
 653
 654                 default:
 655                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 656                             "property '%s'(%d) not defined"), propname, prop);
 657                         break;
 658                 }
 659         }
 660
 661         return (retprops);
 662 error:
 663         nvlist_free(retprops);
 664         return (NULL);
 665 }
 666
 667 /*
 668  * Set zpool property : propname=propval.
 669  */
 670 int
 671 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
 672 {
 673         zfs_cmd_t zc = { 0 };
 674         int ret = -1;
 675         char errbuf[1024];
 676         nvlist_t *nvl = NULL;
 677         nvlist_t *realprops;
 678         uint64_t version;
 679         prop_flags_t flags = { 0 };
 680
 681         (void) snprintf(errbuf, sizeof (errbuf),
 682             dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
 683             zhp->zpool_name);
 684
 685         if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
 686                 return (no_memory(zhp->zpool_hdl));
 687
 688         if (nvlist_add_string(nvl, propname, propval) != 0) {
 689                 nvlist_free(nvl);
 690                 return (no_memory(zhp->zpool_hdl));
 691         }
 692
 693         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
 694         if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
 695             zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
 696                 nvlist_free(nvl);
 697                 return (-1);
 698         }
 699
 700         nvlist_free(nvl);
 701         nvl = realprops;
 702
 703         /*
 704          * Execute the corresponding ioctl() to set this property.
 705          */
 706         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 707
 708         if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
 709                 nvlist_free(nvl);
 710                 return (-1);
 711         }
 712
 713         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
 714
 715         zcmd_free_nvlists(&zc);
 716         nvlist_free(nvl);
 717
 718         if (ret)
 719                 (void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
 720         else
 721                 (void) zpool_props_refresh(zhp);
 722
 723         return (ret);
 724 }
 725
 726 int
 727 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
 728 {
 729         libzfs_handle_t *hdl = zhp->zpool_hdl;
 730         zprop_list_t *entry;
 731         char buf[ZFS_MAXPROPLEN];
 732         nvlist_t *features = NULL;
 733         zprop_list_t **last;
 734         boolean_t firstexpand = (NULL == *plp);
 735
 736         if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
 737                 return (-1);
 738
 739         last = plp;
 740         while (*last != NULL)
 741                 last = &(*last)->pl_next;
 742
 743         if ((*plp)->pl_all)
 744                 features = zpool_get_features(zhp);
 745
 746         if ((*plp)->pl_all && firstexpand) {
 747                 for (int i = 0; i < SPA_FEATURES; i++) {
 748                         zprop_list_t *entry = zfs_alloc(hdl,
 749                             sizeof (zprop_list_t));
 750                         entry->pl_prop = ZPROP_INVAL;
 751                         entry->pl_user_prop = zfs_asprintf(hdl, "feature@%s",
 752                             spa_feature_table[i].fi_uname);
 753                         entry->pl_width = strlen(entry->pl_user_prop);
 754                         entry->pl_all = B_TRUE;
 755
 756                         *last = entry;
 757                         last = &entry->pl_next;
 758                 }
 759         }
 760
 761         /* add any unsupported features */
 762         for (nvpair_t *nvp = nvlist_next_nvpair(features, NULL);
 763             nvp != NULL; nvp = nvlist_next_nvpair(features, nvp)) {
 764                 char *propname;
 765                 boolean_t found;
 766                 zprop_list_t *entry;
 767
 768                 if (zfeature_is_supported(nvpair_name(nvp)))
 769                         continue;
 770
 771                 propname = zfs_asprintf(hdl, "unsupported@%s",
 772                     nvpair_name(nvp));
 773
 774                 /*
 775                  * Before adding the property to the list make sure that no
 776                  * other pool already added the same property.
 777                  */
 778                 found = B_FALSE;
 779                 entry = *plp;
 780                 while (entry != NULL) {
 781                         if (entry->pl_user_prop != NULL &&
 782                             strcmp(propname, entry->pl_user_prop) == 0) {
 783                                 found = B_TRUE;
 784                                 break;
 785                         }
 786                         entry = entry->pl_next;
 787                 }
 788                 if (found) {
 789                         free(propname);
 790                         continue;
 791                 }
 792
 793                 entry = zfs_alloc(hdl, sizeof (zprop_list_t));
 794                 entry->pl_prop = ZPROP_INVAL;
 795                 entry->pl_user_prop = propname;
 796                 entry->pl_width = strlen(entry->pl_user_prop);
 797                 entry->pl_all = B_TRUE;
 798
 799                 *last = entry;
 800                 last = &entry->pl_next;
 801         }
 802
 803         for (entry = *plp; entry != NULL; entry = entry->pl_next) {
 804
 805                 if (entry->pl_fixed)
 806                         continue;
 807
 808                 if (entry->pl_prop != ZPROP_INVAL &&
 809                     zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
 810                     NULL, B_FALSE) == 0) {
 811                         if (strlen(buf) > entry->pl_width)
 812                                 entry->pl_width = strlen(buf);
 813                 }
 814         }
 815
 816         return (0);
 817 }
 818
 819 /*
 820  * Get the state for the given feature on the given ZFS pool.
 821  */
 822 int
 823 zpool_prop_get_feature(zpool_handle_t *zhp, const char *propname, char *buf,
 824     size_t len)
 825 {
 826         uint64_t refcount;
 827         boolean_t found = B_FALSE;
 828         nvlist_t *features = zpool_get_features(zhp);
 829         boolean_t supported;
 830         const char *feature = strchr(propname, '@') + 1;
 831
 832         supported = zpool_prop_feature(propname);
 833         ASSERT(supported || zpool_prop_unsupported(propname));
 834
 835         /*
 836          * Convert from feature name to feature guid. This conversion is
 837          * unecessary for unsupported@... properties because they already
 838          * use guids.
 839          */
 840         if (supported) {
 841                 int ret;
 842                 spa_feature_t fid;
 843
 844                 ret = zfeature_lookup_name(feature, &fid);
 845                 if (ret != 0) {
 846                         (void) strlcpy(buf, "-", len);
 847                         return (ENOTSUP);
 848                 }
 849                 feature = spa_feature_table[fid].fi_guid;
 850         }
 851
 852         if (nvlist_lookup_uint64(features, feature, &refcount) == 0)
 853                 found = B_TRUE;
 854
 855         if (supported) {
 856                 if (!found) {
 857                         (void) strlcpy(buf, ZFS_FEATURE_DISABLED, len);
 858                 } else  {
 859                         if (refcount == 0)
 860                                 (void) strlcpy(buf, ZFS_FEATURE_ENABLED, len);
 861                         else
 862                                 (void) strlcpy(buf, ZFS_FEATURE_ACTIVE, len);
 863                 }
 864         } else {
 865                 if (found) {
 866                         if (refcount == 0) {
 867                                 (void) strcpy(buf, ZFS_UNSUPPORTED_INACTIVE);
 868                         } else {
 869                                 (void) strcpy(buf, ZFS_UNSUPPORTED_READONLY);
 870                         }
 871                 } else {
 872                         (void) strlcpy(buf, "-", len);
 873                         return (ENOTSUP);
 874                 }
 875         }
 876
 877         return (0);
 878 }
 879
 880 /*
 881  * Don't start the slice at the default block of 34; many storage
 882  * devices will use a stripe width of 128k, so start there instead.
 883  */
 884 #define NEW_START_BLOCK 256
 885
 886 /*
 887  * Validate the given pool name, optionally putting an extended error message in
 888  * 'buf'.
 889  */
 890 boolean_t
 891 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
 892 {
 893         namecheck_err_t why;
 894         char what;
 895         int ret;
 896
 897         ret = pool_namecheck(pool, &why, &what);
 898
 899         /*
 900          * The rules for reserved pool names were extended at a later point.
 901          * But we need to support users with existing pools that may now be
 902          * invalid.  So we only check for this expanded set of names during a
 903          * create (or import), and only in userland.
 904          */
 905         if (ret == 0 && !isopen &&
 906             (strncmp(pool, "mirror", 6) == 0 ||
 907             strncmp(pool, "raidz", 5) == 0 ||
 908             strncmp(pool, "spare", 5) == 0 ||
 909             strcmp(pool, "log") == 0)) {
 910                 if (hdl != NULL)
 911                         zfs_error_aux(hdl,
 912                             dgettext(TEXT_DOMAIN, "name is reserved"));
 913                 return (B_FALSE);
 914         }
 915
 916
 917         if (ret != 0) {
 918                 if (hdl != NULL) {
 919                         switch (why) {
 920                         case NAME_ERR_TOOLONG:
 921                                 zfs_error_aux(hdl,
 922                                     dgettext(TEXT_DOMAIN, "name is too long"));
 923                                 break;
 924
 925                         case NAME_ERR_INVALCHAR:
 926                                 zfs_error_aux(hdl,
 927                                     dgettext(TEXT_DOMAIN, "invalid character "
 928                                     "'%c' in pool name"), what);
 929                                 break;
 930
 931                         case NAME_ERR_NOLETTER:
 932                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 933                                     "name must begin with a letter"));
 934                                 break;
 935
 936                         case NAME_ERR_RESERVED:
 937                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 938                                     "name is reserved"));
 939                                 break;
 940
 941                         case NAME_ERR_DISKLIKE:
 942                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 943                                     "pool name is reserved"));
 944                                 break;
 945
 946                         case NAME_ERR_LEADING_SLASH:
 947                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 948                                     "leading slash in name"));
 949                                 break;
 950
 951                         case NAME_ERR_EMPTY_COMPONENT:
 952                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 953                                     "empty component in name"));
 954                                 break;
 955
 956                         case NAME_ERR_TRAILING_SLASH:
 957                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 958                                     "trailing slash in name"));
 959                                 break;
 960
 961                         case NAME_ERR_MULTIPLE_AT:
 962                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 963                                     "multiple '@' delimiters in name"));
 964                                 break;
 965
 966                         default:
 967                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 968                                     "(%d) not defined"), why);
 969                                 break;
 970                         }
 971                 }
 972                 return (B_FALSE);
 973         }
 974
 975         return (B_TRUE);
 976 }
 977
 978 /*
 979  * Open a handle to the given pool, even if the pool is currently in the FAULTED
 980  * state.
 981  */
 982 zpool_handle_t *
 983 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
 984 {
 985         zpool_handle_t *zhp;
 986         boolean_t missing;
 987
 988         /*
 989          * Make sure the pool name is valid.
 990          */
 991         if (!zpool_name_valid(hdl, B_TRUE, pool)) {
 992                 (void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
 993                     dgettext(TEXT_DOMAIN, "cannot open '%s'"),
 994                     pool);
 995                 return (NULL);
 996         }
 997
 998         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
 999                 return (NULL);
1000
1001         zhp->zpool_hdl = hdl;
1002         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1003
1004         if (zpool_refresh_stats(zhp, &missing) != 0) {
1005                 zpool_close(zhp);
1006                 return (NULL);
1007         }
1008
1009         if (missing) {
1010                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
1011                 (void) zfs_error_fmt(hdl, EZFS_NOENT,
1012                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
1013                 zpool_close(zhp);
1014                 return (NULL);
1015         }
1016
1017         return (zhp);
1018 }
1019
1020 /*
1021  * Like the above, but silent on error.  Used when iterating over pools (because
1022  * the configuration cache may be out of date).
1023  */
1024 int
1025 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
1026 {
1027         zpool_handle_t *zhp;
1028         boolean_t missing;
1029
1030         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1031                 return (-1);
1032
1033         zhp->zpool_hdl = hdl;
1034         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1035
1036         if (zpool_refresh_stats(zhp, &missing) != 0) {
1037                 zpool_close(zhp);
1038                 return (-1);
1039         }
1040
1041         if (missing) {
1042                 zpool_close(zhp);
1043                 *ret = NULL;
1044                 return (0);
1045         }
1046
1047         *ret = zhp;
1048         return (0);
1049 }
1050
1051 /*
1052  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
1053  * state.
1054  */
1055 zpool_handle_t *
1056 zpool_open(libzfs_handle_t *hdl, const char *pool)
1057 {
1058         zpool_handle_t *zhp;
1059
1060         if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
1061                 return (NULL);
1062
1063         if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
1064                 (void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
1065                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
1066                 zpool_close(zhp);
1067                 return (NULL);
1068         }
1069
1070         return (zhp);
1071 }
1072
1073 /*
1074  * Close the handle.  Simply frees the memory associated with the handle.
1075  */
1076 void
1077 zpool_close(zpool_handle_t *zhp)
1078 {
1079         nvlist_free(zhp->zpool_config);
1080         nvlist_free(zhp->zpool_old_config);
1081         nvlist_free(zhp->zpool_props);
1082         free(zhp);
1083 }
1084
1085 /*
1086  * Return the name of the pool.
1087  */
1088 const char *
1089 zpool_get_name(zpool_handle_t *zhp)
1090 {
1091         return (zhp->zpool_name);
1092 }
1093
1094
1095 /*
1096  * Return the state of the pool (ACTIVE or UNAVAILABLE)
1097  */
1098 int
1099 zpool_get_state(zpool_handle_t *zhp)
1100 {
1101         return (zhp->zpool_state);
1102 }
1103
1104 /*
1105  * Create the named pool, using the provided vdev list.  It is assumed
1106  * that the consumer has already validated the contents of the nvlist, so we
1107  * don't have to worry about error semantics.
1108  */
1109 int
1110 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
1111     nvlist_t *props, nvlist_t *fsprops)
1112 {
1113         zfs_cmd_t zc = { 0 };
1114         nvlist_t *zc_fsprops = NULL;
1115         nvlist_t *zc_props = NULL;
1116         char msg[1024];
1117         int ret = -1;
1118
1119         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1120             "cannot create '%s'"), pool);
1121
1122         if (!zpool_name_valid(hdl, B_FALSE, pool))
1123                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
1124
1125         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1126                 return (-1);
1127
1128         if (props) {
1129                 prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
1130
1131                 if ((zc_props = zpool_valid_proplist(hdl, pool, props,
1132                     SPA_VERSION_1, flags, msg)) == NULL) {
1133                         goto create_failed;
1134                 }
1135         }
1136
1137         if (fsprops) {
1138                 uint64_t zoned;
1139                 char *zonestr;
1140
1141                 zoned = ((nvlist_lookup_string(fsprops,
1142                     zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
1143                     strcmp(zonestr, "on") == 0);
1144
1145                 if ((zc_fsprops = zfs_valid_proplist(hdl, ZFS_TYPE_FILESYSTEM,
1146                     fsprops, zoned, NULL, NULL, msg)) == NULL) {
1147                         goto create_failed;
1148                 }
1149                 if (!zc_props &&
1150                     (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
1151                         goto create_failed;
1152                 }
1153                 if (nvlist_add_nvlist(zc_props,
1154                     ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
1155                         goto create_failed;
1156                 }
1157         }
1158
1159         if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
1160                 goto create_failed;
1161
1162         (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
1163
1164         if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
1165
1166                 zcmd_free_nvlists(&zc);
1167                 nvlist_free(zc_props);
1168                 nvlist_free(zc_fsprops);
1169
1170                 switch (errno) {
1171                 case EBUSY:
1172                         /*
1173                          * This can happen if the user has specified the same
1174                          * device multiple times.  We can't reliably detect this
1175                          * until we try to add it and see we already have a
1176                          * label.
1177                          */
1178                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1179                             "one or more vdevs refer to the same device"));
1180                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1181
1182                 case ERANGE:
1183                         /*
1184                          * This happens if the record size is smaller or larger
1185                          * than the allowed size range, or not a power of 2.
1186                          *
1187                          * NOTE: although zfs_valid_proplist is called earlier,
1188                          * this case may have slipped through since the
1189                          * pool does not exist yet and it is therefore
1190                          * impossible to read properties e.g. max blocksize
1191                          * from the pool.
1192                          */
1193                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1194                             "record size invalid"));
1195                         return (zfs_error(hdl, EZFS_BADPROP, msg));
1196
1197                 case EOVERFLOW:
1198                         /*
1199                          * This occurs when one of the devices is below
1200                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1201                          * device was the problem device since there's no
1202                          * reliable way to determine device size from userland.
1203                          */
1204                         {
1205                                 char buf[64];
1206
1207                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1208
1209                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1210                                     "one or more devices is less than the "
1211                                     "minimum size (%s)"), buf);
1212                         }
1213                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1214
1215                 case ENOSPC:
1216                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1217                             "one or more devices is out of space"));
1218                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1219
1220                 case ENOTBLK:
1221                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1222                             "cache device must be a disk or disk slice"));
1223                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1224
1225                 default:
1226                         return (zpool_standard_error(hdl, errno, msg));
1227                 }
1228         }
1229
1230 create_failed:
1231         zcmd_free_nvlists(&zc);
1232         nvlist_free(zc_props);
1233         nvlist_free(zc_fsprops);
1234         return (ret);
1235 }
1236
1237 /*
1238  * Destroy the given pool.  It is up to the caller to ensure that there are no
1239  * datasets left in the pool.
1240  */
1241 int
1242 zpool_destroy(zpool_handle_t *zhp, const char *log_str)
1243 {
1244         zfs_cmd_t zc = { 0 };
1245         zfs_handle_t *zfp = NULL;
1246         libzfs_handle_t *hdl = zhp->zpool_hdl;
1247         char msg[1024];
1248
1249         if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1250             (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1251                 return (-1);
1252
1253         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1254         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1255
1256         if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1257                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1258                     "cannot destroy '%s'"), zhp->zpool_name);
1259
1260                 if (errno == EROFS) {
1261                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1262                             "one or more devices is read only"));
1263                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1264                 } else {
1265                         (void) zpool_standard_error(hdl, errno, msg);
1266                 }
1267
1268                 if (zfp)
1269                         zfs_close(zfp);
1270                 return (-1);
1271         }
1272
1273         if (zfp) {
1274                 remove_mountpoint(zfp);
1275                 zfs_close(zfp);
1276         }
1277
1278         return (0);
1279 }
1280
1281 /*
1282  * Add the given vdevs to the pool.  The caller must have already performed the
1283  * necessary verification to ensure that the vdev specification is well-formed.
1284  */
1285 int
1286 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1287 {
1288         zfs_cmd_t zc = { 0 };
1289         int ret;
1290         libzfs_handle_t *hdl = zhp->zpool_hdl;
1291         char msg[1024];
1292         nvlist_t **spares, **l2cache;
1293         uint_t nspares, nl2cache;
1294
1295         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1296             "cannot add to '%s'"), zhp->zpool_name);
1297
1298         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1299             SPA_VERSION_SPARES &&
1300             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1301             &spares, &nspares) == 0) {
1302                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1303                     "upgraded to add hot spares"));
1304                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1305         }
1306
1307         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1308             SPA_VERSION_L2CACHE &&
1309             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1310             &l2cache, &nl2cache) == 0) {
1311                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1312                     "upgraded to add cache devices"));
1313                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1314         }
1315
1316         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1317                 return (-1);
1318         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1319
1320         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1321                 switch (errno) {
1322                 case EBUSY:
1323                         /*
1324                          * This can happen if the user has specified the same
1325                          * device multiple times.  We can't reliably detect this
1326                          * until we try to add it and see we already have a
1327                          * label.
1328                          */
1329                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1330                             "one or more vdevs refer to the same device"));
1331                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1332                         break;
1333
1334                 case EOVERFLOW:
1335                         /*
1336                          * This occurrs when one of the devices is below
1337                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1338                          * device was the problem device since there's no
1339                          * reliable way to determine device size from userland.
1340                          */
1341                         {
1342                                 char buf[64];
1343
1344                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1345
1346                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1347                                     "device is less than the minimum "
1348                                     "size (%s)"), buf);
1349                         }
1350                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1351                         break;
1352
1353                 case ENOTSUP:
1354                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1355                             "pool must be upgraded to add these vdevs"));
1356                         (void) zfs_error(hdl, EZFS_BADVERSION, msg);
1357                         break;
1358
1359                 case EDOM:
1360                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1361                             "root pool can not have multiple vdevs"
1362                             " or separate logs"));
1363                         (void) zfs_error(hdl, EZFS_POOL_NOTSUP, msg);
1364                         break;
1365
1366                 case ENOTBLK:
1367                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1368                             "cache device must be a disk or disk slice"));
1369                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1370                         break;
1371
1372                 default:
1373                         (void) zpool_standard_error(hdl, errno, msg);
1374                 }
1375
1376                 ret = -1;
1377         } else {
1378                 ret = 0;
1379         }
1380
1381         zcmd_free_nvlists(&zc);
1382
1383         return (ret);
1384 }
1385
1386 /*
1387  * Exports the pool from the system.  The caller must ensure that there are no
1388  * mounted datasets in the pool.
1389  */
1390 static int
1391 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce,
1392     const char *log_str)
1393 {
1394         zfs_cmd_t zc = { 0 };
1395         char msg[1024];
1396
1397         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1398             "cannot export '%s'"), zhp->zpool_name);
1399
1400         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1401         zc.zc_cookie = force;
1402         zc.zc_guid = hardforce;
1403         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1404
1405         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1406                 switch (errno) {
1407                 case EXDEV:
1408                         zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1409                             "use '-f' to override the following errors:\n"
1410                             "'%s' has an active shared spare which could be"
1411                             " used by other pools once '%s' is exported."),
1412                             zhp->zpool_name, zhp->zpool_name);
1413                         return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1414                             msg));
1415                 default:
1416                         return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1417                             msg));
1418                 }
1419         }
1420
1421         return (0);
1422 }
1423
1424 int
1425 zpool_export(zpool_handle_t *zhp, boolean_t force, const char *log_str)
1426 {
1427         return (zpool_export_common(zhp, force, B_FALSE, log_str));
1428 }
1429
1430 int
1431 zpool_export_force(zpool_handle_t *zhp, const char *log_str)
1432 {
1433         return (zpool_export_common(zhp, B_TRUE, B_TRUE, log_str));
1434 }
1435
1436 static void
1437 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1438     nvlist_t *config)
1439 {
1440         nvlist_t *nv = NULL;
1441         uint64_t rewindto;
1442         int64_t loss = -1;
1443         struct tm t;
1444         char timestr[128];
1445
1446         if (!hdl->libzfs_printerr || config == NULL)
1447                 return;
1448
1449         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1450             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0) {
1451                 return;
1452         }
1453
1454         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1455                 return;
1456         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1457
1458         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1459             strftime(timestr, 128, 0, &t) != 0) {
1460                 if (dryrun) {
1461                         (void) printf(dgettext(TEXT_DOMAIN,
1462                             "Would be able to return %s "
1463                             "to its state as of %s.\n"),
1464                             name, timestr);
1465                 } else {
1466                         (void) printf(dgettext(TEXT_DOMAIN,
1467                             "Pool %s returned to its state as of %s.\n"),
1468                             name, timestr);
1469                 }
1470                 if (loss > 120) {
1471                         (void) printf(dgettext(TEXT_DOMAIN,
1472                             "%s approximately %lld "),
1473                             dryrun ? "Would discard" : "Discarded",
1474                             (loss + 30) / 60);
1475                         (void) printf(dgettext(TEXT_DOMAIN,
1476                             "minutes of transactions.\n"));
1477                 } else if (loss > 0) {
1478                         (void) printf(dgettext(TEXT_DOMAIN,
1479                             "%s approximately %lld "),
1480                             dryrun ? "Would discard" : "Discarded", loss);
1481                         (void) printf(dgettext(TEXT_DOMAIN,
1482                             "seconds of transactions.\n"));
1483                 }
1484         }
1485 }
1486
1487 void
1488 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1489     nvlist_t *config)
1490 {
1491         nvlist_t *nv = NULL;
1492         int64_t loss = -1;
1493         uint64_t edata = UINT64_MAX;
1494         uint64_t rewindto;
1495         struct tm t;
1496         char timestr[128];
1497
1498         if (!hdl->libzfs_printerr)
1499                 return;
1500
1501         if (reason >= 0)
1502                 (void) printf(dgettext(TEXT_DOMAIN, "action: "));
1503         else
1504                 (void) printf(dgettext(TEXT_DOMAIN, "\t"));
1505
1506         /* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1507         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1508             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0 ||
1509             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1510                 goto no_info;
1511
1512         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1513         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1514             &edata);
1515
1516         (void) printf(dgettext(TEXT_DOMAIN,
1517             "Recovery is possible, but will result in some data loss.\n"));
1518
1519         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1520             strftime(timestr, 128, 0, &t) != 0) {
1521                 (void) printf(dgettext(TEXT_DOMAIN,
1522                     "\tReturning the pool to its state as of %s\n"
1523                     "\tshould correct the problem.  "),
1524                     timestr);
1525         } else {
1526                 (void) printf(dgettext(TEXT_DOMAIN,
1527                     "\tReverting the pool to an earlier state "
1528                     "should correct the problem.\n\t"));
1529         }
1530
1531         if (loss > 120) {
1532                 (void) printf(dgettext(TEXT_DOMAIN,
1533                     "Approximately %lld minutes of data\n"
1534                     "\tmust be discarded, irreversibly.  "), (loss + 30) / 60);
1535         } else if (loss > 0) {
1536                 (void) printf(dgettext(TEXT_DOMAIN,
1537                     "Approximately %lld seconds of data\n"
1538                     "\tmust be discarded, irreversibly.  "), loss);
1539         }
1540         if (edata != 0 && edata != UINT64_MAX) {
1541                 if (edata == 1) {
1542                         (void) printf(dgettext(TEXT_DOMAIN,
1543                             "After rewind, at least\n"
1544                             "\tone persistent user-data error will remain.  "));
1545                 } else {
1546                         (void) printf(dgettext(TEXT_DOMAIN,
1547                             "After rewind, several\n"
1548                             "\tpersistent user-data errors will remain.  "));
1549                 }
1550         }
1551         (void) printf(dgettext(TEXT_DOMAIN,
1552             "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1553             reason >= 0 ? "clear" : "import", name);
1554
1555         (void) printf(dgettext(TEXT_DOMAIN,
1556             "A scrub of the pool\n"
1557             "\tis strongly recommended after recovery.\n"));
1558         return;
1559
1560 no_info:
1561         (void) printf(dgettext(TEXT_DOMAIN,
1562             "Destroy and re-create the pool from\n\ta backup source.\n"));
1563 }
1564
1565 /*
1566  * zpool_import() is a contracted interface. Should be kept the same
1567  * if possible.
1568  *
1569  * Applications should use zpool_import_props() to import a pool with
1570  * new properties value to be set.
1571  */
1572 int
1573 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1574     char *altroot)
1575 {
1576         nvlist_t *props = NULL;
1577         int ret;
1578
1579         if (altroot != NULL) {
1580                 if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1581                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1582                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1583                             newname));
1584                 }
1585
1586                 if (nvlist_add_string(props,
1587                     zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1588                     nvlist_add_string(props,
1589                     zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1590                         nvlist_free(props);
1591                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1592                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1593                             newname));
1594                 }
1595         }
1596
1597         ret = zpool_import_props(hdl, config, newname, props,
1598             ZFS_IMPORT_NORMAL);
1599         nvlist_free(props);
1600         return (ret);
1601 }
1602
1603 static void
1604 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1605     int indent)
1606 {
1607         nvlist_t **child;
1608         uint_t c, children;
1609         char *vname;
1610         uint64_t is_log = 0;
1611
1612         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1613             &is_log);
1614
1615         if (name != NULL)
1616                 (void) printf("\t%*s%s%s\n", indent, "", name,
1617                     is_log ? " [log]" : "");
1618
1619         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1620             &child, &children) != 0)
1621                 return;
1622
1623         for (c = 0; c < children; c++) {
1624                 vname = zpool_vdev_name(hdl, NULL, child[c], B_TRUE);
1625                 print_vdev_tree(hdl, vname, child[c], indent + 2);
1626                 free(vname);
1627         }
1628 }
1629
1630 void
1631 zpool_print_unsup_feat(nvlist_t *config)
1632 {
1633         nvlist_t *nvinfo, *unsup_feat;
1634
1635         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nvinfo) ==
1636             0);
1637         verify(nvlist_lookup_nvlist(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT,
1638             &unsup_feat) == 0);
1639
1640         for (nvpair_t *nvp = nvlist_next_nvpair(unsup_feat, NULL); nvp != NULL;
1641             nvp = nvlist_next_nvpair(unsup_feat, nvp)) {
1642                 char *desc;
1643
1644                 verify(nvpair_type(nvp) == DATA_TYPE_STRING);
1645                 verify(nvpair_value_string(nvp, &desc) == 0);
1646
1647                 if (strlen(desc) > 0)
1648                         (void) printf("\t%s (%s)\n", nvpair_name(nvp), desc);
1649                 else
1650                         (void) printf("\t%s\n", nvpair_name(nvp));
1651         }
1652 }
1653
1654 /*
1655  * Import the given pool using the known configuration and a list of
1656  * properties to be set. The configuration should have come from
1657  * zpool_find_import(). The 'newname' parameters control whether the pool
1658  * is imported with a different name.
1659  */
1660 int
1661 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1662     nvlist_t *props, int flags)
1663 {
1664         zfs_cmd_t zc = { 0 };
1665         zpool_rewind_policy_t policy;
1666         nvlist_t *nv = NULL;
1667         nvlist_t *nvinfo = NULL;
1668         nvlist_t *missing = NULL;
1669         char *thename;
1670         char *origname;
1671         int ret;
1672         int error = 0;
1673         char errbuf[1024];
1674
1675         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1676             &origname) == 0);
1677
1678         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1679             "cannot import pool '%s'"), origname);
1680
1681         if (newname != NULL) {
1682                 if (!zpool_name_valid(hdl, B_FALSE, newname))
1683                         return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1684                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1685                             newname));
1686                 thename = (char *)newname;
1687         } else {
1688                 thename = origname;
1689         }
1690
1691         if (props != NULL) {
1692                 uint64_t version;
1693                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1694
1695                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1696                     &version) == 0);
1697
1698                 if ((props = zpool_valid_proplist(hdl, origname,
1699                     props, version, flags, errbuf)) == NULL)
1700                         return (-1);
1701                 if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1702                         nvlist_free(props);
1703                         return (-1);
1704                 }
1705                 nvlist_free(props);
1706         }
1707
1708         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1709
1710         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1711             &zc.zc_guid) == 0);
1712
1713         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1714                 zcmd_free_nvlists(&zc);
1715                 return (-1);
1716         }
1717         if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1718                 zcmd_free_nvlists(&zc);
1719                 return (-1);
1720         }
1721
1722         zc.zc_cookie = flags;
1723         while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1724             errno == ENOMEM) {
1725                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1726                         zcmd_free_nvlists(&zc);
1727                         return (-1);
1728                 }
1729         }
1730         if (ret != 0)
1731                 error = errno;
1732
1733         (void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1734
1735         zcmd_free_nvlists(&zc);
1736
1737         zpool_get_rewind_policy(config, &policy);
1738
1739         if (error) {
1740                 char desc[1024];
1741
1742                 /*
1743                  * Dry-run failed, but we print out what success
1744                  * looks like if we found a best txg
1745                  */
1746                 if (policy.zrp_request & ZPOOL_TRY_REWIND) {
1747                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1748                             B_TRUE, nv);
1749                         nvlist_free(nv);
1750                         return (-1);
1751                 }
1752
1753                 if (newname == NULL)
1754                         (void) snprintf(desc, sizeof (desc),
1755                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1756                             thename);
1757                 else
1758                         (void) snprintf(desc, sizeof (desc),
1759                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1760                             origname, thename);
1761
1762                 switch (error) {
1763                 case ENOTSUP:
1764                         if (nv != NULL && nvlist_lookup_nvlist(nv,
1765                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1766                             nvlist_exists(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT)) {
1767                                 (void) printf(dgettext(TEXT_DOMAIN, "This "
1768                                     "pool uses the following feature(s) not "
1769                                     "supported by this system:\n"));
1770                                 zpool_print_unsup_feat(nv);
1771                                 if (nvlist_exists(nvinfo,
1772                                     ZPOOL_CONFIG_CAN_RDONLY)) {
1773                                         (void) printf(dgettext(TEXT_DOMAIN,
1774                                             "All unsupported features are only "
1775                                             "required for writing to the pool."
1776                                             "\nThe pool can be imported using "
1777                                             "'-o readonly=on'.\n"));
1778                                 }
1779                         }
1780                         /*
1781                          * Unsupported version.
1782                          */
1783                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1784                         break;
1785
1786                 case EINVAL:
1787                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1788                         break;
1789
1790                 case EROFS:
1791                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1792                             "one or more devices is read only"));
1793                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1794                         break;
1795
1796                 case ENXIO:
1797                         if (nv && nvlist_lookup_nvlist(nv,
1798                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1799                             nvlist_lookup_nvlist(nvinfo,
1800                             ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1801                                 (void) printf(dgettext(TEXT_DOMAIN,
1802                                     "The devices below are missing, use "
1803                                     "'-m' to import the pool anyway:\n"));
1804                                 print_vdev_tree(hdl, NULL, missing, 2);
1805                                 (void) printf("\n");
1806                         }
1807                         (void) zpool_standard_error(hdl, error, desc);
1808                         break;
1809
1810                 case EEXIST:
1811                         (void) zpool_standard_error(hdl, error, desc);
1812                         break;
1813                 case ENAMETOOLONG:
1814                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1815                             "new name of at least one dataset is longer than "
1816                             "the maximum allowable length"));
1817                         (void) zfs_error(hdl, EZFS_NAMETOOLONG, desc);
1818                         break;
1819                 default:
1820                         (void) zpool_standard_error(hdl, error, desc);
1821                         zpool_explain_recover(hdl,
1822                             newname ? origname : thename, -error, nv);
1823                         break;
1824                 }
1825
1826                 nvlist_free(nv);
1827                 ret = -1;
1828         } else {
1829                 zpool_handle_t *zhp;
1830
1831                 /*
1832                  * This should never fail, but play it safe anyway.
1833                  */
1834                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1835                         ret = -1;
1836                 else if (zhp != NULL)
1837                         zpool_close(zhp);
1838                 if (policy.zrp_request &
1839                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1840                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1841                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1842                 }
1843                 nvlist_free(nv);
1844                 return (0);
1845         }
1846
1847         return (ret);
1848 }
1849
1850 /*
1851  * Scan the pool.
1852  */
1853 int
1854 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1855 {
1856         zfs_cmd_t zc = { 0 };
1857         char msg[1024];
1858         libzfs_handle_t *hdl = zhp->zpool_hdl;
1859
1860         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1861         zc.zc_cookie = func;
1862
1863         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1864             (errno == ENOENT && func != POOL_SCAN_NONE))
1865                 return (0);
1866
1867         if (func == POOL_SCAN_SCRUB) {
1868                 (void) snprintf(msg, sizeof (msg),
1869                     dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1870         } else if (func == POOL_SCAN_NONE) {
1871                 (void) snprintf(msg, sizeof (msg),
1872                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1873                     zc.zc_name);
1874         } else {
1875                 assert(!"unexpected result");
1876         }
1877
1878         if (errno == EBUSY) {
1879                 nvlist_t *nvroot;
1880                 pool_scan_stat_t *ps = NULL;
1881                 uint_t psc;
1882
1883                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
1884                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1885                 (void) nvlist_lookup_uint64_array(nvroot,
1886                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1887                 if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1888                         return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1889                 else
1890                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
1891         } else if (errno == ENOENT) {
1892                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1893         } else {
1894                 return (zpool_standard_error(hdl, errno, msg));
1895         }
1896 }
1897
1898 #ifdef illumos
1899 /*
1900  * This provides a very minimal check whether a given string is likely a
1901  * c#t#d# style string.  Users of this are expected to do their own
1902  * verification of the s# part.
1903  */
1904 #define CTD_CHECK(str)  (str && str[0] == 'c' && isdigit(str[1]))
1905
1906 /*
1907  * More elaborate version for ones which may start with "/dev/dsk/"
1908  * and the like.
1909  */
1910 static int
1911 ctd_check_path(char *str)
1912 {
1913         /*
1914          * If it starts with a slash, check the last component.
1915          */
1916         if (str && str[0] == '/') {
1917                 char *tmp = strrchr(str, '/');
1918
1919                 /*
1920                  * If it ends in "/old", check the second-to-last
1921                  * component of the string instead.
1922                  */
1923                 if (tmp != str && strcmp(tmp, "/old") == 0) {
1924                         for (tmp--; *tmp != '/'; tmp--)
1925                                 ;
1926                 }
1927                 str = tmp + 1;
1928         }
1929         return (CTD_CHECK(str));
1930 }
1931 #endif
1932
1933 /*
1934  * Find a vdev that matches the search criteria specified. We use the
1935  * the nvpair name to determine how we should look for the device.
1936  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1937  * spare; but FALSE if its an INUSE spare.
1938  */
1939 static nvlist_t *
1940 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1941     boolean_t *l2cache, boolean_t *log)
1942 {
1943         uint_t c, children;
1944         nvlist_t **child;
1945         nvlist_t *ret;
1946         uint64_t is_log;
1947         char *srchkey;
1948         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
1949
1950         /* Nothing to look for */
1951         if (search == NULL || pair == NULL)
1952                 return (NULL);
1953
1954         /* Obtain the key we will use to search */
1955         srchkey = nvpair_name(pair);
1956
1957         switch (nvpair_type(pair)) {
1958         case DATA_TYPE_UINT64:
1959                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
1960                         uint64_t srchval, theguid;
1961
1962                         verify(nvpair_value_uint64(pair, &srchval) == 0);
1963                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
1964                             &theguid) == 0);
1965                         if (theguid == srchval)
1966                                 return (nv);
1967                 }
1968                 break;
1969
1970         case DATA_TYPE_STRING: {
1971                 char *srchval, *val;
1972
1973                 verify(nvpair_value_string(pair, &srchval) == 0);
1974                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
1975                         break;
1976
1977                 /*
1978                  * Search for the requested value. Special cases:
1979                  *
1980                  * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
1981                  *   "s0" or "s0/old".  The "s0" part is hidden from the user,
1982                  *   but included in the string, so this matches around it.
1983                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
1984                  *
1985                  * Otherwise, all other searches are simple string compares.
1986                  */
1987 #ifdef illumos
1988                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0 &&
1989                     ctd_check_path(val)) {
1990                         uint64_t wholedisk = 0;
1991
1992                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1993                             &wholedisk);
1994                         if (wholedisk) {
1995                                 int slen = strlen(srchval);
1996                                 int vlen = strlen(val);
1997
1998                                 if (slen != vlen - 2)
1999                                         break;
2000
2001                                 /*
2002                                  * make_leaf_vdev() should only set
2003                                  * wholedisk for ZPOOL_CONFIG_PATHs which
2004                                  * will include "/dev/dsk/", giving plenty of
2005                                  * room for the indices used next.
2006                                  */
2007                                 ASSERT(vlen >= 6);
2008
2009                                 /*
2010                                  * strings identical except trailing "s0"
2011                                  */
2012                                 if (strcmp(&val[vlen - 2], "s0") == 0 &&
2013                                     strncmp(srchval, val, slen) == 0)
2014                                         return (nv);
2015
2016                                 /*
2017                                  * strings identical except trailing "s0/old"
2018                                  */
2019                                 if (strcmp(&val[vlen - 6], "s0/old") == 0 &&
2020                                     strcmp(&srchval[slen - 4], "/old") == 0 &&
2021                                     strncmp(srchval, val, slen - 4) == 0)
2022                                         return (nv);
2023
2024                                 break;
2025                         }
2026                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2027 #else
2028                 if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2029 #endif
2030                         char *type, *idx, *end, *p;
2031                         uint64_t id, vdev_id;
2032
2033                         /*
2034                          * Determine our vdev type, keeping in mind
2035                          * that the srchval is composed of a type and
2036                          * vdev id pair (i.e. mirror-4).
2037                          */
2038                         if ((type = strdup(srchval)) == NULL)
2039                                 return (NULL);
2040
2041                         if ((p = strrchr(type, '-')) == NULL) {
2042                                 free(type);
2043                                 break;
2044                         }
2045                         idx = p + 1;
2046                         *p = '\0';
2047
2048                         /*
2049                          * If the types don't match then keep looking.
2050                          */
2051                         if (strncmp(val, type, strlen(val)) != 0) {
2052                                 free(type);
2053                                 break;
2054                         }
2055
2056                         verify(strncmp(type, VDEV_TYPE_RAIDZ,
2057                             strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2058                             strncmp(type, VDEV_TYPE_MIRROR,
2059                             strlen(VDEV_TYPE_MIRROR)) == 0);
2060                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
2061                             &id) == 0);
2062
2063                         errno = 0;
2064                         vdev_id = strtoull(idx, &end, 10);
2065
2066                         free(type);
2067                         if (errno != 0)
2068                                 return (NULL);
2069
2070                         /*
2071                          * Now verify that we have the correct vdev id.
2072                          */
2073                         if (vdev_id == id)
2074                                 return (nv);
2075                 }
2076
2077                 /*
2078                  * Common case
2079                  */
2080                 if (strcmp(srchval, val) == 0)
2081                         return (nv);
2082                 break;
2083         }
2084
2085         default:
2086                 break;
2087         }
2088
2089         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2090             &child, &children) != 0)
2091                 return (NULL);
2092
2093         for (c = 0; c < children; c++) {
2094                 if ((ret = vdev_to_nvlist_iter(child[c], search,
2095                     avail_spare, l2cache, NULL)) != NULL) {
2096                         /*
2097                          * The 'is_log' value is only set for the toplevel
2098                          * vdev, not the leaf vdevs.  So we always lookup the
2099                          * log device from the root of the vdev tree (where
2100                          * 'log' is non-NULL).
2101                          */
2102                         if (log != NULL &&
2103                             nvlist_lookup_uint64(child[c],
2104                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2105                             is_log) {
2106                                 *log = B_TRUE;
2107                         }
2108                         return (ret);
2109                 }
2110         }
2111
2112         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2113             &child, &children) == 0) {
2114                 for (c = 0; c < children; c++) {
2115                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2116                             avail_spare, l2cache, NULL)) != NULL) {
2117                                 *avail_spare = B_TRUE;
2118                                 return (ret);
2119                         }
2120                 }
2121         }
2122
2123         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2124             &child, &children) == 0) {
2125                 for (c = 0; c < children; c++) {
2126                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2127                             avail_spare, l2cache, NULL)) != NULL) {
2128                                 *l2cache = B_TRUE;
2129                                 return (ret);
2130                         }
2131                 }
2132         }
2133
2134         return (NULL);
2135 }
2136
2137 /*
2138  * Given a physical path (minus the "/devices" prefix), find the
2139  * associated vdev.
2140  */
2141 nvlist_t *
2142 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2143     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2144 {
2145         nvlist_t *search, *nvroot, *ret;
2146
2147         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2148         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
2149
2150         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2151             &nvroot) == 0);
2152
2153         *avail_spare = B_FALSE;
2154         *l2cache = B_FALSE;
2155         if (log != NULL)
2156                 *log = B_FALSE;
2157         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2158         nvlist_free(search);
2159
2160         return (ret);
2161 }
2162
2163 /*
2164  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2165  */
2166 boolean_t
2167 zpool_vdev_is_interior(const char *name)
2168 {
2169         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2170             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2171                 return (B_TRUE);
2172         return (B_FALSE);
2173 }
2174
2175 nvlist_t *
2176 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2177     boolean_t *l2cache, boolean_t *log)
2178 {
2179         char buf[MAXPATHLEN];
2180         char *end;
2181         nvlist_t *nvroot, *search, *ret;
2182         uint64_t guid;
2183
2184         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2185
2186         guid = strtoull(path, &end, 10);
2187         if (guid != 0 && *end == '\0') {
2188                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
2189         } else if (zpool_vdev_is_interior(path)) {
2190                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
2191         } else if (path[0] != '/') {
2192                 (void) snprintf(buf, sizeof (buf), "%s%s", _PATH_DEV, path);
2193                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, buf) == 0);
2194         } else {
2195                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
2196         }
2197
2198         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2199             &nvroot) == 0);
2200
2201         *avail_spare = B_FALSE;
2202         *l2cache = B_FALSE;
2203         if (log != NULL)
2204                 *log = B_FALSE;
2205         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2206         nvlist_free(search);
2207
2208         return (ret);
2209 }
2210
2211 static int
2212 vdev_online(nvlist_t *nv)
2213 {
2214         uint64_t ival;
2215
2216         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2217             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2218             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2219                 return (0);
2220
2221         return (1);
2222 }
2223
2224 /*
2225  * Helper function for zpool_get_physpaths().
2226  */
2227 static int
2228 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2229     size_t *bytes_written)
2230 {
2231         size_t bytes_left, pos, rsz;
2232         char *tmppath;
2233         const char *format;
2234
2235         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2236             &tmppath) != 0)
2237                 return (EZFS_NODEVICE);
2238
2239         pos = *bytes_written;
2240         bytes_left = physpath_size - pos;
2241         format = (pos == 0) ? "%s" : " %s";
2242
2243         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2244         *bytes_written += rsz;
2245
2246         if (rsz >= bytes_left) {
2247                 /* if physpath was not copied properly, clear it */
2248                 if (bytes_left != 0) {
2249                         physpath[pos] = 0;
2250                 }
2251                 return (EZFS_NOSPC);
2252         }
2253         return (0);
2254 }
2255
2256 static int
2257 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2258     size_t *rsz, boolean_t is_spare)
2259 {
2260         char *type;
2261         int ret;
2262
2263         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2264                 return (EZFS_INVALCONFIG);
2265
2266         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2267                 /*
2268                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2269                  * For a spare vdev, we only want to boot from the active
2270                  * spare device.
2271                  */
2272                 if (is_spare) {
2273                         uint64_t spare = 0;
2274                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2275                             &spare);
2276                         if (!spare)
2277                                 return (EZFS_INVALCONFIG);
2278                 }
2279
2280                 if (vdev_online(nv)) {
2281                         if ((ret = vdev_get_one_physpath(nv, physpath,
2282                             phypath_size, rsz)) != 0)
2283                                 return (ret);
2284                 }
2285         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2286             strcmp(type, VDEV_TYPE_RAIDZ) == 0 ||
2287             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2288             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2289                 nvlist_t **child;
2290                 uint_t count;
2291                 int i, ret;
2292
2293                 if (nvlist_lookup_nvlist_array(nv,
2294                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2295                         return (EZFS_INVALCONFIG);
2296
2297                 for (i = 0; i < count; i++) {
2298                         ret = vdev_get_physpaths(child[i], physpath,
2299                             phypath_size, rsz, is_spare);
2300                         if (ret == EZFS_NOSPC)
2301                                 return (ret);
2302                 }
2303         }
2304
2305         return (EZFS_POOL_INVALARG);
2306 }
2307
2308 /*
2309  * Get phys_path for a root pool config.
2310  * Return 0 on success; non-zero on failure.
2311  */
2312 static int
2313 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2314 {
2315         size_t rsz;
2316         nvlist_t *vdev_root;
2317         nvlist_t **child;
2318         uint_t count;
2319         char *type;
2320
2321         rsz = 0;
2322
2323         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2324             &vdev_root) != 0)
2325                 return (EZFS_INVALCONFIG);
2326
2327         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2328             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2329             &child, &count) != 0)
2330                 return (EZFS_INVALCONFIG);
2331
2332         /*
2333          * root pool can only have a single top-level vdev.
2334          */
2335         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1)
2336                 return (EZFS_POOL_INVALARG);
2337
2338         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2339             B_FALSE);
2340
2341         /* No online devices */
2342         if (rsz == 0)
2343                 return (EZFS_NODEVICE);
2344
2345         return (0);
2346 }
2347
2348 /*
2349  * Get phys_path for a root pool
2350  * Return 0 on success; non-zero on failure.
2351  */
2352 int
2353 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2354 {
2355         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2356             phypath_size));
2357 }
2358
2359 /*
2360  * If the device has being dynamically expanded then we need to relabel
2361  * the disk to use the new unallocated space.
2362  */
2363 static int
2364 zpool_relabel_disk(libzfs_handle_t *hdl, const char *name)
2365 {
2366 #ifdef illumos
2367         char path[MAXPATHLEN];
2368         char errbuf[1024];
2369         int fd, error;
2370         int (*_efi_use_whole_disk)(int);
2371
2372         if ((_efi_use_whole_disk = (int (*)(int))dlsym(RTLD_DEFAULT,
2373             "efi_use_whole_disk")) == NULL)
2374                 return (-1);
2375
2376         (void) snprintf(path, sizeof (path), "%s/%s", ZFS_RDISK_ROOT, name);
2377
2378         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
2379                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2380                     "relabel '%s': unable to open device"), name);
2381                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
2382         }
2383
2384         /*
2385          * It's possible that we might encounter an error if the device
2386          * does not have any unallocated space left. If so, we simply
2387          * ignore that error and continue on.
2388          */
2389         error = _efi_use_whole_disk(fd);
2390         (void) close(fd);
2391         if (error && error != VT_ENOSPC) {
2392                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2393                     "relabel '%s': unable to read disk capacity"), name);
2394                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
2395         }
2396 #endif  /* illumos */
2397         return (0);
2398 }
2399
2400 /*
2401  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2402  * ZFS_ONLINE_* flags.
2403  */
2404 int
2405 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2406     vdev_state_t *newstate)
2407 {
2408         zfs_cmd_t zc = { 0 };
2409         char msg[1024];
2410         nvlist_t *tgt;
2411         boolean_t avail_spare, l2cache, islog;
2412         libzfs_handle_t *hdl = zhp->zpool_hdl;
2413
2414         if (flags & ZFS_ONLINE_EXPAND) {
2415                 (void) snprintf(msg, sizeof (msg),
2416                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2417         } else {
2418                 (void) snprintf(msg, sizeof (msg),
2419                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2420         }
2421
2422         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2423         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2424             &islog)) == NULL)
2425                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2426
2427         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2428
2429         if (avail_spare)
2430                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2431
2432         if (flags & ZFS_ONLINE_EXPAND ||
2433             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2434                 char *pathname = NULL;
2435                 uint64_t wholedisk = 0;
2436
2437                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2438                     &wholedisk);
2439                 verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
2440                     &pathname) == 0);
2441
2442                 /*
2443                  * XXX - L2ARC 1.0 devices can't support expansion.
2444                  */
2445                 if (l2cache) {
2446                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2447                             "cannot expand cache devices"));
2448                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2449                 }
2450
2451                 if (wholedisk) {
2452                         pathname += strlen(ZFS_DISK_ROOT) + 1;
2453                         (void) zpool_relabel_disk(hdl, pathname);
2454                 }
2455         }
2456
2457         zc.zc_cookie = VDEV_STATE_ONLINE;
2458         zc.zc_obj = flags;
2459
2460         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2461                 if (errno == EINVAL) {
2462                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2463                             "from this pool into a new one.  Use '%s' "
2464                             "instead"), "zpool detach");
2465                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2466                 }
2467                 return (zpool_standard_error(hdl, errno, msg));
2468         }
2469
2470         *newstate = zc.zc_cookie;
2471         return (0);
2472 }
2473
2474 /*
2475  * Take the specified vdev offline
2476  */
2477 int
2478 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2479 {
2480         zfs_cmd_t zc = { 0 };
2481         char msg[1024];
2482         nvlist_t *tgt;
2483         boolean_t avail_spare, l2cache;
2484         libzfs_handle_t *hdl = zhp->zpool_hdl;
2485
2486         (void) snprintf(msg, sizeof (msg),
2487             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2488
2489         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2490         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2491             NULL)) == NULL)
2492                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2493
2494         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2495
2496         if (avail_spare)
2497                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2498
2499         zc.zc_cookie = VDEV_STATE_OFFLINE;
2500         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2501
2502         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2503                 return (0);
2504
2505         switch (errno) {
2506         case EBUSY:
2507
2508                 /*
2509                  * There are no other replicas of this device.
2510                  */
2511                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2512
2513         case EEXIST:
2514                 /*
2515                  * The log device has unplayed logs
2516                  */
2517                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2518
2519         default:
2520                 return (zpool_standard_error(hdl, errno, msg));
2521         }
2522 }
2523
2524 /*
2525  * Mark the given vdev faulted.
2526  */
2527 int
2528 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2529 {
2530         zfs_cmd_t zc = { 0 };
2531         char msg[1024];
2532         libzfs_handle_t *hdl = zhp->zpool_hdl;
2533
2534         (void) snprintf(msg, sizeof (msg),
2535             dgettext(TEXT_DOMAIN, "cannot fault %llu"), guid);
2536
2537         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2538         zc.zc_guid = guid;
2539         zc.zc_cookie = VDEV_STATE_FAULTED;
2540         zc.zc_obj = aux;
2541
2542         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2543                 return (0);
2544
2545         switch (errno) {
2546         case EBUSY:
2547
2548                 /*
2549                  * There are no other replicas of this device.
2550                  */
2551                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2552
2553         default:
2554                 return (zpool_standard_error(hdl, errno, msg));
2555         }
2556
2557 }
2558
2559 /*
2560  * Mark the given vdev degraded.
2561  */
2562 int
2563 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2564 {
2565         zfs_cmd_t zc = { 0 };
2566         char msg[1024];
2567         libzfs_handle_t *hdl = zhp->zpool_hdl;
2568
2569         (void) snprintf(msg, sizeof (msg),
2570             dgettext(TEXT_DOMAIN, "cannot degrade %llu"), guid);
2571
2572         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2573         zc.zc_guid = guid;
2574         zc.zc_cookie = VDEV_STATE_DEGRADED;
2575         zc.zc_obj = aux;
2576
2577         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2578                 return (0);
2579
2580         return (zpool_standard_error(hdl, errno, msg));
2581 }
2582
2583 /*
2584  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2585  * a hot spare.
2586  */
2587 static boolean_t
2588 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2589 {
2590         nvlist_t **child;
2591         uint_t c, children;
2592         char *type;
2593
2594         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2595             &children) == 0) {
2596                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2597                     &type) == 0);
2598
2599                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2600                     children == 2 && child[which] == tgt)
2601                         return (B_TRUE);
2602
2603                 for (c = 0; c < children; c++)
2604                         if (is_replacing_spare(child[c], tgt, which))
2605                                 return (B_TRUE);
2606         }
2607
2608         return (B_FALSE);
2609 }
2610
2611 /*
2612  * Attach new_disk (fully described by nvroot) to old_disk.
2613  * If 'replacing' is specified, the new disk will replace the old one.
2614  */
2615 int
2616 zpool_vdev_attach(zpool_handle_t *zhp,
2617     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2618 {
2619         zfs_cmd_t zc = { 0 };
2620         char msg[1024];
2621         int ret;
2622         nvlist_t *tgt;
2623         boolean_t avail_spare, l2cache, islog;
2624         uint64_t val;
2625         char *newname;
2626         nvlist_t **child;
2627         uint_t children;
2628         nvlist_t *config_root;
2629         libzfs_handle_t *hdl = zhp->zpool_hdl;
2630         boolean_t rootpool = zpool_is_bootable(zhp);
2631
2632         if (replacing)
2633                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2634                     "cannot replace %s with %s"), old_disk, new_disk);
2635         else
2636                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2637                     "cannot attach %s to %s"), new_disk, old_disk);
2638
2639         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2640         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2641             &islog)) == 0)
2642                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2643
2644         if (avail_spare)
2645                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2646
2647         if (l2cache)
2648                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2649
2650         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2651         zc.zc_cookie = replacing;
2652
2653         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2654             &child, &children) != 0 || children != 1) {
2655                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2656                     "new device must be a single disk"));
2657                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2658         }
2659
2660         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2661             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2662
2663         if ((newname = zpool_vdev_name(NULL, NULL, child[0], B_FALSE)) == NULL)
2664                 return (-1);
2665
2666         /*
2667          * If the target is a hot spare that has been swapped in, we can only
2668          * replace it with another hot spare.
2669          */
2670         if (replacing &&
2671             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2672             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2673             NULL) == NULL || !avail_spare) &&
2674             is_replacing_spare(config_root, tgt, 1)) {
2675                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2676                     "can only be replaced by another hot spare"));
2677                 free(newname);
2678                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2679         }
2680
2681         free(newname);
2682
2683         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2684                 return (-1);
2685
2686         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2687
2688         zcmd_free_nvlists(&zc);
2689
2690         if (ret == 0) {
2691                 if (rootpool) {
2692                         /*
2693                          * XXX need a better way to prevent user from
2694                          * booting up a half-baked vdev.
2695                          */
2696                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2697                             "sure to wait until resilver is done "
2698                             "before rebooting.\n"));
2699                         (void) fprintf(stderr, "\n");
2700                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "If "
2701                             "you boot from pool '%s', you may need to update\n"
2702                             "boot code on newly attached disk '%s'.\n\n"
2703                             "Assuming you use GPT partitioning and 'da0' is "
2704                             "your new boot disk\n"
2705                             "you may use the following command:\n\n"
2706                             "\tgpart bootcode -b /boot/pmbr -p "
2707                             "/boot/gptzfsboot -i 1 da0\n\n"),
2708                             zhp->zpool_name, new_disk);
2709                 }
2710                 return (0);
2711         }
2712
2713         switch (errno) {
2714         case ENOTSUP:
2715                 /*
2716                  * Can't attach to or replace this type of vdev.
2717                  */
2718                 if (replacing) {
2719                         uint64_t version = zpool_get_prop_int(zhp,
2720                             ZPOOL_PROP_VERSION, NULL);
2721
2722                         if (islog)
2723                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2724                                     "cannot replace a log with a spare"));
2725                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2726                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2727                                     "already in replacing/spare config; wait "
2728                                     "for completion or use 'zpool detach'"));
2729                         else
2730                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2731                                     "cannot replace a replacing device"));
2732                 } else {
2733                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2734                             "can only attach to mirrors and top-level "
2735                             "disks"));
2736                 }
2737                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2738                 break;
2739
2740         case EINVAL:
2741                 /*
2742                  * The new device must be a single disk.
2743                  */
2744                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2745                     "new device must be a single disk"));
2746                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2747                 break;
2748
2749         case EBUSY:
2750                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2751                     new_disk);
2752                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2753                 break;
2754
2755         case EOVERFLOW:
2756                 /*
2757                  * The new device is too small.
2758                  */
2759                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2760                     "device is too small"));
2761                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2762                 break;
2763
2764         case EDOM:
2765                 /*
2766                  * The new device has a different alignment requirement.
2767                  */
2768                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2769                     "devices have different sector alignment"));
2770                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2771                 break;
2772
2773         case ENAMETOOLONG:
2774                 /*
2775                  * The resulting top-level vdev spec won't fit in the label.
2776                  */
2777                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2778                 break;
2779
2780         default:
2781                 (void) zpool_standard_error(hdl, errno, msg);
2782         }
2783
2784         return (-1);
2785 }
2786
2787 /*
2788  * Detach the specified device.
2789  */
2790 int
2791 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2792 {
2793         zfs_cmd_t zc = { 0 };
2794         char msg[1024];
2795         nvlist_t *tgt;
2796         boolean_t avail_spare, l2cache;
2797         libzfs_handle_t *hdl = zhp->zpool_hdl;
2798
2799         (void) snprintf(msg, sizeof (msg),
2800             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2801
2802         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2803         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2804             NULL)) == 0)
2805                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2806
2807         if (avail_spare)
2808                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2809
2810         if (l2cache)
2811                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2812
2813         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2814
2815         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2816                 return (0);
2817
2818         switch (errno) {
2819
2820         case ENOTSUP:
2821                 /*
2822                  * Can't detach from this type of vdev.
2823                  */
2824                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2825                     "applicable to mirror and replacing vdevs"));
2826                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2827                 break;
2828
2829         case EBUSY:
2830                 /*
2831                  * There are no other replicas of this device.
2832                  */
2833                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2834                 break;
2835
2836         default:
2837                 (void) zpool_standard_error(hdl, errno, msg);
2838         }
2839
2840         return (-1);
2841 }
2842
2843 /*
2844  * Find a mirror vdev in the source nvlist.
2845  *
2846  * The mchild array contains a list of disks in one of the top-level mirrors
2847  * of the source pool.  The schild array contains a list of disks that the
2848  * user specified on the command line.  We loop over the mchild array to
2849  * see if any entry in the schild array matches.
2850  *
2851  * If a disk in the mchild array is found in the schild array, we return
2852  * the index of that entry.  Otherwise we return -1.
2853  */
2854 static int
2855 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2856     nvlist_t **schild, uint_t schildren)
2857 {
2858         uint_t mc;
2859
2860         for (mc = 0; mc < mchildren; mc++) {
2861                 uint_t sc;
2862                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2863                     mchild[mc], B_FALSE);
2864
2865                 for (sc = 0; sc < schildren; sc++) {
2866                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2867                             schild[sc], B_FALSE);
2868                         boolean_t result = (strcmp(mpath, spath) == 0);
2869
2870                         free(spath);
2871                         if (result) {
2872                                 free(mpath);
2873                                 return (mc);
2874                         }
2875                 }
2876
2877                 free(mpath);
2878         }
2879
2880         return (-1);
2881 }
2882
2883 /*
2884  * Split a mirror pool.  If newroot points to null, then a new nvlist
2885  * is generated and it is the responsibility of the caller to free it.
2886  */
2887 int
2888 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2889     nvlist_t *props, splitflags_t flags)
2890 {
2891         zfs_cmd_t zc = { 0 };
2892         char msg[1024];
2893         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2894         nvlist_t **varray = NULL, *zc_props = NULL;
2895         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2896         libzfs_handle_t *hdl = zhp->zpool_hdl;
2897         uint64_t vers;
2898         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2899         int retval = 0;
2900
2901         (void) snprintf(msg, sizeof (msg),
2902             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2903
2904         if (!zpool_name_valid(hdl, B_FALSE, newname))
2905                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2906
2907         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2908                 (void) fprintf(stderr, gettext("Internal error: unable to "
2909                     "retrieve pool configuration\n"));
2910                 return (-1);
2911         }
2912
2913         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2914             == 0);
2915         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2916
2917         if (props) {
2918                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2919                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2920                     props, vers, flags, msg)) == NULL)
2921                         return (-1);
2922         }
2923
2924         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2925             &children) != 0) {
2926                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2927                     "Source pool is missing vdev tree"));
2928                 nvlist_free(zc_props);
2929                 return (-1);
2930         }
2931
2932         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2933         vcount = 0;
2934
2935         if (*newroot == NULL ||
2936             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2937             &newchild, &newchildren) != 0)
2938                 newchildren = 0;
2939
2940         for (c = 0; c < children; c++) {
2941                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2942                 char *type;
2943                 nvlist_t **mchild, *vdev;
2944                 uint_t mchildren;
2945                 int entry;
2946
2947                 /*
2948                  * Unlike cache & spares, slogs are stored in the
2949                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2950                  */
2951                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2952                     &is_log);
2953                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2954                     &is_hole);
2955                 if (is_log || is_hole) {
2956                         /*
2957                          * Create a hole vdev and put it in the config.
2958                          */
2959                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2960                                 goto out;
2961                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2962                             VDEV_TYPE_HOLE) != 0)
2963                                 goto out;
2964                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2965                             1) != 0)
2966                                 goto out;
2967                         if (lastlog == 0)
2968                                 lastlog = vcount;
2969                         varray[vcount++] = vdev;
2970                         continue;
2971                 }
2972                 lastlog = 0;
2973                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
2974                     == 0);
2975                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
2976                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2977                             "Source pool must be composed only of mirrors\n"));
2978                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2979                         goto out;
2980                 }
2981
2982                 verify(nvlist_lookup_nvlist_array(child[c],
2983                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2984
2985                 /* find or add an entry for this top-level vdev */
2986                 if (newchildren > 0 &&
2987                     (entry = find_vdev_entry(zhp, mchild, mchildren,
2988                     newchild, newchildren)) >= 0) {
2989                         /* We found a disk that the user specified. */
2990                         vdev = mchild[entry];
2991                         ++found;
2992                 } else {
2993                         /* User didn't specify a disk for this vdev. */
2994                         vdev = mchild[mchildren - 1];
2995                 }
2996
2997                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
2998                         goto out;
2999         }
3000
3001         /* did we find every disk the user specified? */
3002         if (found != newchildren) {
3003                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
3004                     "include at most one disk from each mirror"));
3005                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3006                 goto out;
3007         }
3008
3009         /* Prepare the nvlist for populating. */
3010         if (*newroot == NULL) {
3011                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
3012                         goto out;
3013                 freelist = B_TRUE;
3014                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
3015                     VDEV_TYPE_ROOT) != 0)
3016                         goto out;
3017         } else {
3018                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
3019         }
3020
3021         /* Add all the children we found */
3022         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
3023             lastlog == 0 ? vcount : lastlog) != 0)
3024                 goto out;
3025
3026         /*
3027          * If we're just doing a dry run, exit now with success.
3028          */
3029         if (flags.dryrun) {
3030                 memory_err = B_FALSE;
3031                 freelist = B_FALSE;
3032                 goto out;
3033         }
3034
3035         /* now build up the config list & call the ioctl */
3036         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3037                 goto out;
3038
3039         if (nvlist_add_nvlist(newconfig,
3040             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3041             nvlist_add_string(newconfig,
3042             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3043             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3044                 goto out;
3045
3046         /*
3047          * The new pool is automatically part of the namespace unless we
3048          * explicitly export it.
3049          */
3050         if (!flags.import)
3051                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3052         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3053         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3054         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
3055                 goto out;
3056         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
3057                 goto out;
3058
3059         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3060                 retval = zpool_standard_error(hdl, errno, msg);
3061                 goto out;
3062         }
3063
3064         freelist = B_FALSE;
3065         memory_err = B_FALSE;
3066
3067 out:
3068         if (varray != NULL) {
3069                 int v;
3070
3071                 for (v = 0; v < vcount; v++)
3072                         nvlist_free(varray[v]);
3073                 free(varray);
3074         }
3075         zcmd_free_nvlists(&zc);
3076         nvlist_free(zc_props);
3077         nvlist_free(newconfig);
3078         if (freelist) {
3079                 nvlist_free(*newroot);
3080                 *newroot = NULL;
3081         }
3082
3083         if (retval != 0)
3084                 return (retval);
3085
3086         if (memory_err)
3087                 return (no_memory(hdl));
3088
3089         return (0);
3090 }
3091
3092 /*
3093  * Remove the given device.  Currently, this is supported only for hot spares
3094  * and level 2 cache devices.
3095  */
3096 int
3097 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3098 {
3099         zfs_cmd_t zc = { 0 };
3100         char msg[1024];
3101         nvlist_t *tgt;
3102         boolean_t avail_spare, l2cache, islog;
3103         libzfs_handle_t *hdl = zhp->zpool_hdl;
3104         uint64_t version;
3105
3106         (void) snprintf(msg, sizeof (msg),
3107             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3108
3109         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3110         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3111             &islog)) == 0)
3112                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3113         /*
3114          * XXX - this should just go away.
3115          */
3116         if (!avail_spare && !l2cache && !islog) {
3117                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3118                     "only inactive hot spares, cache, top-level, "
3119                     "or log devices can be removed"));
3120                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3121         }
3122
3123         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3124         if (islog && version < SPA_VERSION_HOLES) {
3125                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3126                     "pool must be upgrade to support log removal"));
3127                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
3128         }
3129
3130         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
3131
3132         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3133                 return (0);
3134
3135         return (zpool_standard_error(hdl, errno, msg));
3136 }
3137
3138 /*
3139  * Clear the errors for the pool, or the particular device if specified.
3140  */
3141 int
3142 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3143 {
3144         zfs_cmd_t zc = { 0 };
3145         char msg[1024];
3146         nvlist_t *tgt;
3147         zpool_rewind_policy_t policy;
3148         boolean_t avail_spare, l2cache;
3149         libzfs_handle_t *hdl = zhp->zpool_hdl;
3150         nvlist_t *nvi = NULL;
3151         int error;
3152
3153         if (path)
3154                 (void) snprintf(msg, sizeof (msg),
3155                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3156                     path);
3157         else
3158                 (void) snprintf(msg, sizeof (msg),
3159                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3160                     zhp->zpool_name);
3161
3162         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3163         if (path) {
3164                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3165                     &l2cache, NULL)) == 0)
3166                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
3167
3168                 /*
3169                  * Don't allow error clearing for hot spares.  Do allow
3170                  * error clearing for l2cache devices.
3171                  */
3172                 if (avail_spare)
3173                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
3174
3175                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
3176                     &zc.zc_guid) == 0);
3177         }
3178
3179         zpool_get_rewind_policy(rewindnvl, &policy);
3180         zc.zc_cookie = policy.zrp_request;
3181
3182         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
3183                 return (-1);
3184
3185         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
3186                 return (-1);
3187
3188         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3189             errno == ENOMEM) {
3190                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3191                         zcmd_free_nvlists(&zc);
3192                         return (-1);
3193                 }
3194         }
3195
3196         if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
3197             errno != EPERM && errno != EACCES)) {
3198                 if (policy.zrp_request &
3199                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
3200                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3201                         zpool_rewind_exclaim(hdl, zc.zc_name,
3202                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
3203                             nvi);
3204                         nvlist_free(nvi);
3205                 }
3206                 zcmd_free_nvlists(&zc);
3207                 return (0);
3208         }
3209
3210         zcmd_free_nvlists(&zc);
3211         return (zpool_standard_error(hdl, errno, msg));
3212 }
3213
3214 /*
3215  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3216  */
3217 int
3218 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3219 {
3220         zfs_cmd_t zc = { 0 };
3221         char msg[1024];
3222         libzfs_handle_t *hdl = zhp->zpool_hdl;
3223
3224         (void) snprintf(msg, sizeof (msg),
3225             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3226             guid);
3227
3228         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3229         zc.zc_guid = guid;
3230         zc.zc_cookie = ZPOOL_NO_REWIND;
3231
3232         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
3233                 return (0);
3234
3235         return (zpool_standard_error(hdl, errno, msg));
3236 }
3237
3238 /*
3239  * Change the GUID for a pool.
3240  */
3241 int
3242 zpool_reguid(zpool_handle_t *zhp)
3243 {
3244         char msg[1024];
3245         libzfs_handle_t *hdl = zhp->zpool_hdl;
3246         zfs_cmd_t zc = { 0 };
3247
3248         (void) snprintf(msg, sizeof (msg),
3249             dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3250
3251         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3252         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3253                 return (0);
3254
3255         return (zpool_standard_error(hdl, errno, msg));
3256 }
3257
3258 /*
3259  * Reopen the pool.
3260  */
3261 int
3262 zpool_reopen(zpool_handle_t *zhp)
3263 {
3264         zfs_cmd_t zc = { 0 };
3265         char msg[1024];
3266         libzfs_handle_t *hdl = zhp->zpool_hdl;
3267
3268         (void) snprintf(msg, sizeof (msg),
3269             dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3270             zhp->zpool_name);
3271
3272         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3273         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3274                 return (0);
3275         return (zpool_standard_error(hdl, errno, msg));
3276 }
3277
3278 /*
3279  * Convert from a devid string to a path.
3280  */
3281 static char *
3282 devid_to_path(char *devid_str)
3283 {
3284         ddi_devid_t devid;
3285         char *minor;
3286         char *path;
3287         devid_nmlist_t *list = NULL;
3288         int ret;
3289
3290         if (devid_str_decode(devid_str, &devid, &minor) != 0)
3291                 return (NULL);
3292
3293         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3294
3295         devid_str_free(minor);
3296         devid_free(devid);
3297
3298         if (ret != 0)
3299                 return (NULL);
3300
3301         /*
3302          * In a case the strdup() fails, we will just return NULL below.
3303          */
3304         path = strdup(list[0].devname);
3305
3306         devid_free_nmlist(list);
3307
3308         return (path);
3309 }
3310
3311 /*
3312  * Convert from a path to a devid string.
3313  */
3314 static char *
3315 path_to_devid(const char *path)
3316 {
3317 #ifdef have_devid
3318         int fd;
3319         ddi_devid_t devid;
3320         char *minor, *ret;
3321
3322         if ((fd = open(path, O_RDONLY)) < 0)
3323                 return (NULL);
3324
3325         minor = NULL;
3326         ret = NULL;
3327         if (devid_get(fd, &devid) == 0) {
3328                 if (devid_get_minor_name(fd, &minor) == 0)
3329                         ret = devid_str_encode(devid, minor);
3330                 if (minor != NULL)
3331                         devid_str_free(minor);
3332                 devid_free(devid);
3333         }
3334         (void) close(fd);
3335
3336         return (ret);
3337 #else
3338         return (NULL);
3339 #endif
3340 }
3341
3342 /*
3343  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3344  * ignore any failure here, since a common case is for an unprivileged user to
3345  * type 'zpool status', and we'll display the correct information anyway.
3346  */
3347 static void
3348 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3349 {
3350         zfs_cmd_t zc = { 0 };
3351
3352         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3353         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3354         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3355             &zc.zc_guid) == 0);
3356
3357         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3358 }
3359
3360 /*
3361  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3362  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3363  * We also check if this is a whole disk, in which case we strip off the
3364  * trailing 's0' slice name.
3365  *
3366  * This routine is also responsible for identifying when disks have been
3367  * reconfigured in a new location.  The kernel will have opened the device by
3368  * devid, but the path will still refer to the old location.  To catch this, we
3369  * first do a path -> devid translation (which is fast for the common case).  If
3370  * the devid matches, we're done.  If not, we do a reverse devid -> path
3371  * translation and issue the appropriate ioctl() to update the path of the vdev.
3372  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3373  * of these checks.
3374  */
3375 char *
3376 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3377     boolean_t verbose)
3378 {
3379         char *path, *devid;
3380         uint64_t value;
3381         char buf[64];
3382         vdev_stat_t *vs;
3383         uint_t vsc;
3384         int have_stats;
3385         int have_path;
3386
3387         have_stats = nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3388             (uint64_t **)&vs, &vsc) == 0;
3389         have_path = nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0;
3390
3391         /*
3392          * If the device is not currently present, assume it will not
3393          * come back at the same device path.  Display the device by GUID.
3394          */
3395         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &value) == 0 ||
3396             have_path && have_stats && vs->vs_state <= VDEV_STATE_CANT_OPEN) {
3397                 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3398                     &value) == 0);
3399                 (void) snprintf(buf, sizeof (buf), "%llu",
3400                     (u_longlong_t)value);
3401                 path = buf;
3402         } else if (have_path) {
3403
3404                 /*
3405                  * If the device is dead (faulted, offline, etc) then don't
3406                  * bother opening it.  Otherwise we may be forcing the user to
3407                  * open a misbehaving device, which can have undesirable
3408                  * effects.
3409                  */
3410                 if ((have_stats == 0 ||
3411                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3412                     zhp != NULL &&
3413                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3414                         /*
3415                          * Determine if the current path is correct.
3416                          */
3417                         char *newdevid = path_to_devid(path);
3418
3419                         if (newdevid == NULL ||
3420                             strcmp(devid, newdevid) != 0) {
3421                                 char *newpath;
3422
3423                                 if ((newpath = devid_to_path(devid)) != NULL) {
3424                                         /*
3425                                          * Update the path appropriately.
3426                                          */
3427                                         set_path(zhp, nv, newpath);
3428                                         if (nvlist_add_string(nv,
3429                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3430                                                 verify(nvlist_lookup_string(nv,
3431                                                     ZPOOL_CONFIG_PATH,
3432                                                     &path) == 0);
3433                                         free(newpath);
3434                                 }
3435                         }
3436
3437                         if (newdevid)
3438                                 devid_str_free(newdevid);
3439                 }
3440
3441 #ifdef illumos
3442                 if (strncmp(path, ZFS_DISK_ROOTD, strlen(ZFS_DISK_ROOTD)) == 0)
3443                         path += strlen(ZFS_DISK_ROOTD);
3444
3445                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
3446                     &value) == 0 && value) {
3447                         int pathlen = strlen(path);
3448                         char *tmp = zfs_strdup(hdl, path);
3449
3450                         /*
3451                          * If it starts with c#, and ends with "s0", chop
3452                          * the "s0" off, or if it ends with "s0/old", remove
3453                          * the "s0" from the middle.
3454                          */
3455                         if (CTD_CHECK(tmp)) {
3456                                 if (strcmp(&tmp[pathlen - 2], "s0") == 0) {
3457                                         tmp[pathlen - 2] = '\0';
3458                                 } else if (pathlen > 6 &&
3459                                     strcmp(&tmp[pathlen - 6], "s0/old") == 0) {
3460                                         (void) strcpy(&tmp[pathlen - 6],
3461                                             "/old");
3462                                 }
3463                         }
3464                         return (tmp);
3465                 }
3466 #else   /* !illumos */
3467                 if (strncmp(path, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
3468                         path += sizeof(_PATH_DEV) - 1;
3469 #endif  /* illumos */
3470         } else {
3471                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3472
3473                 /*
3474                  * If it's a raidz device, we need to stick in the parity level.
3475                  */
3476                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3477                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3478                             &value) == 0);
3479                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3480                             (u_longlong_t)value);
3481                         path = buf;
3482                 }
3483
3484                 /*
3485                  * We identify each top-level vdev by using a <type-id>
3486                  * naming convention.
3487                  */
3488                 if (verbose) {
3489                         uint64_t id;
3490
3491                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3492                             &id) == 0);
3493                         (void) snprintf(buf, sizeof (buf), "%s-%llu", path,
3494                             (u_longlong_t)id);
3495                         path = buf;
3496                 }
3497         }
3498
3499         return (zfs_strdup(hdl, path));
3500 }
3501
3502 static int
3503 zbookmark_mem_compare(const void *a, const void *b)
3504 {
3505         return (memcmp(a, b, sizeof (zbookmark_phys_t)));
3506 }
3507
3508 /*
3509  * Retrieve the persistent error log, uniquify the members, and return to the
3510  * caller.
3511  */
3512 int
3513 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3514 {
3515         zfs_cmd_t zc = { 0 };
3516         uint64_t count;
3517         zbookmark_phys_t *zb = NULL;
3518         int i;
3519
3520         /*
3521          * Retrieve the raw error list from the kernel.  If the number of errors
3522          * has increased, allocate more space and continue until we get the
3523          * entire list.
3524          */
3525         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3526             &count) == 0);
3527         if (count == 0)
3528                 return (0);
3529         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3530             count * sizeof (zbookmark_phys_t))) == (uintptr_t)NULL)
3531                 return (-1);
3532         zc.zc_nvlist_dst_size = count;
3533         (void) strcpy(zc.zc_name, zhp->zpool_name);
3534         for (;;) {
3535                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3536                     &zc) != 0) {
3537                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3538                         if (errno == ENOMEM) {
3539                                 void *dst;
3540
3541                                 count = zc.zc_nvlist_dst_size;
3542                                 dst = zfs_alloc(zhp->zpool_hdl, count *
3543                                     sizeof (zbookmark_phys_t));
3544                                 if (dst == NULL)
3545                                         return (-1);
3546                                 zc.zc_nvlist_dst = (uintptr_t)dst;
3547                         } else {
3548                                 return (-1);
3549                         }
3550                 } else {
3551                         break;
3552                 }
3553         }
3554
3555         /*
3556          * Sort the resulting bookmarks.  This is a little confusing due to the
3557          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3558          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3559          * _not_ copied as part of the process.  So we point the start of our
3560          * array appropriate and decrement the total number of elements.
3561          */
3562         zb = ((zbookmark_phys_t *)(uintptr_t)zc.zc_nvlist_dst) +
3563             zc.zc_nvlist_dst_size;
3564         count -= zc.zc_nvlist_dst_size;
3565
3566         qsort(zb, count, sizeof (zbookmark_phys_t), zbookmark_mem_compare);
3567
3568         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3569
3570         /*
3571          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3572          */
3573         for (i = 0; i < count; i++) {
3574                 nvlist_t *nv;
3575
3576                 /* ignoring zb_blkid and zb_level for now */
3577                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3578                     zb[i-1].zb_object == zb[i].zb_object)
3579                         continue;
3580
3581                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3582                         goto nomem;
3583                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3584                     zb[i].zb_objset) != 0) {
3585                         nvlist_free(nv);
3586                         goto nomem;
3587                 }
3588                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3589                     zb[i].zb_object) != 0) {
3590                         nvlist_free(nv);
3591                         goto nomem;
3592                 }
3593                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3594                         nvlist_free(nv);
3595                         goto nomem;
3596                 }
3597                 nvlist_free(nv);
3598         }
3599
3600         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3601         return (0);
3602
3603 nomem:
3604         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3605         return (no_memory(zhp->zpool_hdl));
3606 }
3607
3608 /*
3609  * Upgrade a ZFS pool to the latest on-disk version.
3610  */
3611 int
3612 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3613 {
3614         zfs_cmd_t zc = { 0 };
3615         libzfs_handle_t *hdl = zhp->zpool_hdl;
3616
3617         (void) strcpy(zc.zc_name, zhp->zpool_name);
3618         zc.zc_cookie = new_version;
3619
3620         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3621                 return (zpool_standard_error_fmt(hdl, errno,
3622                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3623                     zhp->zpool_name));
3624         return (0);
3625 }
3626
3627 void
3628 zfs_save_arguments(int argc, char **argv, char *string, int len)
3629 {
3630         (void) strlcpy(string, basename(argv[0]), len);
3631         for (int i = 1; i < argc; i++) {
3632                 (void) strlcat(string, " ", len);
3633                 (void) strlcat(string, argv[i], len);
3634         }
3635 }
3636
3637 int
3638 zpool_log_history(libzfs_handle_t *hdl, const char *message)
3639 {
3640         zfs_cmd_t zc = { 0 };
3641         nvlist_t *args;
3642         int err;
3643
3644         args = fnvlist_alloc();
3645         fnvlist_add_string(args, "message", message);
3646         err = zcmd_write_src_nvlist(hdl, &zc, args);
3647         if (err == 0)
3648                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_LOG_HISTORY, &zc);
3649         nvlist_free(args);
3650         zcmd_free_nvlists(&zc);
3651         return (err);
3652 }
3653
3654 /*
3655  * Perform ioctl to get some command history of a pool.
3656  *
3657  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3658  * logical offset of the history buffer to start reading from.
3659  *
3660  * Upon return, 'off' is the next logical offset to read from and
3661  * 'len' is the actual amount of bytes read into 'buf'.
3662  */
3663 static int
3664 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3665 {
3666         zfs_cmd_t zc = { 0 };
3667         libzfs_handle_t *hdl = zhp->zpool_hdl;
3668
3669         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3670
3671         zc.zc_history = (uint64_t)(uintptr_t)buf;
3672         zc.zc_history_len = *len;
3673         zc.zc_history_offset = *off;
3674
3675         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3676                 switch (errno) {
3677                 case EPERM:
3678                         return (zfs_error_fmt(hdl, EZFS_PERM,
3679                             dgettext(TEXT_DOMAIN,
3680                             "cannot show history for pool '%s'"),
3681                             zhp->zpool_name));
3682                 case ENOENT:
3683                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3684                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3685                             "'%s'"), zhp->zpool_name));
3686                 case ENOTSUP:
3687                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3688                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3689                             "'%s', pool must be upgraded"), zhp->zpool_name));
3690                 default:
3691                         return (zpool_standard_error_fmt(hdl, errno,
3692                             dgettext(TEXT_DOMAIN,
3693                             "cannot get history for '%s'"), zhp->zpool_name));
3694                 }
3695         }
3696
3697         *len = zc.zc_history_len;
3698         *off = zc.zc_history_offset;
3699
3700         return (0);
3701 }
3702
3703 /*
3704  * Process the buffer of nvlists, unpacking and storing each nvlist record
3705  * into 'records'.  'leftover' is set to the number of bytes that weren't
3706  * processed as there wasn't a complete record.
3707  */
3708 int
3709 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3710     nvlist_t ***records, uint_t *numrecords)
3711 {
3712         uint64_t reclen;
3713         nvlist_t *nv;
3714         int i;
3715
3716         while (bytes_read > sizeof (reclen)) {
3717
3718                 /* get length of packed record (stored as little endian) */
3719                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3720                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3721
3722                 if (bytes_read < sizeof (reclen) + reclen)
3723                         break;
3724
3725                 /* unpack record */
3726                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3727                         return (ENOMEM);
3728                 bytes_read -= sizeof (reclen) + reclen;
3729                 buf += sizeof (reclen) + reclen;
3730
3731                 /* add record to nvlist array */
3732                 (*numrecords)++;
3733                 if (ISP2(*numrecords + 1)) {
3734                         *records = realloc(*records,
3735                             *numrecords * 2 * sizeof (nvlist_t *));
3736                 }
3737                 (*records)[*numrecords - 1] = nv;
3738         }
3739
3740         *leftover = bytes_read;
3741         return (0);
3742 }
3743
3744 /* from spa_history.c: spa_history_create_obj() */
3745 #define HIS_BUF_LEN_DEF (128 << 10)
3746 #define HIS_BUF_LEN_MAX (1 << 30)
3747
3748 /*
3749  * Retrieve the command history of a pool.
3750  */
3751 int
3752 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3753 {
3754         char *buf;
3755         uint64_t buflen = HIS_BUF_LEN_DEF;
3756         uint64_t off = 0;
3757         nvlist_t **records = NULL;
3758         uint_t numrecords = 0;
3759         int err, i;
3760
3761         buf = malloc(buflen);
3762         if (buf == NULL)
3763                 return (ENOMEM);
3764         do {
3765                 uint64_t bytes_read = buflen;
3766                 uint64_t leftover;
3767
3768                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3769                         break;
3770
3771                 /* if nothing else was read in, we're at EOF, just return */
3772                 if (bytes_read == 0)
3773                         break;
3774
3775                 if ((err = zpool_history_unpack(buf, bytes_read,
3776                     &leftover, &records, &numrecords)) != 0)
3777                         break;
3778                 off -= leftover;
3779                 if (leftover == bytes_read) {
3780                         /*
3781                          * no progress made, because buffer is not big enough
3782                          * to hold this record; resize and retry.
3783                          */
3784                         buflen *= 2;
3785                         free(buf);
3786                         buf = NULL;
3787                         if ((buflen >= HIS_BUF_LEN_MAX) ||
3788                             ((buf = malloc(buflen)) == NULL)) {
3789                                 err = ENOMEM;
3790                                 break;
3791                         }
3792                 }
3793
3794                 /* CONSTCOND */
3795         } while (1);
3796
3797         free(buf);
3798
3799         if (!err) {
3800                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3801                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3802                     records, numrecords) == 0);
3803         }
3804         for (i = 0; i < numrecords; i++)
3805                 nvlist_free(records[i]);
3806         free(records);
3807
3808         return (err);
3809 }
3810
3811 void
3812 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
3813     char *pathname, size_t len)
3814 {
3815         zfs_cmd_t zc = { 0 };
3816         boolean_t mounted = B_FALSE;
3817         char *mntpnt = NULL;
3818         char dsname[ZFS_MAX_DATASET_NAME_LEN];
3819
3820         if (dsobj == 0) {
3821                 /* special case for the MOS */
3822                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>", obj);
3823                 return;
3824         }
3825
3826         /* get the dataset's name */
3827         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3828         zc.zc_obj = dsobj;
3829         if (ioctl(zhp->zpool_hdl->libzfs_fd,
3830             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
3831                 /* just write out a path of two object numbers */
3832                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
3833                     dsobj, obj);
3834                 return;
3835         }
3836         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
3837
3838         /* find out if the dataset is mounted */
3839         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
3840
3841         /* get the corrupted object's path */
3842         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
3843         zc.zc_obj = obj;
3844         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
3845             &zc) == 0) {
3846                 if (mounted) {
3847                         (void) snprintf(pathname, len, "%s%s", mntpnt,
3848                             zc.zc_value);
3849                 } else {
3850                         (void) snprintf(pathname, len, "%s:%s",
3851                             dsname, zc.zc_value);
3852                 }
3853         } else {
3854                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname, obj);
3855         }
3856         free(mntpnt);
3857 }
3858
3859 #ifdef illumos
3860 /*
3861  * Read the EFI label from the config, if a label does not exist then
3862  * pass back the error to the caller. If the caller has passed a non-NULL
3863  * diskaddr argument then we set it to the starting address of the EFI
3864  * partition.
3865  */
3866 static int
3867 read_efi_label(nvlist_t *config, diskaddr_t *sb)
3868 {
3869         char *path;
3870         int fd;
3871         char diskname[MAXPATHLEN];
3872         int err = -1;
3873
3874         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
3875                 return (err);
3876
3877         (void) snprintf(diskname, sizeof (diskname), "%s%s", ZFS_RDISK_ROOT,
3878             strrchr(path, '/'));
3879         if ((fd = open(diskname, O_RDONLY|O_NDELAY)) >= 0) {
3880                 struct dk_gpt *vtoc;
3881
3882                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
3883                         if (sb != NULL)
3884                                 *sb = vtoc->efi_parts[0].p_start;
3885                         efi_free(vtoc);
3886                 }
3887                 (void) close(fd);
3888         }
3889         return (err);
3890 }
3891
3892 /*
3893  * determine where a partition starts on a disk in the current
3894  * configuration
3895  */
3896 static diskaddr_t
3897 find_start_block(nvlist_t *config)
3898 {
3899         nvlist_t **child;
3900         uint_t c, children;
3901         diskaddr_t sb = MAXOFFSET_T;
3902         uint64_t wholedisk;
3903
3904         if (nvlist_lookup_nvlist_array(config,
3905             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
3906                 if (nvlist_lookup_uint64(config,
3907                     ZPOOL_CONFIG_WHOLE_DISK,
3908                     &wholedisk) != 0 || !wholedisk) {
3909                         return (MAXOFFSET_T);
3910                 }
3911                 if (read_efi_label(config, &sb) < 0)
3912                         sb = MAXOFFSET_T;
3913                 return (sb);
3914         }
3915
3916         for (c = 0; c < children; c++) {
3917                 sb = find_start_block(child[c]);
3918                 if (sb != MAXOFFSET_T) {
3919                         return (sb);
3920                 }
3921         }
3922         return (MAXOFFSET_T);
3923 }
3924 #endif /* illumos */
3925
3926 /*
3927  * Label an individual disk.  The name provided is the short name,
3928  * stripped of any leading /dev path.
3929  */
3930 int
3931 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, const char *name)
3932 {
3933 #ifdef illumos
3934         char path[MAXPATHLEN];
3935         struct dk_gpt *vtoc;
3936         int fd;
3937         size_t resv = EFI_MIN_RESV_SIZE;
3938         uint64_t slice_size;
3939         diskaddr_t start_block;
3940         char errbuf[1024];
3941
3942         /* prepare an error message just in case */
3943         (void) snprintf(errbuf, sizeof (errbuf),
3944             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
3945
3946         if (zhp) {
3947                 nvlist_t *nvroot;
3948
3949                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
3950                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3951
3952                 if (zhp->zpool_start_block == 0)
3953                         start_block = find_start_block(nvroot);
3954                 else
3955                         start_block = zhp->zpool_start_block;
3956                 zhp->zpool_start_block = start_block;
3957         } else {
3958                 /* new pool */
3959                 start_block = NEW_START_BLOCK;
3960         }
3961
3962         (void) snprintf(path, sizeof (path), "%s/%s%s", ZFS_RDISK_ROOT, name,
3963             BACKUP_SLICE);
3964
3965         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
3966                 /*
3967                  * This shouldn't happen.  We've long since verified that this
3968                  * is a valid device.
3969                  */
3970                 zfs_error_aux(hdl,
3971                     dgettext(TEXT_DOMAIN, "unable to open device"));
3972                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
3973         }
3974
3975         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
3976                 /*
3977                  * The only way this can fail is if we run out of memory, or we
3978                  * were unable to read the disk's capacity
3979                  */
3980                 if (errno == ENOMEM)
3981                         (void) no_memory(hdl);
3982
3983                 (void) close(fd);
3984                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3985                     "unable to read disk capacity"), name);
3986
3987                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
3988         }
3989
3990         slice_size = vtoc->efi_last_u_lba + 1;
3991         slice_size -= EFI_MIN_RESV_SIZE;
3992         if (start_block == MAXOFFSET_T)
3993                 start_block = NEW_START_BLOCK;
3994         slice_size -= start_block;
3995
3996         vtoc->efi_parts[0].p_start = start_block;
3997         vtoc->efi_parts[0].p_size = slice_size;
3998
3999         /*
4000          * Why we use V_USR: V_BACKUP confuses users, and is considered
4001          * disposable by some EFI utilities (since EFI doesn't have a backup
4002          * slice).  V_UNASSIGNED is supposed to be used only for zero size
4003          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
4004          * etc. were all pretty specific.  V_USR is as close to reality as we
4005          * can get, in the absence of V_OTHER.
4006          */
4007         vtoc->efi_parts[0].p_tag = V_USR;
4008         (void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
4009
4010         vtoc->efi_parts[8].p_start = slice_size + start_block;
4011         vtoc->efi_parts[8].p_size = resv;
4012         vtoc->efi_parts[8].p_tag = V_RESERVED;
4013
4014         if (efi_write(fd, vtoc) != 0) {
4015                 /*
4016                  * Some block drivers (like pcata) may not support EFI
4017                  * GPT labels.  Print out a helpful error message dir-
4018                  * ecting the user to manually label the disk and give
4019                  * a specific slice.
4020                  */
4021                 (void) close(fd);
4022                 efi_free(vtoc);
4023
4024                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4025                     "try using fdisk(1M) and then provide a specific slice"));
4026                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4027         }
4028
4029         (void) close(fd);
4030         efi_free(vtoc);
4031 #endif /* illumos */
4032         return (0);
4033 }
4034
4035 static boolean_t
4036 supported_dump_vdev_type(libzfs_handle_t *hdl, nvlist_t *config, char *errbuf)
4037 {
4038         char *type;
4039         nvlist_t **child;
4040         uint_t children, c;
4041
4042         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_TYPE, &type) == 0);
4043         if (strcmp(type, VDEV_TYPE_FILE) == 0 ||
4044             strcmp(type, VDEV_TYPE_HOLE) == 0 ||
4045             strcmp(type, VDEV_TYPE_MISSING) == 0) {
4046                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4047                     "vdev type '%s' is not supported"), type);
4048                 (void) zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf);
4049                 return (B_FALSE);
4050         }
4051         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
4052             &child, &children) == 0) {
4053                 for (c = 0; c < children; c++) {
4054                         if (!supported_dump_vdev_type(hdl, child[c], errbuf))
4055                                 return (B_FALSE);
4056                 }
4057         }
4058         return (B_TRUE);
4059 }
4060
4061 /*
4062  * Check if this zvol is allowable for use as a dump device; zero if
4063  * it is, > 0 if it isn't, < 0 if it isn't a zvol.
4064  *
4065  * Allowable storage configurations include mirrors, all raidz variants, and
4066  * pools with log, cache, and spare devices.  Pools which are backed by files or
4067  * have missing/hole vdevs are not suitable.
4068  */
4069 int
4070 zvol_check_dump_config(char *arg)
4071 {
4072         zpool_handle_t *zhp = NULL;
4073         nvlist_t *config, *nvroot;
4074         char *p, *volname;
4075         nvlist_t **top;
4076         uint_t toplevels;
4077         libzfs_handle_t *hdl;
4078         char errbuf[1024];
4079         char poolname[ZFS_MAX_DATASET_NAME_LEN];
4080         int pathlen = strlen(ZVOL_FULL_DEV_DIR);
4081         int ret = 1;
4082
4083         if (strncmp(arg, ZVOL_FULL_DEV_DIR, pathlen)) {
4084                 return (-1);
4085         }
4086
4087         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
4088             "dump is not supported on device '%s'"), arg);
4089
4090         if ((hdl = libzfs_init()) == NULL)
4091                 return (1);
4092         libzfs_print_on_error(hdl, B_TRUE);
4093
4094         volname = arg + pathlen;
4095
4096         /* check the configuration of the pool */
4097         if ((p = strchr(volname, '/')) == NULL) {
4098                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4099                     "malformed dataset name"));
4100                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4101                 return (1);
4102         } else if (p - volname >= ZFS_MAX_DATASET_NAME_LEN) {
4103                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4104                     "dataset name is too long"));
4105                 (void) zfs_error(hdl, EZFS_NAMETOOLONG, errbuf);
4106                 return (1);
4107         } else {
4108                 (void) strncpy(poolname, volname, p - volname);
4109                 poolname[p - volname] = '\0';
4110         }
4111
4112         if ((zhp = zpool_open(hdl, poolname)) == NULL) {
4113                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4114                     "could not open pool '%s'"), poolname);
4115                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
4116                 goto out;
4117         }
4118         config = zpool_get_config(zhp, NULL);
4119         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4120             &nvroot) != 0) {
4121                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4122                     "could not obtain vdev configuration for  '%s'"), poolname);
4123                 (void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
4124                 goto out;
4125         }
4126
4127         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4128             &top, &toplevels) == 0);
4129
4130         if (!supported_dump_vdev_type(hdl, top[0], errbuf)) {
4131                 goto out;
4132         }
4133         ret = 0;
4134
4135 out:
4136         if (zhp)
4137                 zpool_close(zhp);
4138         libzfs_fini(hdl);
4139         return (ret);
4140 }
4141
4142 int
4143 zpool_nextboot(libzfs_handle_t *hdl, uint64_t pool_guid, uint64_t dev_guid,
4144     const char *command)
4145 {
4146         zfs_cmd_t zc = { 0 };
4147         nvlist_t *args;
4148         char *packed;
4149         size_t size;
4150         int error;
4151
4152         args = fnvlist_alloc();
4153         fnvlist_add_uint64(args, ZPOOL_CONFIG_POOL_GUID, pool_guid);
4154         fnvlist_add_uint64(args, ZPOOL_CONFIG_GUID, dev_guid);
4155         fnvlist_add_string(args, "command", command);
4156         error = zcmd_write_src_nvlist(hdl, &zc, args);
4157         if (error == 0)
4158                 error = ioctl(hdl->libzfs_fd, ZFS_IOC_NEXTBOOT, &zc);
4159         zcmd_free_nvlists(&zc);
4160         nvlist_free(args);
4161         return (error);
4162 }