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