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, 2017 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  * Copyright (c) 2017 Datto Inc.
  29  */
  30
  31 #include <sys/types.h>
  32 #include <sys/stat.h>
  33 #include <ctype.h>
  34 #include <errno.h>
  35 #include <devid.h>
  36 #include <fcntl.h>
  37 #include <libintl.h>
  38 #include <stdio.h>
  39 #include <stdlib.h>
  40 #include <strings.h>
  41 #include <unistd.h>
  42 #include <libgen.h>
  43 #include <sys/zfs_ioctl.h>
  44 #include <dlfcn.h>
  45
  46 #include "zfs_namecheck.h"
  47 #include "zfs_prop.h"
  48 #include "libzfs_impl.h"
  49 #include "zfs_comutil.h"
  50 #include "zfeature_common.h"
  51
  52 static int read_efi_label(nvlist_t *, diskaddr_t *, boolean_t *);
  53 static boolean_t zpool_vdev_is_interior(const char *name);
  54
  55 #define BACKUP_SLICE    "s2"
  56
  57 typedef struct prop_flags {
  58         int create:1;   /* Validate property on creation */
  59         int import:1;   /* Validate property on import */
  60 } prop_flags_t;
  61
  62 /*
  63  * ====================================================================
  64  *   zpool property functions
  65  * ====================================================================
  66  */
  67
  68 static int
  69 zpool_get_all_props(zpool_handle_t *zhp)
  70 {
  71         zfs_cmd_t zc = { 0 };
  72         libzfs_handle_t *hdl = zhp->zpool_hdl;
  73
  74         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
  75
  76         if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
  77                 return (-1);
  78
  79         while (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
  80                 if (errno == ENOMEM) {
  81                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
  82                                 zcmd_free_nvlists(&zc);
  83                                 return (-1);
  84                         }
  85                 } else {
  86                         zcmd_free_nvlists(&zc);
  87                         return (-1);
  88                 }
  89         }
  90
  91         if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
  92                 zcmd_free_nvlists(&zc);
  93                 return (-1);
  94         }
  95
  96         zcmd_free_nvlists(&zc);
  97
  98         return (0);
  99 }
 100
 101 static int
 102 zpool_props_refresh(zpool_handle_t *zhp)
 103 {
 104         nvlist_t *old_props;
 105
 106         old_props = zhp->zpool_props;
 107
 108         if (zpool_get_all_props(zhp) != 0)
 109                 return (-1);
 110
 111         nvlist_free(old_props);
 112         return (0);
 113 }
 114
 115 static char *
 116 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
 117     zprop_source_t *src)
 118 {
 119         nvlist_t *nv, *nvl;
 120         uint64_t ival;
 121         char *value;
 122         zprop_source_t source;
 123
 124         nvl = zhp->zpool_props;
 125         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 126                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &ival) == 0);
 127                 source = ival;
 128                 verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
 129         } else {
 130                 source = ZPROP_SRC_DEFAULT;
 131                 if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
 132                         value = "-";
 133         }
 134
 135         if (src)
 136                 *src = source;
 137
 138         return (value);
 139 }
 140
 141 uint64_t
 142 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
 143 {
 144         nvlist_t *nv, *nvl;
 145         uint64_t value;
 146         zprop_source_t source;
 147
 148         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
 149                 /*
 150                  * zpool_get_all_props() has most likely failed because
 151                  * the pool is faulted, but if all we need is the top level
 152                  * vdev's guid then get it from the zhp config nvlist.
 153                  */
 154                 if ((prop == ZPOOL_PROP_GUID) &&
 155                     (nvlist_lookup_nvlist(zhp->zpool_config,
 156                     ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
 157                     (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
 158                     == 0)) {
 159                         return (value);
 160                 }
 161                 return (zpool_prop_default_numeric(prop));
 162         }
 163
 164         nvl = zhp->zpool_props;
 165         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 166                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &value) == 0);
 167                 source = value;
 168                 verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
 169         } else {
 170                 source = ZPROP_SRC_DEFAULT;
 171                 value = zpool_prop_default_numeric(prop);
 172         }
 173
 174         if (src)
 175                 *src = source;
 176
 177         return (value);
 178 }
 179
 180 /*
 181  * Map VDEV STATE to printed strings.
 182  */
 183 const char *
 184 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
 185 {
 186         switch (state) {
 187         case VDEV_STATE_CLOSED:
 188         case VDEV_STATE_OFFLINE:
 189                 return (gettext("OFFLINE"));
 190         case VDEV_STATE_REMOVED:
 191                 return (gettext("REMOVED"));
 192         case VDEV_STATE_CANT_OPEN:
 193                 if (aux == VDEV_AUX_CORRUPT_DATA || aux == VDEV_AUX_BAD_LOG)
 194                         return (gettext("FAULTED"));
 195                 else if (aux == VDEV_AUX_SPLIT_POOL)
 196                         return (gettext("SPLIT"));
 197                 else
 198                         return (gettext("UNAVAIL"));
 199         case VDEV_STATE_FAULTED:
 200                 return (gettext("FAULTED"));
 201         case VDEV_STATE_DEGRADED:
 202                 return (gettext("DEGRADED"));
 203         case VDEV_STATE_HEALTHY:
 204                 return (gettext("ONLINE"));
 205
 206         default:
 207                 break;
 208         }
 209
 210         return (gettext("UNKNOWN"));
 211 }
 212
 213 /*
 214  * Map POOL STATE to printed strings.
 215  */
 216 const char *
 217 zpool_pool_state_to_name(pool_state_t state)
 218 {
 219         switch (state) {
 220         case POOL_STATE_ACTIVE:
 221                 return (gettext("ACTIVE"));
 222         case POOL_STATE_EXPORTED:
 223                 return (gettext("EXPORTED"));
 224         case POOL_STATE_DESTROYED:
 225                 return (gettext("DESTROYED"));
 226         case POOL_STATE_SPARE:
 227                 return (gettext("SPARE"));
 228         case POOL_STATE_L2CACHE:
 229                 return (gettext("L2CACHE"));
 230         case POOL_STATE_UNINITIALIZED:
 231                 return (gettext("UNINITIALIZED"));
 232         case POOL_STATE_UNAVAIL:
 233                 return (gettext("UNAVAIL"));
 234         case POOL_STATE_POTENTIALLY_ACTIVE:
 235                 return (gettext("POTENTIALLY_ACTIVE"));
 236         }
 237
 238         return (gettext("UNKNOWN"));
 239 }
 240
 241 /*
 242  * Get a zpool property value for 'prop' and return the value in
 243  * a pre-allocated buffer.
 244  */
 245 int
 246 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf, size_t len,
 247     zprop_source_t *srctype, boolean_t literal)
 248 {
 249         uint64_t intval;
 250         const char *strval;
 251         zprop_source_t src = ZPROP_SRC_NONE;
 252         nvlist_t *nvroot;
 253         vdev_stat_t *vs;
 254         uint_t vsc;
 255
 256         if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
 257                 switch (prop) {
 258                 case ZPOOL_PROP_NAME:
 259                         (void) strlcpy(buf, zpool_get_name(zhp), len);
 260                         break;
 261
 262                 case ZPOOL_PROP_HEALTH:
 263                         (void) strlcpy(buf,
 264                             zpool_pool_state_to_name(POOL_STATE_UNAVAIL), len);
 265                         break;
 266
 267                 case ZPOOL_PROP_GUID:
 268                         intval = zpool_get_prop_int(zhp, prop, &src);
 269                         (void) snprintf(buf, len, "%llu", intval);
 270                         break;
 271
 272                 case ZPOOL_PROP_ALTROOT:
 273                 case ZPOOL_PROP_CACHEFILE:
 274                 case ZPOOL_PROP_COMMENT:
 275                         if (zhp->zpool_props != NULL ||
 276                             zpool_get_all_props(zhp) == 0) {
 277                                 (void) strlcpy(buf,
 278                                     zpool_get_prop_string(zhp, prop, &src),
 279                                     len);
 280                                 break;
 281                         }
 282                         /* FALLTHROUGH */
 283                 default:
 284                         (void) strlcpy(buf, "-", len);
 285                         break;
 286                 }
 287
 288                 if (srctype != NULL)
 289                         *srctype = src;
 290                 return (0);
 291         }
 292
 293         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
 294             prop != ZPOOL_PROP_NAME)
 295                 return (-1);
 296
 297         switch (zpool_prop_get_type(prop)) {
 298         case PROP_TYPE_STRING:
 299                 (void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
 300                     len);
 301                 break;
 302
 303         case PROP_TYPE_NUMBER:
 304                 intval = zpool_get_prop_int(zhp, prop, &src);
 305
 306                 switch (prop) {
 307                 case ZPOOL_PROP_SIZE:
 308                 case ZPOOL_PROP_ALLOCATED:
 309                 case ZPOOL_PROP_FREE:
 310                 case ZPOOL_PROP_FREEING:
 311                 case ZPOOL_PROP_LEAKED:
 312                         if (literal) {
 313                                 (void) snprintf(buf, len, "%llu",
 314                                     (u_longlong_t)intval);
 315                         } else {
 316                                 (void) zfs_nicenum(intval, buf, len);
 317                         }
 318                         break;
 319                 case ZPOOL_PROP_BOOTSIZE:
 320                 case ZPOOL_PROP_EXPANDSZ:
 321                 case ZPOOL_PROP_CHECKPOINT:
 322                         if (intval == 0) {
 323                                 (void) strlcpy(buf, "-", len);
 324                         } else if (literal) {
 325                                 (void) snprintf(buf, len, "%llu",
 326                                     (u_longlong_t)intval);
 327                         } else {
 328                                 (void) zfs_nicenum(intval, buf, len);
 329                         }
 330                         break;
 331                 case ZPOOL_PROP_CAPACITY:
 332                         if (literal) {
 333                                 (void) snprintf(buf, len, "%llu",
 334                                     (u_longlong_t)intval);
 335                         } else {
 336                                 (void) snprintf(buf, len, "%llu%%",
 337                                     (u_longlong_t)intval);
 338                         }
 339                         break;
 340                 case ZPOOL_PROP_FRAGMENTATION:
 341                         if (intval == UINT64_MAX) {
 342                                 (void) strlcpy(buf, "-", len);
 343                         } else {
 344                                 (void) snprintf(buf, len, "%llu%%",
 345                                     (u_longlong_t)intval);
 346                         }
 347                         break;
 348                 case ZPOOL_PROP_DEDUPRATIO:
 349                         (void) snprintf(buf, len, "%llu.%02llux",
 350                             (u_longlong_t)(intval / 100),
 351                             (u_longlong_t)(intval % 100));
 352                         break;
 353                 case ZPOOL_PROP_HEALTH:
 354                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 355                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 356                         verify(nvlist_lookup_uint64_array(nvroot,
 357                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
 358                             == 0);
 359
 360                         (void) strlcpy(buf, zpool_state_to_name(intval,
 361                             vs->vs_aux), len);
 362                         break;
 363                 case ZPOOL_PROP_VERSION:
 364                         if (intval >= SPA_VERSION_FEATURES) {
 365                                 (void) snprintf(buf, len, "-");
 366                                 break;
 367                         }
 368                         /* FALLTHROUGH */
 369                 default:
 370                         (void) snprintf(buf, len, "%llu", intval);
 371                 }
 372                 break;
 373
 374         case PROP_TYPE_INDEX:
 375                 intval = zpool_get_prop_int(zhp, prop, &src);
 376                 if (zpool_prop_index_to_string(prop, intval, &strval)
 377                     != 0)
 378                         return (-1);
 379                 (void) strlcpy(buf, strval, len);
 380                 break;
 381
 382         default:
 383                 abort();
 384         }
 385
 386         if (srctype)
 387                 *srctype = src;
 388
 389         return (0);
 390 }
 391
 392 /*
 393  * Check if the bootfs name has the same pool name as it is set to.
 394  * Assuming bootfs is a valid dataset name.
 395  */
 396 static boolean_t
 397 bootfs_name_valid(const char *pool, char *bootfs)
 398 {
 399         int len = strlen(pool);
 400
 401         if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
 402                 return (B_FALSE);
 403
 404         if (strncmp(pool, bootfs, len) == 0 &&
 405             (bootfs[len] == '/' || bootfs[len] == '\0'))
 406                 return (B_TRUE);
 407
 408         return (B_FALSE);
 409 }
 410
 411 boolean_t
 412 zpool_is_bootable(zpool_handle_t *zhp)
 413 {
 414         char bootfs[ZFS_MAX_DATASET_NAME_LEN];
 415
 416         return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
 417             sizeof (bootfs), NULL, B_FALSE) == 0 && strncmp(bootfs, "-",
 418             sizeof (bootfs)) != 0);
 419 }
 420
 421
 422 /*
 423  * Given an nvlist of zpool properties to be set, validate that they are
 424  * correct, and parse any numeric properties (index, boolean, etc) if they are
 425  * specified as strings.
 426  */
 427 static nvlist_t *
 428 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 429     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
 430 {
 431         nvpair_t *elem;
 432         nvlist_t *retprops;
 433         zpool_prop_t prop;
 434         char *strval;
 435         uint64_t intval;
 436         char *slash, *check;
 437         struct stat64 statbuf;
 438         zpool_handle_t *zhp;
 439
 440         if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
 441                 (void) no_memory(hdl);
 442                 return (NULL);
 443         }
 444
 445         elem = NULL;
 446         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 447                 const char *propname = nvpair_name(elem);
 448
 449                 prop = zpool_name_to_prop(propname);
 450                 if (prop == ZPOOL_PROP_INVAL && zpool_prop_feature(propname)) {
 451                         int err;
 452                         char *fname = strchr(propname, '@') + 1;
 453
 454                         err = zfeature_lookup_name(fname, NULL);
 455                         if (err != 0) {
 456                                 ASSERT3U(err, ==, ENOENT);
 457                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 458                                     "invalid feature '%s'"), fname);
 459                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 460                                 goto error;
 461                         }
 462
 463                         if (nvpair_type(elem) != DATA_TYPE_STRING) {
 464                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 465                                     "'%s' must be a string"), propname);
 466                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 467                                 goto error;
 468                         }
 469
 470                         (void) nvpair_value_string(elem, &strval);
 471                         if (strcmp(strval, ZFS_FEATURE_ENABLED) != 0) {
 472                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 473                                     "property '%s' can only be set to "
 474                                     "'enabled'"), propname);
 475                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 476                                 goto error;
 477                         }
 478
 479                         if (nvlist_add_uint64(retprops, propname, 0) != 0) {
 480                                 (void) no_memory(hdl);
 481                                 goto error;
 482                         }
 483                         continue;
 484                 }
 485
 486                 /*
 487                  * Make sure this property is valid and applies to this type.
 488                  */
 489                 if (prop == ZPOOL_PROP_INVAL) {
 490                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 491                             "invalid property '%s'"), propname);
 492                         (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 493                         goto error;
 494                 }
 495
 496                 if (zpool_prop_readonly(prop)) {
 497                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 498                             "is readonly"), propname);
 499                         (void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
 500                         goto error;
 501                 }
 502
 503                 if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
 504                     &strval, &intval, errbuf) != 0)
 505                         goto error;
 506
 507                 /*
 508                  * Perform additional checking for specific properties.
 509                  */
 510                 switch (prop) {
 511                 case ZPOOL_PROP_VERSION:
 512                         if (intval < version ||
 513                             !SPA_VERSION_IS_SUPPORTED(intval)) {
 514                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 515                                     "property '%s' number %d is invalid."),
 516                                     propname, intval);
 517                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 518                                 goto error;
 519                         }
 520                         break;
 521
 522                 case ZPOOL_PROP_BOOTSIZE:
 523                         if (!flags.create) {
 524                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 525                                     "property '%s' can only be set during pool "
 526                                     "creation"), propname);
 527                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 528                                 goto error;
 529                         }
 530                         break;
 531
 532                 case ZPOOL_PROP_BOOTFS:
 533                         if (flags.create || flags.import) {
 534                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 535                                     "property '%s' cannot be set at creation "
 536                                     "or import time"), propname);
 537                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 538                                 goto error;
 539                         }
 540
 541                         if (version < SPA_VERSION_BOOTFS) {
 542                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 543                                     "pool must be upgraded to support "
 544                                     "'%s' property"), propname);
 545                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 546                                 goto error;
 547                         }
 548
 549                         /*
 550                          * bootfs property value has to be a dataset name and
 551                          * the dataset has to be in the same pool as it sets to.
 552                          */
 553                         if (strval[0] != '\0' && !bootfs_name_valid(poolname,
 554                             strval)) {
 555                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 556                                     "is an invalid name"), strval);
 557                                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
 558                                 goto error;
 559                         }
 560
 561                         if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
 562                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 563                                     "could not open pool '%s'"), poolname);
 564                                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
 565                                 goto error;
 566                         }
 567                         zpool_close(zhp);
 568                         break;
 569
 570                 case ZPOOL_PROP_ALTROOT:
 571                         if (!flags.create && !flags.import) {
 572                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 573                                     "property '%s' can only be set during pool "
 574                                     "creation or import"), propname);
 575                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 576                                 goto error;
 577                         }
 578
 579                         if (strval[0] != '/') {
 580                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 581                                     "bad alternate root '%s'"), strval);
 582                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 583                                 goto error;
 584                         }
 585                         break;
 586
 587                 case ZPOOL_PROP_CACHEFILE:
 588                         if (strval[0] == '\0')
 589                                 break;
 590
 591                         if (strcmp(strval, "none") == 0)
 592                                 break;
 593
 594                         if (strval[0] != '/') {
 595                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 596                                     "property '%s' must be empty, an "
 597                                     "absolute path, or 'none'"), propname);
 598                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 599                                 goto error;
 600                         }
 601
 602                         slash = strrchr(strval, '/');
 603
 604                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 605                             strcmp(slash, "/..") == 0) {
 606                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 607                                     "'%s' is not a valid file"), strval);
 608                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 609                                 goto error;
 610                         }
 611
 612                         *slash = '\0';
 613
 614                         if (strval[0] != '\0' &&
 615                             (stat64(strval, &statbuf) != 0 ||
 616                             !S_ISDIR(statbuf.st_mode))) {
 617                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 618                                     "'%s' is not a valid directory"),
 619                                     strval);
 620                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 621                                 goto error;
 622                         }
 623
 624                         *slash = '/';
 625                         break;
 626
 627                 case ZPOOL_PROP_COMMENT:
 628                         for (check = strval; *check != '\0'; check++) {
 629                                 if (!isprint(*check)) {
 630                                         zfs_error_aux(hdl,
 631                                             dgettext(TEXT_DOMAIN,
 632                                             "comment may only have printable "
 633                                             "characters"));
 634                                         (void) zfs_error(hdl, EZFS_BADPROP,
 635                                             errbuf);
 636                                         goto error;
 637                                 }
 638                         }
 639                         if (strlen(strval) > ZPROP_MAX_COMMENT) {
 640                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 641                                     "comment must not exceed %d characters"),
 642                                     ZPROP_MAX_COMMENT);
 643                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 644                                 goto error;
 645                         }
 646                         break;
 647
 648                 case ZPOOL_PROP_READONLY:
 649                         if (!flags.import) {
 650                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 651                                     "property '%s' can only be set at "
 652                                     "import time"), propname);
 653                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 654                                 goto error;
 655                         }
 656                         break;
 657
 658                 case ZPOOL_PROP_TNAME:
 659                         if (!flags.create) {
 660                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 661                                     "property '%s' can only be set at "
 662                                     "creation time"), propname);
 663                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 664                                 goto error;
 665                         }
 666                         break;
 667
 668                 default:
 669                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 670                             "property '%s'(%d) not defined"), propname, prop);
 671                         break;
 672                 }
 673         }
 674
 675         return (retprops);
 676 error:
 677         nvlist_free(retprops);
 678         return (NULL);
 679 }
 680
 681 /*
 682  * Set zpool property : propname=propval.
 683  */
 684 int
 685 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
 686 {
 687         zfs_cmd_t zc = { 0 };
 688         int ret = -1;
 689         char errbuf[1024];
 690         nvlist_t *nvl = NULL;
 691         nvlist_t *realprops;
 692         uint64_t version;
 693         prop_flags_t flags = { 0 };
 694
 695         (void) snprintf(errbuf, sizeof (errbuf),
 696             dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
 697             zhp->zpool_name);
 698
 699         if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
 700                 return (no_memory(zhp->zpool_hdl));
 701
 702         if (nvlist_add_string(nvl, propname, propval) != 0) {
 703                 nvlist_free(nvl);
 704                 return (no_memory(zhp->zpool_hdl));
 705         }
 706
 707         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
 708         if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
 709             zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
 710                 nvlist_free(nvl);
 711                 return (-1);
 712         }
 713
 714         nvlist_free(nvl);
 715         nvl = realprops;
 716
 717         /*
 718          * Execute the corresponding ioctl() to set this property.
 719          */
 720         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 721
 722         if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
 723                 nvlist_free(nvl);
 724                 return (-1);
 725         }
 726
 727         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
 728
 729         zcmd_free_nvlists(&zc);
 730         nvlist_free(nvl);
 731
 732         if (ret)
 733                 (void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
 734         else
 735                 (void) zpool_props_refresh(zhp);
 736
 737         return (ret);
 738 }
 739
 740 int
 741 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
 742 {
 743         libzfs_handle_t *hdl = zhp->zpool_hdl;
 744         zprop_list_t *entry;
 745         char buf[ZFS_MAXPROPLEN];
 746         nvlist_t *features = NULL;
 747         zprop_list_t **last;
 748         boolean_t firstexpand = (NULL == *plp);
 749
 750         if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
 751                 return (-1);
 752
 753         last = plp;
 754         while (*last != NULL)
 755                 last = &(*last)->pl_next;
 756
 757         if ((*plp)->pl_all)
 758                 features = zpool_get_features(zhp);
 759
 760         if ((*plp)->pl_all && firstexpand) {
 761                 for (int i = 0; i < SPA_FEATURES; i++) {
 762                         zprop_list_t *entry = zfs_alloc(hdl,
 763                             sizeof (zprop_list_t));
 764                         entry->pl_prop = ZPROP_INVAL;
 765                         entry->pl_user_prop = zfs_asprintf(hdl, "feature@%s",
 766                             spa_feature_table[i].fi_uname);
 767                         entry->pl_width = strlen(entry->pl_user_prop);
 768                         entry->pl_all = B_TRUE;
 769
 770                         *last = entry;
 771                         last = &entry->pl_next;
 772                 }
 773         }
 774
 775         /* add any unsupported features */
 776         for (nvpair_t *nvp = nvlist_next_nvpair(features, NULL);
 777             nvp != NULL; nvp = nvlist_next_nvpair(features, nvp)) {
 778                 char *propname;
 779                 boolean_t found;
 780                 zprop_list_t *entry;
 781
 782                 if (zfeature_is_supported(nvpair_name(nvp)))
 783                         continue;
 784
 785                 propname = zfs_asprintf(hdl, "unsupported@%s",
 786                     nvpair_name(nvp));
 787
 788                 /*
 789                  * Before adding the property to the list make sure that no
 790                  * other pool already added the same property.
 791                  */
 792                 found = B_FALSE;
 793                 entry = *plp;
 794                 while (entry != NULL) {
 795                         if (entry->pl_user_prop != NULL &&
 796                             strcmp(propname, entry->pl_user_prop) == 0) {
 797                                 found = B_TRUE;
 798                                 break;
 799                         }
 800                         entry = entry->pl_next;
 801                 }
 802                 if (found) {
 803                         free(propname);
 804                         continue;
 805                 }
 806
 807                 entry = zfs_alloc(hdl, sizeof (zprop_list_t));
 808                 entry->pl_prop = ZPROP_INVAL;
 809                 entry->pl_user_prop = propname;
 810                 entry->pl_width = strlen(entry->pl_user_prop);
 811                 entry->pl_all = B_TRUE;
 812
 813                 *last = entry;
 814                 last = &entry->pl_next;
 815         }
 816
 817         for (entry = *plp; entry != NULL; entry = entry->pl_next) {
 818
 819                 if (entry->pl_fixed)
 820                         continue;
 821
 822                 if (entry->pl_prop != ZPROP_INVAL &&
 823                     zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
 824                     NULL, B_FALSE) == 0) {
 825                         if (strlen(buf) > entry->pl_width)
 826                                 entry->pl_width = strlen(buf);
 827                 }
 828         }
 829
 830         return (0);
 831 }
 832
 833 /*
 834  * Get the state for the given feature on the given ZFS pool.
 835  */
 836 int
 837 zpool_prop_get_feature(zpool_handle_t *zhp, const char *propname, char *buf,
 838     size_t len)
 839 {
 840         uint64_t refcount;
 841         boolean_t found = B_FALSE;
 842         nvlist_t *features = zpool_get_features(zhp);
 843         boolean_t supported;
 844         const char *feature = strchr(propname, '@') + 1;
 845
 846         supported = zpool_prop_feature(propname);
 847         ASSERT(supported || zpool_prop_unsupported(propname));
 848
 849         /*
 850          * Convert from feature name to feature guid. This conversion is
 851          * unecessary for unsupported@... properties because they already
 852          * use guids.
 853          */
 854         if (supported) {
 855                 int ret;
 856                 spa_feature_t fid;
 857
 858                 ret = zfeature_lookup_name(feature, &fid);
 859                 if (ret != 0) {
 860                         (void) strlcpy(buf, "-", len);
 861                         return (ENOTSUP);
 862                 }
 863                 feature = spa_feature_table[fid].fi_guid;
 864         }
 865
 866         if (nvlist_lookup_uint64(features, feature, &refcount) == 0)
 867                 found = B_TRUE;
 868
 869         if (supported) {
 870                 if (!found) {
 871                         (void) strlcpy(buf, ZFS_FEATURE_DISABLED, len);
 872                 } else  {
 873                         if (refcount == 0)
 874                                 (void) strlcpy(buf, ZFS_FEATURE_ENABLED, len);
 875                         else
 876                                 (void) strlcpy(buf, ZFS_FEATURE_ACTIVE, len);
 877                 }
 878         } else {
 879                 if (found) {
 880                         if (refcount == 0) {
 881                                 (void) strcpy(buf, ZFS_UNSUPPORTED_INACTIVE);
 882                         } else {
 883                                 (void) strcpy(buf, ZFS_UNSUPPORTED_READONLY);
 884                         }
 885                 } else {
 886                         (void) strlcpy(buf, "-", len);
 887                         return (ENOTSUP);
 888                 }
 889         }
 890
 891         return (0);
 892 }
 893
 894 /*
 895  * Don't start the slice at the default block of 34; many storage
 896  * devices will use a stripe width of 128k, so start there instead.
 897  */
 898 #define NEW_START_BLOCK 256
 899
 900 /*
 901  * Validate the given pool name, optionally putting an extended error message in
 902  * 'buf'.
 903  */
 904 boolean_t
 905 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
 906 {
 907         namecheck_err_t why;
 908         char what;
 909         int ret;
 910
 911         ret = pool_namecheck(pool, &why, &what);
 912
 913         /*
 914          * The rules for reserved pool names were extended at a later point.
 915          * But we need to support users with existing pools that may now be
 916          * invalid.  So we only check for this expanded set of names during a
 917          * create (or import), and only in userland.
 918          */
 919         if (ret == 0 && !isopen &&
 920             (strncmp(pool, "mirror", 6) == 0 ||
 921             strncmp(pool, "raidz", 5) == 0 ||
 922             strncmp(pool, "spare", 5) == 0 ||
 923             strcmp(pool, "log") == 0)) {
 924                 if (hdl != NULL)
 925                         zfs_error_aux(hdl,
 926                             dgettext(TEXT_DOMAIN, "name is reserved"));
 927                 return (B_FALSE);
 928         }
 929
 930
 931         if (ret != 0) {
 932                 if (hdl != NULL) {
 933                         switch (why) {
 934                         case NAME_ERR_TOOLONG:
 935                                 zfs_error_aux(hdl,
 936                                     dgettext(TEXT_DOMAIN, "name is too long"));
 937                                 break;
 938
 939                         case NAME_ERR_INVALCHAR:
 940                                 zfs_error_aux(hdl,
 941                                     dgettext(TEXT_DOMAIN, "invalid character "
 942                                     "'%c' in pool name"), what);
 943                                 break;
 944
 945                         case NAME_ERR_NOLETTER:
 946                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 947                                     "name must begin with a letter"));
 948                                 break;
 949
 950                         case NAME_ERR_RESERVED:
 951                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 952                                     "name is reserved"));
 953                                 break;
 954
 955                         case NAME_ERR_DISKLIKE:
 956                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 957                                     "pool name is reserved"));
 958                                 break;
 959
 960                         case NAME_ERR_LEADING_SLASH:
 961                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 962                                     "leading slash in name"));
 963                                 break;
 964
 965                         case NAME_ERR_EMPTY_COMPONENT:
 966                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 967                                     "empty component in name"));
 968                                 break;
 969
 970                         case NAME_ERR_TRAILING_SLASH:
 971                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 972                                     "trailing slash in name"));
 973                                 break;
 974
 975                         case NAME_ERR_MULTIPLE_DELIMITERS:
 976                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 977                                     "multiple '@' and/or '#' delimiters in "
 978                                     "name"));
 979                                 break;
 980
 981                         default:
 982                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 983                                     "(%d) not defined"), why);
 984                                 break;
 985                         }
 986                 }
 987                 return (B_FALSE);
 988         }
 989
 990         return (B_TRUE);
 991 }
 992
 993 /*
 994  * Open a handle to the given pool, even if the pool is currently in the FAULTED
 995  * state.
 996  */
 997 zpool_handle_t *
 998 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
 999 {
1000         zpool_handle_t *zhp;
1001         boolean_t missing;
1002
1003         /*
1004          * Make sure the pool name is valid.
1005          */
1006         if (!zpool_name_valid(hdl, B_TRUE, pool)) {
1007                 (void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1008                     dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1009                     pool);
1010                 return (NULL);
1011         }
1012
1013         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1014                 return (NULL);
1015
1016         zhp->zpool_hdl = hdl;
1017         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1018
1019         if (zpool_refresh_stats(zhp, &missing) != 0) {
1020                 zpool_close(zhp);
1021                 return (NULL);
1022         }
1023
1024         if (missing) {
1025                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
1026                 (void) zfs_error_fmt(hdl, EZFS_NOENT,
1027                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
1028                 zpool_close(zhp);
1029                 return (NULL);
1030         }
1031
1032         return (zhp);
1033 }
1034
1035 /*
1036  * Like the above, but silent on error.  Used when iterating over pools (because
1037  * the configuration cache may be out of date).
1038  */
1039 int
1040 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
1041 {
1042         zpool_handle_t *zhp;
1043         boolean_t missing;
1044
1045         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1046                 return (-1);
1047
1048         zhp->zpool_hdl = hdl;
1049         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1050
1051         if (zpool_refresh_stats(zhp, &missing) != 0) {
1052                 zpool_close(zhp);
1053                 return (-1);
1054         }
1055
1056         if (missing) {
1057                 zpool_close(zhp);
1058                 *ret = NULL;
1059                 return (0);
1060         }
1061
1062         *ret = zhp;
1063         return (0);
1064 }
1065
1066 /*
1067  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
1068  * state.
1069  */
1070 zpool_handle_t *
1071 zpool_open(libzfs_handle_t *hdl, const char *pool)
1072 {
1073         zpool_handle_t *zhp;
1074
1075         if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
1076                 return (NULL);
1077
1078         if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
1079                 (void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
1080                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
1081                 zpool_close(zhp);
1082                 return (NULL);
1083         }
1084
1085         return (zhp);
1086 }
1087
1088 /*
1089  * Close the handle.  Simply frees the memory associated with the handle.
1090  */
1091 void
1092 zpool_close(zpool_handle_t *zhp)
1093 {
1094         nvlist_free(zhp->zpool_config);
1095         nvlist_free(zhp->zpool_old_config);
1096         nvlist_free(zhp->zpool_props);
1097         free(zhp);
1098 }
1099
1100 /*
1101  * Return the name of the pool.
1102  */
1103 const char *
1104 zpool_get_name(zpool_handle_t *zhp)
1105 {
1106         return (zhp->zpool_name);
1107 }
1108
1109
1110 /*
1111  * Return the state of the pool (ACTIVE or UNAVAILABLE)
1112  */
1113 int
1114 zpool_get_state(zpool_handle_t *zhp)
1115 {
1116         return (zhp->zpool_state);
1117 }
1118
1119 /*
1120  * Create the named pool, using the provided vdev list.  It is assumed
1121  * that the consumer has already validated the contents of the nvlist, so we
1122  * don't have to worry about error semantics.
1123  */
1124 int
1125 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
1126     nvlist_t *props, nvlist_t *fsprops)
1127 {
1128         zfs_cmd_t zc = { 0 };
1129         nvlist_t *zc_fsprops = NULL;
1130         nvlist_t *zc_props = NULL;
1131         char msg[1024];
1132         int ret = -1;
1133
1134         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1135             "cannot create '%s'"), pool);
1136
1137         if (!zpool_name_valid(hdl, B_FALSE, pool))
1138                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
1139
1140         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1141                 return (-1);
1142
1143         if (props) {
1144                 prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
1145
1146                 if ((zc_props = zpool_valid_proplist(hdl, pool, props,
1147                     SPA_VERSION_1, flags, msg)) == NULL) {
1148                         goto create_failed;
1149                 }
1150         }
1151
1152         if (fsprops) {
1153                 uint64_t zoned;
1154                 char *zonestr;
1155
1156                 zoned = ((nvlist_lookup_string(fsprops,
1157                     zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
1158                     strcmp(zonestr, "on") == 0);
1159
1160                 if ((zc_fsprops = zfs_valid_proplist(hdl, ZFS_TYPE_FILESYSTEM,
1161                     fsprops, zoned, NULL, NULL, msg)) == NULL) {
1162                         goto create_failed;
1163                 }
1164                 if (!zc_props &&
1165                     (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
1166                         goto create_failed;
1167                 }
1168                 if (nvlist_add_nvlist(zc_props,
1169                     ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
1170                         goto create_failed;
1171                 }
1172         }
1173
1174         if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
1175                 goto create_failed;
1176
1177         (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
1178
1179         if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
1180
1181                 zcmd_free_nvlists(&zc);
1182                 nvlist_free(zc_props);
1183                 nvlist_free(zc_fsprops);
1184
1185                 switch (errno) {
1186                 case EBUSY:
1187                         /*
1188                          * This can happen if the user has specified the same
1189                          * device multiple times.  We can't reliably detect this
1190                          * until we try to add it and see we already have a
1191                          * label.
1192                          */
1193                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1194                             "one or more vdevs refer to the same device"));
1195                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1196
1197                 case ERANGE:
1198                         /*
1199                          * This happens if the record size is smaller or larger
1200                          * than the allowed size range, or not a power of 2.
1201                          *
1202                          * NOTE: although zfs_valid_proplist is called earlier,
1203                          * this case may have slipped through since the
1204                          * pool does not exist yet and it is therefore
1205                          * impossible to read properties e.g. max blocksize
1206                          * from the pool.
1207                          */
1208                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1209                             "record size invalid"));
1210                         return (zfs_error(hdl, EZFS_BADPROP, msg));
1211
1212                 case EOVERFLOW:
1213                         /*
1214                          * This occurs when one of the devices is below
1215                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1216                          * device was the problem device since there's no
1217                          * reliable way to determine device size from userland.
1218                          */
1219                         {
1220                                 char buf[64];
1221
1222                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1223
1224                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1225                                     "one or more devices is less than the "
1226                                     "minimum size (%s)"), buf);
1227                         }
1228                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1229
1230                 case ENOSPC:
1231                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1232                             "one or more devices is out of space"));
1233                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1234
1235                 case ENOTBLK:
1236                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1237                             "cache device must be a disk or disk slice"));
1238                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1239
1240                 default:
1241                         return (zpool_standard_error(hdl, errno, msg));
1242                 }
1243         }
1244
1245 create_failed:
1246         zcmd_free_nvlists(&zc);
1247         nvlist_free(zc_props);
1248         nvlist_free(zc_fsprops);
1249         return (ret);
1250 }
1251
1252 /*
1253  * Destroy the given pool.  It is up to the caller to ensure that there are no
1254  * datasets left in the pool.
1255  */
1256 int
1257 zpool_destroy(zpool_handle_t *zhp, const char *log_str)
1258 {
1259         zfs_cmd_t zc = { 0 };
1260         zfs_handle_t *zfp = NULL;
1261         libzfs_handle_t *hdl = zhp->zpool_hdl;
1262         char msg[1024];
1263
1264         if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1265             (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1266                 return (-1);
1267
1268         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1269         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1270
1271         if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1272                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1273                     "cannot destroy '%s'"), zhp->zpool_name);
1274
1275                 if (errno == EROFS) {
1276                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1277                             "one or more devices is read only"));
1278                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1279                 } else {
1280                         (void) zpool_standard_error(hdl, errno, msg);
1281                 }
1282
1283                 if (zfp)
1284                         zfs_close(zfp);
1285                 return (-1);
1286         }
1287
1288         if (zfp) {
1289                 remove_mountpoint(zfp);
1290                 zfs_close(zfp);
1291         }
1292
1293         return (0);
1294 }
1295
1296 /*
1297  * Create a checkpoint in the given pool.
1298  */
1299 int
1300 zpool_checkpoint(zpool_handle_t *zhp)
1301 {
1302         libzfs_handle_t *hdl = zhp->zpool_hdl;
1303         char msg[1024];
1304         int error;
1305
1306         error = lzc_pool_checkpoint(zhp->zpool_name);
1307         if (error != 0) {
1308                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1309                     "cannot checkpoint '%s'"), zhp->zpool_name);
1310                 (void) zpool_standard_error(hdl, error, msg);
1311                 return (-1);
1312         }
1313
1314         return (0);
1315 }
1316
1317 /*
1318  * Discard the checkpoint from the given pool.
1319  */
1320 int
1321 zpool_discard_checkpoint(zpool_handle_t *zhp)
1322 {
1323         libzfs_handle_t *hdl = zhp->zpool_hdl;
1324         char msg[1024];
1325         int error;
1326
1327         error = lzc_pool_checkpoint_discard(zhp->zpool_name);
1328         if (error != 0) {
1329                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1330                     "cannot discard checkpoint in '%s'"), zhp->zpool_name);
1331                 (void) zpool_standard_error(hdl, error, msg);
1332                 return (-1);
1333         }
1334
1335         return (0);
1336 }
1337
1338 /*
1339  * Add the given vdevs to the pool.  The caller must have already performed the
1340  * necessary verification to ensure that the vdev specification is well-formed.
1341  */
1342 int
1343 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1344 {
1345         zfs_cmd_t zc = { 0 };
1346         int ret;
1347         libzfs_handle_t *hdl = zhp->zpool_hdl;
1348         char msg[1024];
1349         nvlist_t **spares, **l2cache;
1350         uint_t nspares, nl2cache;
1351
1352         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1353             "cannot add to '%s'"), zhp->zpool_name);
1354
1355         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1356             SPA_VERSION_SPARES &&
1357             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1358             &spares, &nspares) == 0) {
1359                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1360                     "upgraded to add hot spares"));
1361                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1362         }
1363
1364         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1365             SPA_VERSION_L2CACHE &&
1366             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1367             &l2cache, &nl2cache) == 0) {
1368                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1369                     "upgraded to add cache devices"));
1370                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1371         }
1372
1373         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1374                 return (-1);
1375         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1376
1377         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1378                 switch (errno) {
1379                 case EBUSY:
1380                         /*
1381                          * This can happen if the user has specified the same
1382                          * device multiple times.  We can't reliably detect this
1383                          * until we try to add it and see we already have a
1384                          * label.
1385                          */
1386                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1387                             "one or more vdevs refer to the same device"));
1388                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1389                         break;
1390
1391                 case EINVAL:
1392                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1393                             "invalid config; a pool with removing/removed "
1394                             "vdevs does not support adding raidz vdevs"));
1395                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1396                         break;
1397
1398                 case EOVERFLOW:
1399                         /*
1400                          * This occurrs when one of the devices is below
1401                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1402                          * device was the problem device since there's no
1403                          * reliable way to determine device size from userland.
1404                          */
1405                         {
1406                                 char buf[64];
1407
1408                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1409
1410                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1411                                     "device is less than the minimum "
1412                                     "size (%s)"), buf);
1413                         }
1414                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1415                         break;
1416
1417                 case ENOTSUP:
1418                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1419                             "pool must be upgraded to add these vdevs"));
1420                         (void) zfs_error(hdl, EZFS_BADVERSION, msg);
1421                         break;
1422
1423                 case EDOM:
1424                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1425                             "root pool can not have multiple vdevs"
1426                             " or separate logs"));
1427                         (void) zfs_error(hdl, EZFS_POOL_NOTSUP, msg);
1428                         break;
1429
1430                 case ENOTBLK:
1431                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1432                             "cache device must be a disk or disk slice"));
1433                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1434                         break;
1435
1436                 default:
1437                         (void) zpool_standard_error(hdl, errno, msg);
1438                 }
1439
1440                 ret = -1;
1441         } else {
1442                 ret = 0;
1443         }
1444
1445         zcmd_free_nvlists(&zc);
1446
1447         return (ret);
1448 }
1449
1450 /*
1451  * Exports the pool from the system.  The caller must ensure that there are no
1452  * mounted datasets in the pool.
1453  */
1454 static int
1455 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce,
1456     const char *log_str)
1457 {
1458         zfs_cmd_t zc = { 0 };
1459         char msg[1024];
1460
1461         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1462             "cannot export '%s'"), zhp->zpool_name);
1463
1464         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1465         zc.zc_cookie = force;
1466         zc.zc_guid = hardforce;
1467         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1468
1469         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1470                 switch (errno) {
1471                 case EXDEV:
1472                         zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1473                             "use '-f' to override the following errors:\n"
1474                             "'%s' has an active shared spare which could be"
1475                             " used by other pools once '%s' is exported."),
1476                             zhp->zpool_name, zhp->zpool_name);
1477                         return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1478                             msg));
1479                 default:
1480                         return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1481                             msg));
1482                 }
1483         }
1484
1485         return (0);
1486 }
1487
1488 int
1489 zpool_export(zpool_handle_t *zhp, boolean_t force, const char *log_str)
1490 {
1491         return (zpool_export_common(zhp, force, B_FALSE, log_str));
1492 }
1493
1494 int
1495 zpool_export_force(zpool_handle_t *zhp, const char *log_str)
1496 {
1497         return (zpool_export_common(zhp, B_TRUE, B_TRUE, log_str));
1498 }
1499
1500 static void
1501 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1502     nvlist_t *config)
1503 {
1504         nvlist_t *nv = NULL;
1505         uint64_t rewindto;
1506         int64_t loss = -1;
1507         struct tm t;
1508         char timestr[128];
1509
1510         if (!hdl->libzfs_printerr || config == NULL)
1511                 return;
1512
1513         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1514             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0) {
1515                 return;
1516         }
1517
1518         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1519                 return;
1520         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1521
1522         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1523             strftime(timestr, 128, 0, &t) != 0) {
1524                 if (dryrun) {
1525                         (void) printf(dgettext(TEXT_DOMAIN,
1526                             "Would be able to return %s "
1527                             "to its state as of %s.\n"),
1528                             name, timestr);
1529                 } else {
1530                         (void) printf(dgettext(TEXT_DOMAIN,
1531                             "Pool %s returned to its state as of %s.\n"),
1532                             name, timestr);
1533                 }
1534                 if (loss > 120) {
1535                         (void) printf(dgettext(TEXT_DOMAIN,
1536                             "%s approximately %lld "),
1537                             dryrun ? "Would discard" : "Discarded",
1538                             (loss + 30) / 60);
1539                         (void) printf(dgettext(TEXT_DOMAIN,
1540                             "minutes of transactions.\n"));
1541                 } else if (loss > 0) {
1542                         (void) printf(dgettext(TEXT_DOMAIN,
1543                             "%s approximately %lld "),
1544                             dryrun ? "Would discard" : "Discarded", loss);
1545                         (void) printf(dgettext(TEXT_DOMAIN,
1546                             "seconds of transactions.\n"));
1547                 }
1548         }
1549 }
1550
1551 void
1552 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1553     nvlist_t *config)
1554 {
1555         nvlist_t *nv = NULL;
1556         int64_t loss = -1;
1557         uint64_t edata = UINT64_MAX;
1558         uint64_t rewindto;
1559         struct tm t;
1560         char timestr[128];
1561
1562         if (!hdl->libzfs_printerr)
1563                 return;
1564
1565         if (reason >= 0)
1566                 (void) printf(dgettext(TEXT_DOMAIN, "action: "));
1567         else
1568                 (void) printf(dgettext(TEXT_DOMAIN, "\t"));
1569
1570         /* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1571         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1572             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0 ||
1573             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1574                 goto no_info;
1575
1576         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1577         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1578             &edata);
1579
1580         (void) printf(dgettext(TEXT_DOMAIN,
1581             "Recovery is possible, but will result in some data loss.\n"));
1582
1583         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1584             strftime(timestr, 128, 0, &t) != 0) {
1585                 (void) printf(dgettext(TEXT_DOMAIN,
1586                     "\tReturning the pool to its state as of %s\n"
1587                     "\tshould correct the problem.  "),
1588                     timestr);
1589         } else {
1590                 (void) printf(dgettext(TEXT_DOMAIN,
1591                     "\tReverting the pool to an earlier state "
1592                     "should correct the problem.\n\t"));
1593         }
1594
1595         if (loss > 120) {
1596                 (void) printf(dgettext(TEXT_DOMAIN,
1597                     "Approximately %lld minutes of data\n"
1598                     "\tmust be discarded, irreversibly.  "), (loss + 30) / 60);
1599         } else if (loss > 0) {
1600                 (void) printf(dgettext(TEXT_DOMAIN,
1601                     "Approximately %lld seconds of data\n"
1602                     "\tmust be discarded, irreversibly.  "), loss);
1603         }
1604         if (edata != 0 && edata != UINT64_MAX) {
1605                 if (edata == 1) {
1606                         (void) printf(dgettext(TEXT_DOMAIN,
1607                             "After rewind, at least\n"
1608                             "\tone persistent user-data error will remain.  "));
1609                 } else {
1610                         (void) printf(dgettext(TEXT_DOMAIN,
1611                             "After rewind, several\n"
1612                             "\tpersistent user-data errors will remain.  "));
1613                 }
1614         }
1615         (void) printf(dgettext(TEXT_DOMAIN,
1616             "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1617             reason >= 0 ? "clear" : "import", name);
1618
1619         (void) printf(dgettext(TEXT_DOMAIN,
1620             "A scrub of the pool\n"
1621             "\tis strongly recommended after recovery.\n"));
1622         return;
1623
1624 no_info:
1625         (void) printf(dgettext(TEXT_DOMAIN,
1626             "Destroy and re-create the pool from\n\ta backup source.\n"));
1627 }
1628
1629 /*
1630  * zpool_import() is a contracted interface. Should be kept the same
1631  * if possible.
1632  *
1633  * Applications should use zpool_import_props() to import a pool with
1634  * new properties value to be set.
1635  */
1636 int
1637 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1638     char *altroot)
1639 {
1640         nvlist_t *props = NULL;
1641         int ret;
1642
1643         if (altroot != NULL) {
1644                 if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1645                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1646                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1647                             newname));
1648                 }
1649
1650                 if (nvlist_add_string(props,
1651                     zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1652                     nvlist_add_string(props,
1653                     zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1654                         nvlist_free(props);
1655                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1656                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1657                             newname));
1658                 }
1659         }
1660
1661         ret = zpool_import_props(hdl, config, newname, props,
1662             ZFS_IMPORT_NORMAL);
1663         nvlist_free(props);
1664         return (ret);
1665 }
1666
1667 static void
1668 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1669     int indent)
1670 {
1671         nvlist_t **child;
1672         uint_t c, children;
1673         char *vname;
1674         uint64_t is_log = 0;
1675
1676         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1677             &is_log);
1678
1679         if (name != NULL)
1680                 (void) printf("\t%*s%s%s\n", indent, "", name,
1681                     is_log ? " [log]" : "");
1682
1683         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1684             &child, &children) != 0)
1685                 return;
1686
1687         for (c = 0; c < children; c++) {
1688                 vname = zpool_vdev_name(hdl, NULL, child[c], B_TRUE);
1689                 print_vdev_tree(hdl, vname, child[c], indent + 2);
1690                 free(vname);
1691         }
1692 }
1693
1694 void
1695 zpool_print_unsup_feat(nvlist_t *config)
1696 {
1697         nvlist_t *nvinfo, *unsup_feat;
1698
1699         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nvinfo) ==
1700             0);
1701         verify(nvlist_lookup_nvlist(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT,
1702             &unsup_feat) == 0);
1703
1704         for (nvpair_t *nvp = nvlist_next_nvpair(unsup_feat, NULL); nvp != NULL;
1705             nvp = nvlist_next_nvpair(unsup_feat, nvp)) {
1706                 char *desc;
1707
1708                 verify(nvpair_type(nvp) == DATA_TYPE_STRING);
1709                 verify(nvpair_value_string(nvp, &desc) == 0);
1710
1711                 if (strlen(desc) > 0)
1712                         (void) printf("\t%s (%s)\n", nvpair_name(nvp), desc);
1713                 else
1714                         (void) printf("\t%s\n", nvpair_name(nvp));
1715         }
1716 }
1717
1718 /*
1719  * Import the given pool using the known configuration and a list of
1720  * properties to be set. The configuration should have come from
1721  * zpool_find_import(). The 'newname' parameters control whether the pool
1722  * is imported with a different name.
1723  */
1724 int
1725 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1726     nvlist_t *props, int flags)
1727 {
1728         zfs_cmd_t zc = { 0 };
1729         zpool_load_policy_t policy;
1730         nvlist_t *nv = NULL;
1731         nvlist_t *nvinfo = NULL;
1732         nvlist_t *missing = NULL;
1733         char *thename;
1734         char *origname;
1735         int ret;
1736         int error = 0;
1737         char errbuf[1024];
1738
1739         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1740             &origname) == 0);
1741
1742         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1743             "cannot import pool '%s'"), origname);
1744
1745         if (newname != NULL) {
1746                 if (!zpool_name_valid(hdl, B_FALSE, newname))
1747                         return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1748                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1749                             newname));
1750                 thename = (char *)newname;
1751         } else {
1752                 thename = origname;
1753         }
1754
1755         if (props != NULL) {
1756                 uint64_t version;
1757                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1758
1759                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1760                     &version) == 0);
1761
1762                 if ((props = zpool_valid_proplist(hdl, origname,
1763                     props, version, flags, errbuf)) == NULL)
1764                         return (-1);
1765                 if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1766                         nvlist_free(props);
1767                         return (-1);
1768                 }
1769                 nvlist_free(props);
1770         }
1771
1772         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1773
1774         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1775             &zc.zc_guid) == 0);
1776
1777         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1778                 zcmd_free_nvlists(&zc);
1779                 return (-1);
1780         }
1781         if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1782                 zcmd_free_nvlists(&zc);
1783                 return (-1);
1784         }
1785
1786         zc.zc_cookie = flags;
1787         while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1788             errno == ENOMEM) {
1789                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1790                         zcmd_free_nvlists(&zc);
1791                         return (-1);
1792                 }
1793         }
1794         if (ret != 0)
1795                 error = errno;
1796
1797         (void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1798
1799         zcmd_free_nvlists(&zc);
1800
1801         zpool_get_load_policy(config, &policy);
1802
1803         if (error) {
1804                 char desc[1024];
1805
1806                 /*
1807                  * Dry-run failed, but we print out what success
1808                  * looks like if we found a best txg
1809                  */
1810                 if (policy.zlp_rewind & ZPOOL_TRY_REWIND) {
1811                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1812                             B_TRUE, nv);
1813                         nvlist_free(nv);
1814                         return (-1);
1815                 }
1816
1817                 if (newname == NULL)
1818                         (void) snprintf(desc, sizeof (desc),
1819                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1820                             thename);
1821                 else
1822                         (void) snprintf(desc, sizeof (desc),
1823                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1824                             origname, thename);
1825
1826                 switch (error) {
1827                 case ENOTSUP:
1828                         if (nv != NULL && nvlist_lookup_nvlist(nv,
1829                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1830                             nvlist_exists(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT)) {
1831                                 (void) printf(dgettext(TEXT_DOMAIN, "This "
1832                                     "pool uses the following feature(s) not "
1833                                     "supported by this system:\n"));
1834                                 zpool_print_unsup_feat(nv);
1835                                 if (nvlist_exists(nvinfo,
1836                                     ZPOOL_CONFIG_CAN_RDONLY)) {
1837                                         (void) printf(dgettext(TEXT_DOMAIN,
1838                                             "All unsupported features are only "
1839                                             "required for writing to the pool."
1840                                             "\nThe pool can be imported using "
1841                                             "'-o readonly=on'.\n"));
1842                                 }
1843                         }
1844                         /*
1845                          * Unsupported version.
1846                          */
1847                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1848                         break;
1849
1850                 case EINVAL:
1851                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1852                         break;
1853
1854                 case EROFS:
1855                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1856                             "one or more devices is read only"));
1857                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1858                         break;
1859
1860                 case ENXIO:
1861                         if (nv && nvlist_lookup_nvlist(nv,
1862                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1863                             nvlist_lookup_nvlist(nvinfo,
1864                             ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1865                                 (void) printf(dgettext(TEXT_DOMAIN,
1866                                     "The devices below are missing or "
1867                                     "corrupted, use '-m' to import the pool "
1868                                     "anyway:\n"));
1869                                 print_vdev_tree(hdl, NULL, missing, 2);
1870                                 (void) printf("\n");
1871                         }
1872                         (void) zpool_standard_error(hdl, error, desc);
1873                         break;
1874
1875                 case EEXIST:
1876                         (void) zpool_standard_error(hdl, error, desc);
1877                         break;
1878                 case ENAMETOOLONG:
1879                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1880                             "new name of at least one dataset is longer than "
1881                             "the maximum allowable length"));
1882                         (void) zfs_error(hdl, EZFS_NAMETOOLONG, desc);
1883                         break;
1884                 default:
1885                         (void) zpool_standard_error(hdl, error, desc);
1886                         zpool_explain_recover(hdl,
1887                             newname ? origname : thename, -error, nv);
1888                         break;
1889                 }
1890
1891                 nvlist_free(nv);
1892                 ret = -1;
1893         } else {
1894                 zpool_handle_t *zhp;
1895
1896                 /*
1897                  * This should never fail, but play it safe anyway.
1898                  */
1899                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1900                         ret = -1;
1901                 else if (zhp != NULL)
1902                         zpool_close(zhp);
1903                 if (policy.zlp_rewind &
1904                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1905                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1906                             ((policy.zlp_rewind & ZPOOL_TRY_REWIND) != 0), nv);
1907                 }
1908                 nvlist_free(nv);
1909                 return (0);
1910         }
1911
1912         return (ret);
1913 }
1914
1915 /*
1916  * Scan the pool.
1917  */
1918 int
1919 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func, pool_scrub_cmd_t cmd)
1920 {
1921         zfs_cmd_t zc = { 0 };
1922         char msg[1024];
1923         int err;
1924         libzfs_handle_t *hdl = zhp->zpool_hdl;
1925
1926         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1927         zc.zc_cookie = func;
1928         zc.zc_flags = cmd;
1929
1930         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0)
1931                 return (0);
1932
1933         err = errno;
1934
1935         /* ECANCELED on a scrub means we resumed a paused scrub */
1936         if (err == ECANCELED && func == POOL_SCAN_SCRUB &&
1937             cmd == POOL_SCRUB_NORMAL)
1938                 return (0);
1939
1940         if (err == ENOENT && func != POOL_SCAN_NONE && cmd == POOL_SCRUB_NORMAL)
1941                 return (0);
1942
1943         if (func == POOL_SCAN_SCRUB) {
1944                 if (cmd == POOL_SCRUB_PAUSE) {
1945                         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1946                             "cannot pause scrubbing %s"), zc.zc_name);
1947                 } else {
1948                         assert(cmd == POOL_SCRUB_NORMAL);
1949                         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1950                             "cannot scrub %s"), zc.zc_name);
1951                 }
1952         } else if (func == POOL_SCAN_NONE) {
1953                 (void) snprintf(msg, sizeof (msg),
1954                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1955                     zc.zc_name);
1956         } else {
1957                 assert(!"unexpected result");
1958         }
1959
1960         if (err == EBUSY) {
1961                 nvlist_t *nvroot;
1962                 pool_scan_stat_t *ps = NULL;
1963                 uint_t psc;
1964
1965                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
1966                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1967                 (void) nvlist_lookup_uint64_array(nvroot,
1968                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1969                 if (ps && ps->pss_func == POOL_SCAN_SCRUB) {
1970                         if (cmd == POOL_SCRUB_PAUSE)
1971                                 return (zfs_error(hdl, EZFS_SCRUB_PAUSED, msg));
1972                         else
1973                                 return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1974                 } else {
1975                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
1976                 }
1977         } else if (err == ENOENT) {
1978                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1979         } else {
1980                 return (zpool_standard_error(hdl, err, msg));
1981         }
1982 }
1983
1984 static int
1985 xlate_init_err(int err)
1986 {
1987         switch (err) {
1988         case ENODEV:
1989                 return (EZFS_NODEVICE);
1990         case EINVAL:
1991         case EROFS:
1992                 return (EZFS_BADDEV);
1993         case EBUSY:
1994                 return (EZFS_INITIALIZING);
1995         case ESRCH:
1996                 return (EZFS_NO_INITIALIZE);
1997         }
1998         return (err);
1999 }
2000
2001 /*
2002  * Begin, suspend, or cancel the initialization (initializing of all free
2003  * blocks) for the given vdevs in the given pool.
2004  */
2005 int
2006 zpool_initialize(zpool_handle_t *zhp, pool_initialize_func_t cmd_type,
2007     nvlist_t *vds)
2008 {
2009         char msg[1024];
2010         libzfs_handle_t *hdl = zhp->zpool_hdl;
2011
2012         nvlist_t *errlist;
2013
2014         /* translate vdev names to guids */
2015         nvlist_t *vdev_guids = fnvlist_alloc();
2016         nvlist_t *guids_to_paths = fnvlist_alloc();
2017         boolean_t spare, cache;
2018         nvlist_t *tgt;
2019         nvpair_t *elem;
2020
2021         for (elem = nvlist_next_nvpair(vds, NULL); elem != NULL;
2022             elem = nvlist_next_nvpair(vds, elem)) {
2023                 char *vd_path = nvpair_name(elem);
2024                 tgt = zpool_find_vdev(zhp, vd_path, &spare, &cache, NULL);
2025
2026                 if ((tgt == NULL) || cache || spare) {
2027                         (void) snprintf(msg, sizeof (msg),
2028                             dgettext(TEXT_DOMAIN, "cannot initialize '%s'"),
2029                             vd_path);
2030                         int err = (tgt == NULL) ? EZFS_NODEVICE :
2031                             (spare ? EZFS_ISSPARE : EZFS_ISL2CACHE);
2032                         fnvlist_free(vdev_guids);
2033                         fnvlist_free(guids_to_paths);
2034                         return (zfs_error(hdl, err, msg));
2035                 }
2036
2037                 uint64_t guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
2038                 fnvlist_add_uint64(vdev_guids, vd_path, guid);
2039
2040                 (void) snprintf(msg, sizeof (msg), "%llu", guid);
2041                 fnvlist_add_string(guids_to_paths, msg, vd_path);
2042         }
2043
2044         int err = lzc_initialize(zhp->zpool_name, cmd_type, vdev_guids,
2045             &errlist);
2046         fnvlist_free(vdev_guids);
2047
2048         if (err == 0) {
2049                 fnvlist_free(guids_to_paths);
2050                 return (0);
2051         }
2052
2053         nvlist_t *vd_errlist = NULL;
2054         if (errlist != NULL) {
2055                 vd_errlist = fnvlist_lookup_nvlist(errlist,
2056                     ZPOOL_INITIALIZE_VDEVS);
2057         }
2058
2059         (void) snprintf(msg, sizeof (msg),
2060             dgettext(TEXT_DOMAIN, "operation failed"));
2061
2062         for (elem = nvlist_next_nvpair(vd_errlist, NULL); elem != NULL;
2063             elem = nvlist_next_nvpair(vd_errlist, elem)) {
2064                 int64_t vd_error = xlate_init_err(fnvpair_value_int64(elem));
2065                 char *path = fnvlist_lookup_string(guids_to_paths,
2066                     nvpair_name(elem));
2067                 (void) zfs_error_fmt(hdl, vd_error, "cannot initialize '%s'",
2068                     path);
2069         }
2070
2071         fnvlist_free(guids_to_paths);
2072         if (vd_errlist != NULL)
2073                 return (-1);
2074
2075         return (zpool_standard_error(hdl, err, msg));
2076 }
2077
2078 #ifdef illumos
2079 /*
2080  * This provides a very minimal check whether a given string is likely a
2081  * c#t#d# style string.  Users of this are expected to do their own
2082  * verification of the s# part.
2083  */
2084 #define CTD_CHECK(str)  (str && str[0] == 'c' && isdigit(str[1]))
2085
2086 /*
2087  * More elaborate version for ones which may start with "/dev/dsk/"
2088  * and the like.
2089  */
2090 static int
2091 ctd_check_path(char *str)
2092 {
2093         /*
2094          * If it starts with a slash, check the last component.
2095          */
2096         if (str && str[0] == '/') {
2097                 char *tmp = strrchr(str, '/');
2098
2099                 /*
2100                  * If it ends in "/old", check the second-to-last
2101                  * component of the string instead.
2102                  */
2103                 if (tmp != str && strcmp(tmp, "/old") == 0) {
2104                         for (tmp--; *tmp != '/'; tmp--)
2105                                 ;
2106                 }
2107                 str = tmp + 1;
2108         }
2109         return (CTD_CHECK(str));
2110 }
2111 #endif
2112
2113 /*
2114  * Find a vdev that matches the search criteria specified. We use the
2115  * the nvpair name to determine how we should look for the device.
2116  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
2117  * spare; but FALSE if its an INUSE spare.
2118  */
2119 static nvlist_t *
2120 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
2121     boolean_t *l2cache, boolean_t *log)
2122 {
2123         uint_t c, children;
2124         nvlist_t **child;
2125         nvlist_t *ret;
2126         uint64_t is_log;
2127         char *srchkey;
2128         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
2129
2130         /* Nothing to look for */
2131         if (search == NULL || pair == NULL)
2132                 return (NULL);
2133
2134         /* Obtain the key we will use to search */
2135         srchkey = nvpair_name(pair);
2136
2137         switch (nvpair_type(pair)) {
2138         case DATA_TYPE_UINT64:
2139                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
2140                         uint64_t srchval, theguid;
2141
2142                         verify(nvpair_value_uint64(pair, &srchval) == 0);
2143                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
2144                             &theguid) == 0);
2145                         if (theguid == srchval)
2146                                 return (nv);
2147                 }
2148                 break;
2149
2150         case DATA_TYPE_STRING: {
2151                 char *srchval, *val;
2152
2153                 verify(nvpair_value_string(pair, &srchval) == 0);
2154                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
2155                         break;
2156
2157                 /*
2158                  * Search for the requested value. Special cases:
2159                  *
2160                  * - ZPOOL_CONFIG_PATH for whole disk entries. To support
2161                  *   UEFI boot, these end in "s0" or "s0/old" or "s1" or
2162                  *   "s1/old".   The "s0" or "s1" part is hidden from the user,
2163                  *   but included in the string, so this matches around it.
2164                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
2165                  *
2166                  * Otherwise, all other searches are simple string compares.
2167                  */
2168 #ifdef illumos
2169                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0 &&
2170                     ctd_check_path(val)) {
2171                         uint64_t wholedisk = 0;
2172
2173                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
2174                             &wholedisk);
2175                         if (wholedisk) {
2176                                 int slen = strlen(srchval);
2177                                 int vlen = strlen(val);
2178
2179                                 if (slen != vlen - 2)
2180                                         break;
2181
2182                                 /*
2183                                  * make_leaf_vdev() should only set
2184                                  * wholedisk for ZPOOL_CONFIG_PATHs which
2185                                  * will include "/dev/dsk/", giving plenty of
2186                                  * room for the indices used next.
2187                                  */
2188                                 ASSERT(vlen >= 6);
2189
2190                                 /*
2191                                  * strings identical except trailing "s0"
2192                                  */
2193                                 if ((strcmp(&val[vlen - 2], "s0") == 0 ||
2194                                     strcmp(&val[vlen - 2], "s1") == 0) &&
2195                                     strncmp(srchval, val, slen) == 0)
2196                                         return (nv);
2197
2198                                 /*
2199                                  * strings identical except trailing "s0/old"
2200                                  */
2201                                 if ((strcmp(&val[vlen - 6], "s0/old") == 0 ||
2202                                     strcmp(&val[vlen - 6], "s1/old") == 0) &&
2203                                     strcmp(&srchval[slen - 4], "/old") == 0 &&
2204                                     strncmp(srchval, val, slen - 4) == 0)
2205                                         return (nv);
2206
2207                                 break;
2208                         }
2209                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2210 #else
2211                 if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2212 #endif
2213                         char *type, *idx, *end, *p;
2214                         uint64_t id, vdev_id;
2215
2216                         /*
2217                          * Determine our vdev type, keeping in mind
2218                          * that the srchval is composed of a type and
2219                          * vdev id pair (i.e. mirror-4).
2220                          */
2221                         if ((type = strdup(srchval)) == NULL)
2222                                 return (NULL);
2223
2224                         if ((p = strrchr(type, '-')) == NULL) {
2225                                 free(type);
2226                                 break;
2227                         }
2228                         idx = p + 1;
2229                         *p = '\0';
2230
2231                         /*
2232                          * If the types don't match then keep looking.
2233                          */
2234                         if (strncmp(val, type, strlen(val)) != 0) {
2235                                 free(type);
2236                                 break;
2237                         }
2238
2239                         verify(zpool_vdev_is_interior(type));
2240                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
2241                             &id) == 0);
2242
2243                         errno = 0;
2244                         vdev_id = strtoull(idx, &end, 10);
2245
2246                         free(type);
2247                         if (errno != 0)
2248                                 return (NULL);
2249
2250                         /*
2251                          * Now verify that we have the correct vdev id.
2252                          */
2253                         if (vdev_id == id)
2254                                 return (nv);
2255                 }
2256
2257                 /*
2258                  * Common case
2259                  */
2260                 if (strcmp(srchval, val) == 0)
2261                         return (nv);
2262                 break;
2263         }
2264
2265         default:
2266                 break;
2267         }
2268
2269         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2270             &child, &children) != 0)
2271                 return (NULL);
2272
2273         for (c = 0; c < children; c++) {
2274                 if ((ret = vdev_to_nvlist_iter(child[c], search,
2275                     avail_spare, l2cache, NULL)) != NULL) {
2276                         /*
2277                          * The 'is_log' value is only set for the toplevel
2278                          * vdev, not the leaf vdevs.  So we always lookup the
2279                          * log device from the root of the vdev tree (where
2280                          * 'log' is non-NULL).
2281                          */
2282                         if (log != NULL &&
2283                             nvlist_lookup_uint64(child[c],
2284                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2285                             is_log) {
2286                                 *log = B_TRUE;
2287                         }
2288                         return (ret);
2289                 }
2290         }
2291
2292         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2293             &child, &children) == 0) {
2294                 for (c = 0; c < children; c++) {
2295                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2296                             avail_spare, l2cache, NULL)) != NULL) {
2297                                 *avail_spare = B_TRUE;
2298                                 return (ret);
2299                         }
2300                 }
2301         }
2302
2303         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2304             &child, &children) == 0) {
2305                 for (c = 0; c < children; c++) {
2306                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2307                             avail_spare, l2cache, NULL)) != NULL) {
2308                                 *l2cache = B_TRUE;
2309                                 return (ret);
2310                         }
2311                 }
2312         }
2313
2314         return (NULL);
2315 }
2316
2317 /*
2318  * Given a physical path (minus the "/devices" prefix), find the
2319  * associated vdev.
2320  */
2321 nvlist_t *
2322 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2323     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2324 {
2325         nvlist_t *search, *nvroot, *ret;
2326
2327         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2328         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
2329
2330         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2331             &nvroot) == 0);
2332
2333         *avail_spare = B_FALSE;
2334         *l2cache = B_FALSE;
2335         if (log != NULL)
2336                 *log = B_FALSE;
2337         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2338         nvlist_free(search);
2339
2340         return (ret);
2341 }
2342
2343 /*
2344  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2345  */
2346 static boolean_t
2347 zpool_vdev_is_interior(const char *name)
2348 {
2349         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2350             strncmp(name, VDEV_TYPE_SPARE, strlen(VDEV_TYPE_SPARE)) == 0 ||
2351             strncmp(name,
2352             VDEV_TYPE_REPLACING, strlen(VDEV_TYPE_REPLACING)) == 0 ||
2353             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2354                 return (B_TRUE);
2355         return (B_FALSE);
2356 }
2357
2358 nvlist_t *
2359 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2360     boolean_t *l2cache, boolean_t *log)
2361 {
2362         char buf[MAXPATHLEN];
2363         char *end;
2364         nvlist_t *nvroot, *search, *ret;
2365         uint64_t guid;
2366
2367         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2368
2369         guid = strtoull(path, &end, 10);
2370         if (guid != 0 && *end == '\0') {
2371                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
2372         } else if (zpool_vdev_is_interior(path)) {
2373                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
2374         } else if (path[0] != '/') {
2375                 (void) snprintf(buf, sizeof (buf), "%s%s", _PATH_DEV, path);
2376                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, buf) == 0);
2377         } else {
2378                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
2379         }
2380
2381         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2382             &nvroot) == 0);
2383
2384         *avail_spare = B_FALSE;
2385         *l2cache = B_FALSE;
2386         if (log != NULL)
2387                 *log = B_FALSE;
2388         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2389         nvlist_free(search);
2390
2391         return (ret);
2392 }
2393
2394 static int
2395 vdev_online(nvlist_t *nv)
2396 {
2397         uint64_t ival;
2398
2399         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2400             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2401             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2402                 return (0);
2403
2404         return (1);
2405 }
2406
2407 /*
2408  * Helper function for zpool_get_physpaths().
2409  */
2410 static int
2411 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2412     size_t *bytes_written)
2413 {
2414         size_t bytes_left, pos, rsz;
2415         char *tmppath;
2416         const char *format;
2417
2418         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2419             &tmppath) != 0)
2420                 return (EZFS_NODEVICE);
2421
2422         pos = *bytes_written;
2423         bytes_left = physpath_size - pos;
2424         format = (pos == 0) ? "%s" : " %s";
2425
2426         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2427         *bytes_written += rsz;
2428
2429         if (rsz >= bytes_left) {
2430                 /* if physpath was not copied properly, clear it */
2431                 if (bytes_left != 0) {
2432                         physpath[pos] = 0;
2433                 }
2434                 return (EZFS_NOSPC);
2435         }
2436         return (0);
2437 }
2438
2439 static int
2440 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2441     size_t *rsz, boolean_t is_spare)
2442 {
2443         char *type;
2444         int ret;
2445
2446         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2447                 return (EZFS_INVALCONFIG);
2448
2449         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2450                 /*
2451                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2452                  * For a spare vdev, we only want to boot from the active
2453                  * spare device.
2454                  */
2455                 if (is_spare) {
2456                         uint64_t spare = 0;
2457                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2458                             &spare);
2459                         if (!spare)
2460                                 return (EZFS_INVALCONFIG);
2461                 }
2462
2463                 if (vdev_online(nv)) {
2464                         if ((ret = vdev_get_one_physpath(nv, physpath,
2465                             phypath_size, rsz)) != 0)
2466                                 return (ret);
2467                 }
2468         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2469             strcmp(type, VDEV_TYPE_RAIDZ) == 0 ||
2470             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2471             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2472                 nvlist_t **child;
2473                 uint_t count;
2474                 int i, ret;
2475
2476                 if (nvlist_lookup_nvlist_array(nv,
2477                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2478                         return (EZFS_INVALCONFIG);
2479
2480                 for (i = 0; i < count; i++) {
2481                         ret = vdev_get_physpaths(child[i], physpath,
2482                             phypath_size, rsz, is_spare);
2483                         if (ret == EZFS_NOSPC)
2484                                 return (ret);
2485                 }
2486         }
2487
2488         return (EZFS_POOL_INVALARG);
2489 }
2490
2491 /*
2492  * Get phys_path for a root pool config.
2493  * Return 0 on success; non-zero on failure.
2494  */
2495 static int
2496 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2497 {
2498         size_t rsz;
2499         nvlist_t *vdev_root;
2500         nvlist_t **child;
2501         uint_t count;
2502         char *type;
2503
2504         rsz = 0;
2505
2506         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2507             &vdev_root) != 0)
2508                 return (EZFS_INVALCONFIG);
2509
2510         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2511             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2512             &child, &count) != 0)
2513                 return (EZFS_INVALCONFIG);
2514
2515         /*
2516          * root pool can only have a single top-level vdev.
2517          */
2518         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1)
2519                 return (EZFS_POOL_INVALARG);
2520
2521         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2522             B_FALSE);
2523
2524         /* No online devices */
2525         if (rsz == 0)
2526                 return (EZFS_NODEVICE);
2527
2528         return (0);
2529 }
2530
2531 /*
2532  * Get phys_path for a root pool
2533  * Return 0 on success; non-zero on failure.
2534  */
2535 int
2536 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2537 {
2538         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2539             phypath_size));
2540 }
2541
2542 /*
2543  * If the device has being dynamically expanded then we need to relabel
2544  * the disk to use the new unallocated space.
2545  */
2546 static int
2547 zpool_relabel_disk(libzfs_handle_t *hdl, const char *name)
2548 {
2549 #ifdef illumos
2550         char path[MAXPATHLEN];
2551         char errbuf[1024];
2552         int fd, error;
2553         int (*_efi_use_whole_disk)(int);
2554
2555         if ((_efi_use_whole_disk = (int (*)(int))dlsym(RTLD_DEFAULT,
2556             "efi_use_whole_disk")) == NULL)
2557                 return (-1);
2558
2559         (void) snprintf(path, sizeof (path), "%s/%s", ZFS_RDISK_ROOT, name);
2560
2561         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
2562                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2563                     "relabel '%s': unable to open device"), name);
2564                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
2565         }
2566
2567         /*
2568          * It's possible that we might encounter an error if the device
2569          * does not have any unallocated space left. If so, we simply
2570          * ignore that error and continue on.
2571          */
2572         error = _efi_use_whole_disk(fd);
2573         (void) close(fd);
2574         if (error && error != VT_ENOSPC) {
2575                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2576                     "relabel '%s': unable to read disk capacity"), name);
2577                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
2578         }
2579 #endif  /* illumos */
2580         return (0);
2581 }
2582
2583 /*
2584  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2585  * ZFS_ONLINE_* flags.
2586  */
2587 int
2588 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2589     vdev_state_t *newstate)
2590 {
2591         zfs_cmd_t zc = { 0 };
2592         char msg[1024];
2593         char *pathname;
2594         nvlist_t *tgt;
2595         boolean_t avail_spare, l2cache, islog;
2596         libzfs_handle_t *hdl = zhp->zpool_hdl;
2597
2598         if (flags & ZFS_ONLINE_EXPAND) {
2599                 (void) snprintf(msg, sizeof (msg),
2600                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2601         } else {
2602                 (void) snprintf(msg, sizeof (msg),
2603                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2604         }
2605
2606         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2607         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2608             &islog)) == NULL)
2609                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2610
2611         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2612
2613         if (avail_spare)
2614                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2615
2616         if ((flags & ZFS_ONLINE_EXPAND ||
2617             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) &&
2618             nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH, &pathname) == 0) {
2619                 uint64_t wholedisk = 0;
2620
2621                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2622                     &wholedisk);
2623
2624                 /*
2625                  * XXX - L2ARC 1.0 devices can't support expansion.
2626                  */
2627                 if (l2cache) {
2628                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2629                             "cannot expand cache devices"));
2630                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2631                 }
2632
2633                 if (wholedisk) {
2634                         pathname += strlen(ZFS_DISK_ROOT) + 1;
2635                         (void) zpool_relabel_disk(hdl, pathname);
2636                 }
2637         }
2638
2639         zc.zc_cookie = VDEV_STATE_ONLINE;
2640         zc.zc_obj = flags;
2641
2642         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2643                 if (errno == EINVAL) {
2644                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2645                             "from this pool into a new one.  Use '%s' "
2646                             "instead"), "zpool detach");
2647                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2648                 }
2649                 return (zpool_standard_error(hdl, errno, msg));
2650         }
2651
2652         *newstate = zc.zc_cookie;
2653         return (0);
2654 }
2655
2656 /*
2657  * Take the specified vdev offline
2658  */
2659 int
2660 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2661 {
2662         zfs_cmd_t zc = { 0 };
2663         char msg[1024];
2664         nvlist_t *tgt;
2665         boolean_t avail_spare, l2cache;
2666         libzfs_handle_t *hdl = zhp->zpool_hdl;
2667
2668         (void) snprintf(msg, sizeof (msg),
2669             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2670
2671         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2672         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2673             NULL)) == NULL)
2674                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2675
2676         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2677
2678         if (avail_spare)
2679                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2680
2681         zc.zc_cookie = VDEV_STATE_OFFLINE;
2682         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2683
2684         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2685                 return (0);
2686
2687         switch (errno) {
2688         case EBUSY:
2689
2690                 /*
2691                  * There are no other replicas of this device.
2692                  */
2693                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2694
2695         case EEXIST:
2696                 /*
2697                  * The log device has unplayed logs
2698                  */
2699                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2700
2701         default:
2702                 return (zpool_standard_error(hdl, errno, msg));
2703         }
2704 }
2705
2706 /*
2707  * Mark the given vdev faulted.
2708  */
2709 int
2710 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2711 {
2712         zfs_cmd_t zc = { 0 };
2713         char msg[1024];
2714         libzfs_handle_t *hdl = zhp->zpool_hdl;
2715
2716         (void) snprintf(msg, sizeof (msg),
2717             dgettext(TEXT_DOMAIN, "cannot fault %llu"), guid);
2718
2719         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2720         zc.zc_guid = guid;
2721         zc.zc_cookie = VDEV_STATE_FAULTED;
2722         zc.zc_obj = aux;
2723
2724         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2725                 return (0);
2726
2727         switch (errno) {
2728         case EBUSY:
2729
2730                 /*
2731                  * There are no other replicas of this device.
2732                  */
2733                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2734
2735         default:
2736                 return (zpool_standard_error(hdl, errno, msg));
2737         }
2738
2739 }
2740
2741 /*
2742  * Mark the given vdev degraded.
2743  */
2744 int
2745 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2746 {
2747         zfs_cmd_t zc = { 0 };
2748         char msg[1024];
2749         libzfs_handle_t *hdl = zhp->zpool_hdl;
2750
2751         (void) snprintf(msg, sizeof (msg),
2752             dgettext(TEXT_DOMAIN, "cannot degrade %llu"), guid);
2753
2754         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2755         zc.zc_guid = guid;
2756         zc.zc_cookie = VDEV_STATE_DEGRADED;
2757         zc.zc_obj = aux;
2758
2759         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2760                 return (0);
2761
2762         return (zpool_standard_error(hdl, errno, msg));
2763 }
2764
2765 /*
2766  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2767  * a hot spare.
2768  */
2769 static boolean_t
2770 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2771 {
2772         nvlist_t **child;
2773         uint_t c, children;
2774         char *type;
2775
2776         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2777             &children) == 0) {
2778                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2779                     &type) == 0);
2780
2781                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2782                     children == 2 && child[which] == tgt)
2783                         return (B_TRUE);
2784
2785                 for (c = 0; c < children; c++)
2786                         if (is_replacing_spare(child[c], tgt, which))
2787                                 return (B_TRUE);
2788         }
2789
2790         return (B_FALSE);
2791 }
2792
2793 /*
2794  * Attach new_disk (fully described by nvroot) to old_disk.
2795  * If 'replacing' is specified, the new disk will replace the old one.
2796  */
2797 int
2798 zpool_vdev_attach(zpool_handle_t *zhp,
2799     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2800 {
2801         zfs_cmd_t zc = { 0 };
2802         char msg[1024];
2803         int ret;
2804         nvlist_t *tgt;
2805         boolean_t avail_spare, l2cache, islog;
2806         uint64_t val;
2807         char *newname;
2808         nvlist_t **child;
2809         uint_t children;
2810         nvlist_t *config_root;
2811         libzfs_handle_t *hdl = zhp->zpool_hdl;
2812         boolean_t rootpool = zpool_is_bootable(zhp);
2813
2814         if (replacing)
2815                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2816                     "cannot replace %s with %s"), old_disk, new_disk);
2817         else
2818                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2819                     "cannot attach %s to %s"), new_disk, old_disk);
2820
2821         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2822         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2823             &islog)) == NULL)
2824                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2825
2826         if (avail_spare)
2827                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2828
2829         if (l2cache)
2830                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2831
2832         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2833         zc.zc_cookie = replacing;
2834
2835         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2836             &child, &children) != 0 || children != 1) {
2837                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2838                     "new device must be a single disk"));
2839                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2840         }
2841
2842         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2843             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2844
2845         if ((newname = zpool_vdev_name(NULL, NULL, child[0], B_FALSE)) == NULL)
2846                 return (-1);
2847
2848         /*
2849          * If the target is a hot spare that has been swapped in, we can only
2850          * replace it with another hot spare.
2851          */
2852         if (replacing &&
2853             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2854             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2855             NULL) == NULL || !avail_spare) &&
2856             is_replacing_spare(config_root, tgt, 1)) {
2857                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2858                     "can only be replaced by another hot spare"));
2859                 free(newname);
2860                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2861         }
2862
2863         free(newname);
2864
2865         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2866                 return (-1);
2867
2868         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2869
2870         zcmd_free_nvlists(&zc);
2871
2872         if (ret == 0) {
2873                 if (rootpool) {
2874                         /*
2875                          * XXX need a better way to prevent user from
2876                          * booting up a half-baked vdev.
2877                          */
2878                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2879                             "sure to wait until resilver is done "
2880                             "before rebooting.\n"));
2881                         (void) fprintf(stderr, "\n");
2882                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "If "
2883                             "you boot from pool '%s', you may need to update\n"
2884                             "boot code on newly attached disk '%s'.\n\n"
2885                             "Assuming you use GPT partitioning and 'da0' is "
2886                             "your new boot disk\n"
2887                             "you may use the following command:\n\n"
2888                             "\tgpart bootcode -b /boot/pmbr -p "
2889                             "/boot/gptzfsboot -i 1 da0\n\n"),
2890                             zhp->zpool_name, new_disk);
2891                 }
2892                 return (0);
2893         }
2894
2895         switch (errno) {
2896         case ENOTSUP:
2897                 /*
2898                  * Can't attach to or replace this type of vdev.
2899                  */
2900                 if (replacing) {
2901                         uint64_t version = zpool_get_prop_int(zhp,
2902                             ZPOOL_PROP_VERSION, NULL);
2903
2904                         if (islog)
2905                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2906                                     "cannot replace a log with a spare"));
2907                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2908                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2909                                     "already in replacing/spare config; wait "
2910                                     "for completion or use 'zpool detach'"));
2911                         else
2912                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2913                                     "cannot replace a replacing device"));
2914                 } else {
2915                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2916                             "can only attach to mirrors and top-level "
2917                             "disks"));
2918                 }
2919                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2920                 break;
2921
2922         case EINVAL:
2923                 /*
2924                  * The new device must be a single disk.
2925                  */
2926                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2927                     "new device must be a single disk"));
2928                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2929                 break;
2930
2931         case EBUSY:
2932                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy, "
2933                     "or device removal is in progress"),
2934                     new_disk);
2935                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2936                 break;
2937
2938         case EOVERFLOW:
2939                 /*
2940                  * The new device is too small.
2941                  */
2942                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2943                     "device is too small"));
2944                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2945                 break;
2946
2947         case EDOM:
2948                 /*
2949                  * The new device has a different alignment requirement.
2950                  */
2951                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2952                     "devices have different sector alignment"));
2953                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2954                 break;
2955
2956         case ENAMETOOLONG:
2957                 /*
2958                  * The resulting top-level vdev spec won't fit in the label.
2959                  */
2960                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2961                 break;
2962
2963         default:
2964                 (void) zpool_standard_error(hdl, errno, msg);
2965         }
2966
2967         return (-1);
2968 }
2969
2970 /*
2971  * Detach the specified device.
2972  */
2973 int
2974 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2975 {
2976         zfs_cmd_t zc = { 0 };
2977         char msg[1024];
2978         nvlist_t *tgt;
2979         boolean_t avail_spare, l2cache;
2980         libzfs_handle_t *hdl = zhp->zpool_hdl;
2981
2982         (void) snprintf(msg, sizeof (msg),
2983             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2984
2985         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2986         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2987             NULL)) == NULL)
2988                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2989
2990         if (avail_spare)
2991                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2992
2993         if (l2cache)
2994                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2995
2996         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2997
2998         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2999                 return (0);
3000
3001         switch (errno) {
3002
3003         case ENOTSUP:
3004                 /*
3005                  * Can't detach from this type of vdev.
3006                  */
3007                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
3008                     "applicable to mirror and replacing vdevs"));
3009                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
3010                 break;
3011
3012         case EBUSY:
3013                 /*
3014                  * There are no other replicas of this device.
3015                  */
3016                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
3017                 break;
3018
3019         default:
3020                 (void) zpool_standard_error(hdl, errno, msg);
3021         }
3022
3023         return (-1);
3024 }
3025
3026 /*
3027  * Find a mirror vdev in the source nvlist.
3028  *
3029  * The mchild array contains a list of disks in one of the top-level mirrors
3030  * of the source pool.  The schild array contains a list of disks that the
3031  * user specified on the command line.  We loop over the mchild array to
3032  * see if any entry in the schild array matches.
3033  *
3034  * If a disk in the mchild array is found in the schild array, we return
3035  * the index of that entry.  Otherwise we return -1.
3036  */
3037 static int
3038 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
3039     nvlist_t **schild, uint_t schildren)
3040 {
3041         uint_t mc;
3042
3043         for (mc = 0; mc < mchildren; mc++) {
3044                 uint_t sc;
3045                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
3046                     mchild[mc], B_FALSE);
3047
3048                 for (sc = 0; sc < schildren; sc++) {
3049                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
3050                             schild[sc], B_FALSE);
3051                         boolean_t result = (strcmp(mpath, spath) == 0);
3052
3053                         free(spath);
3054                         if (result) {
3055                                 free(mpath);
3056                                 return (mc);
3057                         }
3058                 }
3059
3060                 free(mpath);
3061         }
3062
3063         return (-1);
3064 }
3065
3066 /*
3067  * Split a mirror pool.  If newroot points to null, then a new nvlist
3068  * is generated and it is the responsibility of the caller to free it.
3069  */
3070 int
3071 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
3072     nvlist_t *props, splitflags_t flags)
3073 {
3074         zfs_cmd_t zc = { 0 };
3075         char msg[1024];
3076         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
3077         nvlist_t **varray = NULL, *zc_props = NULL;
3078         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
3079         libzfs_handle_t *hdl = zhp->zpool_hdl;
3080         uint64_t vers;
3081         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
3082         int retval = 0;
3083
3084         (void) snprintf(msg, sizeof (msg),
3085             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
3086
3087         if (!zpool_name_valid(hdl, B_FALSE, newname))
3088                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
3089
3090         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
3091                 (void) fprintf(stderr, gettext("Internal error: unable to "
3092                     "retrieve pool configuration\n"));
3093                 return (-1);
3094         }
3095
3096         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
3097             == 0);
3098         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
3099
3100         if (props) {
3101                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
3102                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
3103                     props, vers, flags, msg)) == NULL)
3104                         return (-1);
3105         }
3106
3107         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
3108             &children) != 0) {
3109                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3110                     "Source pool is missing vdev tree"));
3111                 nvlist_free(zc_props);
3112                 return (-1);
3113         }
3114
3115         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
3116         vcount = 0;
3117
3118         if (*newroot == NULL ||
3119             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
3120             &newchild, &newchildren) != 0)
3121                 newchildren = 0;
3122
3123         for (c = 0; c < children; c++) {
3124                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
3125                 char *type;
3126                 nvlist_t **mchild, *vdev;
3127                 uint_t mchildren;
3128                 int entry;
3129
3130                 /*
3131                  * Unlike cache & spares, slogs are stored in the
3132                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
3133                  */
3134                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
3135                     &is_log);
3136                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
3137                     &is_hole);
3138                 if (is_log || is_hole) {
3139                         /*
3140                          * Create a hole vdev and put it in the config.
3141                          */
3142                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
3143                                 goto out;
3144                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
3145                             VDEV_TYPE_HOLE) != 0)
3146                                 goto out;
3147                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
3148                             1) != 0)
3149                                 goto out;
3150                         if (lastlog == 0)
3151                                 lastlog = vcount;
3152                         varray[vcount++] = vdev;
3153                         continue;
3154                 }
3155                 lastlog = 0;
3156                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
3157                     == 0);
3158                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
3159                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3160                             "Source pool must be composed only of mirrors\n"));
3161                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3162                         goto out;
3163                 }
3164
3165                 verify(nvlist_lookup_nvlist_array(child[c],
3166                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
3167
3168                 /* find or add an entry for this top-level vdev */
3169                 if (newchildren > 0 &&
3170                     (entry = find_vdev_entry(zhp, mchild, mchildren,
3171                     newchild, newchildren)) >= 0) {
3172                         /* We found a disk that the user specified. */
3173                         vdev = mchild[entry];
3174                         ++found;
3175                 } else {
3176                         /* User didn't specify a disk for this vdev. */
3177                         vdev = mchild[mchildren - 1];
3178                 }
3179
3180                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
3181                         goto out;
3182         }
3183
3184         /* did we find every disk the user specified? */
3185         if (found != newchildren) {
3186                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
3187                     "include at most one disk from each mirror"));
3188                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3189                 goto out;
3190         }
3191
3192         /* Prepare the nvlist for populating. */
3193         if (*newroot == NULL) {
3194                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
3195                         goto out;
3196                 freelist = B_TRUE;
3197                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
3198                     VDEV_TYPE_ROOT) != 0)
3199                         goto out;
3200         } else {
3201                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
3202         }
3203
3204         /* Add all the children we found */
3205         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
3206             lastlog == 0 ? vcount : lastlog) != 0)
3207                 goto out;
3208
3209         /*
3210          * If we're just doing a dry run, exit now with success.
3211          */
3212         if (flags.dryrun) {
3213                 memory_err = B_FALSE;
3214                 freelist = B_FALSE;
3215                 goto out;
3216         }
3217
3218         /* now build up the config list & call the ioctl */
3219         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3220                 goto out;
3221
3222         if (nvlist_add_nvlist(newconfig,
3223             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3224             nvlist_add_string(newconfig,
3225             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3226             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3227                 goto out;
3228
3229         /*
3230          * The new pool is automatically part of the namespace unless we
3231          * explicitly export it.
3232          */
3233         if (!flags.import)
3234                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3235         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3236         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3237         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
3238                 goto out;
3239         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
3240                 goto out;
3241
3242         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3243                 retval = zpool_standard_error(hdl, errno, msg);
3244                 goto out;
3245         }
3246
3247         freelist = B_FALSE;
3248         memory_err = B_FALSE;
3249
3250 out:
3251         if (varray != NULL) {
3252                 int v;
3253
3254                 for (v = 0; v < vcount; v++)
3255                         nvlist_free(varray[v]);
3256                 free(varray);
3257         }
3258         zcmd_free_nvlists(&zc);
3259         nvlist_free(zc_props);
3260         nvlist_free(newconfig);
3261         if (freelist) {
3262                 nvlist_free(*newroot);
3263                 *newroot = NULL;
3264         }
3265
3266         if (retval != 0)
3267                 return (retval);
3268
3269         if (memory_err)
3270                 return (no_memory(hdl));
3271
3272         return (0);
3273 }
3274
3275 /*
3276  * Remove the given device.
3277  */
3278 int
3279 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3280 {
3281         zfs_cmd_t zc = { 0 };
3282         char msg[1024];
3283         nvlist_t *tgt;
3284         boolean_t avail_spare, l2cache, islog;
3285         libzfs_handle_t *hdl = zhp->zpool_hdl;
3286         uint64_t version;
3287
3288         (void) snprintf(msg, sizeof (msg),
3289             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3290
3291         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3292         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3293             &islog)) == NULL)
3294                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3295
3296         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3297         if (islog && version < SPA_VERSION_HOLES) {
3298                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3299                     "pool must be upgraded to support log removal"));
3300                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
3301         }
3302
3303         if (!islog && !avail_spare && !l2cache && zpool_is_bootable(zhp)) {
3304                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3305                     "root pool can not have removed devices, "
3306                     "because GRUB does not understand them"));
3307                 return (zfs_error(hdl, EINVAL, msg));
3308         }
3309
3310         zc.zc_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
3311
3312         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3313                 return (0);
3314
3315         switch (errno) {
3316
3317         case EINVAL:
3318                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3319                     "invalid config; all top-level vdevs must "
3320                     "have the same sector size and not be raidz."));
3321                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
3322                 break;
3323
3324         case EBUSY:
3325                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3326                     "Pool busy; removal may already be in progress"));
3327                 (void) zfs_error(hdl, EZFS_BUSY, msg);
3328                 break;
3329
3330         default:
3331                 (void) zpool_standard_error(hdl, errno, msg);
3332         }
3333         return (-1);
3334 }
3335
3336 int
3337 zpool_vdev_remove_cancel(zpool_handle_t *zhp)
3338 {
3339         zfs_cmd_t zc = { 0 };
3340         char msg[1024];
3341         libzfs_handle_t *hdl = zhp->zpool_hdl;
3342
3343         (void) snprintf(msg, sizeof (msg),
3344             dgettext(TEXT_DOMAIN, "cannot cancel removal"));
3345
3346         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3347         zc.zc_cookie = 1;
3348
3349         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3350                 return (0);
3351
3352         return (zpool_standard_error(hdl, errno, msg));
3353 }
3354
3355 int
3356 zpool_vdev_indirect_size(zpool_handle_t *zhp, const char *path,
3357     uint64_t *sizep)
3358 {
3359         char msg[1024];
3360         nvlist_t *tgt;
3361         boolean_t avail_spare, l2cache, islog;
3362         libzfs_handle_t *hdl = zhp->zpool_hdl;
3363
3364         (void) snprintf(msg, sizeof (msg),
3365             dgettext(TEXT_DOMAIN, "cannot determine indirect size of %s"),
3366             path);
3367
3368         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3369             &islog)) == NULL)
3370                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3371
3372         if (avail_spare || l2cache || islog) {
3373                 *sizep = 0;
3374                 return (0);
3375         }
3376
3377         if (nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_INDIRECT_SIZE, sizep) != 0) {
3378                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3379                     "indirect size not available"));
3380                 return (zfs_error(hdl, EINVAL, msg));
3381         }
3382         return (0);
3383 }
3384
3385 /*
3386  * Clear the errors for the pool, or the particular device if specified.
3387  */
3388 int
3389 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3390 {
3391         zfs_cmd_t zc = { 0 };
3392         char msg[1024];
3393         nvlist_t *tgt;
3394         zpool_load_policy_t policy;
3395         boolean_t avail_spare, l2cache;
3396         libzfs_handle_t *hdl = zhp->zpool_hdl;
3397         nvlist_t *nvi = NULL;
3398         int error;
3399
3400         if (path)
3401                 (void) snprintf(msg, sizeof (msg),
3402                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3403                     path);
3404         else
3405                 (void) snprintf(msg, sizeof (msg),
3406                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3407                     zhp->zpool_name);
3408
3409         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3410         if (path) {
3411                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3412                     &l2cache, NULL)) == NULL)
3413                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
3414
3415                 /*
3416                  * Don't allow error clearing for hot spares.  Do allow
3417                  * error clearing for l2cache devices.
3418                  */
3419                 if (avail_spare)
3420                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
3421
3422                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
3423                     &zc.zc_guid) == 0);
3424         }
3425
3426         zpool_get_load_policy(rewindnvl, &policy);
3427         zc.zc_cookie = policy.zlp_rewind;
3428
3429         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
3430                 return (-1);
3431
3432         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
3433                 return (-1);
3434
3435         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3436             errno == ENOMEM) {
3437                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3438                         zcmd_free_nvlists(&zc);
3439                         return (-1);
3440                 }
3441         }
3442
3443         if (!error || ((policy.zlp_rewind & ZPOOL_TRY_REWIND) &&
3444             errno != EPERM && errno != EACCES)) {
3445                 if (policy.zlp_rewind &
3446                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
3447                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3448                         zpool_rewind_exclaim(hdl, zc.zc_name,
3449                             ((policy.zlp_rewind & ZPOOL_TRY_REWIND) != 0),
3450                             nvi);
3451                         nvlist_free(nvi);
3452                 }
3453                 zcmd_free_nvlists(&zc);
3454                 return (0);
3455         }
3456
3457         zcmd_free_nvlists(&zc);
3458         return (zpool_standard_error(hdl, errno, msg));
3459 }
3460
3461 /*
3462  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3463  */
3464 int
3465 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3466 {
3467         zfs_cmd_t zc = { 0 };
3468         char msg[1024];
3469         libzfs_handle_t *hdl = zhp->zpool_hdl;
3470
3471         (void) snprintf(msg, sizeof (msg),
3472             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3473             guid);
3474
3475         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3476         zc.zc_guid = guid;
3477         zc.zc_cookie = ZPOOL_NO_REWIND;
3478
3479         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
3480                 return (0);
3481
3482         return (zpool_standard_error(hdl, errno, msg));
3483 }
3484
3485 /*
3486  * Change the GUID for a pool.
3487  */
3488 int
3489 zpool_reguid(zpool_handle_t *zhp)
3490 {
3491         char msg[1024];
3492         libzfs_handle_t *hdl = zhp->zpool_hdl;
3493         zfs_cmd_t zc = { 0 };
3494
3495         (void) snprintf(msg, sizeof (msg),
3496             dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3497
3498         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3499         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3500                 return (0);
3501
3502         return (zpool_standard_error(hdl, errno, msg));
3503 }
3504
3505 /*
3506  * Reopen the pool.
3507  */
3508 int
3509 zpool_reopen(zpool_handle_t *zhp)
3510 {
3511         zfs_cmd_t zc = { 0 };
3512         char msg[1024];
3513         libzfs_handle_t *hdl = zhp->zpool_hdl;
3514
3515         (void) snprintf(msg, sizeof (msg),
3516             dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3517             zhp->zpool_name);
3518
3519         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3520         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3521                 return (0);
3522         return (zpool_standard_error(hdl, errno, msg));
3523 }
3524
3525 /*
3526  * Convert from a devid string to a path.
3527  */
3528 static char *
3529 devid_to_path(char *devid_str)
3530 {
3531         ddi_devid_t devid;
3532         char *minor;
3533         char *path;
3534         devid_nmlist_t *list = NULL;
3535         int ret;
3536
3537         if (devid_str_decode(devid_str, &devid, &minor) != 0)
3538                 return (NULL);
3539
3540         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3541
3542         devid_str_free(minor);
3543         devid_free(devid);
3544
3545         if (ret != 0)
3546                 return (NULL);
3547
3548         /*
3549          * In a case the strdup() fails, we will just return NULL below.
3550          */
3551         path = strdup(list[0].devname);
3552
3553         devid_free_nmlist(list);
3554
3555         return (path);
3556 }
3557
3558 /*
3559  * Convert from a path to a devid string.
3560  */
3561 static char *
3562 path_to_devid(const char *path)
3563 {
3564 #ifdef have_devid
3565         int fd;
3566         ddi_devid_t devid;
3567         char *minor, *ret;
3568
3569         if ((fd = open(path, O_RDONLY)) < 0)
3570                 return (NULL);
3571
3572         minor = NULL;
3573         ret = NULL;
3574         if (devid_get(fd, &devid) == 0) {
3575                 if (devid_get_minor_name(fd, &minor) == 0)
3576                         ret = devid_str_encode(devid, minor);
3577                 if (minor != NULL)
3578                         devid_str_free(minor);
3579                 devid_free(devid);
3580         }
3581         (void) close(fd);
3582
3583         return (ret);
3584 #else
3585         return (NULL);
3586 #endif
3587 }
3588
3589 /*
3590  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3591  * ignore any failure here, since a common case is for an unprivileged user to
3592  * type 'zpool status', and we'll display the correct information anyway.
3593  */
3594 static void
3595 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3596 {
3597         zfs_cmd_t zc = { 0 };
3598
3599         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3600         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3601         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3602             &zc.zc_guid) == 0);
3603
3604         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3605 }
3606
3607 /*
3608  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3609  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3610  * We also check if this is a whole disk, in which case we strip off the
3611  * trailing 's0' slice name.
3612  *
3613  * This routine is also responsible for identifying when disks have been
3614  * reconfigured in a new location.  The kernel will have opened the device by
3615  * devid, but the path will still refer to the old location.  To catch this, we
3616  * first do a path -> devid translation (which is fast for the common case).  If
3617  * the devid matches, we're done.  If not, we do a reverse devid -> path
3618  * translation and issue the appropriate ioctl() to update the path of the vdev.
3619  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3620  * of these checks.
3621  */
3622 char *
3623 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3624     boolean_t verbose)
3625 {
3626         char *path, *devid;
3627         uint64_t value;
3628         char buf[64];
3629         vdev_stat_t *vs;
3630         uint_t vsc;
3631         int have_stats;
3632         int have_path;
3633
3634         have_stats = nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3635             (uint64_t **)&vs, &vsc) == 0;
3636         have_path = nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0;
3637
3638         /*
3639          * If the device is not currently present, assume it will not
3640          * come back at the same device path.  Display the device by GUID.
3641          */
3642         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &value) == 0 ||
3643             have_path && have_stats && vs->vs_state <= VDEV_STATE_CANT_OPEN) {
3644                 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3645                     &value) == 0);
3646                 (void) snprintf(buf, sizeof (buf), "%llu",
3647                     (u_longlong_t)value);
3648                 path = buf;
3649         } else if (have_path) {
3650
3651                 /*
3652                  * If the device is dead (faulted, offline, etc) then don't
3653                  * bother opening it.  Otherwise we may be forcing the user to
3654                  * open a misbehaving device, which can have undesirable
3655                  * effects.
3656                  */
3657                 if ((have_stats == 0 ||
3658                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3659                     zhp != NULL &&
3660                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3661                         /*
3662                          * Determine if the current path is correct.
3663                          */
3664                         char *newdevid = path_to_devid(path);
3665
3666                         if (newdevid == NULL ||
3667                             strcmp(devid, newdevid) != 0) {
3668                                 char *newpath;
3669
3670                                 if ((newpath = devid_to_path(devid)) != NULL) {
3671                                         /*
3672                                          * Update the path appropriately.
3673                                          */
3674                                         set_path(zhp, nv, newpath);
3675                                         if (nvlist_add_string(nv,
3676                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3677                                                 verify(nvlist_lookup_string(nv,
3678                                                     ZPOOL_CONFIG_PATH,
3679                                                     &path) == 0);
3680                                         free(newpath);
3681                                 }
3682                         }
3683
3684                         if (newdevid)
3685                                 devid_str_free(newdevid);
3686                 }
3687
3688 #ifdef illumos
3689                 if (strncmp(path, ZFS_DISK_ROOTD, strlen(ZFS_DISK_ROOTD)) == 0)
3690                         path += strlen(ZFS_DISK_ROOTD);
3691
3692                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
3693                     &value) == 0 && value) {
3694                         int pathlen = strlen(path);
3695                         char *tmp = zfs_strdup(hdl, path);
3696
3697                         /*
3698                          * If it starts with c#, and ends with "s0" or "s1",
3699                          * chop the slice off, or if it ends with "s0/old" or
3700                          * "s1/old", remove the slice from the middle.
3701                          */
3702                         if (CTD_CHECK(tmp)) {
3703                                 if (strcmp(&tmp[pathlen - 2], "s0") == 0 ||
3704                                     strcmp(&tmp[pathlen - 2], "s1") == 0) {
3705                                         tmp[pathlen - 2] = '\0';
3706                                 } else if (pathlen > 6 &&
3707                                     (strcmp(&tmp[pathlen - 6], "s0/old") == 0 ||
3708                                     strcmp(&tmp[pathlen - 6], "s1/old") == 0)) {
3709                                         (void) strcpy(&tmp[pathlen - 6],
3710                                             "/old");
3711                                 }
3712                         }
3713                         return (tmp);
3714                 }
3715 #else   /* !illumos */
3716                 if (strncmp(path, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
3717                         path += sizeof(_PATH_DEV) - 1;
3718 #endif  /* illumos */
3719         } else {
3720                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3721
3722                 /*
3723                  * If it's a raidz device, we need to stick in the parity level.
3724                  */
3725                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3726                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3727                             &value) == 0);
3728                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3729                             (u_longlong_t)value);
3730                         path = buf;
3731                 }
3732
3733                 /*
3734                  * We identify each top-level vdev by using a <type-id>
3735                  * naming convention.
3736                  */
3737                 if (verbose) {
3738                         uint64_t id;
3739
3740                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3741                             &id) == 0);
3742                         (void) snprintf(buf, sizeof (buf), "%s-%llu", path,
3743                             (u_longlong_t)id);
3744                         path = buf;
3745                 }
3746         }
3747
3748         return (zfs_strdup(hdl, path));
3749 }
3750
3751 static int
3752 zbookmark_mem_compare(const void *a, const void *b)
3753 {
3754         return (memcmp(a, b, sizeof (zbookmark_phys_t)));
3755 }
3756
3757 /*
3758  * Retrieve the persistent error log, uniquify the members, and return to the
3759  * caller.
3760  */
3761 int
3762 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3763 {
3764         zfs_cmd_t zc = { 0 };
3765         uint64_t count;
3766         zbookmark_phys_t *zb = NULL;
3767         int i;
3768
3769         /*
3770          * Retrieve the raw error list from the kernel.  If the number of errors
3771          * has increased, allocate more space and continue until we get the
3772          * entire list.
3773          */
3774         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3775             &count) == 0);
3776         if (count == 0)
3777                 return (0);
3778         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3779             count * sizeof (zbookmark_phys_t))) == (uintptr_t)NULL)
3780                 return (-1);
3781         zc.zc_nvlist_dst_size = count;
3782         (void) strcpy(zc.zc_name, zhp->zpool_name);
3783         for (;;) {
3784                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3785                     &zc) != 0) {
3786                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3787                         if (errno == ENOMEM) {
3788                                 void *dst;
3789
3790                                 count = zc.zc_nvlist_dst_size;
3791                                 dst = zfs_alloc(zhp->zpool_hdl, count *
3792                                     sizeof (zbookmark_phys_t));
3793                                 if (dst == NULL)
3794                                         return (-1);
3795                                 zc.zc_nvlist_dst = (uintptr_t)dst;
3796                         } else {
3797                                 return (-1);
3798                         }
3799                 } else {
3800                         break;
3801                 }
3802         }
3803
3804         /*
3805          * Sort the resulting bookmarks.  This is a little confusing due to the
3806          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3807          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3808          * _not_ copied as part of the process.  So we point the start of our
3809          * array appropriate and decrement the total number of elements.
3810          */
3811         zb = ((zbookmark_phys_t *)(uintptr_t)zc.zc_nvlist_dst) +
3812             zc.zc_nvlist_dst_size;
3813         count -= zc.zc_nvlist_dst_size;
3814
3815         qsort(zb, count, sizeof (zbookmark_phys_t), zbookmark_mem_compare);
3816
3817         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3818
3819         /*
3820          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3821          */
3822         for (i = 0; i < count; i++) {
3823                 nvlist_t *nv;
3824
3825                 /* ignoring zb_blkid and zb_level for now */
3826                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3827                     zb[i-1].zb_object == zb[i].zb_object)
3828                         continue;
3829
3830                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3831                         goto nomem;
3832                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3833                     zb[i].zb_objset) != 0) {
3834                         nvlist_free(nv);
3835                         goto nomem;
3836                 }
3837                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3838                     zb[i].zb_object) != 0) {
3839                         nvlist_free(nv);
3840                         goto nomem;
3841                 }
3842                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3843                         nvlist_free(nv);
3844                         goto nomem;
3845                 }
3846                 nvlist_free(nv);
3847         }
3848
3849         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3850         return (0);
3851
3852 nomem:
3853         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3854         return (no_memory(zhp->zpool_hdl));
3855 }
3856
3857 /*
3858  * Upgrade a ZFS pool to the latest on-disk version.
3859  */
3860 int
3861 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3862 {
3863         zfs_cmd_t zc = { 0 };
3864         libzfs_handle_t *hdl = zhp->zpool_hdl;
3865
3866         (void) strcpy(zc.zc_name, zhp->zpool_name);
3867         zc.zc_cookie = new_version;
3868
3869         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3870                 return (zpool_standard_error_fmt(hdl, errno,
3871                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3872                     zhp->zpool_name));
3873         return (0);
3874 }
3875
3876 void
3877 zfs_save_arguments(int argc, char **argv, char *string, int len)
3878 {
3879         (void) strlcpy(string, basename(argv[0]), len);
3880         for (int i = 1; i < argc; i++) {
3881                 (void) strlcat(string, " ", len);
3882                 (void) strlcat(string, argv[i], len);
3883         }
3884 }
3885
3886 int
3887 zpool_log_history(libzfs_handle_t *hdl, const char *message)
3888 {
3889         zfs_cmd_t zc = { 0 };
3890         nvlist_t *args;
3891         int err;
3892
3893         args = fnvlist_alloc();
3894         fnvlist_add_string(args, "message", message);
3895         err = zcmd_write_src_nvlist(hdl, &zc, args);
3896         if (err == 0)
3897                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_LOG_HISTORY, &zc);
3898         nvlist_free(args);
3899         zcmd_free_nvlists(&zc);
3900         return (err);
3901 }
3902
3903 /*
3904  * Perform ioctl to get some command history of a pool.
3905  *
3906  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3907  * logical offset of the history buffer to start reading from.
3908  *
3909  * Upon return, 'off' is the next logical offset to read from and
3910  * 'len' is the actual amount of bytes read into 'buf'.
3911  */
3912 static int
3913 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3914 {
3915         zfs_cmd_t zc = { 0 };
3916         libzfs_handle_t *hdl = zhp->zpool_hdl;
3917
3918         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3919
3920         zc.zc_history = (uint64_t)(uintptr_t)buf;
3921         zc.zc_history_len = *len;
3922         zc.zc_history_offset = *off;
3923
3924         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3925                 switch (errno) {
3926                 case EPERM:
3927                         return (zfs_error_fmt(hdl, EZFS_PERM,
3928                             dgettext(TEXT_DOMAIN,
3929                             "cannot show history for pool '%s'"),
3930                             zhp->zpool_name));
3931                 case ENOENT:
3932                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3933                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3934                             "'%s'"), zhp->zpool_name));
3935                 case ENOTSUP:
3936                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3937                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3938                             "'%s', pool must be upgraded"), zhp->zpool_name));
3939                 default:
3940                         return (zpool_standard_error_fmt(hdl, errno,
3941                             dgettext(TEXT_DOMAIN,
3942                             "cannot get history for '%s'"), zhp->zpool_name));
3943                 }
3944         }
3945
3946         *len = zc.zc_history_len;
3947         *off = zc.zc_history_offset;
3948
3949         return (0);
3950 }
3951
3952 /*
3953  * Process the buffer of nvlists, unpacking and storing each nvlist record
3954  * into 'records'.  'leftover' is set to the number of bytes that weren't
3955  * processed as there wasn't a complete record.
3956  */
3957 int
3958 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3959     nvlist_t ***records, uint_t *numrecords)
3960 {
3961         uint64_t reclen;
3962         nvlist_t *nv;
3963         int i;
3964
3965         while (bytes_read > sizeof (reclen)) {
3966
3967                 /* get length of packed record (stored as little endian) */
3968                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3969                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3970
3971                 if (bytes_read < sizeof (reclen) + reclen)
3972                         break;
3973
3974                 /* unpack record */
3975                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3976                         return (ENOMEM);
3977                 bytes_read -= sizeof (reclen) + reclen;
3978                 buf += sizeof (reclen) + reclen;
3979
3980                 /* add record to nvlist array */
3981                 (*numrecords)++;
3982                 if (ISP2(*numrecords + 1)) {
3983                         *records = realloc(*records,
3984                             *numrecords * 2 * sizeof (nvlist_t *));
3985                 }
3986                 (*records)[*numrecords - 1] = nv;
3987         }
3988
3989         *leftover = bytes_read;
3990         return (0);
3991 }
3992
3993 /* from spa_history.c: spa_history_create_obj() */
3994 #define HIS_BUF_LEN_DEF (128 << 10)
3995 #define HIS_BUF_LEN_MAX (1 << 30)
3996
3997 /*
3998  * Retrieve the command history of a pool.
3999  */
4000 int
4001 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
4002 {
4003         char *buf;
4004         uint64_t buflen = HIS_BUF_LEN_DEF;
4005         uint64_t off = 0;
4006         nvlist_t **records = NULL;
4007         uint_t numrecords = 0;
4008         int err, i;
4009
4010         buf = malloc(buflen);
4011         if (buf == NULL)
4012                 return (ENOMEM);
4013         do {
4014                 uint64_t bytes_read = buflen;
4015                 uint64_t leftover;
4016
4017                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
4018                         break;
4019
4020                 /* if nothing else was read in, we're at EOF, just return */
4021                 if (bytes_read == 0)
4022                         break;
4023
4024                 if ((err = zpool_history_unpack(buf, bytes_read,
4025                     &leftover, &records, &numrecords)) != 0)
4026                         break;
4027                 off -= leftover;
4028                 if (leftover == bytes_read) {
4029                         /*
4030                          * no progress made, because buffer is not big enough
4031                          * to hold this record; resize and retry.
4032                          */
4033                         buflen *= 2;
4034                         free(buf);
4035                         buf = NULL;
4036                         if ((buflen >= HIS_BUF_LEN_MAX) ||
4037                             ((buf = malloc(buflen)) == NULL)) {
4038                                 err = ENOMEM;
4039                                 break;
4040                         }
4041                 }
4042
4043                 /* CONSTCOND */
4044         } while (1);
4045
4046         free(buf);
4047
4048         if (!err) {
4049                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
4050                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
4051                     records, numrecords) == 0);
4052         }
4053         for (i = 0; i < numrecords; i++)
4054                 nvlist_free(records[i]);
4055         free(records);
4056
4057         return (err);
4058 }
4059
4060 void
4061 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
4062     char *pathname, size_t len)
4063 {
4064         zfs_cmd_t zc = { 0 };
4065         boolean_t mounted = B_FALSE;
4066         char *mntpnt = NULL;
4067         char dsname[ZFS_MAX_DATASET_NAME_LEN];
4068
4069         if (dsobj == 0) {
4070                 /* special case for the MOS */
4071                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>", obj);
4072                 return;
4073         }
4074
4075         /* get the dataset's name */
4076         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
4077         zc.zc_obj = dsobj;
4078         if (ioctl(zhp->zpool_hdl->libzfs_fd,
4079             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
4080                 /* just write out a path of two object numbers */
4081                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
4082                     dsobj, obj);
4083                 return;
4084         }
4085         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
4086
4087         /* find out if the dataset is mounted */
4088         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
4089
4090         /* get the corrupted object's path */
4091         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
4092         zc.zc_obj = obj;
4093         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
4094             &zc) == 0) {
4095                 if (mounted) {
4096                         (void) snprintf(pathname, len, "%s%s", mntpnt,
4097                             zc.zc_value);
4098                 } else {
4099                         (void) snprintf(pathname, len, "%s:%s",
4100                             dsname, zc.zc_value);
4101                 }
4102         } else {
4103                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname, obj);
4104         }
4105         free(mntpnt);
4106 }
4107
4108 #ifdef illumos
4109 /*
4110  * Read the EFI label from the config, if a label does not exist then
4111  * pass back the error to the caller. If the caller has passed a non-NULL
4112  * diskaddr argument then we set it to the starting address of the EFI
4113  * partition. If the caller has passed a non-NULL boolean argument, then
4114  * we set it to indicate if the disk does have efi system partition.
4115  */
4116 static int
4117 read_efi_label(nvlist_t *config, diskaddr_t *sb, boolean_t *system)
4118 {
4119         char *path;
4120         int fd;
4121         char diskname[MAXPATHLEN];
4122         boolean_t boot = B_FALSE;
4123         int err = -1;
4124         int slice;
4125
4126         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
4127                 return (err);
4128
4129         (void) snprintf(diskname, sizeof (diskname), "%s%s", ZFS_RDISK_ROOT,
4130             strrchr(path, '/'));
4131         if ((fd = open(diskname, O_RDONLY|O_NDELAY)) >= 0) {
4132                 struct dk_gpt *vtoc;
4133
4134                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
4135                         for (slice = 0; slice < vtoc->efi_nparts; slice++) {
4136                                 if (vtoc->efi_parts[slice].p_tag == V_SYSTEM)
4137                                         boot = B_TRUE;
4138                                 if (vtoc->efi_parts[slice].p_tag == V_USR)
4139                                         break;
4140                         }
4141                         if (sb != NULL && vtoc->efi_parts[slice].p_tag == V_USR)
4142                                 *sb = vtoc->efi_parts[slice].p_start;
4143                         if (system != NULL)
4144                                 *system = boot;
4145                         efi_free(vtoc);
4146                 }
4147                 (void) close(fd);
4148         }
4149         return (err);
4150 }
4151
4152 /*
4153  * determine where a partition starts on a disk in the current
4154  * configuration
4155  */
4156 static diskaddr_t
4157 find_start_block(nvlist_t *config)
4158 {
4159         nvlist_t **child;
4160         uint_t c, children;
4161         diskaddr_t sb = MAXOFFSET_T;
4162         uint64_t wholedisk;
4163
4164         if (nvlist_lookup_nvlist_array(config,
4165             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
4166                 if (nvlist_lookup_uint64(config,
4167                     ZPOOL_CONFIG_WHOLE_DISK,
4168                     &wholedisk) != 0 || !wholedisk) {
4169                         return (MAXOFFSET_T);
4170                 }
4171                 if (read_efi_label(config, &sb, NULL) < 0)
4172                         sb = MAXOFFSET_T;
4173                 return (sb);
4174         }
4175
4176         for (c = 0; c < children; c++) {
4177                 sb = find_start_block(child[c]);
4178                 if (sb != MAXOFFSET_T) {
4179                         return (sb);
4180                 }
4181         }
4182         return (MAXOFFSET_T);
4183 }
4184 #endif /* illumos */
4185
4186 /*
4187  * Label an individual disk.  The name provided is the short name,
4188  * stripped of any leading /dev path.
4189  */
4190 int
4191 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, const char *name,
4192     zpool_boot_label_t boot_type, uint64_t boot_size, int *slice)
4193 {
4194 #ifdef illumos
4195         char path[MAXPATHLEN];
4196         struct dk_gpt *vtoc;
4197         int fd;
4198         size_t resv = EFI_MIN_RESV_SIZE;
4199         uint64_t slice_size;
4200         diskaddr_t start_block;
4201         char errbuf[1024];
4202
4203         /* prepare an error message just in case */
4204         (void) snprintf(errbuf, sizeof (errbuf),
4205             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
4206
4207         if (zhp) {
4208                 nvlist_t *nvroot;
4209
4210                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
4211                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4212
4213                 if (zhp->zpool_start_block == 0)
4214                         start_block = find_start_block(nvroot);
4215                 else
4216                         start_block = zhp->zpool_start_block;
4217                 zhp->zpool_start_block = start_block;
4218         } else {
4219                 /* new pool */
4220                 start_block = NEW_START_BLOCK;
4221         }
4222
4223         (void) snprintf(path, sizeof (path), "%s/%s%s", ZFS_RDISK_ROOT, name,
4224             BACKUP_SLICE);
4225
4226         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
4227                 /*
4228                  * This shouldn't happen.  We've long since verified that this
4229                  * is a valid device.
4230                  */
4231                 zfs_error_aux(hdl,
4232                     dgettext(TEXT_DOMAIN, "unable to open device"));
4233                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
4234         }
4235
4236         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
4237                 /*
4238                  * The only way this can fail is if we run out of memory, or we
4239                  * were unable to read the disk's capacity
4240                  */
4241                 if (errno == ENOMEM)
4242                         (void) no_memory(hdl);
4243
4244                 (void) close(fd);
4245                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4246                     "unable to read disk capacity"), name);
4247
4248                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
4249         }
4250
4251         /*
4252          * Why we use V_USR: V_BACKUP confuses users, and is considered
4253          * disposable by some EFI utilities (since EFI doesn't have a backup
4254          * slice).  V_UNASSIGNED is supposed to be used only for zero size
4255          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
4256          * etc. were all pretty specific.  V_USR is as close to reality as we
4257          * can get, in the absence of V_OTHER.
4258          */
4259         /* first fix the partition start block */
4260         if (start_block == MAXOFFSET_T)
4261                 start_block = NEW_START_BLOCK;
4262
4263         /*
4264          * EFI System partition is using slice 0.
4265          * ZFS is on slice 1 and slice 8 is reserved.
4266          * We assume the GPT partition table without system
4267          * partition has zfs p_start == NEW_START_BLOCK.
4268          * If start_block != NEW_START_BLOCK, it means we have
4269          * system partition. Correct solution would be to query/cache vtoc
4270          * from existing vdev member.
4271          */
4272         if (boot_type == ZPOOL_CREATE_BOOT_LABEL) {
4273                 if (boot_size % vtoc->efi_lbasize != 0) {
4274                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4275                             "boot partition size must be a multiple of %d"),
4276                             vtoc->efi_lbasize);
4277                         (void) close(fd);
4278                         efi_free(vtoc);
4279                         return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4280                 }
4281                 /*
4282                  * System partition size checks.
4283                  * Note the 1MB is quite arbitrary value, since we
4284                  * are creating dedicated pool, it should be enough
4285                  * to hold fat + efi bootloader. May need to be
4286                  * adjusted if the bootloader size will grow.
4287                  */
4288                 if (boot_size < 1024 * 1024) {
4289                         char buf[64];
4290                         zfs_nicenum(boot_size, buf, sizeof (buf));
4291                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4292                             "Specified size %s for EFI System partition is too "
4293                             "small, the minimum size is 1MB."), buf);
4294                         (void) close(fd);
4295                         efi_free(vtoc);
4296                         return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4297                 }
4298                 /* 33MB is tested with mkfs -F pcfs */
4299                 if (hdl->libzfs_printerr &&
4300                     ((vtoc->efi_lbasize == 512 &&
4301                     boot_size < 33 * 1024 * 1024) ||
4302                     (vtoc->efi_lbasize == 4096 &&
4303                     boot_size < 256 * 1024 * 1024)))  {
4304                         char buf[64];
4305                         zfs_nicenum(boot_size, buf, sizeof (buf));
4306                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
4307                             "Warning: EFI System partition size %s is "
4308                             "not allowing to create FAT32 file\nsystem, which "
4309                             "may result in unbootable system.\n"), buf);
4310                 }
4311                 /* Adjust zfs partition start by size of system partition. */
4312                 start_block += boot_size / vtoc->efi_lbasize;
4313         }
4314
4315         if (start_block == NEW_START_BLOCK) {
4316                 /*
4317                  * Use default layout.
4318                  * ZFS is on slice 0 and slice 8 is reserved.
4319                  */
4320                 slice_size = vtoc->efi_last_u_lba + 1;
4321                 slice_size -= EFI_MIN_RESV_SIZE;
4322                 slice_size -= start_block;
4323                 if (slice != NULL)
4324                         *slice = 0;
4325
4326                 vtoc->efi_parts[0].p_start = start_block;
4327                 vtoc->efi_parts[0].p_size = slice_size;
4328
4329                 vtoc->efi_parts[0].p_tag = V_USR;
4330                 (void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
4331
4332                 vtoc->efi_parts[8].p_start = slice_size + start_block;
4333                 vtoc->efi_parts[8].p_size = resv;
4334                 vtoc->efi_parts[8].p_tag = V_RESERVED;
4335         } else {
4336                 slice_size = start_block - NEW_START_BLOCK;
4337                 vtoc->efi_parts[0].p_start = NEW_START_BLOCK;
4338                 vtoc->efi_parts[0].p_size = slice_size;
4339                 vtoc->efi_parts[0].p_tag = V_SYSTEM;
4340                 (void) strcpy(vtoc->efi_parts[0].p_name, "loader");
4341                 if (slice != NULL)
4342                         *slice = 1;
4343                 /* prepare slice 1 */
4344                 slice_size = vtoc->efi_last_u_lba + 1 - slice_size;
4345                 slice_size -= resv;
4346                 slice_size -= NEW_START_BLOCK;
4347                 vtoc->efi_parts[1].p_start = start_block;
4348                 vtoc->efi_parts[1].p_size = slice_size;
4349                 vtoc->efi_parts[1].p_tag = V_USR;
4350                 (void) strcpy(vtoc->efi_parts[1].p_name, "zfs");
4351
4352                 vtoc->efi_parts[8].p_start = slice_size + start_block;
4353                 vtoc->efi_parts[8].p_size = resv;
4354                 vtoc->efi_parts[8].p_tag = V_RESERVED;
4355         }
4356
4357         if (efi_write(fd, vtoc) != 0) {
4358                 /*
4359                  * Some block drivers (like pcata) may not support EFI
4360                  * GPT labels.  Print out a helpful error message dir-
4361                  * ecting the user to manually label the disk and give
4362                  * a specific slice.
4363                  */
4364                 (void) close(fd);
4365                 efi_free(vtoc);
4366
4367                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4368                     "try using fdisk(1M) and then provide a specific slice"));
4369                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4370         }
4371
4372         (void) close(fd);
4373         efi_free(vtoc);
4374 #endif /* illumos */
4375         return (0);
4376 }
4377
4378 static boolean_t
4379 supported_dump_vdev_type(libzfs_handle_t *hdl, nvlist_t *config, char *errbuf)
4380 {
4381         char *type;
4382         nvlist_t **child;
4383         uint_t children, c;
4384
4385         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_TYPE, &type) == 0);
4386         if (strcmp(type, VDEV_TYPE_FILE) == 0 ||
4387             strcmp(type, VDEV_TYPE_HOLE) == 0 ||
4388             strcmp(type, VDEV_TYPE_MISSING) == 0) {
4389                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4390                     "vdev type '%s' is not supported"), type);
4391                 (void) zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf);
4392                 return (B_FALSE);
4393         }
4394         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
4395             &child, &children) == 0) {
4396                 for (c = 0; c < children; c++) {
4397                         if (!supported_dump_vdev_type(hdl, child[c], errbuf))
4398                                 return (B_FALSE);
4399                 }
4400         }
4401         return (B_TRUE);
4402 }
4403
4404 /*
4405  * Check if this zvol is allowable for use as a dump device; zero if
4406  * it is, > 0 if it isn't, < 0 if it isn't a zvol.
4407  *
4408  * Allowable storage configurations include mirrors, all raidz variants, and
4409  * pools with log, cache, and spare devices.  Pools which are backed by files or
4410  * have missing/hole vdevs are not suitable.
4411  */
4412 int
4413 zvol_check_dump_config(char *arg)
4414 {
4415         zpool_handle_t *zhp = NULL;
4416         nvlist_t *config, *nvroot;
4417         char *p, *volname;
4418         nvlist_t **top;
4419         uint_t toplevels;
4420         libzfs_handle_t *hdl;
4421         char errbuf[1024];
4422         char poolname[ZFS_MAX_DATASET_NAME_LEN];
4423         int pathlen = strlen(ZVOL_FULL_DEV_DIR);
4424         int ret = 1;
4425
4426         if (strncmp(arg, ZVOL_FULL_DEV_DIR, pathlen)) {
4427                 return (-1);
4428         }
4429
4430         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
4431             "dump is not supported on device '%s'"), arg);
4432
4433         if ((hdl = libzfs_init()) == NULL)
4434                 return (1);
4435         libzfs_print_on_error(hdl, B_TRUE);
4436
4437         volname = arg + pathlen;
4438
4439         /* check the configuration of the pool */
4440         if ((p = strchr(volname, '/')) == NULL) {
4441                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4442                     "malformed dataset name"));
4443                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4444                 return (1);
4445         } else if (p - volname >= ZFS_MAX_DATASET_NAME_LEN) {
4446                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4447                     "dataset name is too long"));
4448                 (void) zfs_error(hdl, EZFS_NAMETOOLONG, errbuf);
4449                 return (1);
4450         } else {
4451                 (void) strncpy(poolname, volname, p - volname);
4452                 poolname[p - volname] = '\0';
4453         }
4454
4455         if ((zhp = zpool_open(hdl, poolname)) == NULL) {
4456                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4457                     "could not open pool '%s'"), poolname);
4458                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
4459                 goto out;
4460         }
4461         config = zpool_get_config(zhp, NULL);
4462         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4463             &nvroot) != 0) {
4464                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4465                     "could not obtain vdev configuration for  '%s'"), poolname);
4466                 (void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
4467                 goto out;
4468         }
4469
4470         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4471             &top, &toplevels) == 0);
4472
4473         if (!supported_dump_vdev_type(hdl, top[0], errbuf)) {
4474                 goto out;
4475         }
4476         ret = 0;
4477
4478 out:
4479         if (zhp)
4480                 zpool_close(zhp);
4481         libzfs_fini(hdl);
4482         return (ret);
4483 }
4484
4485 int
4486 zpool_nextboot(libzfs_handle_t *hdl, uint64_t pool_guid, uint64_t dev_guid,
4487     const char *command)
4488 {
4489         zfs_cmd_t zc = { 0 };
4490         nvlist_t *args;
4491         char *packed;
4492         size_t size;
4493         int error;
4494
4495         args = fnvlist_alloc();
4496         fnvlist_add_uint64(args, ZPOOL_CONFIG_POOL_GUID, pool_guid);
4497         fnvlist_add_uint64(args, ZPOOL_CONFIG_GUID, dev_guid);
4498         fnvlist_add_string(args, "command", command);
4499         error = zcmd_write_src_nvlist(hdl, &zc, args);
4500         if (error == 0)
4501                 error = ioctl(hdl->libzfs_fd, ZFS_IOC_NEXTBOOT, &zc);
4502         zcmd_free_nvlists(&zc);
4503         nvlist_free(args);
4504         return (error);
4505 }