granicus.if.org Git - zfs/blob - lib/libzfs/libzfs_pool.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
  24  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  25  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
  26  * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>
  27  */
  28
  29 #include <ctype.h>
  30 #include <errno.h>
  31 #include <devid.h>
  32 #include <fcntl.h>
  33 #include <libintl.h>
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <strings.h>
  37 #include <unistd.h>
  38 #include <libgen.h>
  39 #include <zone.h>
  40 #include <sys/stat.h>
  41 #include <sys/efi_partition.h>
  42 #include <sys/vtoc.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 *config, diskaddr_t *sb);
  53
  54 typedef struct prop_flags {
  55         int create:1;   /* Validate property on creation */
  56         int import:1;   /* Validate property on import */
  57 } prop_flags_t;
  58
  59 /*
  60  * ====================================================================
  61  *   zpool property functions
  62  * ====================================================================
  63  */
  64
  65 static int
  66 zpool_get_all_props(zpool_handle_t *zhp)
  67 {
  68         zfs_cmd_t zc = {"\0"};
  69         libzfs_handle_t *hdl = zhp->zpool_hdl;
  70
  71         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
  72
  73         if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
  74                 return (-1);
  75
  76         while (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
  77                 if (errno == ENOMEM) {
  78                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
  79                                 zcmd_free_nvlists(&zc);
  80                                 return (-1);
  81                         }
  82                 } else {
  83                         zcmd_free_nvlists(&zc);
  84                         return (-1);
  85                 }
  86         }
  87
  88         if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
  89                 zcmd_free_nvlists(&zc);
  90                 return (-1);
  91         }
  92
  93         zcmd_free_nvlists(&zc);
  94
  95         return (0);
  96 }
  97
  98 static int
  99 zpool_props_refresh(zpool_handle_t *zhp)
 100 {
 101         nvlist_t *old_props;
 102
 103         old_props = zhp->zpool_props;
 104
 105         if (zpool_get_all_props(zhp) != 0)
 106                 return (-1);
 107
 108         nvlist_free(old_props);
 109         return (0);
 110 }
 111
 112 static char *
 113 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
 114     zprop_source_t *src)
 115 {
 116         nvlist_t *nv, *nvl;
 117         uint64_t ival;
 118         char *value;
 119         zprop_source_t source;
 120
 121         nvl = zhp->zpool_props;
 122         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 123                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &ival) == 0);
 124                 source = ival;
 125                 verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
 126         } else {
 127                 source = ZPROP_SRC_DEFAULT;
 128                 if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
 129                         value = "-";
 130         }
 131
 132         if (src)
 133                 *src = source;
 134
 135         return (value);
 136 }
 137
 138 uint64_t
 139 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
 140 {
 141         nvlist_t *nv, *nvl;
 142         uint64_t value;
 143         zprop_source_t source;
 144
 145         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
 146                 /*
 147                  * zpool_get_all_props() has most likely failed because
 148                  * the pool is faulted, but if all we need is the top level
 149                  * vdev's guid then get it from the zhp config nvlist.
 150                  */
 151                 if ((prop == ZPOOL_PROP_GUID) &&
 152                     (nvlist_lookup_nvlist(zhp->zpool_config,
 153                     ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
 154                     (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
 155                     == 0)) {
 156                         return (value);
 157                 }
 158                 return (zpool_prop_default_numeric(prop));
 159         }
 160
 161         nvl = zhp->zpool_props;
 162         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 163                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &value) == 0);
 164                 source = value;
 165                 verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
 166         } else {
 167                 source = ZPROP_SRC_DEFAULT;
 168                 value = zpool_prop_default_numeric(prop);
 169         }
 170
 171         if (src)
 172                 *src = source;
 173
 174         return (value);
 175 }
 176
 177 /*
 178  * Map VDEV STATE to printed strings.
 179  */
 180 char *
 181 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
 182 {
 183         switch (state) {
 184         case VDEV_STATE_CLOSED:
 185         case VDEV_STATE_OFFLINE:
 186                 return (gettext("OFFLINE"));
 187         case VDEV_STATE_REMOVED:
 188                 return (gettext("REMOVED"));
 189         case VDEV_STATE_CANT_OPEN:
 190                 if (aux == VDEV_AUX_CORRUPT_DATA || aux == VDEV_AUX_BAD_LOG)
 191                         return (gettext("FAULTED"));
 192                 else if (aux == VDEV_AUX_SPLIT_POOL)
 193                         return (gettext("SPLIT"));
 194                 else
 195                         return (gettext("UNAVAIL"));
 196         case VDEV_STATE_FAULTED:
 197                 return (gettext("FAULTED"));
 198         case VDEV_STATE_DEGRADED:
 199                 return (gettext("DEGRADED"));
 200         case VDEV_STATE_HEALTHY:
 201                 return (gettext("ONLINE"));
 202
 203         default:
 204                 break;
 205         }
 206
 207         return (gettext("UNKNOWN"));
 208 }
 209
 210 /*
 211  * Map POOL STATE to printed strings.
 212  */
 213 const char *
 214 zpool_pool_state_to_name(pool_state_t state)
 215 {
 216         switch (state) {
 217         default:
 218                 break;
 219         case POOL_STATE_ACTIVE:
 220                 return (gettext("ACTIVE"));
 221         case POOL_STATE_EXPORTED:
 222                 return (gettext("EXPORTED"));
 223         case POOL_STATE_DESTROYED:
 224                 return (gettext("DESTROYED"));
 225         case POOL_STATE_SPARE:
 226                 return (gettext("SPARE"));
 227         case POOL_STATE_L2CACHE:
 228                 return (gettext("L2CACHE"));
 229         case POOL_STATE_UNINITIALIZED:
 230                 return (gettext("UNINITIALIZED"));
 231         case POOL_STATE_UNAVAIL:
 232                 return (gettext("UNAVAIL"));
 233         case POOL_STATE_POTENTIALLY_ACTIVE:
 234                 return (gettext("POTENTIALLY_ACTIVE"));
 235         }
 236
 237         return (gettext("UNKNOWN"));
 238 }
 239
 240 /*
 241  * Get a zpool property value for 'prop' and return the value in
 242  * a pre-allocated buffer.
 243  */
 244 int
 245 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf,
 246     size_t len, zprop_source_t *srctype, boolean_t literal)
 247 {
 248         uint64_t intval;
 249         const char *strval;
 250         zprop_source_t src = ZPROP_SRC_NONE;
 251         nvlist_t *nvroot;
 252         vdev_stat_t *vs;
 253         uint_t vsc;
 254
 255         if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
 256                 switch (prop) {
 257                 case ZPOOL_PROP_NAME:
 258                         (void) strlcpy(buf, zpool_get_name(zhp), len);
 259                         break;
 260
 261                 case ZPOOL_PROP_HEALTH:
 262                         (void) strlcpy(buf, "FAULTED", len);
 263                         break;
 264
 265                 case ZPOOL_PROP_GUID:
 266                         intval = zpool_get_prop_int(zhp, prop, &src);
 267                         (void) snprintf(buf, len, "%llu", (u_longlong_t)intval);
 268                         break;
 269
 270                 case ZPOOL_PROP_ALTROOT:
 271                 case ZPOOL_PROP_CACHEFILE:
 272                 case ZPOOL_PROP_COMMENT:
 273                         if (zhp->zpool_props != NULL ||
 274                             zpool_get_all_props(zhp) == 0) {
 275                                 (void) strlcpy(buf,
 276                                     zpool_get_prop_string(zhp, prop, &src),
 277                                     len);
 278                                 break;
 279                         }
 280                         /* FALLTHROUGH */
 281                 default:
 282                         (void) strlcpy(buf, "-", len);
 283                         break;
 284                 }
 285
 286                 if (srctype != NULL)
 287                         *srctype = src;
 288                 return (0);
 289         }
 290
 291         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
 292             prop != ZPOOL_PROP_NAME)
 293                 return (-1);
 294
 295         switch (zpool_prop_get_type(prop)) {
 296         case PROP_TYPE_STRING:
 297                 (void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
 298                     len);
 299                 break;
 300
 301         case PROP_TYPE_NUMBER:
 302                 intval = zpool_get_prop_int(zhp, prop, &src);
 303
 304                 switch (prop) {
 305                 case ZPOOL_PROP_SIZE:
 306                 case ZPOOL_PROP_ALLOCATED:
 307                 case ZPOOL_PROP_FREE:
 308                 case ZPOOL_PROP_FREEING:
 309                 case ZPOOL_PROP_LEAKED:
 310                 case ZPOOL_PROP_ASHIFT:
 311                         if (literal)
 312                                 (void) snprintf(buf, len, "%llu",
 313                                     (u_longlong_t)intval);
 314                         else
 315                                 (void) zfs_nicenum(intval, buf, len);
 316                         break;
 317
 318                 case ZPOOL_PROP_EXPANDSZ:
 319                         if (intval == 0) {
 320                                 (void) strlcpy(buf, "-", len);
 321                         } else if (literal) {
 322                                 (void) snprintf(buf, len, "%llu",
 323                                     (u_longlong_t)intval);
 324                         } else {
 325                                 (void) zfs_nicebytes(intval, buf, len);
 326                         }
 327                         break;
 328
 329                 case ZPOOL_PROP_CAPACITY:
 330                         if (literal) {
 331                                 (void) snprintf(buf, len, "%llu",
 332                                     (u_longlong_t)intval);
 333                         } else {
 334                                 (void) snprintf(buf, len, "%llu%%",
 335                                     (u_longlong_t)intval);
 336                         }
 337                         break;
 338
 339                 case ZPOOL_PROP_FRAGMENTATION:
 340                         if (intval == UINT64_MAX) {
 341                                 (void) strlcpy(buf, "-", len);
 342                         } else if (literal) {
 343                                 (void) snprintf(buf, len, "%llu",
 344                                     (u_longlong_t)intval);
 345                         } else {
 346                                 (void) snprintf(buf, len, "%llu%%",
 347                                     (u_longlong_t)intval);
 348                         }
 349                         break;
 350
 351                 case ZPOOL_PROP_DEDUPRATIO:
 352                         if (literal)
 353                                 (void) snprintf(buf, len, "%llu.%02llu",
 354                                     (u_longlong_t)(intval / 100),
 355                                     (u_longlong_t)(intval % 100));
 356                         else
 357                                 (void) snprintf(buf, len, "%llu.%02llux",
 358                                     (u_longlong_t)(intval / 100),
 359                                     (u_longlong_t)(intval % 100));
 360                         break;
 361
 362                 case ZPOOL_PROP_HEALTH:
 363                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 364                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 365                         verify(nvlist_lookup_uint64_array(nvroot,
 366                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
 367                             == 0);
 368
 369                         (void) strlcpy(buf, zpool_state_to_name(intval,
 370                             vs->vs_aux), len);
 371                         break;
 372                 case ZPOOL_PROP_VERSION:
 373                         if (intval >= SPA_VERSION_FEATURES) {
 374                                 (void) snprintf(buf, len, "-");
 375                                 break;
 376                         }
 377                         /* FALLTHROUGH */
 378                 default:
 379                         (void) snprintf(buf, len, "%llu", (u_longlong_t)intval);
 380                 }
 381                 break;
 382
 383         case PROP_TYPE_INDEX:
 384                 intval = zpool_get_prop_int(zhp, prop, &src);
 385                 if (zpool_prop_index_to_string(prop, intval, &strval)
 386                     != 0)
 387                         return (-1);
 388                 (void) strlcpy(buf, strval, len);
 389                 break;
 390
 391         default:
 392                 abort();
 393         }
 394
 395         if (srctype)
 396                 *srctype = src;
 397
 398         return (0);
 399 }
 400
 401 /*
 402  * Check if the bootfs name has the same pool name as it is set to.
 403  * Assuming bootfs is a valid dataset name.
 404  */
 405 static boolean_t
 406 bootfs_name_valid(const char *pool, char *bootfs)
 407 {
 408         int len = strlen(pool);
 409
 410         if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
 411                 return (B_FALSE);
 412
 413         if (strncmp(pool, bootfs, len) == 0 &&
 414             (bootfs[len] == '/' || bootfs[len] == '\0'))
 415                 return (B_TRUE);
 416
 417         return (B_FALSE);
 418 }
 419
 420 boolean_t
 421 zpool_is_bootable(zpool_handle_t *zhp)
 422 {
 423         char bootfs[ZFS_MAX_DATASET_NAME_LEN];
 424
 425         return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
 426             sizeof (bootfs), NULL, B_FALSE) == 0 && strncmp(bootfs, "-",
 427             sizeof (bootfs)) != 0);
 428 }
 429
 430
 431 /*
 432  * Given an nvlist of zpool properties to be set, validate that they are
 433  * correct, and parse any numeric properties (index, boolean, etc) if they are
 434  * specified as strings.
 435  */
 436 static nvlist_t *
 437 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 438     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
 439 {
 440         nvpair_t *elem;
 441         nvlist_t *retprops;
 442         zpool_prop_t prop;
 443         char *strval;
 444         uint64_t intval;
 445         char *slash, *check;
 446         struct stat64 statbuf;
 447         zpool_handle_t *zhp;
 448
 449         if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
 450                 (void) no_memory(hdl);
 451                 return (NULL);
 452         }
 453
 454         elem = NULL;
 455         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 456                 const char *propname = nvpair_name(elem);
 457
 458                 prop = zpool_name_to_prop(propname);
 459                 if (prop == ZPROP_INVAL && zpool_prop_feature(propname)) {
 460                         int err;
 461                         char *fname = strchr(propname, '@') + 1;
 462
 463                         err = zfeature_lookup_name(fname, NULL);
 464                         if (err != 0) {
 465                                 ASSERT3U(err, ==, ENOENT);
 466                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 467                                     "invalid feature '%s'"), fname);
 468                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 469                                 goto error;
 470                         }
 471
 472                         if (nvpair_type(elem) != DATA_TYPE_STRING) {
 473                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 474                                     "'%s' must be a string"), propname);
 475                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 476                                 goto error;
 477                         }
 478
 479                         (void) nvpair_value_string(elem, &strval);
 480                         if (strcmp(strval, ZFS_FEATURE_ENABLED) != 0 &&
 481                             strcmp(strval, ZFS_FEATURE_DISABLED) != 0) {
 482                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 483                                     "property '%s' can only be set to "
 484                                     "'enabled' or 'disabled'"), propname);
 485                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 486                                 goto error;
 487                         }
 488
 489                         if (nvlist_add_uint64(retprops, propname, 0) != 0) {
 490                                 (void) no_memory(hdl);
 491                                 goto error;
 492                         }
 493                         continue;
 494                 }
 495
 496                 /*
 497                  * Make sure this property is valid and applies to this type.
 498                  */
 499                 if (prop == ZPROP_INVAL) {
 500                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 501                             "invalid property '%s'"), propname);
 502                         (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 503                         goto error;
 504                 }
 505
 506                 if (zpool_prop_readonly(prop)) {
 507                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 508                             "is readonly"), propname);
 509                         (void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
 510                         goto error;
 511                 }
 512
 513                 if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
 514                     &strval, &intval, errbuf) != 0)
 515                         goto error;
 516
 517                 /*
 518                  * Perform additional checking for specific properties.
 519                  */
 520                 switch (prop) {
 521                 case ZPOOL_PROP_VERSION:
 522                         if (intval < version ||
 523                             !SPA_VERSION_IS_SUPPORTED(intval)) {
 524                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 525                                     "property '%s' number %d is invalid."),
 526                                     propname, intval);
 527                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 528                                 goto error;
 529                         }
 530                         break;
 531
 532                 case ZPOOL_PROP_ASHIFT:
 533                         if (!flags.create) {
 534                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 535                                     "property '%s' can only be set at "
 536                                     "creation time"), propname);
 537                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 538                                 goto error;
 539                         }
 540
 541                         if (intval != 0 &&
 542                             (intval < ASHIFT_MIN || intval > ASHIFT_MAX)) {
 543                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 544                                     "invalid '%s=%d' property: only values "
 545                                     "between %" PRId32 " and %" PRId32 " "
 546                                     "are allowed.\n"),
 547                                     propname, intval, ASHIFT_MIN, ASHIFT_MAX);
 548                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 549                                 goto error;
 550                         }
 551                         break;
 552
 553                 case ZPOOL_PROP_BOOTFS:
 554                         if (flags.create || flags.import) {
 555                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 556                                     "property '%s' cannot be set at creation "
 557                                     "or import time"), propname);
 558                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 559                                 goto error;
 560                         }
 561
 562                         if (version < SPA_VERSION_BOOTFS) {
 563                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 564                                     "pool must be upgraded to support "
 565                                     "'%s' property"), propname);
 566                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 567                                 goto error;
 568                         }
 569
 570                         /*
 571                          * bootfs property value has to be a dataset name and
 572                          * the dataset has to be in the same pool as it sets to.
 573                          */
 574                         if (strval[0] != '\0' && !bootfs_name_valid(poolname,
 575                             strval)) {
 576                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 577                                     "is an invalid name"), strval);
 578                                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
 579                                 goto error;
 580                         }
 581
 582                         if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
 583                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 584                                     "could not open pool '%s'"), poolname);
 585                                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
 586                                 goto error;
 587                         }
 588                         zpool_close(zhp);
 589                         break;
 590
 591                 case ZPOOL_PROP_ALTROOT:
 592                         if (!flags.create && !flags.import) {
 593                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 594                                     "property '%s' can only be set during pool "
 595                                     "creation or import"), propname);
 596                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 597                                 goto error;
 598                         }
 599
 600                         if (strval[0] != '/') {
 601                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 602                                     "bad alternate root '%s'"), strval);
 603                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 604                                 goto error;
 605                         }
 606                         break;
 607
 608                 case ZPOOL_PROP_CACHEFILE:
 609                         if (strval[0] == '\0')
 610                                 break;
 611
 612                         if (strcmp(strval, "none") == 0)
 613                                 break;
 614
 615                         if (strval[0] != '/') {
 616                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 617                                     "property '%s' must be empty, an "
 618                                     "absolute path, or 'none'"), propname);
 619                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 620                                 goto error;
 621                         }
 622
 623                         slash = strrchr(strval, '/');
 624
 625                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 626                             strcmp(slash, "/..") == 0) {
 627                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 628                                     "'%s' is not a valid file"), strval);
 629                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 630                                 goto error;
 631                         }
 632
 633                         *slash = '\0';
 634
 635                         if (strval[0] != '\0' &&
 636                             (stat64(strval, &statbuf) != 0 ||
 637                             !S_ISDIR(statbuf.st_mode))) {
 638                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 639                                     "'%s' is not a valid directory"),
 640                                     strval);
 641                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 642                                 goto error;
 643                         }
 644
 645                         *slash = '/';
 646                         break;
 647
 648                 case ZPOOL_PROP_COMMENT:
 649                         for (check = strval; *check != '\0'; check++) {
 650                                 if (!isprint(*check)) {
 651                                         zfs_error_aux(hdl,
 652                                             dgettext(TEXT_DOMAIN,
 653                                             "comment may only have printable "
 654                                             "characters"));
 655                                         (void) zfs_error(hdl, EZFS_BADPROP,
 656                                             errbuf);
 657                                         goto error;
 658                                 }
 659                         }
 660                         if (strlen(strval) > ZPROP_MAX_COMMENT) {
 661                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 662                                     "comment must not exceed %d characters"),
 663                                     ZPROP_MAX_COMMENT);
 664                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 665                                 goto error;
 666                         }
 667                         break;
 668                 case ZPOOL_PROP_READONLY:
 669                         if (!flags.import) {
 670                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 671                                     "property '%s' can only be set at "
 672                                     "import time"), propname);
 673                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 674                                 goto error;
 675                         }
 676                         break;
 677                 case ZPOOL_PROP_TNAME:
 678                         if (!flags.create) {
 679                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 680                                     "property '%s' can only be set at "
 681                                     "creation time"), propname);
 682                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 683                                 goto error;
 684                         }
 685                         break;
 686
 687                 default:
 688                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 689                             "property '%s'(%d) not defined"), propname, prop);
 690                         break;
 691                 }
 692         }
 693
 694         return (retprops);
 695 error:
 696         nvlist_free(retprops);
 697         return (NULL);
 698 }
 699
 700 /*
 701  * Set zpool property : propname=propval.
 702  */
 703 int
 704 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
 705 {
 706         zfs_cmd_t zc = {"\0"};
 707         int ret = -1;
 708         char errbuf[1024];
 709         nvlist_t *nvl = NULL;
 710         nvlist_t *realprops;
 711         uint64_t version;
 712         prop_flags_t flags = { 0 };
 713
 714         (void) snprintf(errbuf, sizeof (errbuf),
 715             dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
 716             zhp->zpool_name);
 717
 718         if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
 719                 return (no_memory(zhp->zpool_hdl));
 720
 721         if (nvlist_add_string(nvl, propname, propval) != 0) {
 722                 nvlist_free(nvl);
 723                 return (no_memory(zhp->zpool_hdl));
 724         }
 725
 726         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
 727         if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
 728             zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
 729                 nvlist_free(nvl);
 730                 return (-1);
 731         }
 732
 733         nvlist_free(nvl);
 734         nvl = realprops;
 735
 736         /*
 737          * Execute the corresponding ioctl() to set this property.
 738          */
 739         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 740
 741         if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
 742                 nvlist_free(nvl);
 743                 return (-1);
 744         }
 745
 746         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
 747
 748         zcmd_free_nvlists(&zc);
 749         nvlist_free(nvl);
 750
 751         if (ret)
 752                 (void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
 753         else
 754                 (void) zpool_props_refresh(zhp);
 755
 756         return (ret);
 757 }
 758
 759 int
 760 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
 761 {
 762         libzfs_handle_t *hdl = zhp->zpool_hdl;
 763         zprop_list_t *entry;
 764         char buf[ZFS_MAXPROPLEN];
 765         nvlist_t *features = NULL;
 766         nvpair_t *nvp;
 767         zprop_list_t **last;
 768         boolean_t firstexpand = (NULL == *plp);
 769         int i;
 770
 771         if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
 772                 return (-1);
 773
 774         last = plp;
 775         while (*last != NULL)
 776                 last = &(*last)->pl_next;
 777
 778         if ((*plp)->pl_all)
 779                 features = zpool_get_features(zhp);
 780
 781         if ((*plp)->pl_all && firstexpand) {
 782                 for (i = 0; i < SPA_FEATURES; i++) {
 783                         zprop_list_t *entry = zfs_alloc(hdl,
 784                             sizeof (zprop_list_t));
 785                         entry->pl_prop = ZPROP_INVAL;
 786                         entry->pl_user_prop = zfs_asprintf(hdl, "feature@%s",
 787                             spa_feature_table[i].fi_uname);
 788                         entry->pl_width = strlen(entry->pl_user_prop);
 789                         entry->pl_all = B_TRUE;
 790
 791                         *last = entry;
 792                         last = &entry->pl_next;
 793                 }
 794         }
 795
 796         /* add any unsupported features */
 797         for (nvp = nvlist_next_nvpair(features, NULL);
 798             nvp != NULL; nvp = nvlist_next_nvpair(features, nvp)) {
 799                 char *propname;
 800                 boolean_t found;
 801                 zprop_list_t *entry;
 802
 803                 if (zfeature_is_supported(nvpair_name(nvp)))
 804                         continue;
 805
 806                 propname = zfs_asprintf(hdl, "unsupported@%s",
 807                     nvpair_name(nvp));
 808
 809                 /*
 810                  * Before adding the property to the list make sure that no
 811                  * other pool already added the same property.
 812                  */
 813                 found = B_FALSE;
 814                 entry = *plp;
 815                 while (entry != NULL) {
 816                         if (entry->pl_user_prop != NULL &&
 817                             strcmp(propname, entry->pl_user_prop) == 0) {
 818                                 found = B_TRUE;
 819                                 break;
 820                         }
 821                         entry = entry->pl_next;
 822                 }
 823                 if (found) {
 824                         free(propname);
 825                         continue;
 826                 }
 827
 828                 entry = zfs_alloc(hdl, sizeof (zprop_list_t));
 829                 entry->pl_prop = ZPROP_INVAL;
 830                 entry->pl_user_prop = propname;
 831                 entry->pl_width = strlen(entry->pl_user_prop);
 832                 entry->pl_all = B_TRUE;
 833
 834                 *last = entry;
 835                 last = &entry->pl_next;
 836         }
 837
 838         for (entry = *plp; entry != NULL; entry = entry->pl_next) {
 839
 840                 if (entry->pl_fixed)
 841                         continue;
 842
 843                 if (entry->pl_prop != ZPROP_INVAL &&
 844                     zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
 845                     NULL, B_FALSE) == 0) {
 846                         if (strlen(buf) > entry->pl_width)
 847                                 entry->pl_width = strlen(buf);
 848                 }
 849         }
 850
 851         return (0);
 852 }
 853
 854 /*
 855  * Get the state for the given feature on the given ZFS pool.
 856  */
 857 int
 858 zpool_prop_get_feature(zpool_handle_t *zhp, const char *propname, char *buf,
 859     size_t len)
 860 {
 861         uint64_t refcount;
 862         boolean_t found = B_FALSE;
 863         nvlist_t *features = zpool_get_features(zhp);
 864         boolean_t supported;
 865         const char *feature = strchr(propname, '@') + 1;
 866
 867         supported = zpool_prop_feature(propname);
 868         ASSERT(supported || zpool_prop_unsupported(propname));
 869
 870         /*
 871          * Convert from feature name to feature guid. This conversion is
 872          * unnecessary for unsupported@... properties because they already
 873          * use guids.
 874          */
 875         if (supported) {
 876                 int ret;
 877                 spa_feature_t fid;
 878
 879                 ret = zfeature_lookup_name(feature, &fid);
 880                 if (ret != 0) {
 881                         (void) strlcpy(buf, "-", len);
 882                         return (ENOTSUP);
 883                 }
 884                 feature = spa_feature_table[fid].fi_guid;
 885         }
 886
 887         if (nvlist_lookup_uint64(features, feature, &refcount) == 0)
 888                 found = B_TRUE;
 889
 890         if (supported) {
 891                 if (!found) {
 892                         (void) strlcpy(buf, ZFS_FEATURE_DISABLED, len);
 893                 } else  {
 894                         if (refcount == 0)
 895                                 (void) strlcpy(buf, ZFS_FEATURE_ENABLED, len);
 896                         else
 897                                 (void) strlcpy(buf, ZFS_FEATURE_ACTIVE, len);
 898                 }
 899         } else {
 900                 if (found) {
 901                         if (refcount == 0) {
 902                                 (void) strcpy(buf, ZFS_UNSUPPORTED_INACTIVE);
 903                         } else {
 904                                 (void) strcpy(buf, ZFS_UNSUPPORTED_READONLY);
 905                         }
 906                 } else {
 907                         (void) strlcpy(buf, "-", len);
 908                         return (ENOTSUP);
 909                 }
 910         }
 911
 912         return (0);
 913 }
 914
 915 /*
 916  * Don't start the slice at the default block of 34; many storage
 917  * devices will use a stripe width of 128k, other vendors prefer a 1m
 918  * alignment.  It is best to play it safe and ensure a 1m alignment
 919  * given 512B blocks.  When the block size is larger by a power of 2
 920  * we will still be 1m aligned.  Some devices are sensitive to the
 921  * partition ending alignment as well.
 922  */
 923 #define NEW_START_BLOCK         2048
 924 #define PARTITION_END_ALIGNMENT 2048
 925
 926 /*
 927  * Validate the given pool name, optionally putting an extended error message in
 928  * 'buf'.
 929  */
 930 boolean_t
 931 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
 932 {
 933         namecheck_err_t why;
 934         char what;
 935         int ret;
 936
 937         ret = pool_namecheck(pool, &why, &what);
 938
 939         /*
 940          * The rules for reserved pool names were extended at a later point.
 941          * But we need to support users with existing pools that may now be
 942          * invalid.  So we only check for this expanded set of names during a
 943          * create (or import), and only in userland.
 944          */
 945         if (ret == 0 && !isopen &&
 946             (strncmp(pool, "mirror", 6) == 0 ||
 947             strncmp(pool, "raidz", 5) == 0 ||
 948             strncmp(pool, "spare", 5) == 0 ||
 949             strcmp(pool, "log") == 0)) {
 950                 if (hdl != NULL)
 951                         zfs_error_aux(hdl,
 952                             dgettext(TEXT_DOMAIN, "name is reserved"));
 953                 return (B_FALSE);
 954         }
 955
 956
 957         if (ret != 0) {
 958                 if (hdl != NULL) {
 959                         switch (why) {
 960                         case NAME_ERR_TOOLONG:
 961                                 zfs_error_aux(hdl,
 962                                     dgettext(TEXT_DOMAIN, "name is too long"));
 963                                 break;
 964
 965                         case NAME_ERR_INVALCHAR:
 966                                 zfs_error_aux(hdl,
 967                                     dgettext(TEXT_DOMAIN, "invalid character "
 968                                     "'%c' in pool name"), what);
 969                                 break;
 970
 971                         case NAME_ERR_NOLETTER:
 972                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 973                                     "name must begin with a letter"));
 974                                 break;
 975
 976                         case NAME_ERR_RESERVED:
 977                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 978                                     "name is reserved"));
 979                                 break;
 980
 981                         case NAME_ERR_DISKLIKE:
 982                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 983                                     "pool name is reserved"));
 984                                 break;
 985
 986                         case NAME_ERR_LEADING_SLASH:
 987                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 988                                     "leading slash in name"));
 989                                 break;
 990
 991                         case NAME_ERR_EMPTY_COMPONENT:
 992                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 993                                     "empty component in name"));
 994                                 break;
 995
 996                         case NAME_ERR_TRAILING_SLASH:
 997                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 998                                     "trailing slash in name"));
 999                                 break;
1000
1001                         case NAME_ERR_MULTIPLE_DELIMITERS:
1002                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1003                                     "multiple '@' and/or '#' delimiters in "
1004                                     "name"));
1005                                 break;
1006
1007                         case NAME_ERR_NO_AT:
1008                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1009                                     "permission set is missing '@'"));
1010                                 break;
1011
1012                         default:
1013                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1014                                     "(%d) not defined"), why);
1015                                 break;
1016                         }
1017                 }
1018                 return (B_FALSE);
1019         }
1020
1021         return (B_TRUE);
1022 }
1023
1024 /*
1025  * Open a handle to the given pool, even if the pool is currently in the FAULTED
1026  * state.
1027  */
1028 zpool_handle_t *
1029 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
1030 {
1031         zpool_handle_t *zhp;
1032         boolean_t missing;
1033
1034         /*
1035          * Make sure the pool name is valid.
1036          */
1037         if (!zpool_name_valid(hdl, B_TRUE, pool)) {
1038                 (void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1039                     dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1040                     pool);
1041                 return (NULL);
1042         }
1043
1044         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1045                 return (NULL);
1046
1047         zhp->zpool_hdl = hdl;
1048         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1049
1050         if (zpool_refresh_stats(zhp, &missing) != 0) {
1051                 zpool_close(zhp);
1052                 return (NULL);
1053         }
1054
1055         if (missing) {
1056                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
1057                 (void) zfs_error_fmt(hdl, EZFS_NOENT,
1058                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
1059                 zpool_close(zhp);
1060                 return (NULL);
1061         }
1062
1063         return (zhp);
1064 }
1065
1066 /*
1067  * Like the above, but silent on error.  Used when iterating over pools (because
1068  * the configuration cache may be out of date).
1069  */
1070 int
1071 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
1072 {
1073         zpool_handle_t *zhp;
1074         boolean_t missing;
1075
1076         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1077                 return (-1);
1078
1079         zhp->zpool_hdl = hdl;
1080         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1081
1082         if (zpool_refresh_stats(zhp, &missing) != 0) {
1083                 zpool_close(zhp);
1084                 return (-1);
1085         }
1086
1087         if (missing) {
1088                 zpool_close(zhp);
1089                 *ret = NULL;
1090                 return (0);
1091         }
1092
1093         *ret = zhp;
1094         return (0);
1095 }
1096
1097 /*
1098  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
1099  * state.
1100  */
1101 zpool_handle_t *
1102 zpool_open(libzfs_handle_t *hdl, const char *pool)
1103 {
1104         zpool_handle_t *zhp;
1105
1106         if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
1107                 return (NULL);
1108
1109         if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
1110                 (void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
1111                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
1112                 zpool_close(zhp);
1113                 return (NULL);
1114         }
1115
1116         return (zhp);
1117 }
1118
1119 /*
1120  * Close the handle.  Simply frees the memory associated with the handle.
1121  */
1122 void
1123 zpool_close(zpool_handle_t *zhp)
1124 {
1125         nvlist_free(zhp->zpool_config);
1126         nvlist_free(zhp->zpool_old_config);
1127         nvlist_free(zhp->zpool_props);
1128         free(zhp);
1129 }
1130
1131 /*
1132  * Return the name of the pool.
1133  */
1134 const char *
1135 zpool_get_name(zpool_handle_t *zhp)
1136 {
1137         return (zhp->zpool_name);
1138 }
1139
1140
1141 /*
1142  * Return the state of the pool (ACTIVE or UNAVAILABLE)
1143  */
1144 int
1145 zpool_get_state(zpool_handle_t *zhp)
1146 {
1147         return (zhp->zpool_state);
1148 }
1149
1150 /*
1151  * Create the named pool, using the provided vdev list.  It is assumed
1152  * that the consumer has already validated the contents of the nvlist, so we
1153  * don't have to worry about error semantics.
1154  */
1155 int
1156 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
1157     nvlist_t *props, nvlist_t *fsprops)
1158 {
1159         zfs_cmd_t zc = {"\0"};
1160         nvlist_t *zc_fsprops = NULL;
1161         nvlist_t *zc_props = NULL;
1162         char msg[1024];
1163         int ret = -1;
1164
1165         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1166             "cannot create '%s'"), pool);
1167
1168         if (!zpool_name_valid(hdl, B_FALSE, pool))
1169                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
1170
1171         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1172                 return (-1);
1173
1174         if (props) {
1175                 prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
1176
1177                 if ((zc_props = zpool_valid_proplist(hdl, pool, props,
1178                     SPA_VERSION_1, flags, msg)) == NULL) {
1179                         goto create_failed;
1180                 }
1181         }
1182
1183         if (fsprops) {
1184                 uint64_t zoned;
1185                 char *zonestr;
1186
1187                 zoned = ((nvlist_lookup_string(fsprops,
1188                     zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
1189                     strcmp(zonestr, "on") == 0);
1190
1191                 if ((zc_fsprops = zfs_valid_proplist(hdl, ZFS_TYPE_FILESYSTEM,
1192                     fsprops, zoned, NULL, NULL, msg)) == NULL) {
1193                         goto create_failed;
1194                 }
1195                 if (!zc_props &&
1196                     (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
1197                         goto create_failed;
1198                 }
1199                 if (nvlist_add_nvlist(zc_props,
1200                     ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
1201                         goto create_failed;
1202                 }
1203         }
1204
1205         if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
1206                 goto create_failed;
1207
1208         (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
1209
1210         if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
1211
1212                 zcmd_free_nvlists(&zc);
1213                 nvlist_free(zc_props);
1214                 nvlist_free(zc_fsprops);
1215
1216                 switch (errno) {
1217                 case EBUSY:
1218                         /*
1219                          * This can happen if the user has specified the same
1220                          * device multiple times.  We can't reliably detect this
1221                          * until we try to add it and see we already have a
1222                          * label.  This can also happen under if the device is
1223                          * part of an active md or lvm device.
1224                          */
1225                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1226                             "one or more vdevs refer to the same device, or "
1227                             "one of\nthe devices is part of an active md or "
1228                             "lvm device"));
1229                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1230
1231                 case ERANGE:
1232                         /*
1233                          * This happens if the record size is smaller or larger
1234                          * than the allowed size range, or not a power of 2.
1235                          *
1236                          * NOTE: although zfs_valid_proplist is called earlier,
1237                          * this case may have slipped through since the
1238                          * pool does not exist yet and it is therefore
1239                          * impossible to read properties e.g. max blocksize
1240                          * from the pool.
1241                          */
1242                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1243                             "record size invalid"));
1244                         return (zfs_error(hdl, EZFS_BADPROP, msg));
1245
1246                 case EOVERFLOW:
1247                         /*
1248                          * This occurs when one of the devices is below
1249                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1250                          * device was the problem device since there's no
1251                          * reliable way to determine device size from userland.
1252                          */
1253                         {
1254                                 char buf[64];
1255
1256                                 zfs_nicebytes(SPA_MINDEVSIZE, buf,
1257                                     sizeof (buf));
1258
1259                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1260                                     "one or more devices is less than the "
1261                                     "minimum size (%s)"), buf);
1262                         }
1263                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1264
1265                 case ENOSPC:
1266                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1267                             "one or more devices is out of space"));
1268                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1269
1270                 case ENOTBLK:
1271                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1272                             "cache device must be a disk or disk slice"));
1273                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1274
1275                 default:
1276                         return (zpool_standard_error(hdl, errno, msg));
1277                 }
1278         }
1279
1280 create_failed:
1281         zcmd_free_nvlists(&zc);
1282         nvlist_free(zc_props);
1283         nvlist_free(zc_fsprops);
1284         return (ret);
1285 }
1286
1287 /*
1288  * Destroy the given pool.  It is up to the caller to ensure that there are no
1289  * datasets left in the pool.
1290  */
1291 int
1292 zpool_destroy(zpool_handle_t *zhp, const char *log_str)
1293 {
1294         zfs_cmd_t zc = {"\0"};
1295         zfs_handle_t *zfp = NULL;
1296         libzfs_handle_t *hdl = zhp->zpool_hdl;
1297         char msg[1024];
1298
1299         if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1300             (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1301                 return (-1);
1302
1303         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1304         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1305
1306         if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1307                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1308                     "cannot destroy '%s'"), zhp->zpool_name);
1309
1310                 if (errno == EROFS) {
1311                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1312                             "one or more devices is read only"));
1313                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1314                 } else {
1315                         (void) zpool_standard_error(hdl, errno, msg);
1316                 }
1317
1318                 if (zfp)
1319                         zfs_close(zfp);
1320                 return (-1);
1321         }
1322
1323         if (zfp) {
1324                 remove_mountpoint(zfp);
1325                 zfs_close(zfp);
1326         }
1327
1328         return (0);
1329 }
1330
1331 /*
1332  * Add the given vdevs to the pool.  The caller must have already performed the
1333  * necessary verification to ensure that the vdev specification is well-formed.
1334  */
1335 int
1336 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1337 {
1338         zfs_cmd_t zc = {"\0"};
1339         int ret;
1340         libzfs_handle_t *hdl = zhp->zpool_hdl;
1341         char msg[1024];
1342         nvlist_t **spares, **l2cache;
1343         uint_t nspares, nl2cache;
1344
1345         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1346             "cannot add to '%s'"), zhp->zpool_name);
1347
1348         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1349             SPA_VERSION_SPARES &&
1350             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1351             &spares, &nspares) == 0) {
1352                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1353                     "upgraded to add hot spares"));
1354                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1355         }
1356
1357         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1358             SPA_VERSION_L2CACHE &&
1359             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1360             &l2cache, &nl2cache) == 0) {
1361                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1362                     "upgraded to add cache devices"));
1363                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1364         }
1365
1366         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1367                 return (-1);
1368         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1369
1370         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1371                 switch (errno) {
1372                 case EBUSY:
1373                         /*
1374                          * This can happen if the user has specified the same
1375                          * device multiple times.  We can't reliably detect this
1376                          * until we try to add it and see we already have a
1377                          * label.
1378                          */
1379                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1380                             "one or more vdevs refer to the same device"));
1381                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1382                         break;
1383
1384                 case EOVERFLOW:
1385                         /*
1386                          * This occurrs when one of the devices is below
1387                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1388                          * device was the problem device since there's no
1389                          * reliable way to determine device size from userland.
1390                          */
1391                         {
1392                                 char buf[64];
1393
1394                                 zfs_nicebytes(SPA_MINDEVSIZE, buf,
1395                                     sizeof (buf));
1396
1397                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1398                                     "device is less than the minimum "
1399                                     "size (%s)"), buf);
1400                         }
1401                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1402                         break;
1403
1404                 case ENOTSUP:
1405                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1406                             "pool must be upgraded to add these vdevs"));
1407                         (void) zfs_error(hdl, EZFS_BADVERSION, msg);
1408                         break;
1409
1410                 case ENOTBLK:
1411                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1412                             "cache device must be a disk or disk slice"));
1413                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1414                         break;
1415
1416                 default:
1417                         (void) zpool_standard_error(hdl, errno, msg);
1418                 }
1419
1420                 ret = -1;
1421         } else {
1422                 ret = 0;
1423         }
1424
1425         zcmd_free_nvlists(&zc);
1426
1427         return (ret);
1428 }
1429
1430 /*
1431  * Exports the pool from the system.  The caller must ensure that there are no
1432  * mounted datasets in the pool.
1433  */
1434 static int
1435 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce,
1436     const char *log_str)
1437 {
1438         zfs_cmd_t zc = {"\0"};
1439         char msg[1024];
1440
1441         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1442             "cannot export '%s'"), zhp->zpool_name);
1443
1444         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1445         zc.zc_cookie = force;
1446         zc.zc_guid = hardforce;
1447         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1448
1449         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1450                 switch (errno) {
1451                 case EXDEV:
1452                         zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1453                             "use '-f' to override the following errors:\n"
1454                             "'%s' has an active shared spare which could be"
1455                             " used by other pools once '%s' is exported."),
1456                             zhp->zpool_name, zhp->zpool_name);
1457                         return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1458                             msg));
1459                 default:
1460                         return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1461                             msg));
1462                 }
1463         }
1464
1465         return (0);
1466 }
1467
1468 int
1469 zpool_export(zpool_handle_t *zhp, boolean_t force, const char *log_str)
1470 {
1471         return (zpool_export_common(zhp, force, B_FALSE, log_str));
1472 }
1473
1474 int
1475 zpool_export_force(zpool_handle_t *zhp, const char *log_str)
1476 {
1477         return (zpool_export_common(zhp, B_TRUE, B_TRUE, log_str));
1478 }
1479
1480 static void
1481 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1482     nvlist_t *config)
1483 {
1484         nvlist_t *nv = NULL;
1485         uint64_t rewindto;
1486         int64_t loss = -1;
1487         struct tm t;
1488         char timestr[128];
1489
1490         if (!hdl->libzfs_printerr || config == NULL)
1491                 return;
1492
1493         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1494             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0) {
1495                 return;
1496         }
1497
1498         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1499                 return;
1500         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1501
1502         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1503             strftime(timestr, 128, "%c", &t) != 0) {
1504                 if (dryrun) {
1505                         (void) printf(dgettext(TEXT_DOMAIN,
1506                             "Would be able to return %s "
1507                             "to its state as of %s.\n"),
1508                             name, timestr);
1509                 } else {
1510                         (void) printf(dgettext(TEXT_DOMAIN,
1511                             "Pool %s returned to its state as of %s.\n"),
1512                             name, timestr);
1513                 }
1514                 if (loss > 120) {
1515                         (void) printf(dgettext(TEXT_DOMAIN,
1516                             "%s approximately %lld "),
1517                             dryrun ? "Would discard" : "Discarded",
1518                             ((longlong_t)loss + 30) / 60);
1519                         (void) printf(dgettext(TEXT_DOMAIN,
1520                             "minutes of transactions.\n"));
1521                 } else if (loss > 0) {
1522                         (void) printf(dgettext(TEXT_DOMAIN,
1523                             "%s approximately %lld "),
1524                             dryrun ? "Would discard" : "Discarded",
1525                             (longlong_t)loss);
1526                         (void) printf(dgettext(TEXT_DOMAIN,
1527                             "seconds of transactions.\n"));
1528                 }
1529         }
1530 }
1531
1532 void
1533 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1534     nvlist_t *config)
1535 {
1536         nvlist_t *nv = NULL;
1537         int64_t loss = -1;
1538         uint64_t edata = UINT64_MAX;
1539         uint64_t rewindto;
1540         struct tm t;
1541         char timestr[128];
1542
1543         if (!hdl->libzfs_printerr)
1544                 return;
1545
1546         if (reason >= 0)
1547                 (void) printf(dgettext(TEXT_DOMAIN, "action: "));
1548         else
1549                 (void) printf(dgettext(TEXT_DOMAIN, "\t"));
1550
1551         /* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1552         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1553             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0 ||
1554             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1555                 goto no_info;
1556
1557         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1558         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1559             &edata);
1560
1561         (void) printf(dgettext(TEXT_DOMAIN,
1562             "Recovery is possible, but will result in some data loss.\n"));
1563
1564         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1565             strftime(timestr, 128, "%c", &t) != 0) {
1566                 (void) printf(dgettext(TEXT_DOMAIN,
1567                     "\tReturning the pool to its state as of %s\n"
1568                     "\tshould correct the problem.  "),
1569                     timestr);
1570         } else {
1571                 (void) printf(dgettext(TEXT_DOMAIN,
1572                     "\tReverting the pool to an earlier state "
1573                     "should correct the problem.\n\t"));
1574         }
1575
1576         if (loss > 120) {
1577                 (void) printf(dgettext(TEXT_DOMAIN,
1578                     "Approximately %lld minutes of data\n"
1579                     "\tmust be discarded, irreversibly.  "),
1580                     ((longlong_t)loss + 30) / 60);
1581         } else if (loss > 0) {
1582                 (void) printf(dgettext(TEXT_DOMAIN,
1583                     "Approximately %lld seconds of data\n"
1584                     "\tmust be discarded, irreversibly.  "),
1585                     (longlong_t)loss);
1586         }
1587         if (edata != 0 && edata != UINT64_MAX) {
1588                 if (edata == 1) {
1589                         (void) printf(dgettext(TEXT_DOMAIN,
1590                             "After rewind, at least\n"
1591                             "\tone persistent user-data error will remain.  "));
1592                 } else {
1593                         (void) printf(dgettext(TEXT_DOMAIN,
1594                             "After rewind, several\n"
1595                             "\tpersistent user-data errors will remain.  "));
1596                 }
1597         }
1598         (void) printf(dgettext(TEXT_DOMAIN,
1599             "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1600             reason >= 0 ? "clear" : "import", name);
1601
1602         (void) printf(dgettext(TEXT_DOMAIN,
1603             "A scrub of the pool\n"
1604             "\tis strongly recommended after recovery.\n"));
1605         return;
1606
1607 no_info:
1608         (void) printf(dgettext(TEXT_DOMAIN,
1609             "Destroy and re-create the pool from\n\ta backup source.\n"));
1610 }
1611
1612 /*
1613  * zpool_import() is a contracted interface. Should be kept the same
1614  * if possible.
1615  *
1616  * Applications should use zpool_import_props() to import a pool with
1617  * new properties value to be set.
1618  */
1619 int
1620 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1621     char *altroot)
1622 {
1623         nvlist_t *props = NULL;
1624         int ret;
1625
1626         if (altroot != NULL) {
1627                 if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1628                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1629                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1630                             newname));
1631                 }
1632
1633                 if (nvlist_add_string(props,
1634                     zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1635                     nvlist_add_string(props,
1636                     zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1637                         nvlist_free(props);
1638                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1639                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1640                             newname));
1641                 }
1642         }
1643
1644         ret = zpool_import_props(hdl, config, newname, props,
1645             ZFS_IMPORT_NORMAL);
1646         nvlist_free(props);
1647         return (ret);
1648 }
1649
1650 static void
1651 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1652     int indent)
1653 {
1654         nvlist_t **child;
1655         uint_t c, children;
1656         char *vname;
1657         uint64_t is_log = 0;
1658
1659         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1660             &is_log);
1661
1662         if (name != NULL)
1663                 (void) printf("\t%*s%s%s\n", indent, "", name,
1664                     is_log ? " [log]" : "");
1665
1666         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1667             &child, &children) != 0)
1668                 return;
1669
1670         for (c = 0; c < children; c++) {
1671                 vname = zpool_vdev_name(hdl, NULL, child[c], VDEV_NAME_TYPE_ID);
1672                 print_vdev_tree(hdl, vname, child[c], indent + 2);
1673                 free(vname);
1674         }
1675 }
1676
1677 void
1678 zpool_print_unsup_feat(nvlist_t *config)
1679 {
1680         nvlist_t *nvinfo, *unsup_feat;
1681         nvpair_t *nvp;
1682
1683         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nvinfo) ==
1684             0);
1685         verify(nvlist_lookup_nvlist(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT,
1686             &unsup_feat) == 0);
1687
1688         for (nvp = nvlist_next_nvpair(unsup_feat, NULL); nvp != NULL;
1689             nvp = nvlist_next_nvpair(unsup_feat, nvp)) {
1690                 char *desc;
1691
1692                 verify(nvpair_type(nvp) == DATA_TYPE_STRING);
1693                 verify(nvpair_value_string(nvp, &desc) == 0);
1694
1695                 if (strlen(desc) > 0)
1696                         (void) printf("\t%s (%s)\n", nvpair_name(nvp), desc);
1697                 else
1698                         (void) printf("\t%s\n", nvpair_name(nvp));
1699         }
1700 }
1701
1702 /*
1703  * Import the given pool using the known configuration and a list of
1704  * properties to be set. The configuration should have come from
1705  * zpool_find_import(). The 'newname' parameters control whether the pool
1706  * is imported with a different name.
1707  */
1708 int
1709 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1710     nvlist_t *props, int flags)
1711 {
1712         zfs_cmd_t zc = {"\0"};
1713         zpool_rewind_policy_t policy;
1714         nvlist_t *nv = NULL;
1715         nvlist_t *nvinfo = NULL;
1716         nvlist_t *missing = NULL;
1717         char *thename;
1718         char *origname;
1719         int ret;
1720         int error = 0;
1721         char errbuf[1024];
1722
1723         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1724             &origname) == 0);
1725
1726         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1727             "cannot import pool '%s'"), origname);
1728
1729         if (newname != NULL) {
1730                 if (!zpool_name_valid(hdl, B_FALSE, newname))
1731                         return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1732                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1733                             newname));
1734                 thename = (char *)newname;
1735         } else {
1736                 thename = origname;
1737         }
1738
1739         if (props != NULL) {
1740                 uint64_t version;
1741                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1742
1743                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1744                     &version) == 0);
1745
1746                 if ((props = zpool_valid_proplist(hdl, origname,
1747                     props, version, flags, errbuf)) == NULL)
1748                         return (-1);
1749                 if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1750                         nvlist_free(props);
1751                         return (-1);
1752                 }
1753                 nvlist_free(props);
1754         }
1755
1756         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1757
1758         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1759             &zc.zc_guid) == 0);
1760
1761         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1762                 zcmd_free_nvlists(&zc);
1763                 return (-1);
1764         }
1765         if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1766                 zcmd_free_nvlists(&zc);
1767                 return (-1);
1768         }
1769
1770         zc.zc_cookie = flags;
1771         while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1772             errno == ENOMEM) {
1773                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1774                         zcmd_free_nvlists(&zc);
1775                         return (-1);
1776                 }
1777         }
1778         if (ret != 0)
1779                 error = errno;
1780
1781         (void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1782
1783         zcmd_free_nvlists(&zc);
1784
1785         zpool_get_rewind_policy(config, &policy);
1786
1787         if (error) {
1788                 char desc[1024];
1789
1790                 /*
1791                  * Dry-run failed, but we print out what success
1792                  * looks like if we found a best txg
1793                  */
1794                 if (policy.zrp_request & ZPOOL_TRY_REWIND) {
1795                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1796                             B_TRUE, nv);
1797                         nvlist_free(nv);
1798                         return (-1);
1799                 }
1800
1801                 if (newname == NULL)
1802                         (void) snprintf(desc, sizeof (desc),
1803                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1804                             thename);
1805                 else
1806                         (void) snprintf(desc, sizeof (desc),
1807                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1808                             origname, thename);
1809
1810                 switch (error) {
1811                 case ENOTSUP:
1812                         if (nv != NULL && nvlist_lookup_nvlist(nv,
1813                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1814                             nvlist_exists(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT)) {
1815                                 (void) printf(dgettext(TEXT_DOMAIN, "This "
1816                                     "pool uses the following feature(s) not "
1817                                     "supported by this system:\n"));
1818                                 zpool_print_unsup_feat(nv);
1819                                 if (nvlist_exists(nvinfo,
1820                                     ZPOOL_CONFIG_CAN_RDONLY)) {
1821                                         (void) printf(dgettext(TEXT_DOMAIN,
1822                                             "All unsupported features are only "
1823                                             "required for writing to the pool."
1824                                             "\nThe pool can be imported using "
1825                                             "'-o readonly=on'.\n"));
1826                                 }
1827                         }
1828                         /*
1829                          * Unsupported version.
1830                          */
1831                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1832                         break;
1833
1834                 case EINVAL:
1835                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1836                         break;
1837
1838                 case EROFS:
1839                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1840                             "one or more devices is read only"));
1841                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1842                         break;
1843
1844                 case ENXIO:
1845                         if (nv && nvlist_lookup_nvlist(nv,
1846                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1847                             nvlist_lookup_nvlist(nvinfo,
1848                             ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1849                                 (void) printf(dgettext(TEXT_DOMAIN,
1850                                     "The devices below are missing, use "
1851                                     "'-m' to import the pool anyway:\n"));
1852                                 print_vdev_tree(hdl, NULL, missing, 2);
1853                                 (void) printf("\n");
1854                         }
1855                         (void) zpool_standard_error(hdl, error, desc);
1856                         break;
1857
1858                 case EEXIST:
1859                         (void) zpool_standard_error(hdl, error, desc);
1860                         break;
1861
1862                 case EBUSY:
1863                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1864                             "one or more devices are already in use\n"));
1865                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1866                         break;
1867                 case ENAMETOOLONG:
1868                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1869                             "new name of at least one dataset is longer than "
1870                             "the maximum allowable length"));
1871                         (void) zfs_error(hdl, EZFS_NAMETOOLONG, desc);
1872                         break;
1873                 default:
1874                         (void) zpool_standard_error(hdl, error, desc);
1875                         zpool_explain_recover(hdl,
1876                             newname ? origname : thename, -error, nv);
1877                         break;
1878                 }
1879
1880                 nvlist_free(nv);
1881                 ret = -1;
1882         } else {
1883                 zpool_handle_t *zhp;
1884
1885                 /*
1886                  * This should never fail, but play it safe anyway.
1887                  */
1888                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1889                         ret = -1;
1890                 else if (zhp != NULL)
1891                         zpool_close(zhp);
1892                 if (policy.zrp_request &
1893                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1894                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1895                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1896                 }
1897                 nvlist_free(nv);
1898                 return (0);
1899         }
1900
1901         return (ret);
1902 }
1903
1904 /*
1905  * Scan the pool.
1906  */
1907 int
1908 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1909 {
1910         zfs_cmd_t zc = {"\0"};
1911         char msg[1024];
1912         libzfs_handle_t *hdl = zhp->zpool_hdl;
1913
1914         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1915         zc.zc_cookie = func;
1916
1917         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1918             (errno == ENOENT && func != POOL_SCAN_NONE))
1919                 return (0);
1920
1921         if (func == POOL_SCAN_SCRUB) {
1922                 (void) snprintf(msg, sizeof (msg),
1923                     dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1924         } else if (func == POOL_SCAN_NONE) {
1925                 (void) snprintf(msg, sizeof (msg),
1926                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1927                     zc.zc_name);
1928         } else {
1929                 assert(!"unexpected result");
1930         }
1931
1932         if (errno == EBUSY) {
1933                 nvlist_t *nvroot;
1934                 pool_scan_stat_t *ps = NULL;
1935                 uint_t psc;
1936
1937                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
1938                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1939                 (void) nvlist_lookup_uint64_array(nvroot,
1940                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1941                 if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1942                         return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1943                 else
1944                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
1945         } else if (errno == ENOENT) {
1946                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1947         } else {
1948                 return (zpool_standard_error(hdl, errno, msg));
1949         }
1950 }
1951
1952 /*
1953  * Find a vdev that matches the search criteria specified. We use the
1954  * the nvpair name to determine how we should look for the device.
1955  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1956  * spare; but FALSE if its an INUSE spare.
1957  */
1958 static nvlist_t *
1959 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1960     boolean_t *l2cache, boolean_t *log)
1961 {
1962         uint_t c, children;
1963         nvlist_t **child;
1964         nvlist_t *ret;
1965         uint64_t is_log;
1966         char *srchkey;
1967         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
1968
1969         /* Nothing to look for */
1970         if (search == NULL || pair == NULL)
1971                 return (NULL);
1972
1973         /* Obtain the key we will use to search */
1974         srchkey = nvpair_name(pair);
1975
1976         switch (nvpair_type(pair)) {
1977         case DATA_TYPE_UINT64:
1978                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
1979                         uint64_t srchval, theguid;
1980
1981                         verify(nvpair_value_uint64(pair, &srchval) == 0);
1982                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
1983                             &theguid) == 0);
1984                         if (theguid == srchval)
1985                                 return (nv);
1986                 }
1987                 break;
1988
1989         case DATA_TYPE_STRING: {
1990                 char *srchval, *val;
1991
1992                 verify(nvpair_value_string(pair, &srchval) == 0);
1993                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
1994                         break;
1995
1996                 /*
1997                  * Search for the requested value. Special cases:
1998                  *
1999                  * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
2000                  *   "-part1", or "p1".  The suffix is hidden from the user,
2001                  *   but included in the string, so this matches around it.
2002                  * - ZPOOL_CONFIG_PATH for short names zfs_strcmp_shortname()
2003                  *   is used to check all possible expanded paths.
2004                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
2005                  *
2006                  * Otherwise, all other searches are simple string compares.
2007                  */
2008                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0) {
2009                         uint64_t wholedisk = 0;
2010
2011                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
2012                             &wholedisk);
2013                         if (zfs_strcmp_pathname(srchval, val, wholedisk) == 0)
2014                                 return (nv);
2015
2016                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2017                         char *type, *idx, *end, *p;
2018                         uint64_t id, vdev_id;
2019
2020                         /*
2021                          * Determine our vdev type, keeping in mind
2022                          * that the srchval is composed of a type and
2023                          * vdev id pair (i.e. mirror-4).
2024                          */
2025                         if ((type = strdup(srchval)) == NULL)
2026                                 return (NULL);
2027
2028                         if ((p = strrchr(type, '-')) == NULL) {
2029                                 free(type);
2030                                 break;
2031                         }
2032                         idx = p + 1;
2033                         *p = '\0';
2034
2035                         /*
2036                          * If the types don't match then keep looking.
2037                          */
2038                         if (strncmp(val, type, strlen(val)) != 0) {
2039                                 free(type);
2040                                 break;
2041                         }
2042
2043                         verify(strncmp(type, VDEV_TYPE_RAIDZ,
2044                             strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2045                             strncmp(type, VDEV_TYPE_MIRROR,
2046                             strlen(VDEV_TYPE_MIRROR)) == 0);
2047                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
2048                             &id) == 0);
2049
2050                         errno = 0;
2051                         vdev_id = strtoull(idx, &end, 10);
2052
2053                         free(type);
2054                         if (errno != 0)
2055                                 return (NULL);
2056
2057                         /*
2058                          * Now verify that we have the correct vdev id.
2059                          */
2060                         if (vdev_id == id)
2061                                 return (nv);
2062                 }
2063
2064                 /*
2065                  * Common case
2066                  */
2067                 if (strcmp(srchval, val) == 0)
2068                         return (nv);
2069                 break;
2070         }
2071
2072         default:
2073                 break;
2074         }
2075
2076         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2077             &child, &children) != 0)
2078                 return (NULL);
2079
2080         for (c = 0; c < children; c++) {
2081                 if ((ret = vdev_to_nvlist_iter(child[c], search,
2082                     avail_spare, l2cache, NULL)) != NULL) {
2083                         /*
2084                          * The 'is_log' value is only set for the toplevel
2085                          * vdev, not the leaf vdevs.  So we always lookup the
2086                          * log device from the root of the vdev tree (where
2087                          * 'log' is non-NULL).
2088                          */
2089                         if (log != NULL &&
2090                             nvlist_lookup_uint64(child[c],
2091                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2092                             is_log) {
2093                                 *log = B_TRUE;
2094                         }
2095                         return (ret);
2096                 }
2097         }
2098
2099         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2100             &child, &children) == 0) {
2101                 for (c = 0; c < children; c++) {
2102                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2103                             avail_spare, l2cache, NULL)) != NULL) {
2104                                 *avail_spare = B_TRUE;
2105                                 return (ret);
2106                         }
2107                 }
2108         }
2109
2110         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2111             &child, &children) == 0) {
2112                 for (c = 0; c < children; c++) {
2113                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2114                             avail_spare, l2cache, NULL)) != NULL) {
2115                                 *l2cache = B_TRUE;
2116                                 return (ret);
2117                         }
2118                 }
2119         }
2120
2121         return (NULL);
2122 }
2123
2124 /*
2125  * Given a physical path (minus the "/devices" prefix), find the
2126  * associated vdev.
2127  */
2128 nvlist_t *
2129 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2130     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2131 {
2132         nvlist_t *search, *nvroot, *ret;
2133
2134         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2135         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
2136
2137         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2138             &nvroot) == 0);
2139
2140         *avail_spare = B_FALSE;
2141         *l2cache = B_FALSE;
2142         if (log != NULL)
2143                 *log = B_FALSE;
2144         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2145         nvlist_free(search);
2146
2147         return (ret);
2148 }
2149
2150 /*
2151  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2152  */
2153 boolean_t
2154 zpool_vdev_is_interior(const char *name)
2155 {
2156         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2157             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2158                 return (B_TRUE);
2159         return (B_FALSE);
2160 }
2161
2162 nvlist_t *
2163 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2164     boolean_t *l2cache, boolean_t *log)
2165 {
2166         char *end;
2167         nvlist_t *nvroot, *search, *ret;
2168         uint64_t guid;
2169
2170         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2171
2172         guid = strtoull(path, &end, 0);
2173         if (guid != 0 && *end == '\0') {
2174                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
2175         } else if (zpool_vdev_is_interior(path)) {
2176                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
2177         } else {
2178                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
2179         }
2180
2181         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2182             &nvroot) == 0);
2183
2184         *avail_spare = B_FALSE;
2185         *l2cache = B_FALSE;
2186         if (log != NULL)
2187                 *log = B_FALSE;
2188         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2189         nvlist_free(search);
2190
2191         return (ret);
2192 }
2193
2194 static int
2195 vdev_online(nvlist_t *nv)
2196 {
2197         uint64_t ival;
2198
2199         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2200             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2201             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2202                 return (0);
2203
2204         return (1);
2205 }
2206
2207 /*
2208  * Helper function for zpool_get_physpaths().
2209  */
2210 static int
2211 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2212     size_t *bytes_written)
2213 {
2214         size_t bytes_left, pos, rsz;
2215         char *tmppath;
2216         const char *format;
2217
2218         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2219             &tmppath) != 0)
2220                 return (EZFS_NODEVICE);
2221
2222         pos = *bytes_written;
2223         bytes_left = physpath_size - pos;
2224         format = (pos == 0) ? "%s" : " %s";
2225
2226         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2227         *bytes_written += rsz;
2228
2229         if (rsz >= bytes_left) {
2230                 /* if physpath was not copied properly, clear it */
2231                 if (bytes_left != 0) {
2232                         physpath[pos] = 0;
2233                 }
2234                 return (EZFS_NOSPC);
2235         }
2236         return (0);
2237 }
2238
2239 static int
2240 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2241     size_t *rsz, boolean_t is_spare)
2242 {
2243         char *type;
2244         int ret;
2245
2246         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2247                 return (EZFS_INVALCONFIG);
2248
2249         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2250                 /*
2251                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2252                  * For a spare vdev, we only want to boot from the active
2253                  * spare device.
2254                  */
2255                 if (is_spare) {
2256                         uint64_t spare = 0;
2257                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2258                             &spare);
2259                         if (!spare)
2260                                 return (EZFS_INVALCONFIG);
2261                 }
2262
2263                 if (vdev_online(nv)) {
2264                         if ((ret = vdev_get_one_physpath(nv, physpath,
2265                             phypath_size, rsz)) != 0)
2266                                 return (ret);
2267                 }
2268         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2269             strcmp(type, VDEV_TYPE_RAIDZ) == 0 ||
2270             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2271             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2272                 nvlist_t **child;
2273                 uint_t count;
2274                 int i, ret;
2275
2276                 if (nvlist_lookup_nvlist_array(nv,
2277                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2278                         return (EZFS_INVALCONFIG);
2279
2280                 for (i = 0; i < count; i++) {
2281                         ret = vdev_get_physpaths(child[i], physpath,
2282                             phypath_size, rsz, is_spare);
2283                         if (ret == EZFS_NOSPC)
2284                                 return (ret);
2285                 }
2286         }
2287
2288         return (EZFS_POOL_INVALARG);
2289 }
2290
2291 /*
2292  * Get phys_path for a root pool config.
2293  * Return 0 on success; non-zero on failure.
2294  */
2295 static int
2296 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2297 {
2298         size_t rsz;
2299         nvlist_t *vdev_root;
2300         nvlist_t **child;
2301         uint_t count;
2302         char *type;
2303
2304         rsz = 0;
2305
2306         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2307             &vdev_root) != 0)
2308                 return (EZFS_INVALCONFIG);
2309
2310         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2311             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2312             &child, &count) != 0)
2313                 return (EZFS_INVALCONFIG);
2314
2315         /*
2316          * root pool can only have a single top-level vdev.
2317          */
2318         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1)
2319                 return (EZFS_POOL_INVALARG);
2320
2321         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2322             B_FALSE);
2323
2324         /* No online devices */
2325         if (rsz == 0)
2326                 return (EZFS_NODEVICE);
2327
2328         return (0);
2329 }
2330
2331 /*
2332  * Get phys_path for a root pool
2333  * Return 0 on success; non-zero on failure.
2334  */
2335 int
2336 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2337 {
2338         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2339             phypath_size));
2340 }
2341
2342 /*
2343  * If the device has being dynamically expanded then we need to relabel
2344  * the disk to use the new unallocated space.
2345  */
2346 static int
2347 zpool_relabel_disk(libzfs_handle_t *hdl, const char *path, const char *msg)
2348 {
2349         int fd, error;
2350
2351         if ((fd = open(path, O_RDWR|O_DIRECT)) < 0) {
2352                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2353                     "relabel '%s': unable to open device: %d"), path, errno);
2354                 return (zfs_error(hdl, EZFS_OPENFAILED, msg));
2355         }
2356
2357         /*
2358          * It's possible that we might encounter an error if the device
2359          * does not have any unallocated space left. If so, we simply
2360          * ignore that error and continue on.
2361          *
2362          * Also, we don't call efi_rescan() - that would just return EBUSY.
2363          * The module will do it for us in vdev_disk_open().
2364          */
2365         error = efi_use_whole_disk(fd);
2366
2367         /* Flush the buffers to disk and invalidate the page cache. */
2368         (void) fsync(fd);
2369         (void) ioctl(fd, BLKFLSBUF);
2370
2371         (void) close(fd);
2372         if (error && error != VT_ENOSPC) {
2373                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2374                     "relabel '%s': unable to read disk capacity"), path);
2375                 return (zfs_error(hdl, EZFS_NOCAP, msg));
2376         }
2377
2378         return (0);
2379 }
2380
2381 /*
2382  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2383  * ZFS_ONLINE_* flags.
2384  */
2385 int
2386 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2387     vdev_state_t *newstate)
2388 {
2389         zfs_cmd_t zc = {"\0"};
2390         char msg[1024];
2391         nvlist_t *tgt;
2392         boolean_t avail_spare, l2cache, islog;
2393         libzfs_handle_t *hdl = zhp->zpool_hdl;
2394         int error;
2395
2396         if (flags & ZFS_ONLINE_EXPAND) {
2397                 (void) snprintf(msg, sizeof (msg),
2398                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2399         } else {
2400                 (void) snprintf(msg, sizeof (msg),
2401                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2402         }
2403
2404         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2405         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2406             &islog)) == NULL)
2407                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2408
2409         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2410
2411         if (avail_spare)
2412                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2413
2414         if (flags & ZFS_ONLINE_EXPAND ||
2415             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2416                 uint64_t wholedisk = 0;
2417
2418                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2419                     &wholedisk);
2420
2421                 /*
2422                  * XXX - L2ARC 1.0 devices can't support expansion.
2423                  */
2424                 if (l2cache) {
2425                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2426                             "cannot expand cache devices"));
2427                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2428                 }
2429
2430                 if (wholedisk) {
2431                         const char *fullpath = path;
2432                         char buf[MAXPATHLEN];
2433
2434                         if (path[0] != '/') {
2435                                 error = zfs_resolve_shortname(path, buf,
2436                                     sizeof (buf));
2437                                 if (error != 0)
2438                                         return (zfs_error(hdl, EZFS_NODEVICE,
2439                                             msg));
2440
2441                                 fullpath = buf;
2442                         }
2443
2444                         error = zpool_relabel_disk(hdl, fullpath, msg);
2445                         if (error != 0)
2446                                 return (error);
2447                 }
2448         }
2449
2450         zc.zc_cookie = VDEV_STATE_ONLINE;
2451         zc.zc_obj = flags;
2452
2453         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2454                 if (errno == EINVAL) {
2455                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2456                             "from this pool into a new one.  Use '%s' "
2457                             "instead"), "zpool detach");
2458                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2459                 }
2460                 return (zpool_standard_error(hdl, errno, msg));
2461         }
2462
2463         *newstate = zc.zc_cookie;
2464         return (0);
2465 }
2466
2467 /*
2468  * Take the specified vdev offline
2469  */
2470 int
2471 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2472 {
2473         zfs_cmd_t zc = {"\0"};
2474         char msg[1024];
2475         nvlist_t *tgt;
2476         boolean_t avail_spare, l2cache;
2477         libzfs_handle_t *hdl = zhp->zpool_hdl;
2478
2479         (void) snprintf(msg, sizeof (msg),
2480             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2481
2482         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2483         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2484             NULL)) == NULL)
2485                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2486
2487         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2488
2489         if (avail_spare)
2490                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2491
2492         zc.zc_cookie = VDEV_STATE_OFFLINE;
2493         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2494
2495         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2496                 return (0);
2497
2498         switch (errno) {
2499         case EBUSY:
2500
2501                 /*
2502                  * There are no other replicas of this device.
2503                  */
2504                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2505
2506         case EEXIST:
2507                 /*
2508                  * The log device has unplayed logs
2509                  */
2510                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2511
2512         default:
2513                 return (zpool_standard_error(hdl, errno, msg));
2514         }
2515 }
2516
2517 /*
2518  * Mark the given vdev faulted.
2519  */
2520 int
2521 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2522 {
2523         zfs_cmd_t zc = {"\0"};
2524         char msg[1024];
2525         libzfs_handle_t *hdl = zhp->zpool_hdl;
2526
2527         (void) snprintf(msg, sizeof (msg),
2528             dgettext(TEXT_DOMAIN, "cannot fault %llu"), (u_longlong_t)guid);
2529
2530         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2531         zc.zc_guid = guid;
2532         zc.zc_cookie = VDEV_STATE_FAULTED;
2533         zc.zc_obj = aux;
2534
2535         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2536                 return (0);
2537
2538         switch (errno) {
2539         case EBUSY:
2540
2541                 /*
2542                  * There are no other replicas of this device.
2543                  */
2544                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2545
2546         default:
2547                 return (zpool_standard_error(hdl, errno, msg));
2548         }
2549
2550 }
2551
2552 /*
2553  * Mark the given vdev degraded.
2554  */
2555 int
2556 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2557 {
2558         zfs_cmd_t zc = {"\0"};
2559         char msg[1024];
2560         libzfs_handle_t *hdl = zhp->zpool_hdl;
2561
2562         (void) snprintf(msg, sizeof (msg),
2563             dgettext(TEXT_DOMAIN, "cannot degrade %llu"), (u_longlong_t)guid);
2564
2565         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2566         zc.zc_guid = guid;
2567         zc.zc_cookie = VDEV_STATE_DEGRADED;
2568         zc.zc_obj = aux;
2569
2570         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2571                 return (0);
2572
2573         return (zpool_standard_error(hdl, errno, msg));
2574 }
2575
2576 /*
2577  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2578  * a hot spare.
2579  */
2580 static boolean_t
2581 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2582 {
2583         nvlist_t **child;
2584         uint_t c, children;
2585         char *type;
2586
2587         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2588             &children) == 0) {
2589                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2590                     &type) == 0);
2591
2592                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2593                     children == 2 && child[which] == tgt)
2594                         return (B_TRUE);
2595
2596                 for (c = 0; c < children; c++)
2597                         if (is_replacing_spare(child[c], tgt, which))
2598                                 return (B_TRUE);
2599         }
2600
2601         return (B_FALSE);
2602 }
2603
2604 /*
2605  * Attach new_disk (fully described by nvroot) to old_disk.
2606  * If 'replacing' is specified, the new disk will replace the old one.
2607  */
2608 int
2609 zpool_vdev_attach(zpool_handle_t *zhp,
2610     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2611 {
2612         zfs_cmd_t zc = {"\0"};
2613         char msg[1024];
2614         int ret;
2615         nvlist_t *tgt;
2616         boolean_t avail_spare, l2cache, islog;
2617         uint64_t val;
2618         char *newname;
2619         nvlist_t **child;
2620         uint_t children;
2621         nvlist_t *config_root;
2622         libzfs_handle_t *hdl = zhp->zpool_hdl;
2623         boolean_t rootpool = zpool_is_bootable(zhp);
2624
2625         if (replacing)
2626                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2627                     "cannot replace %s with %s"), old_disk, new_disk);
2628         else
2629                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2630                     "cannot attach %s to %s"), new_disk, old_disk);
2631
2632         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2633         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2634             &islog)) == 0)
2635                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2636
2637         if (avail_spare)
2638                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2639
2640         if (l2cache)
2641                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2642
2643         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2644         zc.zc_cookie = replacing;
2645
2646         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2647             &child, &children) != 0 || children != 1) {
2648                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2649                     "new device must be a single disk"));
2650                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2651         }
2652
2653         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2654             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2655
2656         if ((newname = zpool_vdev_name(NULL, NULL, child[0], 0)) == NULL)
2657                 return (-1);
2658
2659         /*
2660          * If the target is a hot spare that has been swapped in, we can only
2661          * replace it with another hot spare.
2662          */
2663         if (replacing &&
2664             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2665             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2666             NULL) == NULL || !avail_spare) &&
2667             is_replacing_spare(config_root, tgt, 1)) {
2668                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2669                     "can only be replaced by another hot spare"));
2670                 free(newname);
2671                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2672         }
2673
2674         free(newname);
2675
2676         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2677                 return (-1);
2678
2679         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2680
2681         zcmd_free_nvlists(&zc);
2682
2683         if (ret == 0) {
2684                 if (rootpool) {
2685                         /*
2686                          * XXX need a better way to prevent user from
2687                          * booting up a half-baked vdev.
2688                          */
2689                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2690                             "sure to wait until resilver is done "
2691                             "before rebooting.\n"));
2692                 }
2693                 return (0);
2694         }
2695
2696         switch (errno) {
2697         case ENOTSUP:
2698                 /*
2699                  * Can't attach to or replace this type of vdev.
2700                  */
2701                 if (replacing) {
2702                         uint64_t version = zpool_get_prop_int(zhp,
2703                             ZPOOL_PROP_VERSION, NULL);
2704
2705                         if (islog)
2706                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2707                                     "cannot replace a log with a spare"));
2708                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2709                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2710                                     "already in replacing/spare config; wait "
2711                                     "for completion or use 'zpool detach'"));
2712                         else
2713                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2714                                     "cannot replace a replacing device"));
2715                 } else {
2716                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2717                             "can only attach to mirrors and top-level "
2718                             "disks"));
2719                 }
2720                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2721                 break;
2722
2723         case EINVAL:
2724                 /*
2725                  * The new device must be a single disk.
2726                  */
2727                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2728                     "new device must be a single disk"));
2729                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2730                 break;
2731
2732         case EBUSY:
2733                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2734                     new_disk);
2735                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2736                 break;
2737
2738         case EOVERFLOW:
2739                 /*
2740                  * The new device is too small.
2741                  */
2742                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2743                     "device is too small"));
2744                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2745                 break;
2746
2747         case EDOM:
2748                 /*
2749                  * The new device has a different optimal sector size.
2750                  */
2751                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2752                     "new device has a different optimal sector size; use the "
2753                     "option '-o ashift=N' to override the optimal size"));
2754                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2755                 break;
2756
2757         case ENAMETOOLONG:
2758                 /*
2759                  * The resulting top-level vdev spec won't fit in the label.
2760                  */
2761                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2762                 break;
2763
2764         default:
2765                 (void) zpool_standard_error(hdl, errno, msg);
2766         }
2767
2768         return (-1);
2769 }
2770
2771 /*
2772  * Detach the specified device.
2773  */
2774 int
2775 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2776 {
2777         zfs_cmd_t zc = {"\0"};
2778         char msg[1024];
2779         nvlist_t *tgt;
2780         boolean_t avail_spare, l2cache;
2781         libzfs_handle_t *hdl = zhp->zpool_hdl;
2782
2783         (void) snprintf(msg, sizeof (msg),
2784             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2785
2786         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2787         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2788             NULL)) == 0)
2789                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2790
2791         if (avail_spare)
2792                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2793
2794         if (l2cache)
2795                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2796
2797         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2798
2799         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2800                 return (0);
2801
2802         switch (errno) {
2803
2804         case ENOTSUP:
2805                 /*
2806                  * Can't detach from this type of vdev.
2807                  */
2808                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2809                     "applicable to mirror and replacing vdevs"));
2810                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2811                 break;
2812
2813         case EBUSY:
2814                 /*
2815                  * There are no other replicas of this device.
2816                  */
2817                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2818                 break;
2819
2820         default:
2821                 (void) zpool_standard_error(hdl, errno, msg);
2822         }
2823
2824         return (-1);
2825 }
2826
2827 /*
2828  * Find a mirror vdev in the source nvlist.
2829  *
2830  * The mchild array contains a list of disks in one of the top-level mirrors
2831  * of the source pool.  The schild array contains a list of disks that the
2832  * user specified on the command line.  We loop over the mchild array to
2833  * see if any entry in the schild array matches.
2834  *
2835  * If a disk in the mchild array is found in the schild array, we return
2836  * the index of that entry.  Otherwise we return -1.
2837  */
2838 static int
2839 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2840     nvlist_t **schild, uint_t schildren)
2841 {
2842         uint_t mc;
2843
2844         for (mc = 0; mc < mchildren; mc++) {
2845                 uint_t sc;
2846                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2847                     mchild[mc], 0);
2848
2849                 for (sc = 0; sc < schildren; sc++) {
2850                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2851                             schild[sc], 0);
2852                         boolean_t result = (strcmp(mpath, spath) == 0);
2853
2854                         free(spath);
2855                         if (result) {
2856                                 free(mpath);
2857                                 return (mc);
2858                         }
2859                 }
2860
2861                 free(mpath);
2862         }
2863
2864         return (-1);
2865 }
2866
2867 /*
2868  * Split a mirror pool.  If newroot points to null, then a new nvlist
2869  * is generated and it is the responsibility of the caller to free it.
2870  */
2871 int
2872 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2873     nvlist_t *props, splitflags_t flags)
2874 {
2875         zfs_cmd_t zc = {"\0"};
2876         char msg[1024];
2877         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2878         nvlist_t **varray = NULL, *zc_props = NULL;
2879         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2880         libzfs_handle_t *hdl = zhp->zpool_hdl;
2881         uint64_t vers;
2882         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2883         int retval = 0;
2884
2885         (void) snprintf(msg, sizeof (msg),
2886             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2887
2888         if (!zpool_name_valid(hdl, B_FALSE, newname))
2889                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2890
2891         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2892                 (void) fprintf(stderr, gettext("Internal error: unable to "
2893                     "retrieve pool configuration\n"));
2894                 return (-1);
2895         }
2896
2897         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2898             == 0);
2899         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2900
2901         if (props) {
2902                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2903                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2904                     props, vers, flags, msg)) == NULL)
2905                         return (-1);
2906         }
2907
2908         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2909             &children) != 0) {
2910                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2911                     "Source pool is missing vdev tree"));
2912                 nvlist_free(zc_props);
2913                 return (-1);
2914         }
2915
2916         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2917         vcount = 0;
2918
2919         if (*newroot == NULL ||
2920             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2921             &newchild, &newchildren) != 0)
2922                 newchildren = 0;
2923
2924         for (c = 0; c < children; c++) {
2925                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2926                 char *type;
2927                 nvlist_t **mchild, *vdev;
2928                 uint_t mchildren;
2929                 int entry;
2930
2931                 /*
2932                  * Unlike cache & spares, slogs are stored in the
2933                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2934                  */
2935                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2936                     &is_log);
2937                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2938                     &is_hole);
2939                 if (is_log || is_hole) {
2940                         /*
2941                          * Create a hole vdev and put it in the config.
2942                          */
2943                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2944                                 goto out;
2945                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2946                             VDEV_TYPE_HOLE) != 0)
2947                                 goto out;
2948                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2949                             1) != 0)
2950                                 goto out;
2951                         if (lastlog == 0)
2952                                 lastlog = vcount;
2953                         varray[vcount++] = vdev;
2954                         continue;
2955                 }
2956                 lastlog = 0;
2957                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
2958                     == 0);
2959                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
2960                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2961                             "Source pool must be composed only of mirrors\n"));
2962                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2963                         goto out;
2964                 }
2965
2966                 verify(nvlist_lookup_nvlist_array(child[c],
2967                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2968
2969                 /* find or add an entry for this top-level vdev */
2970                 if (newchildren > 0 &&
2971                     (entry = find_vdev_entry(zhp, mchild, mchildren,
2972                     newchild, newchildren)) >= 0) {
2973                         /* We found a disk that the user specified. */
2974                         vdev = mchild[entry];
2975                         ++found;
2976                 } else {
2977                         /* User didn't specify a disk for this vdev. */
2978                         vdev = mchild[mchildren - 1];
2979                 }
2980
2981                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
2982                         goto out;
2983         }
2984
2985         /* did we find every disk the user specified? */
2986         if (found != newchildren) {
2987                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
2988                     "include at most one disk from each mirror"));
2989                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2990                 goto out;
2991         }
2992
2993         /* Prepare the nvlist for populating. */
2994         if (*newroot == NULL) {
2995                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
2996                         goto out;
2997                 freelist = B_TRUE;
2998                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
2999                     VDEV_TYPE_ROOT) != 0)
3000                         goto out;
3001         } else {
3002                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
3003         }
3004
3005         /* Add all the children we found */
3006         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
3007             lastlog == 0 ? vcount : lastlog) != 0)
3008                 goto out;
3009
3010         /*
3011          * If we're just doing a dry run, exit now with success.
3012          */
3013         if (flags.dryrun) {
3014                 memory_err = B_FALSE;
3015                 freelist = B_FALSE;
3016                 goto out;
3017         }
3018
3019         /* now build up the config list & call the ioctl */
3020         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3021                 goto out;
3022
3023         if (nvlist_add_nvlist(newconfig,
3024             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3025             nvlist_add_string(newconfig,
3026             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3027             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3028                 goto out;
3029
3030         /*
3031          * The new pool is automatically part of the namespace unless we
3032          * explicitly export it.
3033          */
3034         if (!flags.import)
3035                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3036         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3037         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3038         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
3039                 goto out;
3040         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
3041                 goto out;
3042
3043         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3044                 retval = zpool_standard_error(hdl, errno, msg);
3045                 goto out;
3046         }
3047
3048         freelist = B_FALSE;
3049         memory_err = B_FALSE;
3050
3051 out:
3052         if (varray != NULL) {
3053                 int v;
3054
3055                 for (v = 0; v < vcount; v++)
3056                         nvlist_free(varray[v]);
3057                 free(varray);
3058         }
3059         zcmd_free_nvlists(&zc);
3060         nvlist_free(zc_props);
3061         nvlist_free(newconfig);
3062         if (freelist) {
3063                 nvlist_free(*newroot);
3064                 *newroot = NULL;
3065         }
3066
3067         if (retval != 0)
3068                 return (retval);
3069
3070         if (memory_err)
3071                 return (no_memory(hdl));
3072
3073         return (0);
3074 }
3075
3076 /*
3077  * Remove the given device.  Currently, this is supported only for hot spares,
3078  * cache, and log devices.
3079  */
3080 int
3081 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3082 {
3083         zfs_cmd_t zc = {"\0"};
3084         char msg[1024];
3085         nvlist_t *tgt;
3086         boolean_t avail_spare, l2cache, islog;
3087         libzfs_handle_t *hdl = zhp->zpool_hdl;
3088         uint64_t version;
3089
3090         (void) snprintf(msg, sizeof (msg),
3091             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3092
3093         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3094         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3095             &islog)) == 0)
3096                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3097         /*
3098          * XXX - this should just go away.
3099          */
3100         if (!avail_spare && !l2cache && !islog) {
3101                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3102                     "only inactive hot spares, cache, "
3103                     "or log devices can be removed"));
3104                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3105         }
3106
3107         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3108         if (islog && version < SPA_VERSION_HOLES) {
3109                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3110                     "pool must be upgrade to support log removal"));
3111                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
3112         }
3113
3114         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
3115
3116         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3117                 return (0);
3118
3119         return (zpool_standard_error(hdl, errno, msg));
3120 }
3121
3122 /*
3123  * Clear the errors for the pool, or the particular device if specified.
3124  */
3125 int
3126 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3127 {
3128         zfs_cmd_t zc = {"\0"};
3129         char msg[1024];
3130         nvlist_t *tgt;
3131         zpool_rewind_policy_t policy;
3132         boolean_t avail_spare, l2cache;
3133         libzfs_handle_t *hdl = zhp->zpool_hdl;
3134         nvlist_t *nvi = NULL;
3135         int error;
3136
3137         if (path)
3138                 (void) snprintf(msg, sizeof (msg),
3139                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3140                     path);
3141         else
3142                 (void) snprintf(msg, sizeof (msg),
3143                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3144                     zhp->zpool_name);
3145
3146         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3147         if (path) {
3148                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3149                     &l2cache, NULL)) == 0)
3150                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
3151
3152                 /*
3153                  * Don't allow error clearing for hot spares.  Do allow
3154                  * error clearing for l2cache devices.
3155                  */
3156                 if (avail_spare)
3157                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
3158
3159                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
3160                     &zc.zc_guid) == 0);
3161         }
3162
3163         zpool_get_rewind_policy(rewindnvl, &policy);
3164         zc.zc_cookie = policy.zrp_request;
3165
3166         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
3167                 return (-1);
3168
3169         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
3170                 return (-1);
3171
3172         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3173             errno == ENOMEM) {
3174                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3175                         zcmd_free_nvlists(&zc);
3176                         return (-1);
3177                 }
3178         }
3179
3180         if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
3181             errno != EPERM && errno != EACCES)) {
3182                 if (policy.zrp_request &
3183                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
3184                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3185                         zpool_rewind_exclaim(hdl, zc.zc_name,
3186                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
3187                             nvi);
3188                         nvlist_free(nvi);
3189                 }
3190                 zcmd_free_nvlists(&zc);
3191                 return (0);
3192         }
3193
3194         zcmd_free_nvlists(&zc);
3195         return (zpool_standard_error(hdl, errno, msg));
3196 }
3197
3198 /*
3199  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3200  */
3201 int
3202 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3203 {
3204         zfs_cmd_t zc = {"\0"};
3205         char msg[1024];
3206         libzfs_handle_t *hdl = zhp->zpool_hdl;
3207
3208         (void) snprintf(msg, sizeof (msg),
3209             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3210             (u_longlong_t)guid);
3211
3212         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3213         zc.zc_guid = guid;
3214         zc.zc_cookie = ZPOOL_NO_REWIND;
3215
3216         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
3217                 return (0);
3218
3219         return (zpool_standard_error(hdl, errno, msg));
3220 }
3221
3222 /*
3223  * Change the GUID for a pool.
3224  */
3225 int
3226 zpool_reguid(zpool_handle_t *zhp)
3227 {
3228         char msg[1024];
3229         libzfs_handle_t *hdl = zhp->zpool_hdl;
3230         zfs_cmd_t zc = {"\0"};
3231
3232         (void) snprintf(msg, sizeof (msg),
3233             dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3234
3235         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3236         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3237                 return (0);
3238
3239         return (zpool_standard_error(hdl, errno, msg));
3240 }
3241
3242 /*
3243  * Reopen the pool.
3244  */
3245 int
3246 zpool_reopen(zpool_handle_t *zhp)
3247 {
3248         zfs_cmd_t zc = {"\0"};
3249         char msg[1024];
3250         libzfs_handle_t *hdl = zhp->zpool_hdl;
3251
3252         (void) snprintf(msg, sizeof (msg),
3253             dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3254             zhp->zpool_name);
3255
3256         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3257         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3258                 return (0);
3259         return (zpool_standard_error(hdl, errno, msg));
3260 }
3261
3262 #if defined(__sun__) || defined(__sun)
3263 /*
3264  * Convert from a devid string to a path.
3265  */
3266 static char *
3267 devid_to_path(char *devid_str)
3268 {
3269         ddi_devid_t devid;
3270         char *minor;
3271         char *path;
3272         devid_nmlist_t *list = NULL;
3273         int ret;
3274
3275         if (devid_str_decode(devid_str, &devid, &minor) != 0)
3276                 return (NULL);
3277
3278         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3279
3280         devid_str_free(minor);
3281         devid_free(devid);
3282
3283         if (ret != 0)
3284                 return (NULL);
3285
3286         /*
3287          * In a case the strdup() fails, we will just return NULL below.
3288          */
3289         path = strdup(list[0].devname);
3290
3291         devid_free_nmlist(list);
3292
3293         return (path);
3294 }
3295
3296 /*
3297  * Convert from a path to a devid string.
3298  */
3299 static char *
3300 path_to_devid(const char *path)
3301 {
3302         int fd;
3303         ddi_devid_t devid;
3304         char *minor, *ret;
3305
3306         if ((fd = open(path, O_RDONLY)) < 0)
3307                 return (NULL);
3308
3309         minor = NULL;
3310         ret = NULL;
3311         if (devid_get(fd, &devid) == 0) {
3312                 if (devid_get_minor_name(fd, &minor) == 0)
3313                         ret = devid_str_encode(devid, minor);
3314                 if (minor != NULL)
3315                         devid_str_free(minor);
3316                 devid_free(devid);
3317         }
3318         (void) close(fd);
3319
3320         return (ret);
3321 }
3322
3323 /*
3324  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3325  * ignore any failure here, since a common case is for an unprivileged user to
3326  * type 'zpool status', and we'll display the correct information anyway.
3327  */
3328 static void
3329 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3330 {
3331         zfs_cmd_t zc = {"\0"};
3332
3333         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3334         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3335         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3336             &zc.zc_guid) == 0);
3337
3338         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3339 }
3340 #endif /* sun */
3341
3342 /*
3343  * Remove partition suffix from a vdev path.  Partition suffixes may take three
3344  * forms: "-partX", "pX", or "X", where X is a string of digits.  The second
3345  * case only occurs when the suffix is preceded by a digit, i.e. "md0p0" The
3346  * third case only occurs when preceded by a string matching the regular
3347  * expression "^([hsv]|xv)d[a-z]+", i.e. a scsi, ide, virtio or xen disk.
3348  *
3349  * caller must free the returned string
3350  */
3351 char *
3352 zfs_strip_partition(char *path)
3353 {
3354         char *tmp = strdup(path);
3355         char *part = NULL, *d = NULL;
3356         if (!tmp)
3357                 return (NULL);
3358
3359         if ((part = strstr(tmp, "-part")) && part != tmp) {
3360                 d = part + 5;
3361         } else if ((part = strrchr(tmp, 'p')) &&
3362             part > tmp + 1 && isdigit(*(part-1))) {
3363                 d = part + 1;
3364         } else if ((tmp[0] == 'h' || tmp[0] == 's' || tmp[0] == 'v') &&
3365             tmp[1] == 'd') {
3366                 for (d = &tmp[2]; isalpha(*d); part = ++d) { }
3367         } else if (strncmp("xvd", tmp, 3) == 0) {
3368                 for (d = &tmp[3]; isalpha(*d); part = ++d) { }
3369         }
3370         if (part && d && *d != '\0') {
3371                 for (; isdigit(*d); d++) { }
3372                 if (*d == '\0')
3373                         *part = '\0';
3374         }
3375
3376         return (tmp);
3377 }
3378
3379 /*
3380  * Same as zfs_strip_partition, but allows "/dev/" to be in the pathname
3381  *
3382  * path:        /dev/sda1
3383  * returns:     /dev/sda
3384  *
3385  * Returned string must be freed.
3386  */
3387 char *
3388 zfs_strip_partition_path(char *path)
3389 {
3390         char *newpath = strdup(path);
3391         char *sd_offset;
3392         char *new_sd;
3393
3394         if (!newpath)
3395                 return (NULL);
3396
3397         /* Point to "sda1" part of "/dev/sda1" */
3398         sd_offset = strrchr(newpath, '/') + 1;
3399
3400         /* Get our new name "sda" */
3401         new_sd = zfs_strip_partition(sd_offset);
3402         if (!new_sd) {
3403                 free(newpath);
3404                 return (NULL);
3405         }
3406
3407         /* Paste the "sda" where "sda1" was */
3408         strlcpy(sd_offset, new_sd, strlen(sd_offset) + 1);
3409
3410         /* Free temporary "sda" */
3411         free(new_sd);
3412
3413         return (newpath);
3414 }
3415
3416 #define PATH_BUF_LEN    64
3417
3418 /*
3419  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3420  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3421  * We also check if this is a whole disk, in which case we strip off the
3422  * trailing 's0' slice name.
3423  *
3424  * This routine is also responsible for identifying when disks have been
3425  * reconfigured in a new location.  The kernel will have opened the device by
3426  * devid, but the path will still refer to the old location.  To catch this, we
3427  * first do a path -> devid translation (which is fast for the common case).  If
3428  * the devid matches, we're done.  If not, we do a reverse devid -> path
3429  * translation and issue the appropriate ioctl() to update the path of the vdev.
3430  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3431  * of these checks.
3432  */
3433 char *
3434 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3435     int name_flags)
3436 {
3437         char *path, *type, *env;
3438         uint64_t value;
3439         char buf[PATH_BUF_LEN];
3440         char tmpbuf[PATH_BUF_LEN];
3441
3442         env = getenv("ZPOOL_VDEV_NAME_PATH");
3443         if (env && (strtoul(env, NULL, 0) > 0 ||
3444             !strncasecmp(env, "YES", 3) || !strncasecmp(env, "ON", 2)))
3445                 name_flags |= VDEV_NAME_PATH;
3446
3447         env = getenv("ZPOOL_VDEV_NAME_GUID");
3448         if (env && (strtoul(env, NULL, 0) > 0 ||
3449             !strncasecmp(env, "YES", 3) || !strncasecmp(env, "ON", 2)))
3450                 name_flags |= VDEV_NAME_GUID;
3451
3452         env = getenv("ZPOOL_VDEV_NAME_FOLLOW_LINKS");
3453         if (env && (strtoul(env, NULL, 0) > 0 ||
3454             !strncasecmp(env, "YES", 3) || !strncasecmp(env, "ON", 2)))
3455                 name_flags |= VDEV_NAME_FOLLOW_LINKS;
3456
3457         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &value) == 0 ||
3458             name_flags & VDEV_NAME_GUID) {
3459                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value);
3460                 (void) snprintf(buf, sizeof (buf), "%llu", (u_longlong_t)value);
3461                 path = buf;
3462         } else if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0) {
3463 #if defined(__sun__) || defined(__sun)
3464                 /*
3465                  * Live VDEV path updates to a kernel VDEV during a
3466                  * zpool_vdev_name lookup are not supported on Linux.
3467                  */
3468                 char *devid;
3469                 vdev_stat_t *vs;
3470                 uint_t vsc;
3471
3472                 /*
3473                  * If the device is dead (faulted, offline, etc) then don't
3474                  * bother opening it.  Otherwise we may be forcing the user to
3475                  * open a misbehaving device, which can have undesirable
3476                  * effects.
3477                  */
3478                 if ((nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3479                     (uint64_t **)&vs, &vsc) != 0 ||
3480                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3481                     zhp != NULL &&
3482                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3483                         /*
3484                          * Determine if the current path is correct.
3485                          */
3486                         char *newdevid = path_to_devid(path);
3487
3488                         if (newdevid == NULL ||
3489                             strcmp(devid, newdevid) != 0) {
3490                                 char *newpath;
3491
3492                                 if ((newpath = devid_to_path(devid)) != NULL) {
3493                                         /*
3494                                          * Update the path appropriately.
3495                                          */
3496                                         set_path(zhp, nv, newpath);
3497                                         if (nvlist_add_string(nv,
3498                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3499                                                 verify(nvlist_lookup_string(nv,
3500                                                     ZPOOL_CONFIG_PATH,
3501                                                     &path) == 0);
3502                                         free(newpath);
3503                                 }
3504                         }
3505
3506                         if (newdevid)
3507                                 devid_str_free(newdevid);
3508                 }
3509 #endif /* sun */
3510
3511                 if (name_flags & VDEV_NAME_FOLLOW_LINKS) {
3512                         char *rp = realpath(path, NULL);
3513                         if (rp) {
3514                                 strlcpy(buf, rp, sizeof (buf));
3515                                 path = buf;
3516                                 free(rp);
3517                         }
3518                 }
3519
3520                 /*
3521                  * For a block device only use the name.
3522                  */
3523                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
3524                 if ((strcmp(type, VDEV_TYPE_DISK) == 0) &&
3525                     !(name_flags & VDEV_NAME_PATH)) {
3526                         path = strrchr(path, '/');
3527                         path++;
3528                 }
3529
3530                 /*
3531                  * Remove the partition from the path it this is a whole disk.
3532                  */
3533                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, &value)
3534                     == 0 && value && !(name_flags & VDEV_NAME_PATH)) {
3535                         return (zfs_strip_partition(path));
3536                 }
3537         } else {
3538                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3539
3540                 /*
3541                  * If it's a raidz device, we need to stick in the parity level.
3542                  */
3543                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3544                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3545                             &value) == 0);
3546                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3547                             (u_longlong_t)value);
3548                         path = buf;
3549                 }
3550
3551                 /*
3552                  * We identify each top-level vdev by using a <type-id>
3553                  * naming convention.
3554                  */
3555                 if (name_flags & VDEV_NAME_TYPE_ID) {
3556                         uint64_t id;
3557                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3558                             &id) == 0);
3559                         (void) snprintf(tmpbuf, sizeof (tmpbuf), "%s-%llu",
3560                             path, (u_longlong_t)id);
3561                         path = tmpbuf;
3562                 }
3563         }
3564
3565         return (zfs_strdup(hdl, path));
3566 }
3567
3568 static int
3569 zbookmark_mem_compare(const void *a, const void *b)
3570 {
3571         return (memcmp(a, b, sizeof (zbookmark_phys_t)));
3572 }
3573
3574 /*
3575  * Retrieve the persistent error log, uniquify the members, and return to the
3576  * caller.
3577  */
3578 int
3579 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3580 {
3581         zfs_cmd_t zc = {"\0"};
3582         libzfs_handle_t *hdl = zhp->zpool_hdl;
3583         uint64_t count;
3584         zbookmark_phys_t *zb = NULL;
3585         int i;
3586
3587         /*
3588          * Retrieve the raw error list from the kernel.  If the number of errors
3589          * has increased, allocate more space and continue until we get the
3590          * entire list.
3591          */
3592         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3593             &count) == 0);
3594         if (count == 0)
3595                 return (0);
3596         zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3597             count * sizeof (zbookmark_phys_t));
3598         zc.zc_nvlist_dst_size = count;
3599         (void) strcpy(zc.zc_name, zhp->zpool_name);
3600         for (;;) {
3601                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3602                     &zc) != 0) {
3603                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3604                         if (errno == ENOMEM) {
3605                                 void *dst;
3606
3607                                 count = zc.zc_nvlist_dst_size;
3608                                 dst = zfs_alloc(zhp->zpool_hdl, count *
3609                                     sizeof (zbookmark_phys_t));
3610                                 zc.zc_nvlist_dst = (uintptr_t)dst;
3611                         } else {
3612                                 return (zpool_standard_error_fmt(hdl, errno,
3613                                     dgettext(TEXT_DOMAIN, "errors: List of "
3614                                     "errors unavailable")));
3615                         }
3616                 } else {
3617                         break;
3618                 }
3619         }
3620
3621         /*
3622          * Sort the resulting bookmarks.  This is a little confusing due to the
3623          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3624          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3625          * _not_ copied as part of the process.  So we point the start of our
3626          * array appropriate and decrement the total number of elements.
3627          */
3628         zb = ((zbookmark_phys_t *)(uintptr_t)zc.zc_nvlist_dst) +
3629             zc.zc_nvlist_dst_size;
3630         count -= zc.zc_nvlist_dst_size;
3631
3632         qsort(zb, count, sizeof (zbookmark_phys_t), zbookmark_mem_compare);
3633
3634         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3635
3636         /*
3637          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3638          */
3639         for (i = 0; i < count; i++) {
3640                 nvlist_t *nv;
3641
3642                 /* ignoring zb_blkid and zb_level for now */
3643                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3644                     zb[i-1].zb_object == zb[i].zb_object)
3645                         continue;
3646
3647                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3648                         goto nomem;
3649                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3650                     zb[i].zb_objset) != 0) {
3651                         nvlist_free(nv);
3652                         goto nomem;
3653                 }
3654                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3655                     zb[i].zb_object) != 0) {
3656                         nvlist_free(nv);
3657                         goto nomem;
3658                 }
3659                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3660                         nvlist_free(nv);
3661                         goto nomem;
3662                 }
3663                 nvlist_free(nv);
3664         }
3665
3666         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3667         return (0);
3668
3669 nomem:
3670         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3671         return (no_memory(zhp->zpool_hdl));
3672 }
3673
3674 /*
3675  * Upgrade a ZFS pool to the latest on-disk version.
3676  */
3677 int
3678 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3679 {
3680         zfs_cmd_t zc = {"\0"};
3681         libzfs_handle_t *hdl = zhp->zpool_hdl;
3682
3683         (void) strcpy(zc.zc_name, zhp->zpool_name);
3684         zc.zc_cookie = new_version;
3685
3686         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3687                 return (zpool_standard_error_fmt(hdl, errno,
3688                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3689                     zhp->zpool_name));
3690         return (0);
3691 }
3692
3693 void
3694 zfs_save_arguments(int argc, char **argv, char *string, int len)
3695 {
3696         int i;
3697
3698         (void) strlcpy(string, basename(argv[0]), len);
3699         for (i = 1; i < argc; i++) {
3700                 (void) strlcat(string, " ", len);
3701                 (void) strlcat(string, argv[i], len);
3702         }
3703 }
3704
3705 int
3706 zpool_log_history(libzfs_handle_t *hdl, const char *message)
3707 {
3708         zfs_cmd_t zc = {"\0"};
3709         nvlist_t *args;
3710         int err;
3711
3712         args = fnvlist_alloc();
3713         fnvlist_add_string(args, "message", message);
3714         err = zcmd_write_src_nvlist(hdl, &zc, args);
3715         if (err == 0)
3716                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_LOG_HISTORY, &zc);
3717         nvlist_free(args);
3718         zcmd_free_nvlists(&zc);
3719         return (err);
3720 }
3721
3722 /*
3723  * Perform ioctl to get some command history of a pool.
3724  *
3725  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3726  * logical offset of the history buffer to start reading from.
3727  *
3728  * Upon return, 'off' is the next logical offset to read from and
3729  * 'len' is the actual amount of bytes read into 'buf'.
3730  */
3731 static int
3732 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3733 {
3734         zfs_cmd_t zc = {"\0"};
3735         libzfs_handle_t *hdl = zhp->zpool_hdl;
3736
3737         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3738
3739         zc.zc_history = (uint64_t)(uintptr_t)buf;
3740         zc.zc_history_len = *len;
3741         zc.zc_history_offset = *off;
3742
3743         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3744                 switch (errno) {
3745                 case EPERM:
3746                         return (zfs_error_fmt(hdl, EZFS_PERM,
3747                             dgettext(TEXT_DOMAIN,
3748                             "cannot show history for pool '%s'"),
3749                             zhp->zpool_name));
3750                 case ENOENT:
3751                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3752                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3753                             "'%s'"), zhp->zpool_name));
3754                 case ENOTSUP:
3755                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3756                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3757                             "'%s', pool must be upgraded"), zhp->zpool_name));
3758                 default:
3759                         return (zpool_standard_error_fmt(hdl, errno,
3760                             dgettext(TEXT_DOMAIN,
3761                             "cannot get history for '%s'"), zhp->zpool_name));
3762                 }
3763         }
3764
3765         *len = zc.zc_history_len;
3766         *off = zc.zc_history_offset;
3767
3768         return (0);
3769 }
3770
3771 /*
3772  * Process the buffer of nvlists, unpacking and storing each nvlist record
3773  * into 'records'.  'leftover' is set to the number of bytes that weren't
3774  * processed as there wasn't a complete record.
3775  */
3776 int
3777 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3778     nvlist_t ***records, uint_t *numrecords)
3779 {
3780         uint64_t reclen;
3781         nvlist_t *nv;
3782         int i;
3783         void *tmp;
3784
3785         while (bytes_read > sizeof (reclen)) {
3786
3787                 /* get length of packed record (stored as little endian) */
3788                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3789                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3790
3791                 if (bytes_read < sizeof (reclen) + reclen)
3792                         break;
3793
3794                 /* unpack record */
3795                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3796                         return (ENOMEM);
3797                 bytes_read -= sizeof (reclen) + reclen;
3798                 buf += sizeof (reclen) + reclen;
3799
3800                 /* add record to nvlist array */
3801                 (*numrecords)++;
3802                 if (ISP2(*numrecords + 1)) {
3803                         tmp = realloc(*records,
3804                             *numrecords * 2 * sizeof (nvlist_t *));
3805                         if (tmp == NULL) {
3806                                 nvlist_free(nv);
3807                                 (*numrecords)--;
3808                                 return (ENOMEM);
3809                         }
3810                         *records = tmp;
3811                 }
3812                 (*records)[*numrecords - 1] = nv;
3813         }
3814
3815         *leftover = bytes_read;
3816         return (0);
3817 }
3818
3819 /*
3820  * Retrieve the command history of a pool.
3821  */
3822 int
3823 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3824 {
3825         char *buf;
3826         int buflen = 128 * 1024;
3827         uint64_t off = 0;
3828         nvlist_t **records = NULL;
3829         uint_t numrecords = 0;
3830         int err, i;
3831
3832         buf = malloc(buflen);
3833         if (buf == NULL)
3834                 return (ENOMEM);
3835         do {
3836                 uint64_t bytes_read = buflen;
3837                 uint64_t leftover;
3838
3839                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3840                         break;
3841
3842                 /* if nothing else was read in, we're at EOF, just return */
3843                 if (!bytes_read)
3844                         break;
3845
3846                 if ((err = zpool_history_unpack(buf, bytes_read,
3847                     &leftover, &records, &numrecords)) != 0)
3848                         break;
3849                 off -= leftover;
3850                 if (leftover == bytes_read) {
3851                         /*
3852                          * no progress made, because buffer is not big enough
3853                          * to hold this record; resize and retry.
3854                          */
3855                         buflen *= 2;
3856                         free(buf);
3857                         buf = malloc(buflen);
3858                         if (buf == NULL)
3859                                 return (ENOMEM);
3860                 }
3861
3862                 /* CONSTCOND */
3863         } while (1);
3864
3865         free(buf);
3866
3867         if (!err) {
3868                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3869                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3870                     records, numrecords) == 0);
3871         }
3872         for (i = 0; i < numrecords; i++)
3873                 nvlist_free(records[i]);
3874         free(records);
3875
3876         return (err);
3877 }
3878
3879 /*
3880  * Retrieve the next event given the passed 'zevent_fd' file descriptor.
3881  * If there is a new event available 'nvp' will contain a newly allocated
3882  * nvlist and 'dropped' will be set to the number of missed events since
3883  * the last call to this function.  When 'nvp' is set to NULL it indicates
3884  * no new events are available.  In either case the function returns 0 and
3885  * it is up to the caller to free 'nvp'.  In the case of a fatal error the
3886  * function will return a non-zero value.  When the function is called in
3887  * blocking mode (the default, unless the ZEVENT_NONBLOCK flag is passed),
3888  * it will not return until a new event is available.
3889  */
3890 int
3891 zpool_events_next(libzfs_handle_t *hdl, nvlist_t **nvp,
3892     int *dropped, unsigned flags, int zevent_fd)
3893 {
3894         zfs_cmd_t zc = {"\0"};
3895         int error = 0;
3896
3897         *nvp = NULL;
3898         *dropped = 0;
3899         zc.zc_cleanup_fd = zevent_fd;
3900
3901         if (flags & ZEVENT_NONBLOCK)
3902                 zc.zc_guid = ZEVENT_NONBLOCK;
3903
3904         if (zcmd_alloc_dst_nvlist(hdl, &zc, ZEVENT_SIZE) != 0)
3905                 return (-1);
3906
3907 retry:
3908         if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_NEXT, &zc) != 0) {
3909                 switch (errno) {
3910                 case ESHUTDOWN:
3911                         error = zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
3912                             dgettext(TEXT_DOMAIN, "zfs shutdown"));
3913                         goto out;
3914                 case ENOENT:
3915                         /* Blocking error case should not occur */
3916                         if (!(flags & ZEVENT_NONBLOCK))
3917                                 error = zpool_standard_error_fmt(hdl, errno,
3918                                     dgettext(TEXT_DOMAIN, "cannot get event"));
3919
3920                         goto out;
3921                 case ENOMEM:
3922                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3923                                 error = zfs_error_fmt(hdl, EZFS_NOMEM,
3924                                     dgettext(TEXT_DOMAIN, "cannot get event"));
3925                                 goto out;
3926                         } else {
3927                                 goto retry;
3928                         }
3929                 default:
3930                         error = zpool_standard_error_fmt(hdl, errno,
3931                             dgettext(TEXT_DOMAIN, "cannot get event"));
3932                         goto out;
3933                 }
3934         }
3935
3936         error = zcmd_read_dst_nvlist(hdl, &zc, nvp);
3937         if (error != 0)
3938                 goto out;
3939
3940         *dropped = (int)zc.zc_cookie;
3941 out:
3942         zcmd_free_nvlists(&zc);
3943
3944         return (error);
3945 }
3946
3947 /*
3948  * Clear all events.
3949  */
3950 int
3951 zpool_events_clear(libzfs_handle_t *hdl, int *count)
3952 {
3953         zfs_cmd_t zc = {"\0"};
3954         char msg[1024];
3955
3956         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
3957             "cannot clear events"));
3958
3959         if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_CLEAR, &zc) != 0)
3960                 return (zpool_standard_error_fmt(hdl, errno, msg));
3961
3962         if (count != NULL)
3963                 *count = (int)zc.zc_cookie; /* # of events cleared */
3964
3965         return (0);
3966 }
3967
3968 /*
3969  * Seek to a specific EID, ZEVENT_SEEK_START, or ZEVENT_SEEK_END for
3970  * the passed zevent_fd file handle.  On success zero is returned,
3971  * otherwise -1 is returned and hdl->libzfs_error is set to the errno.
3972  */
3973 int
3974 zpool_events_seek(libzfs_handle_t *hdl, uint64_t eid, int zevent_fd)
3975 {
3976         zfs_cmd_t zc = {"\0"};
3977         int error = 0;
3978
3979         zc.zc_guid = eid;
3980         zc.zc_cleanup_fd = zevent_fd;
3981
3982         if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_SEEK, &zc) != 0) {
3983                 switch (errno) {
3984                 case ENOENT:
3985                         error = zfs_error_fmt(hdl, EZFS_NOENT,
3986                             dgettext(TEXT_DOMAIN, "cannot get event"));
3987                         break;
3988
3989                 case ENOMEM:
3990                         error = zfs_error_fmt(hdl, EZFS_NOMEM,
3991                             dgettext(TEXT_DOMAIN, "cannot get event"));
3992                         break;
3993
3994                 default:
3995                         error = zpool_standard_error_fmt(hdl, errno,
3996                             dgettext(TEXT_DOMAIN, "cannot get event"));
3997                         break;
3998                 }
3999         }
4000
4001         return (error);
4002 }
4003
4004 void
4005 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
4006     char *pathname, size_t len)
4007 {
4008         zfs_cmd_t zc = {"\0"};
4009         boolean_t mounted = B_FALSE;
4010         char *mntpnt = NULL;
4011         char dsname[ZFS_MAX_DATASET_NAME_LEN];
4012
4013         if (dsobj == 0) {
4014                 /* special case for the MOS */
4015                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>",
4016                     (longlong_t)obj);
4017                 return;
4018         }
4019
4020         /* get the dataset's name */
4021         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
4022         zc.zc_obj = dsobj;
4023         if (ioctl(zhp->zpool_hdl->libzfs_fd,
4024             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
4025                 /* just write out a path of two object numbers */
4026                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
4027                     (longlong_t)dsobj, (longlong_t)obj);
4028                 return;
4029         }
4030         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
4031
4032         /* find out if the dataset is mounted */
4033         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
4034
4035         /* get the corrupted object's path */
4036         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
4037         zc.zc_obj = obj;
4038         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
4039             &zc) == 0) {
4040                 if (mounted) {
4041                         (void) snprintf(pathname, len, "%s%s", mntpnt,
4042                             zc.zc_value);
4043                 } else {
4044                         (void) snprintf(pathname, len, "%s:%s",
4045                             dsname, zc.zc_value);
4046                 }
4047         } else {
4048                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname,
4049                     (longlong_t)obj);
4050         }
4051         free(mntpnt);
4052 }
4053
4054 /*
4055  * Read the EFI label from the config, if a label does not exist then
4056  * pass back the error to the caller. If the caller has passed a non-NULL
4057  * diskaddr argument then we set it to the starting address of the EFI
4058  * partition.
4059  */
4060 static int
4061 read_efi_label(nvlist_t *config, diskaddr_t *sb)
4062 {
4063         char *path;
4064         int fd;
4065         char diskname[MAXPATHLEN];
4066         int err = -1;
4067
4068         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
4069                 return (err);
4070
4071         (void) snprintf(diskname, sizeof (diskname), "%s%s", DISK_ROOT,
4072             strrchr(path, '/'));
4073         if ((fd = open(diskname, O_RDONLY|O_DIRECT)) >= 0) {
4074                 struct dk_gpt *vtoc;
4075
4076                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
4077                         if (sb != NULL)
4078                                 *sb = vtoc->efi_parts[0].p_start;
4079                         efi_free(vtoc);
4080                 }
4081                 (void) close(fd);
4082         }
4083         return (err);
4084 }
4085
4086 /*
4087  * determine where a partition starts on a disk in the current
4088  * configuration
4089  */
4090 static diskaddr_t
4091 find_start_block(nvlist_t *config)
4092 {
4093         nvlist_t **child;
4094         uint_t c, children;
4095         diskaddr_t sb = MAXOFFSET_T;
4096         uint64_t wholedisk;
4097
4098         if (nvlist_lookup_nvlist_array(config,
4099             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
4100                 if (nvlist_lookup_uint64(config,
4101                     ZPOOL_CONFIG_WHOLE_DISK,
4102                     &wholedisk) != 0 || !wholedisk) {
4103                         return (MAXOFFSET_T);
4104                 }
4105                 if (read_efi_label(config, &sb) < 0)
4106                         sb = MAXOFFSET_T;
4107                 return (sb);
4108         }
4109
4110         for (c = 0; c < children; c++) {
4111                 sb = find_start_block(child[c]);
4112                 if (sb != MAXOFFSET_T) {
4113                         return (sb);
4114                 }
4115         }
4116         return (MAXOFFSET_T);
4117 }
4118
4119 static int
4120 zpool_label_disk_check(char *path)
4121 {
4122         struct dk_gpt *vtoc;
4123         int fd, err;
4124
4125         if ((fd = open(path, O_RDONLY|O_DIRECT)) < 0)
4126                 return (errno);
4127
4128         if ((err = efi_alloc_and_read(fd, &vtoc)) != 0) {
4129                 (void) close(fd);
4130                 return (err);
4131         }
4132
4133         if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
4134                 efi_free(vtoc);
4135                 (void) close(fd);
4136                 return (EIDRM);
4137         }
4138
4139         efi_free(vtoc);
4140         (void) close(fd);
4141         return (0);
4142 }
4143
4144 /*
4145  * Generate a unique partition name for the ZFS member.  Partitions must
4146  * have unique names to ensure udev will be able to create symlinks under
4147  * /dev/disk/by-partlabel/ for all pool members.  The partition names are
4148  * of the form <pool>-<unique-id>.
4149  */
4150 static void
4151 zpool_label_name(char *label_name, int label_size)
4152 {
4153         uint64_t id = 0;
4154         int fd;
4155
4156         fd = open("/dev/urandom", O_RDONLY);
4157         if (fd >= 0) {
4158                 if (read(fd, &id, sizeof (id)) != sizeof (id))
4159                         id = 0;
4160
4161                 close(fd);
4162         }
4163
4164         if (id == 0)
4165                 id = (((uint64_t)rand()) << 32) | (uint64_t)rand();
4166
4167         snprintf(label_name, label_size, "zfs-%016llx", (u_longlong_t)id);
4168 }
4169
4170 /*
4171  * Label an individual disk.  The name provided is the short name,
4172  * stripped of any leading /dev path.
4173  */
4174 int
4175 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, char *name)
4176 {
4177         char path[MAXPATHLEN];
4178         struct dk_gpt *vtoc;
4179         int rval, fd;
4180         size_t resv = EFI_MIN_RESV_SIZE;
4181         uint64_t slice_size;
4182         diskaddr_t start_block;
4183         char errbuf[1024];
4184
4185         /* prepare an error message just in case */
4186         (void) snprintf(errbuf, sizeof (errbuf),
4187             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
4188
4189         if (zhp) {
4190                 nvlist_t *nvroot;
4191
4192                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
4193                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4194
4195                 if (zhp->zpool_start_block == 0)
4196                         start_block = find_start_block(nvroot);
4197                 else
4198                         start_block = zhp->zpool_start_block;
4199                 zhp->zpool_start_block = start_block;
4200         } else {
4201                 /* new pool */
4202                 start_block = NEW_START_BLOCK;
4203         }
4204
4205         (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT, name);
4206
4207         if ((fd = open(path, O_RDWR|O_DIRECT|O_EXCL)) < 0) {
4208                 /*
4209                  * This shouldn't happen.  We've long since verified that this
4210                  * is a valid device.
4211                  */
4212                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
4213                     "label '%s': unable to open device: %d"), path, errno);
4214                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
4215         }
4216
4217         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
4218                 /*
4219                  * The only way this can fail is if we run out of memory, or we
4220                  * were unable to read the disk's capacity
4221                  */
4222                 if (errno == ENOMEM)
4223                         (void) no_memory(hdl);
4224
4225                 (void) close(fd);
4226                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
4227                     "label '%s': unable to read disk capacity"), path);
4228
4229                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
4230         }
4231
4232         slice_size = vtoc->efi_last_u_lba + 1;
4233         slice_size -= EFI_MIN_RESV_SIZE;
4234         if (start_block == MAXOFFSET_T)
4235                 start_block = NEW_START_BLOCK;
4236         slice_size -= start_block;
4237         slice_size = P2ALIGN(slice_size, PARTITION_END_ALIGNMENT);
4238
4239         vtoc->efi_parts[0].p_start = start_block;
4240         vtoc->efi_parts[0].p_size = slice_size;
4241
4242         /*
4243          * Why we use V_USR: V_BACKUP confuses users, and is considered
4244          * disposable by some EFI utilities (since EFI doesn't have a backup
4245          * slice).  V_UNASSIGNED is supposed to be used only for zero size
4246          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
4247          * etc. were all pretty specific.  V_USR is as close to reality as we
4248          * can get, in the absence of V_OTHER.
4249          */
4250         vtoc->efi_parts[0].p_tag = V_USR;
4251         zpool_label_name(vtoc->efi_parts[0].p_name, EFI_PART_NAME_LEN);
4252
4253         vtoc->efi_parts[8].p_start = slice_size + start_block;
4254         vtoc->efi_parts[8].p_size = resv;
4255         vtoc->efi_parts[8].p_tag = V_RESERVED;
4256
4257         rval = efi_write(fd, vtoc);
4258
4259         /* Flush the buffers to disk and invalidate the page cache. */
4260         (void) fsync(fd);
4261         (void) ioctl(fd, BLKFLSBUF);
4262
4263         if (rval == 0)
4264                 rval = efi_rescan(fd);
4265
4266         /*
4267          * Some block drivers (like pcata) may not support EFI GPT labels.
4268          * Print out a helpful error message directing the user to manually
4269          * label the disk and give a specific slice.
4270          */
4271         if (rval != 0) {
4272                 (void) close(fd);
4273                 efi_free(vtoc);
4274
4275                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "try using "
4276                     "parted(8) and then provide a specific slice: %d"), rval);
4277                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4278         }
4279
4280         (void) close(fd);
4281         efi_free(vtoc);
4282
4283         (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT, name);
4284         (void) zfs_append_partition(path, MAXPATHLEN);
4285
4286         /* Wait to udev to signal use the device has settled. */
4287         rval = zpool_label_disk_wait(path, DISK_LABEL_WAIT);
4288         if (rval) {
4289                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "failed to "
4290                     "detect device partitions on '%s': %d"), path, rval);
4291                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4292         }
4293
4294         /* We can't be to paranoid.  Read the label back and verify it. */
4295         (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT, name);
4296         rval = zpool_label_disk_check(path);
4297         if (rval) {
4298                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "freshly written "
4299                     "EFI label on '%s' is damaged.  Ensure\nthis device "
4300                     "is not in in use, and is functioning properly: %d"),
4301                     path, rval);
4302                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4303         }
4304
4305         return (0);
4306 }
4307
4308 /*
4309  * Allocate and return the underlying device name for a device mapper device.
4310  * If a device mapper device maps to multiple devices, return the first device.
4311  *
4312  * For example, dm_name = "/dev/dm-0" could return "/dev/sda". Symlinks to a
4313  * DM device (like /dev/disk/by-vdev/A0) are also allowed.
4314  *
4315  * Returns device name, or NULL on error or no match.  If dm_name is not a DM
4316  * device then return NULL.
4317  *
4318  * NOTE: The returned name string must be *freed*.
4319  */
4320 char *
4321 dm_get_underlying_path(char *dm_name)
4322 {
4323         DIR *dp = NULL;
4324         struct dirent *ep;
4325         char *realp;
4326         char *tmp = NULL;
4327         char *path = NULL;
4328         char *dev_str;
4329         int size;
4330
4331         if (dm_name == NULL)
4332                 return (NULL);
4333
4334         /* dm name may be a symlink (like /dev/disk/by-vdev/A0) */
4335         realp = realpath(dm_name, NULL);
4336         if (realp == NULL)
4337                 return (NULL);
4338
4339         /*
4340          * If they preface 'dev' with a path (like "/dev") then strip it off.
4341          * We just want the 'dm-N' part.
4342          */
4343         tmp = strrchr(realp, '/');
4344         if (tmp != NULL)
4345                 dev_str = tmp + 1;    /* +1 since we want the chr after '/' */
4346         else
4347                 dev_str = tmp;
4348
4349         size = asprintf(&tmp, "/sys/block/%s/slaves/", dev_str);
4350         if (size == -1 || !tmp)
4351                 goto end;
4352
4353         dp = opendir(tmp);
4354         if (dp == NULL)
4355                 goto end;
4356
4357         /* Return first sd* entry in /sys/block/dm-N/slaves/ */
4358         while ((ep = readdir(dp))) {
4359                 if (ep->d_type != DT_DIR) {     /* skip "." and ".." dirs */
4360                         size = asprintf(&path, "/dev/%s", ep->d_name);
4361                         break;
4362                 }
4363         }
4364
4365 end:
4366         if (dp != NULL)
4367                 closedir(dp);
4368         free(tmp);
4369         free(realp);
4370         return (path);
4371 }
4372
4373 /*
4374  * Return 1 if device is a device mapper or multipath device.
4375  * Return 0 if not.
4376  */
4377 int
4378 zfs_dev_is_dm(char *dev_name)
4379 {
4380
4381         char *tmp;
4382         tmp = dm_get_underlying_path(dev_name);
4383         if (tmp == NULL)
4384                 return (0);
4385
4386         free(tmp);
4387         return (1);
4388 }
4389
4390 /*
4391  * By "whole disk" we mean an entire physical disk (something we can
4392  * label, toggle the write cache on, etc.) as opposed to the full
4393  * capacity of a pseudo-device such as lofi or did.  We act as if we
4394  * are labeling the disk, which should be a pretty good test of whether
4395  * it's a viable device or not.  Returns B_TRUE if it is and B_FALSE if
4396  * it isn't.
4397  */
4398 int
4399 zfs_dev_is_whole_disk(char *dev_name)
4400 {
4401         struct dk_gpt *label;
4402         int fd;
4403
4404         if ((fd = open(dev_name, O_RDONLY | O_DIRECT)) < 0)
4405                 return (0);
4406
4407         if (efi_alloc_and_init(fd, EFI_NUMPAR, &label) != 0) {
4408                 (void) close(fd);
4409                 return (0);
4410         }
4411
4412         efi_free(label);
4413         (void) close(fd);
4414
4415         return (1);
4416 }
4417
4418 /*
4419  * Lookup the underlying device for a device name
4420  *
4421  * Often you'll have a symlink to a device, a partition device,
4422  * or a multipath device, and want to look up the underlying device.
4423  * This function returns the underlying device name.  If the device
4424  * name is already the underlying device, then just return the same
4425  * name.  If the device is a DM device with multiple underlying devices
4426  * then return the first one.
4427  *
4428  * For example:
4429  *
4430  * 1. /dev/disk/by-id/ata-QEMU_HARDDISK_QM00001 -> ../../sda
4431  * dev_name:    /dev/disk/by-id/ata-QEMU_HARDDISK_QM00001
4432  * returns:     /dev/sda
4433  *
4434  * 2. /dev/mapper/mpatha (made up of /dev/sda and /dev/sdb)
4435  * dev_name:    /dev/mapper/mpatha
4436  * returns:     /dev/sda (first device)
4437  *
4438  * 3. /dev/sda (already the underlying device)
4439  * dev_name:    /dev/sda
4440  * returns:     /dev/sda
4441  *
4442  * 4. /dev/dm-3 (mapped to /dev/sda)
4443  * dev_name:    /dev/dm-3
4444  * returns:     /dev/sda
4445  *
4446  * 5. /dev/disk/by-id/scsi-0QEMU_drive-scsi0-0-0-0-part9 -> ../../sdb9
4447  * dev_name:    /dev/disk/by-id/scsi-0QEMU_drive-scsi0-0-0-0-part9
4448  * returns:     /dev/sdb
4449  *
4450  * 6. /dev/disk/by-uuid/5df030cf-3cd9-46e4-8e99-3ccb462a4e9a -> ../dev/sda2
4451  * dev_name:    /dev/disk/by-uuid/5df030cf-3cd9-46e4-8e99-3ccb462a4e9a
4452  * returns:     /dev/sda
4453  *
4454  * Returns underlying device name, or NULL on error or no match.
4455  *
4456  * NOTE: The returned name string must be *freed*.
4457  */
4458 char *
4459 zfs_get_underlying_path(char *dev_name)
4460 {
4461         char *name = NULL;
4462         char *tmp;
4463
4464         if (dev_name == NULL)
4465                 return (NULL);
4466
4467         tmp = dm_get_underlying_path(dev_name);
4468
4469         /* dev_name not a DM device, so just un-symlinkize it */
4470         if (tmp == NULL)
4471                 tmp = realpath(dev_name, NULL);
4472
4473         if (tmp != NULL) {
4474                 name = zfs_strip_partition_path(tmp);
4475                 free(tmp);
4476         }
4477
4478         return (name);
4479 }
4480
4481 /*
4482  * Given a dev name like "sda", return the full enclosure sysfs path to
4483  * the disk.  You can also pass in the name with "/dev" prepended
4484  * to it (like /dev/sda).
4485  *
4486  * For example, disk "sda" in enclosure slot 1:
4487  *     dev:            "sda"
4488  *     returns:        "/sys/class/enclosure/1:0:3:0/Slot 1"
4489  *
4490  * 'dev' must be a non-devicemapper device.
4491  *
4492  * Returned string must be freed.
4493  */
4494 char *
4495 zfs_get_enclosure_sysfs_path(char *dev_name)
4496 {
4497         DIR *dp = NULL;
4498         struct dirent *ep;
4499         char buf[MAXPATHLEN];
4500         char *tmp1 = NULL;
4501         char *tmp2 = NULL;
4502         char *tmp3 = NULL;
4503         char *path = NULL;
4504         size_t size;
4505         int tmpsize;
4506
4507         if (dev_name == NULL)
4508                 return (NULL);
4509
4510         /* If they preface 'dev' with a path (like "/dev") then strip it off */
4511         tmp1 = strrchr(dev_name, '/');
4512         if (tmp1 != NULL)
4513                 dev_name = tmp1 + 1;    /* +1 since we want the chr after '/' */
4514
4515         tmpsize = asprintf(&tmp1, "/sys/block/%s/device", dev_name);
4516         if (tmpsize == -1 || tmp1 == NULL) {
4517                 tmp1 = NULL;
4518                 goto end;
4519         }
4520
4521         dp = opendir(tmp1);
4522         if (dp == NULL) {
4523                 tmp1 = NULL;    /* To make free() at the end a NOP */
4524                 goto end;
4525         }
4526
4527         /*
4528          * Look though all sysfs entries in /sys/block/<dev>/device for
4529          * the enclosure symlink.
4530          */
4531         while ((ep = readdir(dp))) {
4532                 /* Ignore everything that's not our enclosure_device link */
4533                 if (strstr(ep->d_name, "enclosure_device") == NULL)
4534                         continue;
4535
4536                 if (asprintf(&tmp2, "%s/%s", tmp1, ep->d_name) == -1 ||
4537                     tmp2 == NULL)
4538                         break;
4539
4540                 size = readlink(tmp2, buf, sizeof (buf));
4541
4542                 /* Did readlink fail or crop the link name? */
4543                 if (size == -1 || size >= sizeof (buf)) {
4544                         free(tmp2);
4545                         tmp2 = NULL;    /* To make free() at the end a NOP */
4546                         break;
4547                 }
4548
4549                 /*
4550                  * We got a valid link.  readlink() doesn't terminate strings
4551                  * so we have to do it.
4552                  */
4553                 buf[size] = '\0';
4554
4555                 /*
4556                  * Our link will look like:
4557                  *
4558                  * "../../../../port-11:1:2/..STUFF../enclosure/1:0:3:0/SLOT 1"
4559                  *
4560                  * We want to grab the "enclosure/1:0:3:0/SLOT 1" part
4561                  */
4562                 tmp3 = strstr(buf, "enclosure");
4563                 if (tmp3 == NULL)
4564                         break;
4565
4566                 if (asprintf(&path, "/sys/class/%s", tmp3) == -1) {
4567                         /* If asprintf() fails, 'path' is undefined */
4568                         path = NULL;
4569                         break;
4570                 }
4571
4572                 if (path == NULL)
4573                         break;
4574         }
4575
4576 end:
4577         free(tmp2);
4578         free(tmp1);
4579
4580         if (dp != NULL)
4581                 closedir(dp);
4582
4583         return (path);
4584 }