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

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
  25  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  26  * Copyright 2016 Nexenta Systems, Inc.
  27  * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>
  28  * Copyright (c) 2017 Datto Inc.
  29  * Copyright (c) 2017, Intel Corporation.
  30  */
  31
  32 #include <sys/types.h>
  33 #include <sys/stat.h>
  34 #include <ctype.h>
  35 #include <errno.h>
  36 #include <devid.h>
  37 #include <fcntl.h>
  38 #include <libintl.h>
  39 #include <stdio.h>
  40 #include <stdlib.h>
  41 #include <strings.h>
  42 #include <unistd.h>
  43 #include <libgen.h>
  44 #include <sys/zfs_ioctl.h>
  45 #include <dlfcn.h>
  46
  47 #include "zfs_namecheck.h"
  48 #include "zfs_prop.h"
  49 #include "libzfs_impl.h"
  50 #include "zfs_comutil.h"
  51 #include "zfeature_common.h"
  52
  53 static int read_efi_label(nvlist_t *, diskaddr_t *, boolean_t *);
  54 static boolean_t zpool_vdev_is_interior(const char *name);
  55
  56 #define BACKUP_SLICE    "s2"
  57
  58 typedef struct prop_flags {
  59         int create:1;   /* Validate property on creation */
  60         int import:1;   /* Validate property on import */
  61 } prop_flags_t;
  62
  63 /*
  64  * ====================================================================
  65  *   zpool property functions
  66  * ====================================================================
  67  */
  68
  69 static int
  70 zpool_get_all_props(zpool_handle_t *zhp)
  71 {
  72         zfs_cmd_t zc = { 0 };
  73         libzfs_handle_t *hdl = zhp->zpool_hdl;
  74
  75         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
  76
  77         if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
  78                 return (-1);
  79
  80         while (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
  81                 if (errno == ENOMEM) {
  82                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
  83                                 zcmd_free_nvlists(&zc);
  84                                 return (-1);
  85                         }
  86                 } else {
  87                         zcmd_free_nvlists(&zc);
  88                         return (-1);
  89                 }
  90         }
  91
  92         if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
  93                 zcmd_free_nvlists(&zc);
  94                 return (-1);
  95         }
  96
  97         zcmd_free_nvlists(&zc);
  98
  99         return (0);
 100 }
 101
 102 static int
 103 zpool_props_refresh(zpool_handle_t *zhp)
 104 {
 105         nvlist_t *old_props;
 106
 107         old_props = zhp->zpool_props;
 108
 109         if (zpool_get_all_props(zhp) != 0)
 110                 return (-1);
 111
 112         nvlist_free(old_props);
 113         return (0);
 114 }
 115
 116 static char *
 117 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
 118     zprop_source_t *src)
 119 {
 120         nvlist_t *nv, *nvl;
 121         uint64_t ival;
 122         char *value;
 123         zprop_source_t source;
 124
 125         nvl = zhp->zpool_props;
 126         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 127                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &ival) == 0);
 128                 source = ival;
 129                 verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
 130         } else {
 131                 source = ZPROP_SRC_DEFAULT;
 132                 if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
 133                         value = "-";
 134         }
 135
 136         if (src)
 137                 *src = source;
 138
 139         return (value);
 140 }
 141
 142 uint64_t
 143 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
 144 {
 145         nvlist_t *nv, *nvl;
 146         uint64_t value;
 147         zprop_source_t source;
 148
 149         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
 150                 /*
 151                  * zpool_get_all_props() has most likely failed because
 152                  * the pool is faulted, but if all we need is the top level
 153                  * vdev's guid then get it from the zhp config nvlist.
 154                  */
 155                 if ((prop == ZPOOL_PROP_GUID) &&
 156                     (nvlist_lookup_nvlist(zhp->zpool_config,
 157                     ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
 158                     (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
 159                     == 0)) {
 160                         return (value);
 161                 }
 162                 return (zpool_prop_default_numeric(prop));
 163         }
 164
 165         nvl = zhp->zpool_props;
 166         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 167                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &value) == 0);
 168                 source = value;
 169                 verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
 170         } else {
 171                 source = ZPROP_SRC_DEFAULT;
 172                 value = zpool_prop_default_numeric(prop);
 173         }
 174
 175         if (src)
 176                 *src = source;
 177
 178         return (value);
 179 }
 180
 181 /*
 182  * Map VDEV STATE to printed strings.
 183  */
 184 const char *
 185 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
 186 {
 187         switch (state) {
 188         case VDEV_STATE_CLOSED:
 189         case VDEV_STATE_OFFLINE:
 190                 return (gettext("OFFLINE"));
 191         case VDEV_STATE_REMOVED:
 192                 return (gettext("REMOVED"));
 193         case VDEV_STATE_CANT_OPEN:
 194                 if (aux == VDEV_AUX_CORRUPT_DATA || aux == VDEV_AUX_BAD_LOG)
 195                         return (gettext("FAULTED"));
 196                 else if (aux == VDEV_AUX_SPLIT_POOL)
 197                         return (gettext("SPLIT"));
 198                 else
 199                         return (gettext("UNAVAIL"));
 200         case VDEV_STATE_FAULTED:
 201                 return (gettext("FAULTED"));
 202         case VDEV_STATE_DEGRADED:
 203                 return (gettext("DEGRADED"));
 204         case VDEV_STATE_HEALTHY:
 205                 return (gettext("ONLINE"));
 206
 207         default:
 208                 break;
 209         }
 210
 211         return (gettext("UNKNOWN"));
 212 }
 213
 214 /*
 215  * Map POOL STATE to printed strings.
 216  */
 217 const char *
 218 zpool_pool_state_to_name(pool_state_t state)
 219 {
 220         switch (state) {
 221         case POOL_STATE_ACTIVE:
 222                 return (gettext("ACTIVE"));
 223         case POOL_STATE_EXPORTED:
 224                 return (gettext("EXPORTED"));
 225         case POOL_STATE_DESTROYED:
 226                 return (gettext("DESTROYED"));
 227         case POOL_STATE_SPARE:
 228                 return (gettext("SPARE"));
 229         case POOL_STATE_L2CACHE:
 230                 return (gettext("L2CACHE"));
 231         case POOL_STATE_UNINITIALIZED:
 232                 return (gettext("UNINITIALIZED"));
 233         case POOL_STATE_UNAVAIL:
 234                 return (gettext("UNAVAIL"));
 235         case POOL_STATE_POTENTIALLY_ACTIVE:
 236                 return (gettext("POTENTIALLY_ACTIVE"));
 237         }
 238
 239         return (gettext("UNKNOWN"));
 240 }
 241
 242 /*
 243  * Get a zpool property value for 'prop' and return the value in
 244  * a pre-allocated buffer.
 245  */
 246 int
 247 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf, size_t len,
 248     zprop_source_t *srctype, boolean_t literal)
 249 {
 250         uint64_t intval;
 251         const char *strval;
 252         zprop_source_t src = ZPROP_SRC_NONE;
 253         nvlist_t *nvroot;
 254         vdev_stat_t *vs;
 255         uint_t vsc;
 256
 257         if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
 258                 switch (prop) {
 259                 case ZPOOL_PROP_NAME:
 260                         (void) strlcpy(buf, zpool_get_name(zhp), len);
 261                         break;
 262
 263                 case ZPOOL_PROP_HEALTH:
 264                         (void) strlcpy(buf,
 265                             zpool_pool_state_to_name(POOL_STATE_UNAVAIL), len);
 266                         break;
 267
 268                 case ZPOOL_PROP_GUID:
 269                         intval = zpool_get_prop_int(zhp, prop, &src);
 270                         (void) snprintf(buf, len, "%llu", intval);
 271                         break;
 272
 273                 case ZPOOL_PROP_ALTROOT:
 274                 case ZPOOL_PROP_CACHEFILE:
 275                 case ZPOOL_PROP_COMMENT:
 276                         if (zhp->zpool_props != NULL ||
 277                             zpool_get_all_props(zhp) == 0) {
 278                                 (void) strlcpy(buf,
 279                                     zpool_get_prop_string(zhp, prop, &src),
 280                                     len);
 281                                 break;
 282                         }
 283                         /* FALLTHROUGH */
 284                 default:
 285                         (void) strlcpy(buf, "-", len);
 286                         break;
 287                 }
 288
 289                 if (srctype != NULL)
 290                         *srctype = src;
 291                 return (0);
 292         }
 293
 294         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
 295             prop != ZPOOL_PROP_NAME)
 296                 return (-1);
 297
 298         switch (zpool_prop_get_type(prop)) {
 299         case PROP_TYPE_STRING:
 300                 (void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
 301                     len);
 302                 break;
 303
 304         case PROP_TYPE_NUMBER:
 305                 intval = zpool_get_prop_int(zhp, prop, &src);
 306
 307                 switch (prop) {
 308                 case ZPOOL_PROP_SIZE:
 309                 case ZPOOL_PROP_ALLOCATED:
 310                 case ZPOOL_PROP_FREE:
 311                 case ZPOOL_PROP_FREEING:
 312                 case ZPOOL_PROP_LEAKED:
 313                         if (literal) {
 314                                 (void) snprintf(buf, len, "%llu",
 315                                     (u_longlong_t)intval);
 316                         } else {
 317                                 (void) zfs_nicenum(intval, buf, len);
 318                         }
 319                         break;
 320                 case ZPOOL_PROP_BOOTSIZE:
 321                 case ZPOOL_PROP_EXPANDSZ:
 322                 case ZPOOL_PROP_CHECKPOINT:
 323                         if (intval == 0) {
 324                                 (void) strlcpy(buf, "-", len);
 325                         } else if (literal) {
 326                                 (void) snprintf(buf, len, "%llu",
 327                                     (u_longlong_t)intval);
 328                         } else {
 329                                 (void) zfs_nicenum(intval, buf, len);
 330                         }
 331                         break;
 332                 case ZPOOL_PROP_CAPACITY:
 333                         if (literal) {
 334                                 (void) snprintf(buf, len, "%llu",
 335                                     (u_longlong_t)intval);
 336                         } else {
 337                                 (void) snprintf(buf, len, "%llu%%",
 338                                     (u_longlong_t)intval);
 339                         }
 340                         break;
 341                 case ZPOOL_PROP_FRAGMENTATION:
 342                         if (intval == UINT64_MAX) {
 343                                 (void) strlcpy(buf, "-", len);
 344                         } else {
 345                                 (void) snprintf(buf, len, "%llu%%",
 346                                     (u_longlong_t)intval);
 347                         }
 348                         break;
 349                 case ZPOOL_PROP_DEDUPRATIO:
 350                         (void) snprintf(buf, len, "%llu.%02llux",
 351                             (u_longlong_t)(intval / 100),
 352                             (u_longlong_t)(intval % 100));
 353                         break;
 354                 case ZPOOL_PROP_HEALTH:
 355                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 356                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 357                         verify(nvlist_lookup_uint64_array(nvroot,
 358                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
 359                             == 0);
 360
 361                         (void) strlcpy(buf, zpool_state_to_name(intval,
 362                             vs->vs_aux), len);
 363                         break;
 364                 case ZPOOL_PROP_VERSION:
 365                         if (intval >= SPA_VERSION_FEATURES) {
 366                                 (void) snprintf(buf, len, "-");
 367                                 break;
 368                         }
 369                         /* FALLTHROUGH */
 370                 default:
 371                         (void) snprintf(buf, len, "%llu", intval);
 372                 }
 373                 break;
 374
 375         case PROP_TYPE_INDEX:
 376                 intval = zpool_get_prop_int(zhp, prop, &src);
 377                 if (zpool_prop_index_to_string(prop, intval, &strval)
 378                     != 0)
 379                         return (-1);
 380                 (void) strlcpy(buf, strval, len);
 381                 break;
 382
 383         default:
 384                 abort();
 385         }
 386
 387         if (srctype)
 388                 *srctype = src;
 389
 390         return (0);
 391 }
 392
 393 /*
 394  * Check if the bootfs name has the same pool name as it is set to.
 395  * Assuming bootfs is a valid dataset name.
 396  */
 397 static boolean_t
 398 bootfs_name_valid(const char *pool, char *bootfs)
 399 {
 400         int len = strlen(pool);
 401
 402         if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
 403                 return (B_FALSE);
 404
 405         if (strncmp(pool, bootfs, len) == 0 &&
 406             (bootfs[len] == '/' || bootfs[len] == '\0'))
 407                 return (B_TRUE);
 408
 409         return (B_FALSE);
 410 }
 411
 412 boolean_t
 413 zpool_is_bootable(zpool_handle_t *zhp)
 414 {
 415         char bootfs[ZFS_MAX_DATASET_NAME_LEN];
 416
 417         return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
 418             sizeof (bootfs), NULL, B_FALSE) == 0 && strncmp(bootfs, "-",
 419             sizeof (bootfs)) != 0);
 420 }
 421
 422
 423 /*
 424  * Given an nvlist of zpool properties to be set, validate that they are
 425  * correct, and parse any numeric properties (index, boolean, etc) if they are
 426  * specified as strings.
 427  */
 428 static nvlist_t *
 429 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 430     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
 431 {
 432         nvpair_t *elem;
 433         nvlist_t *retprops;
 434         zpool_prop_t prop;
 435         char *strval;
 436         uint64_t intval;
 437         char *slash, *check;
 438         struct stat64 statbuf;
 439         zpool_handle_t *zhp;
 440
 441         if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
 442                 (void) no_memory(hdl);
 443                 return (NULL);
 444         }
 445
 446         elem = NULL;
 447         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 448                 const char *propname = nvpair_name(elem);
 449
 450                 prop = zpool_name_to_prop(propname);
 451                 if (prop == ZPOOL_PROP_INVAL && zpool_prop_feature(propname)) {
 452                         int err;
 453                         char *fname = strchr(propname, '@') + 1;
 454
 455                         err = zfeature_lookup_name(fname, NULL);
 456                         if (err != 0) {
 457                                 ASSERT3U(err, ==, ENOENT);
 458                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 459                                     "invalid feature '%s'"), fname);
 460                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 461                                 goto error;
 462                         }
 463
 464                         if (nvpair_type(elem) != DATA_TYPE_STRING) {
 465                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 466                                     "'%s' must be a string"), propname);
 467                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 468                                 goto error;
 469                         }
 470
 471                         (void) nvpair_value_string(elem, &strval);
 472                         if (strcmp(strval, ZFS_FEATURE_ENABLED) != 0) {
 473                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 474                                     "property '%s' can only be set to "
 475                                     "'enabled'"), propname);
 476                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 477                                 goto error;
 478                         }
 479
 480                         if (nvlist_add_uint64(retprops, propname, 0) != 0) {
 481                                 (void) no_memory(hdl);
 482                                 goto error;
 483                         }
 484                         continue;
 485                 }
 486
 487                 /*
 488                  * Make sure this property is valid and applies to this type.
 489                  */
 490                 if (prop == ZPOOL_PROP_INVAL) {
 491                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 492                             "invalid property '%s'"), propname);
 493                         (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 494                         goto error;
 495                 }
 496
 497                 if (zpool_prop_readonly(prop)) {
 498                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 499                             "is readonly"), propname);
 500                         (void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
 501                         goto error;
 502                 }
 503
 504                 if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
 505                     &strval, &intval, errbuf) != 0)
 506                         goto error;
 507
 508                 /*
 509                  * Perform additional checking for specific properties.
 510                  */
 511                 switch (prop) {
 512                 case ZPOOL_PROP_VERSION:
 513                         if (intval < version ||
 514                             !SPA_VERSION_IS_SUPPORTED(intval)) {
 515                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 516                                     "property '%s' number %d is invalid."),
 517                                     propname, intval);
 518                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 519                                 goto error;
 520                         }
 521                         break;
 522
 523                 case ZPOOL_PROP_BOOTSIZE:
 524                         if (!flags.create) {
 525                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 526                                     "property '%s' can only be set during pool "
 527                                     "creation"), propname);
 528                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 529                                 goto error;
 530                         }
 531                         break;
 532
 533                 case ZPOOL_PROP_BOOTFS:
 534                         if (flags.create || flags.import) {
 535                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 536                                     "property '%s' cannot be set at creation "
 537                                     "or import time"), propname);
 538                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 539                                 goto error;
 540                         }
 541
 542                         if (version < SPA_VERSION_BOOTFS) {
 543                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 544                                     "pool must be upgraded to support "
 545                                     "'%s' property"), propname);
 546                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 547                                 goto error;
 548                         }
 549
 550                         /*
 551                          * bootfs property value has to be a dataset name and
 552                          * the dataset has to be in the same pool as it sets to.
 553                          */
 554                         if (strval[0] != '\0' && !bootfs_name_valid(poolname,
 555                             strval)) {
 556                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 557                                     "is an invalid name"), strval);
 558                                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
 559                                 goto error;
 560                         }
 561
 562                         if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
 563                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 564                                     "could not open pool '%s'"), poolname);
 565                                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
 566                                 goto error;
 567                         }
 568                         zpool_close(zhp);
 569                         break;
 570
 571                 case ZPOOL_PROP_ALTROOT:
 572                         if (!flags.create && !flags.import) {
 573                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 574                                     "property '%s' can only be set during pool "
 575                                     "creation or import"), propname);
 576                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 577                                 goto error;
 578                         }
 579
 580                         if (strval[0] != '/') {
 581                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 582                                     "bad alternate root '%s'"), strval);
 583                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 584                                 goto error;
 585                         }
 586                         break;
 587
 588                 case ZPOOL_PROP_CACHEFILE:
 589                         if (strval[0] == '\0')
 590                                 break;
 591
 592                         if (strcmp(strval, "none") == 0)
 593                                 break;
 594
 595                         if (strval[0] != '/') {
 596                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 597                                     "property '%s' must be empty, an "
 598                                     "absolute path, or 'none'"), propname);
 599                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 600                                 goto error;
 601                         }
 602
 603                         slash = strrchr(strval, '/');
 604
 605                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 606                             strcmp(slash, "/..") == 0) {
 607                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 608                                     "'%s' is not a valid file"), strval);
 609                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 610                                 goto error;
 611                         }
 612
 613                         *slash = '\0';
 614
 615                         if (strval[0] != '\0' &&
 616                             (stat64(strval, &statbuf) != 0 ||
 617                             !S_ISDIR(statbuf.st_mode))) {
 618                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 619                                     "'%s' is not a valid directory"),
 620                                     strval);
 621                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 622                                 goto error;
 623                         }
 624
 625                         *slash = '/';
 626                         break;
 627
 628                 case ZPOOL_PROP_COMMENT:
 629                         for (check = strval; *check != '\0'; check++) {
 630                                 if (!isprint(*check)) {
 631                                         zfs_error_aux(hdl,
 632                                             dgettext(TEXT_DOMAIN,
 633                                             "comment may only have printable "
 634                                             "characters"));
 635                                         (void) zfs_error(hdl, EZFS_BADPROP,
 636                                             errbuf);
 637                                         goto error;
 638                                 }
 639                         }
 640                         if (strlen(strval) > ZPROP_MAX_COMMENT) {
 641                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 642                                     "comment must not exceed %d characters"),
 643                                     ZPROP_MAX_COMMENT);
 644                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 645                                 goto error;
 646                         }
 647                         break;
 648
 649                 case ZPOOL_PROP_READONLY:
 650                         if (!flags.import) {
 651                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 652                                     "property '%s' can only be set at "
 653                                     "import time"), propname);
 654                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 655                                 goto error;
 656                         }
 657                         break;
 658
 659                 case ZPOOL_PROP_TNAME:
 660                         if (!flags.create) {
 661                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 662                                     "property '%s' can only be set at "
 663                                     "creation time"), propname);
 664                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 665                                 goto error;
 666                         }
 667                         break;
 668
 669                 case ZPOOL_PROP_MULTIHOST:
 670                         if (get_system_hostid() == 0) {
 671                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 672                                     "requires a non-zero system hostid"));
 673                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 674                                 goto error;
 675                         }
 676                         break;
 677
 678                 default:
 679                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 680                             "property '%s'(%d) not defined"), propname, prop);
 681                         break;
 682                 }
 683         }
 684
 685         return (retprops);
 686 error:
 687         nvlist_free(retprops);
 688         return (NULL);
 689 }
 690
 691 /*
 692  * Set zpool property : propname=propval.
 693  */
 694 int
 695 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
 696 {
 697         zfs_cmd_t zc = { 0 };
 698         int ret = -1;
 699         char errbuf[1024];
 700         nvlist_t *nvl = NULL;
 701         nvlist_t *realprops;
 702         uint64_t version;
 703         prop_flags_t flags = { 0 };
 704
 705         (void) snprintf(errbuf, sizeof (errbuf),
 706             dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
 707             zhp->zpool_name);
 708
 709         if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
 710                 return (no_memory(zhp->zpool_hdl));
 711
 712         if (nvlist_add_string(nvl, propname, propval) != 0) {
 713                 nvlist_free(nvl);
 714                 return (no_memory(zhp->zpool_hdl));
 715         }
 716
 717         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
 718         if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
 719             zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
 720                 nvlist_free(nvl);
 721                 return (-1);
 722         }
 723
 724         nvlist_free(nvl);
 725         nvl = realprops;
 726
 727         /*
 728          * Execute the corresponding ioctl() to set this property.
 729          */
 730         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 731
 732         if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
 733                 nvlist_free(nvl);
 734                 return (-1);
 735         }
 736
 737         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
 738
 739         zcmd_free_nvlists(&zc);
 740         nvlist_free(nvl);
 741
 742         if (ret)
 743                 (void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
 744         else
 745                 (void) zpool_props_refresh(zhp);
 746
 747         return (ret);
 748 }
 749
 750 int
 751 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
 752 {
 753         libzfs_handle_t *hdl = zhp->zpool_hdl;
 754         zprop_list_t *entry;
 755         char buf[ZFS_MAXPROPLEN];
 756         nvlist_t *features = NULL;
 757         zprop_list_t **last;
 758         boolean_t firstexpand = (NULL == *plp);
 759
 760         if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
 761                 return (-1);
 762
 763         last = plp;
 764         while (*last != NULL)
 765                 last = &(*last)->pl_next;
 766
 767         if ((*plp)->pl_all)
 768                 features = zpool_get_features(zhp);
 769
 770         if ((*plp)->pl_all && firstexpand) {
 771                 for (int i = 0; i < SPA_FEATURES; i++) {
 772                         zprop_list_t *entry = zfs_alloc(hdl,
 773                             sizeof (zprop_list_t));
 774                         entry->pl_prop = ZPROP_INVAL;
 775                         entry->pl_user_prop = zfs_asprintf(hdl, "feature@%s",
 776                             spa_feature_table[i].fi_uname);
 777                         entry->pl_width = strlen(entry->pl_user_prop);
 778                         entry->pl_all = B_TRUE;
 779
 780                         *last = entry;
 781                         last = &entry->pl_next;
 782                 }
 783         }
 784
 785         /* add any unsupported features */
 786         for (nvpair_t *nvp = nvlist_next_nvpair(features, NULL);
 787             nvp != NULL; nvp = nvlist_next_nvpair(features, nvp)) {
 788                 char *propname;
 789                 boolean_t found;
 790                 zprop_list_t *entry;
 791
 792                 if (zfeature_is_supported(nvpair_name(nvp)))
 793                         continue;
 794
 795                 propname = zfs_asprintf(hdl, "unsupported@%s",
 796                     nvpair_name(nvp));
 797
 798                 /*
 799                  * Before adding the property to the list make sure that no
 800                  * other pool already added the same property.
 801                  */
 802                 found = B_FALSE;
 803                 entry = *plp;
 804                 while (entry != NULL) {
 805                         if (entry->pl_user_prop != NULL &&
 806                             strcmp(propname, entry->pl_user_prop) == 0) {
 807                                 found = B_TRUE;
 808                                 break;
 809                         }
 810                         entry = entry->pl_next;
 811                 }
 812                 if (found) {
 813                         free(propname);
 814                         continue;
 815                 }
 816
 817                 entry = zfs_alloc(hdl, sizeof (zprop_list_t));
 818                 entry->pl_prop = ZPROP_INVAL;
 819                 entry->pl_user_prop = propname;
 820                 entry->pl_width = strlen(entry->pl_user_prop);
 821                 entry->pl_all = B_TRUE;
 822
 823                 *last = entry;
 824                 last = &entry->pl_next;
 825         }
 826
 827         for (entry = *plp; entry != NULL; entry = entry->pl_next) {
 828
 829                 if (entry->pl_fixed)
 830                         continue;
 831
 832                 if (entry->pl_prop != ZPROP_INVAL &&
 833                     zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
 834                     NULL, B_FALSE) == 0) {
 835                         if (strlen(buf) > entry->pl_width)
 836                                 entry->pl_width = strlen(buf);
 837                 }
 838         }
 839
 840         return (0);
 841 }
 842
 843 /*
 844  * Get the state for the given feature on the given ZFS pool.
 845  */
 846 int
 847 zpool_prop_get_feature(zpool_handle_t *zhp, const char *propname, char *buf,
 848     size_t len)
 849 {
 850         uint64_t refcount;
 851         boolean_t found = B_FALSE;
 852         nvlist_t *features = zpool_get_features(zhp);
 853         boolean_t supported;
 854         const char *feature = strchr(propname, '@') + 1;
 855
 856         supported = zpool_prop_feature(propname);
 857         ASSERT(supported || zpool_prop_unsupported(propname));
 858
 859         /*
 860          * Convert from feature name to feature guid. This conversion is
 861          * unecessary for unsupported@... properties because they already
 862          * use guids.
 863          */
 864         if (supported) {
 865                 int ret;
 866                 spa_feature_t fid;
 867
 868                 ret = zfeature_lookup_name(feature, &fid);
 869                 if (ret != 0) {
 870                         (void) strlcpy(buf, "-", len);
 871                         return (ENOTSUP);
 872                 }
 873                 feature = spa_feature_table[fid].fi_guid;
 874         }
 875
 876         if (nvlist_lookup_uint64(features, feature, &refcount) == 0)
 877                 found = B_TRUE;
 878
 879         if (supported) {
 880                 if (!found) {
 881                         (void) strlcpy(buf, ZFS_FEATURE_DISABLED, len);
 882                 } else  {
 883                         if (refcount == 0)
 884                                 (void) strlcpy(buf, ZFS_FEATURE_ENABLED, len);
 885                         else
 886                                 (void) strlcpy(buf, ZFS_FEATURE_ACTIVE, len);
 887                 }
 888         } else {
 889                 if (found) {
 890                         if (refcount == 0) {
 891                                 (void) strcpy(buf, ZFS_UNSUPPORTED_INACTIVE);
 892                         } else {
 893                                 (void) strcpy(buf, ZFS_UNSUPPORTED_READONLY);
 894                         }
 895                 } else {
 896                         (void) strlcpy(buf, "-", len);
 897                         return (ENOTSUP);
 898                 }
 899         }
 900
 901         return (0);
 902 }
 903
 904 /*
 905  * Don't start the slice at the default block of 34; many storage
 906  * devices will use a stripe width of 128k, so start there instead.
 907  */
 908 #define NEW_START_BLOCK 256
 909
 910 /*
 911  * Validate the given pool name, optionally putting an extended error message in
 912  * 'buf'.
 913  */
 914 boolean_t
 915 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
 916 {
 917         namecheck_err_t why;
 918         char what;
 919         int ret;
 920
 921         ret = pool_namecheck(pool, &why, &what);
 922
 923         /*
 924          * The rules for reserved pool names were extended at a later point.
 925          * But we need to support users with existing pools that may now be
 926          * invalid.  So we only check for this expanded set of names during a
 927          * create (or import), and only in userland.
 928          */
 929         if (ret == 0 && !isopen &&
 930             (strncmp(pool, "mirror", 6) == 0 ||
 931             strncmp(pool, "raidz", 5) == 0 ||
 932             strncmp(pool, "spare", 5) == 0 ||
 933             strcmp(pool, "log") == 0)) {
 934                 if (hdl != NULL)
 935                         zfs_error_aux(hdl,
 936                             dgettext(TEXT_DOMAIN, "name is reserved"));
 937                 return (B_FALSE);
 938         }
 939
 940
 941         if (ret != 0) {
 942                 if (hdl != NULL) {
 943                         switch (why) {
 944                         case NAME_ERR_TOOLONG:
 945                                 zfs_error_aux(hdl,
 946                                     dgettext(TEXT_DOMAIN, "name is too long"));
 947                                 break;
 948
 949                         case NAME_ERR_INVALCHAR:
 950                                 zfs_error_aux(hdl,
 951                                     dgettext(TEXT_DOMAIN, "invalid character "
 952                                     "'%c' in pool name"), what);
 953                                 break;
 954
 955                         case NAME_ERR_NOLETTER:
 956                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 957                                     "name must begin with a letter"));
 958                                 break;
 959
 960                         case NAME_ERR_RESERVED:
 961                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 962                                     "name is reserved"));
 963                                 break;
 964
 965                         case NAME_ERR_DISKLIKE:
 966                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 967                                     "pool name is reserved"));
 968                                 break;
 969
 970                         case NAME_ERR_LEADING_SLASH:
 971                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 972                                     "leading slash in name"));
 973                                 break;
 974
 975                         case NAME_ERR_EMPTY_COMPONENT:
 976                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 977                                     "empty component in name"));
 978                                 break;
 979
 980                         case NAME_ERR_TRAILING_SLASH:
 981                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 982                                     "trailing slash in name"));
 983                                 break;
 984
 985                         case NAME_ERR_MULTIPLE_DELIMITERS:
 986                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 987                                     "multiple '@' and/or '#' delimiters in "
 988                                     "name"));
 989                                 break;
 990
 991                         default:
 992                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 993                                     "(%d) not defined"), why);
 994                                 break;
 995                         }
 996                 }
 997                 return (B_FALSE);
 998         }
 999
1000         return (B_TRUE);
1001 }
1002
1003 /*
1004  * Open a handle to the given pool, even if the pool is currently in the FAULTED
1005  * state.
1006  */
1007 zpool_handle_t *
1008 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
1009 {
1010         zpool_handle_t *zhp;
1011         boolean_t missing;
1012
1013         /*
1014          * Make sure the pool name is valid.
1015          */
1016         if (!zpool_name_valid(hdl, B_TRUE, pool)) {
1017                 (void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1018                     dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1019                     pool);
1020                 return (NULL);
1021         }
1022
1023         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1024                 return (NULL);
1025
1026         zhp->zpool_hdl = hdl;
1027         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1028
1029         if (zpool_refresh_stats(zhp, &missing) != 0) {
1030                 zpool_close(zhp);
1031                 return (NULL);
1032         }
1033
1034         if (missing) {
1035                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
1036                 (void) zfs_error_fmt(hdl, EZFS_NOENT,
1037                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
1038                 zpool_close(zhp);
1039                 return (NULL);
1040         }
1041
1042         return (zhp);
1043 }
1044
1045 /*
1046  * Like the above, but silent on error.  Used when iterating over pools (because
1047  * the configuration cache may be out of date).
1048  */
1049 int
1050 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
1051 {
1052         zpool_handle_t *zhp;
1053         boolean_t missing;
1054
1055         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1056                 return (-1);
1057
1058         zhp->zpool_hdl = hdl;
1059         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1060
1061         if (zpool_refresh_stats(zhp, &missing) != 0) {
1062                 zpool_close(zhp);
1063                 return (-1);
1064         }
1065
1066         if (missing) {
1067                 zpool_close(zhp);
1068                 *ret = NULL;
1069                 return (0);
1070         }
1071
1072         *ret = zhp;
1073         return (0);
1074 }
1075
1076 /*
1077  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
1078  * state.
1079  */
1080 zpool_handle_t *
1081 zpool_open(libzfs_handle_t *hdl, const char *pool)
1082 {
1083         zpool_handle_t *zhp;
1084
1085         if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
1086                 return (NULL);
1087
1088         if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
1089                 (void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
1090                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
1091                 zpool_close(zhp);
1092                 return (NULL);
1093         }
1094
1095         return (zhp);
1096 }
1097
1098 /*
1099  * Close the handle.  Simply frees the memory associated with the handle.
1100  */
1101 void
1102 zpool_close(zpool_handle_t *zhp)
1103 {
1104         nvlist_free(zhp->zpool_config);
1105         nvlist_free(zhp->zpool_old_config);
1106         nvlist_free(zhp->zpool_props);
1107         free(zhp);
1108 }
1109
1110 /*
1111  * Return the name of the pool.
1112  */
1113 const char *
1114 zpool_get_name(zpool_handle_t *zhp)
1115 {
1116         return (zhp->zpool_name);
1117 }
1118
1119
1120 /*
1121  * Return the state of the pool (ACTIVE or UNAVAILABLE)
1122  */
1123 int
1124 zpool_get_state(zpool_handle_t *zhp)
1125 {
1126         return (zhp->zpool_state);
1127 }
1128
1129 /*
1130  * Check if vdev list contains a special vdev
1131  */
1132 static boolean_t
1133 zpool_has_special_vdev(nvlist_t *nvroot)
1134 {
1135         nvlist_t **child;
1136         uint_t children;
1137
1138         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN, &child,
1139             &children) == 0) {
1140                 for (uint_t c = 0; c < children; c++) {
1141                         char *bias;
1142
1143                         if (nvlist_lookup_string(child[c],
1144                             ZPOOL_CONFIG_ALLOCATION_BIAS, &bias) == 0 &&
1145                             strcmp(bias, VDEV_ALLOC_BIAS_SPECIAL) == 0) {
1146                                 return (B_TRUE);
1147                         }
1148                 }
1149         }
1150         return (B_FALSE);
1151 }
1152
1153 /*
1154  * Create the named pool, using the provided vdev list.  It is assumed
1155  * that the consumer has already validated the contents of the nvlist, so we
1156  * don't have to worry about error semantics.
1157  */
1158 int
1159 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
1160     nvlist_t *props, nvlist_t *fsprops)
1161 {
1162         zfs_cmd_t zc = { 0 };
1163         nvlist_t *zc_fsprops = NULL;
1164         nvlist_t *zc_props = NULL;
1165         char msg[1024];
1166         int ret = -1;
1167
1168         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1169             "cannot create '%s'"), pool);
1170
1171         if (!zpool_name_valid(hdl, B_FALSE, pool))
1172                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
1173
1174         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1175                 return (-1);
1176
1177         if (props) {
1178                 prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
1179
1180                 if ((zc_props = zpool_valid_proplist(hdl, pool, props,
1181                     SPA_VERSION_1, flags, msg)) == NULL) {
1182                         goto create_failed;
1183                 }
1184         }
1185
1186         if (fsprops) {
1187                 uint64_t zoned;
1188                 char *zonestr;
1189
1190                 zoned = ((nvlist_lookup_string(fsprops,
1191                     zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
1192                     strcmp(zonestr, "on") == 0);
1193
1194                 if ((zc_fsprops = zfs_valid_proplist(hdl, ZFS_TYPE_FILESYSTEM,
1195                     fsprops, zoned, NULL, NULL, msg)) == NULL) {
1196                         goto create_failed;
1197                 }
1198
1199                 if (nvlist_exists(zc_fsprops,
1200                     zfs_prop_to_name(ZFS_PROP_SPECIAL_SMALL_BLOCKS)) &&
1201                     !zpool_has_special_vdev(nvroot)) {
1202                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1203                             "%s property requires a special vdev"),
1204                             zfs_prop_to_name(ZFS_PROP_SPECIAL_SMALL_BLOCKS));
1205                         (void) zfs_error(hdl, EZFS_BADPROP, msg);
1206                         goto create_failed;
1207                 }
1208
1209                 if (!zc_props &&
1210                     (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
1211                         goto create_failed;
1212                 }
1213                 if (nvlist_add_nvlist(zc_props,
1214                     ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
1215                         goto create_failed;
1216                 }
1217         }
1218
1219         if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
1220                 goto create_failed;
1221
1222         (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
1223
1224         if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
1225
1226                 zcmd_free_nvlists(&zc);
1227                 nvlist_free(zc_props);
1228                 nvlist_free(zc_fsprops);
1229
1230                 switch (errno) {
1231                 case EBUSY:
1232                         /*
1233                          * This can happen if the user has specified the same
1234                          * device multiple times.  We can't reliably detect this
1235                          * until we try to add it and see we already have a
1236                          * label.
1237                          */
1238                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1239                             "one or more vdevs refer to the same device"));
1240                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1241
1242                 case ERANGE:
1243                         /*
1244                          * This happens if the record size is smaller or larger
1245                          * than the allowed size range, or not a power of 2.
1246                          *
1247                          * NOTE: although zfs_valid_proplist is called earlier,
1248                          * this case may have slipped through since the
1249                          * pool does not exist yet and it is therefore
1250                          * impossible to read properties e.g. max blocksize
1251                          * from the pool.
1252                          */
1253                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1254                             "record size invalid"));
1255                         return (zfs_error(hdl, EZFS_BADPROP, msg));
1256
1257                 case EOVERFLOW:
1258                         /*
1259                          * This occurs when one of the devices is below
1260                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1261                          * device was the problem device since there's no
1262                          * reliable way to determine device size from userland.
1263                          */
1264                         {
1265                                 char buf[64];
1266
1267                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1268
1269                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1270                                     "one or more devices is less than the "
1271                                     "minimum size (%s)"), buf);
1272                         }
1273                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1274
1275                 case ENOSPC:
1276                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1277                             "one or more devices is out of space"));
1278                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1279
1280                 case ENOTBLK:
1281                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1282                             "cache device must be a disk or disk slice"));
1283                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1284
1285                 default:
1286                         return (zpool_standard_error(hdl, errno, msg));
1287                 }
1288         }
1289
1290 create_failed:
1291         zcmd_free_nvlists(&zc);
1292         nvlist_free(zc_props);
1293         nvlist_free(zc_fsprops);
1294         return (ret);
1295 }
1296
1297 /*
1298  * Destroy the given pool.  It is up to the caller to ensure that there are no
1299  * datasets left in the pool.
1300  */
1301 int
1302 zpool_destroy(zpool_handle_t *zhp, const char *log_str)
1303 {
1304         zfs_cmd_t zc = { 0 };
1305         zfs_handle_t *zfp = NULL;
1306         libzfs_handle_t *hdl = zhp->zpool_hdl;
1307         char msg[1024];
1308
1309         if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1310             (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1311                 return (-1);
1312
1313         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1314         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1315
1316         if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1317                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1318                     "cannot destroy '%s'"), zhp->zpool_name);
1319
1320                 if (errno == EROFS) {
1321                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1322                             "one or more devices is read only"));
1323                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1324                 } else {
1325                         (void) zpool_standard_error(hdl, errno, msg);
1326                 }
1327
1328                 if (zfp)
1329                         zfs_close(zfp);
1330                 return (-1);
1331         }
1332
1333         if (zfp) {
1334                 remove_mountpoint(zfp);
1335                 zfs_close(zfp);
1336         }
1337
1338         return (0);
1339 }
1340
1341 /*
1342  * Create a checkpoint in the given pool.
1343  */
1344 int
1345 zpool_checkpoint(zpool_handle_t *zhp)
1346 {
1347         libzfs_handle_t *hdl = zhp->zpool_hdl;
1348         char msg[1024];
1349         int error;
1350
1351         error = lzc_pool_checkpoint(zhp->zpool_name);
1352         if (error != 0) {
1353                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1354                     "cannot checkpoint '%s'"), zhp->zpool_name);
1355                 (void) zpool_standard_error(hdl, error, msg);
1356                 return (-1);
1357         }
1358
1359         return (0);
1360 }
1361
1362 /*
1363  * Discard the checkpoint from the given pool.
1364  */
1365 int
1366 zpool_discard_checkpoint(zpool_handle_t *zhp)
1367 {
1368         libzfs_handle_t *hdl = zhp->zpool_hdl;
1369         char msg[1024];
1370         int error;
1371
1372         error = lzc_pool_checkpoint_discard(zhp->zpool_name);
1373         if (error != 0) {
1374                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1375                     "cannot discard checkpoint in '%s'"), zhp->zpool_name);
1376                 (void) zpool_standard_error(hdl, error, msg);
1377                 return (-1);
1378         }
1379
1380         return (0);
1381 }
1382
1383 /*
1384  * Add the given vdevs to the pool.  The caller must have already performed the
1385  * necessary verification to ensure that the vdev specification is well-formed.
1386  */
1387 int
1388 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1389 {
1390         zfs_cmd_t zc = { 0 };
1391         int ret;
1392         libzfs_handle_t *hdl = zhp->zpool_hdl;
1393         char msg[1024];
1394         nvlist_t **spares, **l2cache;
1395         uint_t nspares, nl2cache;
1396
1397         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1398             "cannot add to '%s'"), zhp->zpool_name);
1399
1400         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1401             SPA_VERSION_SPARES &&
1402             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1403             &spares, &nspares) == 0) {
1404                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1405                     "upgraded to add hot spares"));
1406                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1407         }
1408
1409         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1410             SPA_VERSION_L2CACHE &&
1411             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1412             &l2cache, &nl2cache) == 0) {
1413                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1414                     "upgraded to add cache devices"));
1415                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1416         }
1417
1418         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1419                 return (-1);
1420         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1421
1422         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1423                 switch (errno) {
1424                 case EBUSY:
1425                         /*
1426                          * This can happen if the user has specified the same
1427                          * device multiple times.  We can't reliably detect this
1428                          * until we try to add it and see we already have a
1429                          * label.
1430                          */
1431                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1432                             "one or more vdevs refer to the same device"));
1433                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1434                         break;
1435
1436                 case EINVAL:
1437                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1438                             "invalid config; a pool with removing/removed "
1439                             "vdevs does not support adding raidz vdevs"));
1440                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1441                         break;
1442
1443                 case EOVERFLOW:
1444                         /*
1445                          * This occurrs when one of the devices is below
1446                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1447                          * device was the problem device since there's no
1448                          * reliable way to determine device size from userland.
1449                          */
1450                         {
1451                                 char buf[64];
1452
1453                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1454
1455                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1456                                     "device is less than the minimum "
1457                                     "size (%s)"), buf);
1458                         }
1459                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1460                         break;
1461
1462                 case ENOTSUP:
1463                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1464                             "pool must be upgraded to add these vdevs"));
1465                         (void) zfs_error(hdl, EZFS_BADVERSION, msg);
1466                         break;
1467
1468                 case EDOM:
1469                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1470                             "root pool can not have multiple vdevs"
1471                             " or separate logs"));
1472                         (void) zfs_error(hdl, EZFS_POOL_NOTSUP, msg);
1473                         break;
1474
1475                 case ENOTBLK:
1476                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1477                             "cache device must be a disk or disk slice"));
1478                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1479                         break;
1480
1481                 default:
1482                         (void) zpool_standard_error(hdl, errno, msg);
1483                 }
1484
1485                 ret = -1;
1486         } else {
1487                 ret = 0;
1488         }
1489
1490         zcmd_free_nvlists(&zc);
1491
1492         return (ret);
1493 }
1494
1495 /*
1496  * Exports the pool from the system.  The caller must ensure that there are no
1497  * mounted datasets in the pool.
1498  */
1499 static int
1500 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce,
1501     const char *log_str)
1502 {
1503         zfs_cmd_t zc = { 0 };
1504         char msg[1024];
1505
1506         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1507             "cannot export '%s'"), zhp->zpool_name);
1508
1509         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1510         zc.zc_cookie = force;
1511         zc.zc_guid = hardforce;
1512         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1513
1514         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1515                 switch (errno) {
1516                 case EXDEV:
1517                         zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1518                             "use '-f' to override the following errors:\n"
1519                             "'%s' has an active shared spare which could be"
1520                             " used by other pools once '%s' is exported."),
1521                             zhp->zpool_name, zhp->zpool_name);
1522                         return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1523                             msg));
1524                 default:
1525                         return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1526                             msg));
1527                 }
1528         }
1529
1530         return (0);
1531 }
1532
1533 int
1534 zpool_export(zpool_handle_t *zhp, boolean_t force, const char *log_str)
1535 {
1536         return (zpool_export_common(zhp, force, B_FALSE, log_str));
1537 }
1538
1539 int
1540 zpool_export_force(zpool_handle_t *zhp, const char *log_str)
1541 {
1542         return (zpool_export_common(zhp, B_TRUE, B_TRUE, log_str));
1543 }
1544
1545 static void
1546 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1547     nvlist_t *config)
1548 {
1549         nvlist_t *nv = NULL;
1550         uint64_t rewindto;
1551         int64_t loss = -1;
1552         struct tm t;
1553         char timestr[128];
1554
1555         if (!hdl->libzfs_printerr || config == NULL)
1556                 return;
1557
1558         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1559             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0) {
1560                 return;
1561         }
1562
1563         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1564                 return;
1565         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1566
1567         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1568             strftime(timestr, 128, 0, &t) != 0) {
1569                 if (dryrun) {
1570                         (void) printf(dgettext(TEXT_DOMAIN,
1571                             "Would be able to return %s "
1572                             "to its state as of %s.\n"),
1573                             name, timestr);
1574                 } else {
1575                         (void) printf(dgettext(TEXT_DOMAIN,
1576                             "Pool %s returned to its state as of %s.\n"),
1577                             name, timestr);
1578                 }
1579                 if (loss > 120) {
1580                         (void) printf(dgettext(TEXT_DOMAIN,
1581                             "%s approximately %lld "),
1582                             dryrun ? "Would discard" : "Discarded",
1583                             (loss + 30) / 60);
1584                         (void) printf(dgettext(TEXT_DOMAIN,
1585                             "minutes of transactions.\n"));
1586                 } else if (loss > 0) {
1587                         (void) printf(dgettext(TEXT_DOMAIN,
1588                             "%s approximately %lld "),
1589                             dryrun ? "Would discard" : "Discarded", loss);
1590                         (void) printf(dgettext(TEXT_DOMAIN,
1591                             "seconds of transactions.\n"));
1592                 }
1593         }
1594 }
1595
1596 void
1597 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1598     nvlist_t *config)
1599 {
1600         nvlist_t *nv = NULL;
1601         int64_t loss = -1;
1602         uint64_t edata = UINT64_MAX;
1603         uint64_t rewindto;
1604         struct tm t;
1605         char timestr[128];
1606
1607         if (!hdl->libzfs_printerr)
1608                 return;
1609
1610         if (reason >= 0)
1611                 (void) printf(dgettext(TEXT_DOMAIN, "action: "));
1612         else
1613                 (void) printf(dgettext(TEXT_DOMAIN, "\t"));
1614
1615         /* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1616         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1617             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0 ||
1618             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1619                 goto no_info;
1620
1621         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1622         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1623             &edata);
1624
1625         (void) printf(dgettext(TEXT_DOMAIN,
1626             "Recovery is possible, but will result in some data loss.\n"));
1627
1628         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1629             strftime(timestr, 128, 0, &t) != 0) {
1630                 (void) printf(dgettext(TEXT_DOMAIN,
1631                     "\tReturning the pool to its state as of %s\n"
1632                     "\tshould correct the problem.  "),
1633                     timestr);
1634         } else {
1635                 (void) printf(dgettext(TEXT_DOMAIN,
1636                     "\tReverting the pool to an earlier state "
1637                     "should correct the problem.\n\t"));
1638         }
1639
1640         if (loss > 120) {
1641                 (void) printf(dgettext(TEXT_DOMAIN,
1642                     "Approximately %lld minutes of data\n"
1643                     "\tmust be discarded, irreversibly.  "), (loss + 30) / 60);
1644         } else if (loss > 0) {
1645                 (void) printf(dgettext(TEXT_DOMAIN,
1646                     "Approximately %lld seconds of data\n"
1647                     "\tmust be discarded, irreversibly.  "), loss);
1648         }
1649         if (edata != 0 && edata != UINT64_MAX) {
1650                 if (edata == 1) {
1651                         (void) printf(dgettext(TEXT_DOMAIN,
1652                             "After rewind, at least\n"
1653                             "\tone persistent user-data error will remain.  "));
1654                 } else {
1655                         (void) printf(dgettext(TEXT_DOMAIN,
1656                             "After rewind, several\n"
1657                             "\tpersistent user-data errors will remain.  "));
1658                 }
1659         }
1660         (void) printf(dgettext(TEXT_DOMAIN,
1661             "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1662             reason >= 0 ? "clear" : "import", name);
1663
1664         (void) printf(dgettext(TEXT_DOMAIN,
1665             "A scrub of the pool\n"
1666             "\tis strongly recommended after recovery.\n"));
1667         return;
1668
1669 no_info:
1670         (void) printf(dgettext(TEXT_DOMAIN,
1671             "Destroy and re-create the pool from\n\ta backup source.\n"));
1672 }
1673
1674 /*
1675  * zpool_import() is a contracted interface. Should be kept the same
1676  * if possible.
1677  *
1678  * Applications should use zpool_import_props() to import a pool with
1679  * new properties value to be set.
1680  */
1681 int
1682 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1683     char *altroot)
1684 {
1685         nvlist_t *props = NULL;
1686         int ret;
1687
1688         if (altroot != NULL) {
1689                 if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1690                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1691                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1692                             newname));
1693                 }
1694
1695                 if (nvlist_add_string(props,
1696                     zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1697                     nvlist_add_string(props,
1698                     zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1699                         nvlist_free(props);
1700                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1701                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1702                             newname));
1703                 }
1704         }
1705
1706         ret = zpool_import_props(hdl, config, newname, props,
1707             ZFS_IMPORT_NORMAL);
1708         nvlist_free(props);
1709         return (ret);
1710 }
1711
1712 static void
1713 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1714     int indent)
1715 {
1716         nvlist_t **child;
1717         uint_t c, children;
1718         char *vname;
1719         uint64_t is_log = 0;
1720
1721         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1722             &is_log);
1723
1724         if (name != NULL)
1725                 (void) printf("\t%*s%s%s\n", indent, "", name,
1726                     is_log ? " [log]" : "");
1727
1728         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1729             &child, &children) != 0)
1730                 return;
1731
1732         for (c = 0; c < children; c++) {
1733                 vname = zpool_vdev_name(hdl, NULL, child[c], VDEV_NAME_TYPE_ID);
1734                 print_vdev_tree(hdl, vname, child[c], indent + 2);
1735                 free(vname);
1736         }
1737 }
1738
1739 void
1740 zpool_print_unsup_feat(nvlist_t *config)
1741 {
1742         nvlist_t *nvinfo, *unsup_feat;
1743
1744         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nvinfo) ==
1745             0);
1746         verify(nvlist_lookup_nvlist(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT,
1747             &unsup_feat) == 0);
1748
1749         for (nvpair_t *nvp = nvlist_next_nvpair(unsup_feat, NULL); nvp != NULL;
1750             nvp = nvlist_next_nvpair(unsup_feat, nvp)) {
1751                 char *desc;
1752
1753                 verify(nvpair_type(nvp) == DATA_TYPE_STRING);
1754                 verify(nvpair_value_string(nvp, &desc) == 0);
1755
1756                 if (strlen(desc) > 0)
1757                         (void) printf("\t%s (%s)\n", nvpair_name(nvp), desc);
1758                 else
1759                         (void) printf("\t%s\n", nvpair_name(nvp));
1760         }
1761 }
1762
1763 /*
1764  * Import the given pool using the known configuration and a list of
1765  * properties to be set. The configuration should have come from
1766  * zpool_find_import(). The 'newname' parameters control whether the pool
1767  * is imported with a different name.
1768  */
1769 int
1770 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1771     nvlist_t *props, int flags)
1772 {
1773         zfs_cmd_t zc = { 0 };
1774         zpool_load_policy_t policy;
1775         nvlist_t *nv = NULL;
1776         nvlist_t *nvinfo = NULL;
1777         nvlist_t *missing = NULL;
1778         char *thename;
1779         char *origname;
1780         int ret;
1781         int error = 0;
1782         char errbuf[1024];
1783
1784         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1785             &origname) == 0);
1786
1787         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1788             "cannot import pool '%s'"), origname);
1789
1790         if (newname != NULL) {
1791                 if (!zpool_name_valid(hdl, B_FALSE, newname))
1792                         return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1793                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1794                             newname));
1795                 thename = (char *)newname;
1796         } else {
1797                 thename = origname;
1798         }
1799
1800         if (props != NULL) {
1801                 uint64_t version;
1802                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1803
1804                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1805                     &version) == 0);
1806
1807                 if ((props = zpool_valid_proplist(hdl, origname,
1808                     props, version, flags, errbuf)) == NULL)
1809                         return (-1);
1810                 if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1811                         nvlist_free(props);
1812                         return (-1);
1813                 }
1814                 nvlist_free(props);
1815         }
1816
1817         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1818
1819         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1820             &zc.zc_guid) == 0);
1821
1822         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1823                 zcmd_free_nvlists(&zc);
1824                 return (-1);
1825         }
1826         if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1827                 zcmd_free_nvlists(&zc);
1828                 return (-1);
1829         }
1830
1831         zc.zc_cookie = flags;
1832         while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1833             errno == ENOMEM) {
1834                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1835                         zcmd_free_nvlists(&zc);
1836                         return (-1);
1837                 }
1838         }
1839         if (ret != 0)
1840                 error = errno;
1841
1842         (void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1843
1844         zcmd_free_nvlists(&zc);
1845
1846         zpool_get_load_policy(config, &policy);
1847
1848         if (error) {
1849                 char desc[1024];
1850                 char aux[256];
1851
1852                 /*
1853                  * Dry-run failed, but we print out what success
1854                  * looks like if we found a best txg
1855                  */
1856                 if (policy.zlp_rewind & ZPOOL_TRY_REWIND) {
1857                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1858                             B_TRUE, nv);
1859                         nvlist_free(nv);
1860                         return (-1);
1861                 }
1862
1863                 if (newname == NULL)
1864                         (void) snprintf(desc, sizeof (desc),
1865                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1866                             thename);
1867                 else
1868                         (void) snprintf(desc, sizeof (desc),
1869                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1870                             origname, thename);
1871
1872                 switch (error) {
1873                 case ENOTSUP:
1874                         if (nv != NULL && nvlist_lookup_nvlist(nv,
1875                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1876                             nvlist_exists(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT)) {
1877                                 (void) printf(dgettext(TEXT_DOMAIN, "This "
1878                                     "pool uses the following feature(s) not "
1879                                     "supported by this system:\n"));
1880                                 zpool_print_unsup_feat(nv);
1881                                 if (nvlist_exists(nvinfo,
1882                                     ZPOOL_CONFIG_CAN_RDONLY)) {
1883                                         (void) printf(dgettext(TEXT_DOMAIN,
1884                                             "All unsupported features are only "
1885                                             "required for writing to the pool."
1886                                             "\nThe pool can be imported using "
1887                                             "'-o readonly=on'.\n"));
1888                                 }
1889                         }
1890                         /*
1891                          * Unsupported version.
1892                          */
1893                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1894                         break;
1895
1896                 case EREMOTEIO:
1897                         if (nv != NULL && nvlist_lookup_nvlist(nv,
1898                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0) {
1899                                 char *hostname = "<unknown>";
1900                                 uint64_t hostid = 0;
1901                                 mmp_state_t mmp_state;
1902
1903                                 mmp_state = fnvlist_lookup_uint64(nvinfo,
1904                                     ZPOOL_CONFIG_MMP_STATE);
1905
1906                                 if (nvlist_exists(nvinfo,
1907                                     ZPOOL_CONFIG_MMP_HOSTNAME))
1908                                         hostname = fnvlist_lookup_string(nvinfo,
1909                                             ZPOOL_CONFIG_MMP_HOSTNAME);
1910
1911                                 if (nvlist_exists(nvinfo,
1912                                     ZPOOL_CONFIG_MMP_HOSTID))
1913                                         hostid = fnvlist_lookup_uint64(nvinfo,
1914                                             ZPOOL_CONFIG_MMP_HOSTID);
1915
1916                                 if (mmp_state == MMP_STATE_ACTIVE) {
1917                                         (void) snprintf(aux, sizeof (aux),
1918                                             dgettext(TEXT_DOMAIN, "pool is imp"
1919                                             "orted on host '%s' (hostid=%lx).\n"
1920                                             "Export the pool on the other "
1921                                             "system, then run 'zpool import'."),
1922                                             hostname, (unsigned long) hostid);
1923                                 } else if (mmp_state == MMP_STATE_NO_HOSTID) {
1924                                         (void) snprintf(aux, sizeof (aux),
1925                                             dgettext(TEXT_DOMAIN, "pool has "
1926                                             "the multihost property on and "
1927                                             "the\nsystem's hostid is not "
1928                                             "set.\n"));
1929                                 }
1930
1931                                 (void) zfs_error_aux(hdl, aux);
1932                         }
1933                         (void) zfs_error(hdl, EZFS_ACTIVE_POOL, desc);
1934                         break;
1935
1936                 case EINVAL:
1937                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1938                         break;
1939
1940                 case EROFS:
1941                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1942                             "one or more devices is read only"));
1943                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1944                         break;
1945
1946                 case ENXIO:
1947                         if (nv && nvlist_lookup_nvlist(nv,
1948                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1949                             nvlist_lookup_nvlist(nvinfo,
1950                             ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1951                                 (void) printf(dgettext(TEXT_DOMAIN,
1952                                     "The devices below are missing or "
1953                                     "corrupted, use '-m' to import the pool "
1954                                     "anyway:\n"));
1955                                 print_vdev_tree(hdl, NULL, missing, 2);
1956                                 (void) printf("\n");
1957                         }
1958                         (void) zpool_standard_error(hdl, error, desc);
1959                         break;
1960
1961                 case EEXIST:
1962                         (void) zpool_standard_error(hdl, error, desc);
1963                         break;
1964                 case ENAMETOOLONG:
1965                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1966                             "new name of at least one dataset is longer than "
1967                             "the maximum allowable length"));
1968                         (void) zfs_error(hdl, EZFS_NAMETOOLONG, desc);
1969                         break;
1970                 default:
1971                         (void) zpool_standard_error(hdl, error, desc);
1972                         zpool_explain_recover(hdl,
1973                             newname ? origname : thename, -error, nv);
1974                         break;
1975                 }
1976
1977                 nvlist_free(nv);
1978                 ret = -1;
1979         } else {
1980                 zpool_handle_t *zhp;
1981
1982                 /*
1983                  * This should never fail, but play it safe anyway.
1984                  */
1985                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1986                         ret = -1;
1987                 else if (zhp != NULL)
1988                         zpool_close(zhp);
1989                 if (policy.zlp_rewind &
1990                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1991                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1992                             ((policy.zlp_rewind & ZPOOL_TRY_REWIND) != 0), nv);
1993                 }
1994                 nvlist_free(nv);
1995                 return (0);
1996         }
1997
1998         return (ret);
1999 }
2000
2001 /*
2002  * Scan the pool.
2003  */
2004 int
2005 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func, pool_scrub_cmd_t cmd)
2006 {
2007         zfs_cmd_t zc = { 0 };
2008         char msg[1024];
2009         int err;
2010         libzfs_handle_t *hdl = zhp->zpool_hdl;
2011
2012         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2013         zc.zc_cookie = func;
2014         zc.zc_flags = cmd;
2015
2016         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0)
2017                 return (0);
2018
2019         err = errno;
2020
2021         /* ECANCELED on a scrub means we resumed a paused scrub */
2022         if (err == ECANCELED && func == POOL_SCAN_SCRUB &&
2023             cmd == POOL_SCRUB_NORMAL)
2024                 return (0);
2025
2026         if (err == ENOENT && func != POOL_SCAN_NONE && cmd == POOL_SCRUB_NORMAL)
2027                 return (0);
2028
2029         if (func == POOL_SCAN_SCRUB) {
2030                 if (cmd == POOL_SCRUB_PAUSE) {
2031                         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2032                             "cannot pause scrubbing %s"), zc.zc_name);
2033                 } else {
2034                         assert(cmd == POOL_SCRUB_NORMAL);
2035                         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2036                             "cannot scrub %s"), zc.zc_name);
2037                 }
2038         } else if (func == POOL_SCAN_NONE) {
2039                 (void) snprintf(msg, sizeof (msg),
2040                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
2041                     zc.zc_name);
2042         } else {
2043                 assert(!"unexpected result");
2044         }
2045
2046         if (err == EBUSY) {
2047                 nvlist_t *nvroot;
2048                 pool_scan_stat_t *ps = NULL;
2049                 uint_t psc;
2050
2051                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
2052                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2053                 (void) nvlist_lookup_uint64_array(nvroot,
2054                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
2055                 if (ps && ps->pss_func == POOL_SCAN_SCRUB) {
2056                         if (cmd == POOL_SCRUB_PAUSE)
2057                                 return (zfs_error(hdl, EZFS_SCRUB_PAUSED, msg));
2058                         else
2059                                 return (zfs_error(hdl, EZFS_SCRUBBING, msg));
2060                 } else {
2061                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
2062                 }
2063         } else if (err == ENOENT) {
2064                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
2065         } else {
2066                 return (zpool_standard_error(hdl, err, msg));
2067         }
2068 }
2069
2070 static int
2071 xlate_init_err(int err)
2072 {
2073         switch (err) {
2074         case ENODEV:
2075                 return (EZFS_NODEVICE);
2076         case EINVAL:
2077         case EROFS:
2078                 return (EZFS_BADDEV);
2079         case EBUSY:
2080                 return (EZFS_INITIALIZING);
2081         case ESRCH:
2082                 return (EZFS_NO_INITIALIZE);
2083         }
2084         return (err);
2085 }
2086
2087 /*
2088  * Begin, suspend, or cancel the initialization (initializing of all free
2089  * blocks) for the given vdevs in the given pool.
2090  */
2091 int
2092 zpool_initialize(zpool_handle_t *zhp, pool_initialize_func_t cmd_type,
2093     nvlist_t *vds)
2094 {
2095         char msg[1024];
2096         libzfs_handle_t *hdl = zhp->zpool_hdl;
2097
2098         nvlist_t *errlist;
2099
2100         /* translate vdev names to guids */
2101         nvlist_t *vdev_guids = fnvlist_alloc();
2102         nvlist_t *guids_to_paths = fnvlist_alloc();
2103         boolean_t spare, cache;
2104         nvlist_t *tgt;
2105         nvpair_t *elem;
2106
2107         for (elem = nvlist_next_nvpair(vds, NULL); elem != NULL;
2108             elem = nvlist_next_nvpair(vds, elem)) {
2109                 char *vd_path = nvpair_name(elem);
2110                 tgt = zpool_find_vdev(zhp, vd_path, &spare, &cache, NULL);
2111
2112                 if ((tgt == NULL) || cache || spare) {
2113                         (void) snprintf(msg, sizeof (msg),
2114                             dgettext(TEXT_DOMAIN, "cannot initialize '%s'"),
2115                             vd_path);
2116                         int err = (tgt == NULL) ? EZFS_NODEVICE :
2117                             (spare ? EZFS_ISSPARE : EZFS_ISL2CACHE);
2118                         fnvlist_free(vdev_guids);
2119                         fnvlist_free(guids_to_paths);
2120                         return (zfs_error(hdl, err, msg));
2121                 }
2122
2123                 uint64_t guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
2124                 fnvlist_add_uint64(vdev_guids, vd_path, guid);
2125
2126                 (void) snprintf(msg, sizeof (msg), "%llu", guid);
2127                 fnvlist_add_string(guids_to_paths, msg, vd_path);
2128         }
2129
2130         int err = lzc_initialize(zhp->zpool_name, cmd_type, vdev_guids,
2131             &errlist);
2132         fnvlist_free(vdev_guids);
2133
2134         if (err == 0) {
2135                 fnvlist_free(guids_to_paths);
2136                 return (0);
2137         }
2138
2139         nvlist_t *vd_errlist = NULL;
2140         if (errlist != NULL) {
2141                 vd_errlist = fnvlist_lookup_nvlist(errlist,
2142                     ZPOOL_INITIALIZE_VDEVS);
2143         }
2144
2145         (void) snprintf(msg, sizeof (msg),
2146             dgettext(TEXT_DOMAIN, "operation failed"));
2147
2148         for (elem = nvlist_next_nvpair(vd_errlist, NULL); elem != NULL;
2149             elem = nvlist_next_nvpair(vd_errlist, elem)) {
2150                 int64_t vd_error = xlate_init_err(fnvpair_value_int64(elem));
2151                 char *path = fnvlist_lookup_string(guids_to_paths,
2152                     nvpair_name(elem));
2153                 (void) zfs_error_fmt(hdl, vd_error, "cannot initialize '%s'",
2154                     path);
2155         }
2156
2157         fnvlist_free(guids_to_paths);
2158         if (vd_errlist != NULL)
2159                 return (-1);
2160
2161         return (zpool_standard_error(hdl, err, msg));
2162 }
2163
2164 #ifdef illumos
2165 /*
2166  * This provides a very minimal check whether a given string is likely a
2167  * c#t#d# style string.  Users of this are expected to do their own
2168  * verification of the s# part.
2169  */
2170 #define CTD_CHECK(str)  (str && str[0] == 'c' && isdigit(str[1]))
2171
2172 /*
2173  * More elaborate version for ones which may start with "/dev/dsk/"
2174  * and the like.
2175  */
2176 static int
2177 ctd_check_path(char *str)
2178 {
2179         /*
2180          * If it starts with a slash, check the last component.
2181          */
2182         if (str && str[0] == '/') {
2183                 char *tmp = strrchr(str, '/');
2184
2185                 /*
2186                  * If it ends in "/old", check the second-to-last
2187                  * component of the string instead.
2188                  */
2189                 if (tmp != str && strcmp(tmp, "/old") == 0) {
2190                         for (tmp--; *tmp != '/'; tmp--)
2191                                 ;
2192                 }
2193                 str = tmp + 1;
2194         }
2195         return (CTD_CHECK(str));
2196 }
2197 #endif
2198
2199 /*
2200  * Find a vdev that matches the search criteria specified. We use the
2201  * the nvpair name to determine how we should look for the device.
2202  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
2203  * spare; but FALSE if its an INUSE spare.
2204  */
2205 static nvlist_t *
2206 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
2207     boolean_t *l2cache, boolean_t *log)
2208 {
2209         uint_t c, children;
2210         nvlist_t **child;
2211         nvlist_t *ret;
2212         uint64_t is_log;
2213         char *srchkey;
2214         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
2215
2216         /* Nothing to look for */
2217         if (search == NULL || pair == NULL)
2218                 return (NULL);
2219
2220         /* Obtain the key we will use to search */
2221         srchkey = nvpair_name(pair);
2222
2223         switch (nvpair_type(pair)) {
2224         case DATA_TYPE_UINT64:
2225                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
2226                         uint64_t srchval, theguid;
2227
2228                         verify(nvpair_value_uint64(pair, &srchval) == 0);
2229                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
2230                             &theguid) == 0);
2231                         if (theguid == srchval)
2232                                 return (nv);
2233                 }
2234                 break;
2235
2236         case DATA_TYPE_STRING: {
2237                 char *srchval, *val;
2238
2239                 verify(nvpair_value_string(pair, &srchval) == 0);
2240                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
2241                         break;
2242
2243                 /*
2244                  * Search for the requested value. Special cases:
2245                  *
2246                  * - ZPOOL_CONFIG_PATH for whole disk entries. To support
2247                  *   UEFI boot, these end in "s0" or "s0/old" or "s1" or
2248                  *   "s1/old".   The "s0" or "s1" part is hidden from the user,
2249                  *   but included in the string, so this matches around it.
2250                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
2251                  *
2252                  * Otherwise, all other searches are simple string compares.
2253                  */
2254 #ifdef illumos
2255                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0 &&
2256                     ctd_check_path(val)) {
2257                         uint64_t wholedisk = 0;
2258
2259                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
2260                             &wholedisk);
2261                         if (wholedisk) {
2262                                 int slen = strlen(srchval);
2263                                 int vlen = strlen(val);
2264
2265                                 if (slen != vlen - 2)
2266                                         break;
2267
2268                                 /*
2269                                  * make_leaf_vdev() should only set
2270                                  * wholedisk for ZPOOL_CONFIG_PATHs which
2271                                  * will include "/dev/dsk/", giving plenty of
2272                                  * room for the indices used next.
2273                                  */
2274                                 ASSERT(vlen >= 6);
2275
2276                                 /*
2277                                  * strings identical except trailing "s0"
2278                                  */
2279                                 if ((strcmp(&val[vlen - 2], "s0") == 0 ||
2280                                     strcmp(&val[vlen - 2], "s1") == 0) &&
2281                                     strncmp(srchval, val, slen) == 0)
2282                                         return (nv);
2283
2284                                 /*
2285                                  * strings identical except trailing "s0/old"
2286                                  */
2287                                 if ((strcmp(&val[vlen - 6], "s0/old") == 0 ||
2288                                     strcmp(&val[vlen - 6], "s1/old") == 0) &&
2289                                     strcmp(&srchval[slen - 4], "/old") == 0 &&
2290                                     strncmp(srchval, val, slen - 4) == 0)
2291                                         return (nv);
2292
2293                                 break;
2294                         }
2295                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2296 #else
2297                 if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2298 #endif
2299                         char *type, *idx, *end, *p;
2300                         uint64_t id, vdev_id;
2301
2302                         /*
2303                          * Determine our vdev type, keeping in mind
2304                          * that the srchval is composed of a type and
2305                          * vdev id pair (i.e. mirror-4).
2306                          */
2307                         if ((type = strdup(srchval)) == NULL)
2308                                 return (NULL);
2309
2310                         if ((p = strrchr(type, '-')) == NULL) {
2311                                 free(type);
2312                                 break;
2313                         }
2314                         idx = p + 1;
2315                         *p = '\0';
2316
2317                         /*
2318                          * If the types don't match then keep looking.
2319                          */
2320                         if (strncmp(val, type, strlen(val)) != 0) {
2321                                 free(type);
2322                                 break;
2323                         }
2324
2325                         verify(zpool_vdev_is_interior(type));
2326                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
2327                             &id) == 0);
2328
2329                         errno = 0;
2330                         vdev_id = strtoull(idx, &end, 10);
2331
2332                         free(type);
2333                         if (errno != 0)
2334                                 return (NULL);
2335
2336                         /*
2337                          * Now verify that we have the correct vdev id.
2338                          */
2339                         if (vdev_id == id)
2340                                 return (nv);
2341                 }
2342
2343                 /*
2344                  * Common case
2345                  */
2346                 if (strcmp(srchval, val) == 0)
2347                         return (nv);
2348                 break;
2349         }
2350
2351         default:
2352                 break;
2353         }
2354
2355         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2356             &child, &children) != 0)
2357                 return (NULL);
2358
2359         for (c = 0; c < children; c++) {
2360                 if ((ret = vdev_to_nvlist_iter(child[c], search,
2361                     avail_spare, l2cache, NULL)) != NULL) {
2362                         /*
2363                          * The 'is_log' value is only set for the toplevel
2364                          * vdev, not the leaf vdevs.  So we always lookup the
2365                          * log device from the root of the vdev tree (where
2366                          * 'log' is non-NULL).
2367                          */
2368                         if (log != NULL &&
2369                             nvlist_lookup_uint64(child[c],
2370                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2371                             is_log) {
2372                                 *log = B_TRUE;
2373                         }
2374                         return (ret);
2375                 }
2376         }
2377
2378         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2379             &child, &children) == 0) {
2380                 for (c = 0; c < children; c++) {
2381                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2382                             avail_spare, l2cache, NULL)) != NULL) {
2383                                 *avail_spare = B_TRUE;
2384                                 return (ret);
2385                         }
2386                 }
2387         }
2388
2389         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2390             &child, &children) == 0) {
2391                 for (c = 0; c < children; c++) {
2392                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2393                             avail_spare, l2cache, NULL)) != NULL) {
2394                                 *l2cache = B_TRUE;
2395                                 return (ret);
2396                         }
2397                 }
2398         }
2399
2400         return (NULL);
2401 }
2402
2403 /*
2404  * Given a physical path (minus the "/devices" prefix), find the
2405  * associated vdev.
2406  */
2407 nvlist_t *
2408 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2409     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2410 {
2411         nvlist_t *search, *nvroot, *ret;
2412
2413         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2414         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
2415
2416         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2417             &nvroot) == 0);
2418
2419         *avail_spare = B_FALSE;
2420         *l2cache = B_FALSE;
2421         if (log != NULL)
2422                 *log = B_FALSE;
2423         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2424         nvlist_free(search);
2425
2426         return (ret);
2427 }
2428
2429 /*
2430  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2431  */
2432 static boolean_t
2433 zpool_vdev_is_interior(const char *name)
2434 {
2435         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2436             strncmp(name, VDEV_TYPE_SPARE, strlen(VDEV_TYPE_SPARE)) == 0 ||
2437             strncmp(name,
2438             VDEV_TYPE_REPLACING, strlen(VDEV_TYPE_REPLACING)) == 0 ||
2439             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2440                 return (B_TRUE);
2441         return (B_FALSE);
2442 }
2443
2444 nvlist_t *
2445 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2446     boolean_t *l2cache, boolean_t *log)
2447 {
2448         char buf[MAXPATHLEN];
2449         char *end;
2450         nvlist_t *nvroot, *search, *ret;
2451         uint64_t guid;
2452
2453         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2454
2455         guid = strtoull(path, &end, 10);
2456         if (guid != 0 && *end == '\0') {
2457                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
2458         } else if (zpool_vdev_is_interior(path)) {
2459                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
2460         } else if (path[0] != '/') {
2461                 (void) snprintf(buf, sizeof (buf), "%s%s", _PATH_DEV, path);
2462                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, buf) == 0);
2463         } else {
2464                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
2465         }
2466
2467         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2468             &nvroot) == 0);
2469
2470         *avail_spare = B_FALSE;
2471         *l2cache = B_FALSE;
2472         if (log != NULL)
2473                 *log = B_FALSE;
2474         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2475         nvlist_free(search);
2476
2477         return (ret);
2478 }
2479
2480 static int
2481 vdev_is_online(nvlist_t *nv)
2482 {
2483         uint64_t ival;
2484
2485         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2486             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2487             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2488                 return (0);
2489
2490         return (1);
2491 }
2492
2493 /*
2494  * Helper function for zpool_get_physpaths().
2495  */
2496 static int
2497 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2498     size_t *bytes_written)
2499 {
2500         size_t bytes_left, pos, rsz;
2501         char *tmppath;
2502         const char *format;
2503
2504         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2505             &tmppath) != 0)
2506                 return (EZFS_NODEVICE);
2507
2508         pos = *bytes_written;
2509         bytes_left = physpath_size - pos;
2510         format = (pos == 0) ? "%s" : " %s";
2511
2512         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2513         *bytes_written += rsz;
2514
2515         if (rsz >= bytes_left) {
2516                 /* if physpath was not copied properly, clear it */
2517                 if (bytes_left != 0) {
2518                         physpath[pos] = 0;
2519                 }
2520                 return (EZFS_NOSPC);
2521         }
2522         return (0);
2523 }
2524
2525 static int
2526 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2527     size_t *rsz, boolean_t is_spare)
2528 {
2529         char *type;
2530         int ret;
2531
2532         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2533                 return (EZFS_INVALCONFIG);
2534
2535         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2536                 /*
2537                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2538                  * For a spare vdev, we only want to boot from the active
2539                  * spare device.
2540                  */
2541                 if (is_spare) {
2542                         uint64_t spare = 0;
2543                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2544                             &spare);
2545                         if (!spare)
2546                                 return (EZFS_INVALCONFIG);
2547                 }
2548
2549                 if (vdev_is_online(nv)) {
2550                         if ((ret = vdev_get_one_physpath(nv, physpath,
2551                             phypath_size, rsz)) != 0)
2552                                 return (ret);
2553                 }
2554         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2555             strcmp(type, VDEV_TYPE_RAIDZ) == 0 ||
2556             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2557             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2558                 nvlist_t **child;
2559                 uint_t count;
2560                 int i, ret;
2561
2562                 if (nvlist_lookup_nvlist_array(nv,
2563                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2564                         return (EZFS_INVALCONFIG);
2565
2566                 for (i = 0; i < count; i++) {
2567                         ret = vdev_get_physpaths(child[i], physpath,
2568                             phypath_size, rsz, is_spare);
2569                         if (ret == EZFS_NOSPC)
2570                                 return (ret);
2571                 }
2572         }
2573
2574         return (EZFS_POOL_INVALARG);
2575 }
2576
2577 /*
2578  * Get phys_path for a root pool config.
2579  * Return 0 on success; non-zero on failure.
2580  */
2581 static int
2582 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2583 {
2584         size_t rsz;
2585         nvlist_t *vdev_root;
2586         nvlist_t **child;
2587         uint_t count;
2588         char *type;
2589
2590         rsz = 0;
2591
2592         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2593             &vdev_root) != 0)
2594                 return (EZFS_INVALCONFIG);
2595
2596         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2597             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2598             &child, &count) != 0)
2599                 return (EZFS_INVALCONFIG);
2600
2601         /*
2602          * root pool can only have a single top-level vdev.
2603          */
2604         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1)
2605                 return (EZFS_POOL_INVALARG);
2606
2607         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2608             B_FALSE);
2609
2610         /* No online devices */
2611         if (rsz == 0)
2612                 return (EZFS_NODEVICE);
2613
2614         return (0);
2615 }
2616
2617 /*
2618  * Get phys_path for a root pool
2619  * Return 0 on success; non-zero on failure.
2620  */
2621 int
2622 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2623 {
2624         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2625             phypath_size));
2626 }
2627
2628 /*
2629  * If the device has being dynamically expanded then we need to relabel
2630  * the disk to use the new unallocated space.
2631  */
2632 static int
2633 zpool_relabel_disk(libzfs_handle_t *hdl, const char *name)
2634 {
2635 #ifdef illumos
2636         char path[MAXPATHLEN];
2637         char errbuf[1024];
2638         int fd, error;
2639         int (*_efi_use_whole_disk)(int);
2640
2641         if ((_efi_use_whole_disk = (int (*)(int))dlsym(RTLD_DEFAULT,
2642             "efi_use_whole_disk")) == NULL)
2643                 return (-1);
2644
2645         (void) snprintf(path, sizeof (path), "%s/%s", ZFS_RDISK_ROOT, name);
2646
2647         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
2648                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2649                     "relabel '%s': unable to open device"), name);
2650                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
2651         }
2652
2653         /*
2654          * It's possible that we might encounter an error if the device
2655          * does not have any unallocated space left. If so, we simply
2656          * ignore that error and continue on.
2657          */
2658         error = _efi_use_whole_disk(fd);
2659         (void) close(fd);
2660         if (error && error != VT_ENOSPC) {
2661                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2662                     "relabel '%s': unable to read disk capacity"), name);
2663                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
2664         }
2665 #endif  /* illumos */
2666         return (0);
2667 }
2668
2669 /*
2670  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2671  * ZFS_ONLINE_* flags.
2672  */
2673 int
2674 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2675     vdev_state_t *newstate)
2676 {
2677         zfs_cmd_t zc = { 0 };
2678         char msg[1024];
2679         char *pathname;
2680         nvlist_t *tgt;
2681         boolean_t avail_spare, l2cache, islog;
2682         libzfs_handle_t *hdl = zhp->zpool_hdl;
2683
2684         if (flags & ZFS_ONLINE_EXPAND) {
2685                 (void) snprintf(msg, sizeof (msg),
2686                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2687         } else {
2688                 (void) snprintf(msg, sizeof (msg),
2689                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2690         }
2691
2692         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2693         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2694             &islog)) == NULL)
2695                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2696
2697         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2698
2699         if (avail_spare)
2700                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2701
2702         if ((flags & ZFS_ONLINE_EXPAND ||
2703             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) &&
2704             nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH, &pathname) == 0) {
2705                 uint64_t wholedisk = 0;
2706
2707                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2708                     &wholedisk);
2709
2710                 /*
2711                  * XXX - L2ARC 1.0 devices can't support expansion.
2712                  */
2713                 if (l2cache) {
2714                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2715                             "cannot expand cache devices"));
2716                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2717                 }
2718
2719                 if (wholedisk) {
2720                         pathname += strlen(ZFS_DISK_ROOT) + 1;
2721                         (void) zpool_relabel_disk(hdl, pathname);
2722                 }
2723         }
2724
2725         zc.zc_cookie = VDEV_STATE_ONLINE;
2726         zc.zc_obj = flags;
2727
2728         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2729                 if (errno == EINVAL) {
2730                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2731                             "from this pool into a new one.  Use '%s' "
2732                             "instead"), "zpool detach");
2733                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2734                 }
2735                 return (zpool_standard_error(hdl, errno, msg));
2736         }
2737
2738         *newstate = zc.zc_cookie;
2739         return (0);
2740 }
2741
2742 /*
2743  * Take the specified vdev offline
2744  */
2745 int
2746 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2747 {
2748         zfs_cmd_t zc = { 0 };
2749         char msg[1024];
2750         nvlist_t *tgt;
2751         boolean_t avail_spare, l2cache;
2752         libzfs_handle_t *hdl = zhp->zpool_hdl;
2753
2754         (void) snprintf(msg, sizeof (msg),
2755             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2756
2757         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2758         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2759             NULL)) == NULL)
2760                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2761
2762         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2763
2764         if (avail_spare)
2765                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2766
2767         zc.zc_cookie = VDEV_STATE_OFFLINE;
2768         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2769
2770         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2771                 return (0);
2772
2773         switch (errno) {
2774         case EBUSY:
2775
2776                 /*
2777                  * There are no other replicas of this device.
2778                  */
2779                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2780
2781         case EEXIST:
2782                 /*
2783                  * The log device has unplayed logs
2784                  */
2785                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2786
2787         default:
2788                 return (zpool_standard_error(hdl, errno, msg));
2789         }
2790 }
2791
2792 /*
2793  * Mark the given vdev faulted.
2794  */
2795 int
2796 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2797 {
2798         zfs_cmd_t zc = { 0 };
2799         char msg[1024];
2800         libzfs_handle_t *hdl = zhp->zpool_hdl;
2801
2802         (void) snprintf(msg, sizeof (msg),
2803             dgettext(TEXT_DOMAIN, "cannot fault %llu"), guid);
2804
2805         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2806         zc.zc_guid = guid;
2807         zc.zc_cookie = VDEV_STATE_FAULTED;
2808         zc.zc_obj = aux;
2809
2810         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2811                 return (0);
2812
2813         switch (errno) {
2814         case EBUSY:
2815
2816                 /*
2817                  * There are no other replicas of this device.
2818                  */
2819                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2820
2821         default:
2822                 return (zpool_standard_error(hdl, errno, msg));
2823         }
2824
2825 }
2826
2827 /*
2828  * Mark the given vdev degraded.
2829  */
2830 int
2831 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2832 {
2833         zfs_cmd_t zc = { 0 };
2834         char msg[1024];
2835         libzfs_handle_t *hdl = zhp->zpool_hdl;
2836
2837         (void) snprintf(msg, sizeof (msg),
2838             dgettext(TEXT_DOMAIN, "cannot degrade %llu"), guid);
2839
2840         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2841         zc.zc_guid = guid;
2842         zc.zc_cookie = VDEV_STATE_DEGRADED;
2843         zc.zc_obj = aux;
2844
2845         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2846                 return (0);
2847
2848         return (zpool_standard_error(hdl, errno, msg));
2849 }
2850
2851 /*
2852  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2853  * a hot spare.
2854  */
2855 static boolean_t
2856 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2857 {
2858         nvlist_t **child;
2859         uint_t c, children;
2860         char *type;
2861
2862         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2863             &children) == 0) {
2864                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2865                     &type) == 0);
2866
2867                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2868                     children == 2 && child[which] == tgt)
2869                         return (B_TRUE);
2870
2871                 for (c = 0; c < children; c++)
2872                         if (is_replacing_spare(child[c], tgt, which))
2873                                 return (B_TRUE);
2874         }
2875
2876         return (B_FALSE);
2877 }
2878
2879 /*
2880  * Attach new_disk (fully described by nvroot) to old_disk.
2881  * If 'replacing' is specified, the new disk will replace the old one.
2882  */
2883 int
2884 zpool_vdev_attach(zpool_handle_t *zhp,
2885     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2886 {
2887         zfs_cmd_t zc = { 0 };
2888         char msg[1024];
2889         int ret;
2890         nvlist_t *tgt;
2891         boolean_t avail_spare, l2cache, islog;
2892         uint64_t val;
2893         char *newname;
2894         nvlist_t **child;
2895         uint_t children;
2896         nvlist_t *config_root;
2897         libzfs_handle_t *hdl = zhp->zpool_hdl;
2898         boolean_t rootpool = zpool_is_bootable(zhp);
2899
2900         if (replacing)
2901                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2902                     "cannot replace %s with %s"), old_disk, new_disk);
2903         else
2904                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2905                     "cannot attach %s to %s"), new_disk, old_disk);
2906
2907         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2908         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2909             &islog)) == NULL)
2910                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2911
2912         if (avail_spare)
2913                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2914
2915         if (l2cache)
2916                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2917
2918         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2919         zc.zc_cookie = replacing;
2920
2921         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2922             &child, &children) != 0 || children != 1) {
2923                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2924                     "new device must be a single disk"));
2925                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2926         }
2927
2928         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2929             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2930
2931         if ((newname = zpool_vdev_name(NULL, NULL, child[0], 0)) == NULL)
2932                 return (-1);
2933
2934         /*
2935          * If the target is a hot spare that has been swapped in, we can only
2936          * replace it with another hot spare.
2937          */
2938         if (replacing &&
2939             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2940             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2941             NULL) == NULL || !avail_spare) &&
2942             is_replacing_spare(config_root, tgt, 1)) {
2943                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2944                     "can only be replaced by another hot spare"));
2945                 free(newname);
2946                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2947         }
2948
2949         free(newname);
2950
2951         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2952                 return (-1);
2953
2954         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2955
2956         zcmd_free_nvlists(&zc);
2957
2958         if (ret == 0) {
2959                 if (rootpool) {
2960                         /*
2961                          * XXX need a better way to prevent user from
2962                          * booting up a half-baked vdev.
2963                          */
2964                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2965                             "sure to wait until resilver is done "
2966                             "before rebooting.\n"));
2967                         (void) fprintf(stderr, "\n");
2968                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "If "
2969                             "you boot from pool '%s', you may need to update\n"
2970                             "boot code on newly attached disk '%s'.\n\n"
2971                             "Assuming you use GPT partitioning and 'da0' is "
2972                             "your new boot disk\n"
2973                             "you may use the following command:\n\n"
2974                             "\tgpart bootcode -b /boot/pmbr -p "
2975                             "/boot/gptzfsboot -i 1 da0\n\n"),
2976                             zhp->zpool_name, new_disk);
2977                 }
2978                 return (0);
2979         }
2980
2981         switch (errno) {
2982         case ENOTSUP:
2983                 /*
2984                  * Can't attach to or replace this type of vdev.
2985                  */
2986                 if (replacing) {
2987                         uint64_t version = zpool_get_prop_int(zhp,
2988                             ZPOOL_PROP_VERSION, NULL);
2989
2990                         if (islog)
2991                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2992                                     "cannot replace a log with a spare"));
2993                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2994                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2995                                     "already in replacing/spare config; wait "
2996                                     "for completion or use 'zpool detach'"));
2997                         else
2998                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2999                                     "cannot replace a replacing device"));
3000                 } else {
3001                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3002                             "can only attach to mirrors and top-level "
3003                             "disks"));
3004                 }
3005                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
3006                 break;
3007
3008         case EINVAL:
3009                 /*
3010                  * The new device must be a single disk.
3011                  */
3012                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3013                     "new device must be a single disk"));
3014                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
3015                 break;
3016
3017         case EBUSY:
3018                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy, "
3019                     "or device removal is in progress"),
3020                     new_disk);
3021                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
3022                 break;
3023
3024         case EOVERFLOW:
3025                 /*
3026                  * The new device is too small.
3027                  */
3028                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3029                     "device is too small"));
3030                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
3031                 break;
3032
3033         case EDOM:
3034                 /*
3035                  * The new device has a different alignment requirement.
3036                  */
3037                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3038                     "devices have different sector alignment"));
3039                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
3040                 break;
3041
3042         case ENAMETOOLONG:
3043                 /*
3044                  * The resulting top-level vdev spec won't fit in the label.
3045                  */
3046                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
3047                 break;
3048
3049         default:
3050                 (void) zpool_standard_error(hdl, errno, msg);
3051         }
3052
3053         return (-1);
3054 }
3055
3056 /*
3057  * Detach the specified device.
3058  */
3059 int
3060 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
3061 {
3062         zfs_cmd_t zc = { 0 };
3063         char msg[1024];
3064         nvlist_t *tgt;
3065         boolean_t avail_spare, l2cache;
3066         libzfs_handle_t *hdl = zhp->zpool_hdl;
3067
3068         (void) snprintf(msg, sizeof (msg),
3069             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
3070
3071         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3072         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3073             NULL)) == NULL)
3074                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3075
3076         if (avail_spare)
3077                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
3078
3079         if (l2cache)
3080                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
3081
3082         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
3083
3084         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
3085                 return (0);
3086
3087         switch (errno) {
3088
3089         case ENOTSUP:
3090                 /*
3091                  * Can't detach from this type of vdev.
3092                  */
3093                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
3094                     "applicable to mirror and replacing vdevs"));
3095                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
3096                 break;
3097
3098         case EBUSY:
3099                 /*
3100                  * There are no other replicas of this device.
3101                  */
3102                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
3103                 break;
3104
3105         default:
3106                 (void) zpool_standard_error(hdl, errno, msg);
3107         }
3108
3109         return (-1);
3110 }
3111
3112 /*
3113  * Find a mirror vdev in the source nvlist.
3114  *
3115  * The mchild array contains a list of disks in one of the top-level mirrors
3116  * of the source pool.  The schild array contains a list of disks that the
3117  * user specified on the command line.  We loop over the mchild array to
3118  * see if any entry in the schild array matches.
3119  *
3120  * If a disk in the mchild array is found in the schild array, we return
3121  * the index of that entry.  Otherwise we return -1.
3122  */
3123 static int
3124 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
3125     nvlist_t **schild, uint_t schildren)
3126 {
3127         uint_t mc;
3128
3129         for (mc = 0; mc < mchildren; mc++) {
3130                 uint_t sc;
3131                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
3132                     mchild[mc], 0);
3133
3134                 for (sc = 0; sc < schildren; sc++) {
3135                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
3136                             schild[sc], 0);
3137                         boolean_t result = (strcmp(mpath, spath) == 0);
3138
3139                         free(spath);
3140                         if (result) {
3141                                 free(mpath);
3142                                 return (mc);
3143                         }
3144                 }
3145
3146                 free(mpath);
3147         }
3148
3149         return (-1);
3150 }
3151
3152 /*
3153  * Split a mirror pool.  If newroot points to null, then a new nvlist
3154  * is generated and it is the responsibility of the caller to free it.
3155  */
3156 int
3157 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
3158     nvlist_t *props, splitflags_t flags)
3159 {
3160         zfs_cmd_t zc = { 0 };
3161         char msg[1024];
3162         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
3163         nvlist_t **varray = NULL, *zc_props = NULL;
3164         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
3165         libzfs_handle_t *hdl = zhp->zpool_hdl;
3166         uint64_t vers;
3167         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
3168         int retval = 0;
3169
3170         (void) snprintf(msg, sizeof (msg),
3171             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
3172
3173         if (!zpool_name_valid(hdl, B_FALSE, newname))
3174                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
3175
3176         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
3177                 (void) fprintf(stderr, gettext("Internal error: unable to "
3178                     "retrieve pool configuration\n"));
3179                 return (-1);
3180         }
3181
3182         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
3183             == 0);
3184         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
3185
3186         if (props) {
3187                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
3188                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
3189                     props, vers, flags, msg)) == NULL)
3190                         return (-1);
3191         }
3192
3193         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
3194             &children) != 0) {
3195                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3196                     "Source pool is missing vdev tree"));
3197                 nvlist_free(zc_props);
3198                 return (-1);
3199         }
3200
3201         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
3202         vcount = 0;
3203
3204         if (*newroot == NULL ||
3205             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
3206             &newchild, &newchildren) != 0)
3207                 newchildren = 0;
3208
3209         for (c = 0; c < children; c++) {
3210                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
3211                 char *type;
3212                 nvlist_t **mchild, *vdev;
3213                 uint_t mchildren;
3214                 int entry;
3215
3216                 /*
3217                  * Unlike cache & spares, slogs are stored in the
3218                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
3219                  */
3220                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
3221                     &is_log);
3222                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
3223                     &is_hole);
3224                 if (is_log || is_hole) {
3225                         /*
3226                          * Create a hole vdev and put it in the config.
3227                          */
3228                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
3229                                 goto out;
3230                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
3231                             VDEV_TYPE_HOLE) != 0)
3232                                 goto out;
3233                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
3234                             1) != 0)
3235                                 goto out;
3236                         if (lastlog == 0)
3237                                 lastlog = vcount;
3238                         varray[vcount++] = vdev;
3239                         continue;
3240                 }
3241                 lastlog = 0;
3242                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
3243                     == 0);
3244                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
3245                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3246                             "Source pool must be composed only of mirrors\n"));
3247                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3248                         goto out;
3249                 }
3250
3251                 verify(nvlist_lookup_nvlist_array(child[c],
3252                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
3253
3254                 /* find or add an entry for this top-level vdev */
3255                 if (newchildren > 0 &&
3256                     (entry = find_vdev_entry(zhp, mchild, mchildren,
3257                     newchild, newchildren)) >= 0) {
3258                         /* We found a disk that the user specified. */
3259                         vdev = mchild[entry];
3260                         ++found;
3261                 } else {
3262                         /* User didn't specify a disk for this vdev. */
3263                         vdev = mchild[mchildren - 1];
3264                 }
3265
3266                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
3267                         goto out;
3268         }
3269
3270         /* did we find every disk the user specified? */
3271         if (found != newchildren) {
3272                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
3273                     "include at most one disk from each mirror"));
3274                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3275                 goto out;
3276         }
3277
3278         /* Prepare the nvlist for populating. */
3279         if (*newroot == NULL) {
3280                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
3281                         goto out;
3282                 freelist = B_TRUE;
3283                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
3284                     VDEV_TYPE_ROOT) != 0)
3285                         goto out;
3286         } else {
3287                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
3288         }
3289
3290         /* Add all the children we found */
3291         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
3292             lastlog == 0 ? vcount : lastlog) != 0)
3293                 goto out;
3294
3295         /*
3296          * If we're just doing a dry run, exit now with success.
3297          */
3298         if (flags.dryrun) {
3299                 memory_err = B_FALSE;
3300                 freelist = B_FALSE;
3301                 goto out;
3302         }
3303
3304         /* now build up the config list & call the ioctl */
3305         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3306                 goto out;
3307
3308         if (nvlist_add_nvlist(newconfig,
3309             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3310             nvlist_add_string(newconfig,
3311             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3312             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3313                 goto out;
3314
3315         /*
3316          * The new pool is automatically part of the namespace unless we
3317          * explicitly export it.
3318          */
3319         if (!flags.import)
3320                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3321         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3322         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3323         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
3324                 goto out;
3325         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
3326                 goto out;
3327
3328         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3329                 retval = zpool_standard_error(hdl, errno, msg);
3330                 goto out;
3331         }
3332
3333         freelist = B_FALSE;
3334         memory_err = B_FALSE;
3335
3336 out:
3337         if (varray != NULL) {
3338                 int v;
3339
3340                 for (v = 0; v < vcount; v++)
3341                         nvlist_free(varray[v]);
3342                 free(varray);
3343         }
3344         zcmd_free_nvlists(&zc);
3345         nvlist_free(zc_props);
3346         nvlist_free(newconfig);
3347         if (freelist) {
3348                 nvlist_free(*newroot);
3349                 *newroot = NULL;
3350         }
3351
3352         if (retval != 0)
3353                 return (retval);
3354
3355         if (memory_err)
3356                 return (no_memory(hdl));
3357
3358         return (0);
3359 }
3360
3361 /*
3362  * Remove the given device.
3363  */
3364 int
3365 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3366 {
3367         zfs_cmd_t zc = { 0 };
3368         char msg[1024];
3369         nvlist_t *tgt;
3370         boolean_t avail_spare, l2cache, islog;
3371         libzfs_handle_t *hdl = zhp->zpool_hdl;
3372         uint64_t version;
3373
3374         (void) snprintf(msg, sizeof (msg),
3375             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3376
3377         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3378         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3379             &islog)) == NULL)
3380                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3381
3382         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3383         if (islog && version < SPA_VERSION_HOLES) {
3384                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3385                     "pool must be upgraded to support log removal"));
3386                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
3387         }
3388
3389         zc.zc_guid = fnvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID);
3390
3391         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3392                 return (0);
3393
3394         switch (errno) {
3395
3396         case EINVAL:
3397                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3398                     "invalid config; all top-level vdevs must "
3399                     "have the same sector size and not be raidz."));
3400                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
3401                 break;
3402
3403         case EBUSY:
3404                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3405                     "Pool busy; removal may already be in progress"));
3406                 (void) zfs_error(hdl, EZFS_BUSY, msg);
3407                 break;
3408
3409         default:
3410                 (void) zpool_standard_error(hdl, errno, msg);
3411         }
3412         return (-1);
3413 }
3414
3415 int
3416 zpool_vdev_remove_cancel(zpool_handle_t *zhp)
3417 {
3418         zfs_cmd_t zc = { 0 };
3419         char msg[1024];
3420         libzfs_handle_t *hdl = zhp->zpool_hdl;
3421
3422         (void) snprintf(msg, sizeof (msg),
3423             dgettext(TEXT_DOMAIN, "cannot cancel removal"));
3424
3425         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3426         zc.zc_cookie = 1;
3427
3428         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3429                 return (0);
3430
3431         return (zpool_standard_error(hdl, errno, msg));
3432 }
3433
3434 int
3435 zpool_vdev_indirect_size(zpool_handle_t *zhp, const char *path,
3436     uint64_t *sizep)
3437 {
3438         char msg[1024];
3439         nvlist_t *tgt;
3440         boolean_t avail_spare, l2cache, islog;
3441         libzfs_handle_t *hdl = zhp->zpool_hdl;
3442
3443         (void) snprintf(msg, sizeof (msg),
3444             dgettext(TEXT_DOMAIN, "cannot determine indirect size of %s"),
3445             path);
3446
3447         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3448             &islog)) == NULL)
3449                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3450
3451         if (avail_spare || l2cache || islog) {
3452                 *sizep = 0;
3453                 return (0);
3454         }
3455
3456         if (nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_INDIRECT_SIZE, sizep) != 0) {
3457                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3458                     "indirect size not available"));
3459                 return (zfs_error(hdl, EINVAL, msg));
3460         }
3461         return (0);
3462 }
3463
3464 /*
3465  * Clear the errors for the pool, or the particular device if specified.
3466  */
3467 int
3468 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3469 {
3470         zfs_cmd_t zc = { 0 };
3471         char msg[1024];
3472         nvlist_t *tgt;
3473         zpool_load_policy_t policy;
3474         boolean_t avail_spare, l2cache;
3475         libzfs_handle_t *hdl = zhp->zpool_hdl;
3476         nvlist_t *nvi = NULL;
3477         int error;
3478
3479         if (path)
3480                 (void) snprintf(msg, sizeof (msg),
3481                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3482                     path);
3483         else
3484                 (void) snprintf(msg, sizeof (msg),
3485                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3486                     zhp->zpool_name);
3487
3488         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3489         if (path) {
3490                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3491                     &l2cache, NULL)) == NULL)
3492                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
3493
3494                 /*
3495                  * Don't allow error clearing for hot spares.  Do allow
3496                  * error clearing for l2cache devices.
3497                  */
3498                 if (avail_spare)
3499                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
3500
3501                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
3502                     &zc.zc_guid) == 0);
3503         }
3504
3505         zpool_get_load_policy(rewindnvl, &policy);
3506         zc.zc_cookie = policy.zlp_rewind;
3507
3508         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
3509                 return (-1);
3510
3511         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
3512                 return (-1);
3513
3514         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3515             errno == ENOMEM) {
3516                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3517                         zcmd_free_nvlists(&zc);
3518                         return (-1);
3519                 }
3520         }
3521
3522         if (!error || ((policy.zlp_rewind & ZPOOL_TRY_REWIND) &&
3523             errno != EPERM && errno != EACCES)) {
3524                 if (policy.zlp_rewind &
3525                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
3526                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3527                         zpool_rewind_exclaim(hdl, zc.zc_name,
3528                             ((policy.zlp_rewind & ZPOOL_TRY_REWIND) != 0),
3529                             nvi);
3530                         nvlist_free(nvi);
3531                 }
3532                 zcmd_free_nvlists(&zc);
3533                 return (0);
3534         }
3535
3536         zcmd_free_nvlists(&zc);
3537         return (zpool_standard_error(hdl, errno, msg));
3538 }
3539
3540 /*
3541  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3542  */
3543 int
3544 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3545 {
3546         zfs_cmd_t zc = { 0 };
3547         char msg[1024];
3548         libzfs_handle_t *hdl = zhp->zpool_hdl;
3549
3550         (void) snprintf(msg, sizeof (msg),
3551             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3552             guid);
3553
3554         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3555         zc.zc_guid = guid;
3556         zc.zc_cookie = ZPOOL_NO_REWIND;
3557
3558         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
3559                 return (0);
3560
3561         return (zpool_standard_error(hdl, errno, msg));
3562 }
3563
3564 /*
3565  * Change the GUID for a pool.
3566  */
3567 int
3568 zpool_reguid(zpool_handle_t *zhp)
3569 {
3570         char msg[1024];
3571         libzfs_handle_t *hdl = zhp->zpool_hdl;
3572         zfs_cmd_t zc = { 0 };
3573
3574         (void) snprintf(msg, sizeof (msg),
3575             dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3576
3577         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3578         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3579                 return (0);
3580
3581         return (zpool_standard_error(hdl, errno, msg));
3582 }
3583
3584 /*
3585  * Reopen the pool.
3586  */
3587 int
3588 zpool_reopen(zpool_handle_t *zhp)
3589 {
3590         zfs_cmd_t zc = { 0 };
3591         char msg[1024];
3592         libzfs_handle_t *hdl = zhp->zpool_hdl;
3593
3594         (void) snprintf(msg, sizeof (msg),
3595             dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3596             zhp->zpool_name);
3597
3598         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3599         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3600                 return (0);
3601         return (zpool_standard_error(hdl, errno, msg));
3602 }
3603
3604 /* call into libzfs_core to execute the sync IOCTL per pool */
3605 int
3606 zpool_sync_one(zpool_handle_t *zhp, void *data)
3607 {
3608         int ret;
3609         libzfs_handle_t *hdl = zpool_get_handle(zhp);
3610         const char *pool_name = zpool_get_name(zhp);
3611         boolean_t *force = data;
3612         nvlist_t *innvl = fnvlist_alloc();
3613
3614         fnvlist_add_boolean_value(innvl, "force", *force);
3615         if ((ret = lzc_sync(pool_name, innvl, NULL)) != 0) {
3616                 nvlist_free(innvl);
3617                 return (zpool_standard_error_fmt(hdl, ret,
3618                     dgettext(TEXT_DOMAIN, "sync '%s' failed"), pool_name));
3619         }
3620         nvlist_free(innvl);
3621
3622         return (0);
3623 }
3624
3625 /*
3626  * Convert from a devid string to a path.
3627  */
3628 static char *
3629 devid_to_path(char *devid_str)
3630 {
3631         ddi_devid_t devid;
3632         char *minor;
3633         char *path;
3634         devid_nmlist_t *list = NULL;
3635         int ret;
3636
3637         if (devid_str_decode(devid_str, &devid, &minor) != 0)
3638                 return (NULL);
3639
3640         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3641
3642         devid_str_free(minor);
3643         devid_free(devid);
3644
3645         if (ret != 0)
3646                 return (NULL);
3647
3648         /*
3649          * In a case the strdup() fails, we will just return NULL below.
3650          */
3651         path = strdup(list[0].devname);
3652
3653         devid_free_nmlist(list);
3654
3655         return (path);
3656 }
3657
3658 /*
3659  * Convert from a path to a devid string.
3660  */
3661 static char *
3662 path_to_devid(const char *path)
3663 {
3664 #ifdef have_devid
3665         int fd;
3666         ddi_devid_t devid;
3667         char *minor, *ret;
3668
3669         if ((fd = open(path, O_RDONLY)) < 0)
3670                 return (NULL);
3671
3672         minor = NULL;
3673         ret = NULL;
3674         if (devid_get(fd, &devid) == 0) {
3675                 if (devid_get_minor_name(fd, &minor) == 0)
3676                         ret = devid_str_encode(devid, minor);
3677                 if (minor != NULL)
3678                         devid_str_free(minor);
3679                 devid_free(devid);
3680         }
3681         (void) close(fd);
3682
3683         return (ret);
3684 #else
3685         return (NULL);
3686 #endif
3687 }
3688
3689 /*
3690  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3691  * ignore any failure here, since a common case is for an unprivileged user to
3692  * type 'zpool status', and we'll display the correct information anyway.
3693  */
3694 static void
3695 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3696 {
3697         zfs_cmd_t zc = { 0 };
3698
3699         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3700         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3701         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3702             &zc.zc_guid) == 0);
3703
3704         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3705 }
3706
3707 /*
3708  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3709  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3710  * We also check if this is a whole disk, in which case we strip off the
3711  * trailing 's0' slice name.
3712  *
3713  * This routine is also responsible for identifying when disks have been
3714  * reconfigured in a new location.  The kernel will have opened the device by
3715  * devid, but the path will still refer to the old location.  To catch this, we
3716  * first do a path -> devid translation (which is fast for the common case).  If
3717  * the devid matches, we're done.  If not, we do a reverse devid -> path
3718  * translation and issue the appropriate ioctl() to update the path of the vdev.
3719  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3720  * of these checks.
3721  */
3722 char *
3723 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3724     int name_flags)
3725 {
3726         char *path, *devid, *env;
3727         uint64_t value;
3728         char buf[64];
3729         vdev_stat_t *vs;
3730         uint_t vsc;
3731         int have_stats;
3732         int have_path;
3733
3734         env = getenv("ZPOOL_VDEV_NAME_PATH");
3735         if (env && (strtoul(env, NULL, 0) > 0 ||
3736             !strncasecmp(env, "YES", 3) || !strncasecmp(env, "ON", 2)))
3737                 name_flags |= VDEV_NAME_PATH;
3738
3739         env = getenv("ZPOOL_VDEV_NAME_GUID");
3740         if (env && (strtoul(env, NULL, 0) > 0 ||
3741             !strncasecmp(env, "YES", 3) || !strncasecmp(env, "ON", 2)))
3742                 name_flags |= VDEV_NAME_GUID;
3743
3744         env = getenv("ZPOOL_VDEV_NAME_FOLLOW_LINKS");
3745         if (env && (strtoul(env, NULL, 0) > 0 ||
3746             !strncasecmp(env, "YES", 3) || !strncasecmp(env, "ON", 2)))
3747                 name_flags |= VDEV_NAME_FOLLOW_LINKS;
3748
3749         have_stats = nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3750             (uint64_t **)&vs, &vsc) == 0;
3751         have_path = nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0;
3752
3753         /*
3754          * If the device is not currently present, assume it will not
3755          * come back at the same device path.  Display the device by GUID.
3756          */
3757         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &value) == 0 ||
3758             (name_flags & VDEV_NAME_GUID) != 0 ||
3759             have_path && have_stats && vs->vs_state <= VDEV_STATE_CANT_OPEN) {
3760                 nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value);
3761                 (void) snprintf(buf, sizeof (buf), "%llu", (u_longlong_t)value);
3762                 path = buf;
3763         } else if (have_path) {
3764
3765                 /*
3766                  * If the device is dead (faulted, offline, etc) then don't
3767                  * bother opening it.  Otherwise we may be forcing the user to
3768                  * open a misbehaving device, which can have undesirable
3769                  * effects.
3770                  */
3771                 if ((have_stats == 0 ||
3772                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3773                     zhp != NULL &&
3774                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3775                         /*
3776                          * Determine if the current path is correct.
3777                          */
3778                         char *newdevid = path_to_devid(path);
3779
3780                         if (newdevid == NULL ||
3781                             strcmp(devid, newdevid) != 0) {
3782                                 char *newpath;
3783
3784                                 if ((newpath = devid_to_path(devid)) != NULL) {
3785                                         /*
3786                                          * Update the path appropriately.
3787                                          */
3788                                         set_path(zhp, nv, newpath);
3789                                         if (nvlist_add_string(nv,
3790                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3791                                                 verify(nvlist_lookup_string(nv,
3792                                                     ZPOOL_CONFIG_PATH,
3793                                                     &path) == 0);
3794                                         free(newpath);
3795                                 }
3796                         }
3797
3798                         if (newdevid)
3799                                 devid_str_free(newdevid);
3800                 }
3801
3802 #ifdef illumos
3803                 if (name_flags & VDEV_NAME_FOLLOW_LINKS) {
3804                         char *rp = realpath(path, NULL);
3805                         if (rp) {
3806                                 strlcpy(buf, rp, sizeof (buf));
3807                                 path = buf;
3808                                 free(rp);
3809                         }
3810                 }
3811
3812                 if (strncmp(path, ZFS_DISK_ROOTD, strlen(ZFS_DISK_ROOTD)) == 0)
3813                         path += strlen(ZFS_DISK_ROOTD);
3814
3815                 /*
3816                  * Remove the partition from the path it this is a whole disk.
3817                  */
3818                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, &value)
3819                     == 0 && value && !(name_flags & VDEV_NAME_PATH)) {
3820                         int pathlen = strlen(path);
3821                         char *tmp = zfs_strdup(hdl, path);
3822
3823                         /*
3824                          * If it starts with c#, and ends with "s0" or "s1",
3825                          * chop the slice off, or if it ends with "s0/old" or
3826                          * "s1/old", remove the slice from the middle.
3827                          */
3828                         if (CTD_CHECK(tmp)) {
3829                                 if (strcmp(&tmp[pathlen - 2], "s0") == 0 ||
3830                                     strcmp(&tmp[pathlen - 2], "s1") == 0) {
3831                                         tmp[pathlen - 2] = '\0';
3832                                 } else if (pathlen > 6 &&
3833                                     (strcmp(&tmp[pathlen - 6], "s0/old") == 0 ||
3834                                     strcmp(&tmp[pathlen - 6], "s1/old") == 0)) {
3835                                         (void) strcpy(&tmp[pathlen - 6],
3836                                             "/old");
3837                                 }
3838                         }
3839                         return (tmp);
3840                 }
3841 #else   /* !illumos */
3842                 if (strncmp(path, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
3843                         path += sizeof(_PATH_DEV) - 1;
3844 #endif  /* illumos */
3845         } else {
3846                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3847
3848                 /*
3849                  * If it's a raidz device, we need to stick in the parity level.
3850                  */
3851                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3852                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3853                             &value) == 0);
3854                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3855                             (u_longlong_t)value);
3856                         path = buf;
3857                 }
3858
3859                 /*
3860                  * We identify each top-level vdev by using a <type-id>
3861                  * naming convention.
3862                  */
3863                 if (name_flags & VDEV_NAME_TYPE_ID) {
3864                         uint64_t id;
3865
3866                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3867                             &id) == 0);
3868                         (void) snprintf(buf, sizeof (buf), "%s-%llu", path,
3869                             (u_longlong_t)id);
3870                         path = buf;
3871                 }
3872         }
3873
3874         return (zfs_strdup(hdl, path));
3875 }
3876
3877 static int
3878 zbookmark_mem_compare(const void *a, const void *b)
3879 {
3880         return (memcmp(a, b, sizeof (zbookmark_phys_t)));
3881 }
3882
3883 /*
3884  * Retrieve the persistent error log, uniquify the members, and return to the
3885  * caller.
3886  */
3887 int
3888 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3889 {
3890         zfs_cmd_t zc = { 0 };
3891         uint64_t count;
3892         zbookmark_phys_t *zb = NULL;
3893         int i;
3894
3895         /*
3896          * Retrieve the raw error list from the kernel.  If the number of errors
3897          * has increased, allocate more space and continue until we get the
3898          * entire list.
3899          */
3900         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3901             &count) == 0);
3902         if (count == 0)
3903                 return (0);
3904         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3905             count * sizeof (zbookmark_phys_t))) == (uintptr_t)NULL)
3906                 return (-1);
3907         zc.zc_nvlist_dst_size = count;
3908         (void) strcpy(zc.zc_name, zhp->zpool_name);
3909         for (;;) {
3910                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3911                     &zc) != 0) {
3912                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3913                         if (errno == ENOMEM) {
3914                                 void *dst;
3915
3916                                 count = zc.zc_nvlist_dst_size;
3917                                 dst = zfs_alloc(zhp->zpool_hdl, count *
3918                                     sizeof (zbookmark_phys_t));
3919                                 if (dst == NULL)
3920                                         return (-1);
3921                                 zc.zc_nvlist_dst = (uintptr_t)dst;
3922                         } else {
3923                                 return (-1);
3924                         }
3925                 } else {
3926                         break;
3927                 }
3928         }
3929
3930         /*
3931          * Sort the resulting bookmarks.  This is a little confusing due to the
3932          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3933          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3934          * _not_ copied as part of the process.  So we point the start of our
3935          * array appropriate and decrement the total number of elements.
3936          */
3937         zb = ((zbookmark_phys_t *)(uintptr_t)zc.zc_nvlist_dst) +
3938             zc.zc_nvlist_dst_size;
3939         count -= zc.zc_nvlist_dst_size;
3940
3941         qsort(zb, count, sizeof (zbookmark_phys_t), zbookmark_mem_compare);
3942
3943         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3944
3945         /*
3946          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3947          */
3948         for (i = 0; i < count; i++) {
3949                 nvlist_t *nv;
3950
3951                 /* ignoring zb_blkid and zb_level for now */
3952                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3953                     zb[i-1].zb_object == zb[i].zb_object)
3954                         continue;
3955
3956                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3957                         goto nomem;
3958                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3959                     zb[i].zb_objset) != 0) {
3960                         nvlist_free(nv);
3961                         goto nomem;
3962                 }
3963                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3964                     zb[i].zb_object) != 0) {
3965                         nvlist_free(nv);
3966                         goto nomem;
3967                 }
3968                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3969                         nvlist_free(nv);
3970                         goto nomem;
3971                 }
3972                 nvlist_free(nv);
3973         }
3974
3975         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3976         return (0);
3977
3978 nomem:
3979         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3980         return (no_memory(zhp->zpool_hdl));
3981 }
3982
3983 /*
3984  * Upgrade a ZFS pool to the latest on-disk version.
3985  */
3986 int
3987 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3988 {
3989         zfs_cmd_t zc = { 0 };
3990         libzfs_handle_t *hdl = zhp->zpool_hdl;
3991
3992         (void) strcpy(zc.zc_name, zhp->zpool_name);
3993         zc.zc_cookie = new_version;
3994
3995         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3996                 return (zpool_standard_error_fmt(hdl, errno,
3997                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3998                     zhp->zpool_name));
3999         return (0);
4000 }
4001
4002 void
4003 zfs_save_arguments(int argc, char **argv, char *string, int len)
4004 {
4005         (void) strlcpy(string, basename(argv[0]), len);
4006         for (int i = 1; i < argc; i++) {
4007                 (void) strlcat(string, " ", len);
4008                 (void) strlcat(string, argv[i], len);
4009         }
4010 }
4011
4012 int
4013 zpool_log_history(libzfs_handle_t *hdl, const char *message)
4014 {
4015         zfs_cmd_t zc = { 0 };
4016         nvlist_t *args;
4017         int err;
4018
4019         args = fnvlist_alloc();
4020         fnvlist_add_string(args, "message", message);
4021         err = zcmd_write_src_nvlist(hdl, &zc, args);
4022         if (err == 0)
4023                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_LOG_HISTORY, &zc);
4024         nvlist_free(args);
4025         zcmd_free_nvlists(&zc);
4026         return (err);
4027 }
4028
4029 /*
4030  * Perform ioctl to get some command history of a pool.
4031  *
4032  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
4033  * logical offset of the history buffer to start reading from.
4034  *
4035  * Upon return, 'off' is the next logical offset to read from and
4036  * 'len' is the actual amount of bytes read into 'buf'.
4037  */
4038 static int
4039 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
4040 {
4041         zfs_cmd_t zc = { 0 };
4042         libzfs_handle_t *hdl = zhp->zpool_hdl;
4043
4044         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
4045
4046         zc.zc_history = (uint64_t)(uintptr_t)buf;
4047         zc.zc_history_len = *len;
4048         zc.zc_history_offset = *off;
4049
4050         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
4051                 switch (errno) {
4052                 case EPERM:
4053                         return (zfs_error_fmt(hdl, EZFS_PERM,
4054                             dgettext(TEXT_DOMAIN,
4055                             "cannot show history for pool '%s'"),
4056                             zhp->zpool_name));
4057                 case ENOENT:
4058                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
4059                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
4060                             "'%s'"), zhp->zpool_name));
4061                 case ENOTSUP:
4062                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
4063                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
4064                             "'%s', pool must be upgraded"), zhp->zpool_name));
4065                 default:
4066                         return (zpool_standard_error_fmt(hdl, errno,
4067                             dgettext(TEXT_DOMAIN,
4068                             "cannot get history for '%s'"), zhp->zpool_name));
4069                 }
4070         }
4071
4072         *len = zc.zc_history_len;
4073         *off = zc.zc_history_offset;
4074
4075         return (0);
4076 }
4077
4078 /*
4079  * Process the buffer of nvlists, unpacking and storing each nvlist record
4080  * into 'records'.  'leftover' is set to the number of bytes that weren't
4081  * processed as there wasn't a complete record.
4082  */
4083 int
4084 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
4085     nvlist_t ***records, uint_t *numrecords)
4086 {
4087         uint64_t reclen;
4088         nvlist_t *nv;
4089         int i;
4090
4091         while (bytes_read > sizeof (reclen)) {
4092
4093                 /* get length of packed record (stored as little endian) */
4094                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
4095                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
4096
4097                 if (bytes_read < sizeof (reclen) + reclen)
4098                         break;
4099
4100                 /* unpack record */
4101                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
4102                         return (ENOMEM);
4103                 bytes_read -= sizeof (reclen) + reclen;
4104                 buf += sizeof (reclen) + reclen;
4105
4106                 /* add record to nvlist array */
4107                 (*numrecords)++;
4108                 if (ISP2(*numrecords + 1)) {
4109                         *records = realloc(*records,
4110                             *numrecords * 2 * sizeof (nvlist_t *));
4111                 }
4112                 (*records)[*numrecords - 1] = nv;
4113         }
4114
4115         *leftover = bytes_read;
4116         return (0);
4117 }
4118
4119 /* from spa_history.c: spa_history_create_obj() */
4120 #define HIS_BUF_LEN_DEF (128 << 10)
4121 #define HIS_BUF_LEN_MAX (1 << 30)
4122
4123 /*
4124  * Retrieve the command history of a pool.
4125  */
4126 int
4127 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
4128 {
4129         char *buf;
4130         uint64_t buflen = HIS_BUF_LEN_DEF;
4131         uint64_t off = 0;
4132         nvlist_t **records = NULL;
4133         uint_t numrecords = 0;
4134         int err, i;
4135
4136         buf = malloc(buflen);
4137         if (buf == NULL)
4138                 return (ENOMEM);
4139         do {
4140                 uint64_t bytes_read = buflen;
4141                 uint64_t leftover;
4142
4143                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
4144                         break;
4145
4146                 /* if nothing else was read in, we're at EOF, just return */
4147                 if (bytes_read == 0)
4148                         break;
4149
4150                 if ((err = zpool_history_unpack(buf, bytes_read,
4151                     &leftover, &records, &numrecords)) != 0)
4152                         break;
4153                 off -= leftover;
4154                 if (leftover == bytes_read) {
4155                         /*
4156                          * no progress made, because buffer is not big enough
4157                          * to hold this record; resize and retry.
4158                          */
4159                         buflen *= 2;
4160                         free(buf);
4161                         buf = NULL;
4162                         if ((buflen >= HIS_BUF_LEN_MAX) ||
4163                             ((buf = malloc(buflen)) == NULL)) {
4164                                 err = ENOMEM;
4165                                 break;
4166                         }
4167                 }
4168
4169                 /* CONSTCOND */
4170         } while (1);
4171
4172         free(buf);
4173
4174         if (!err) {
4175                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
4176                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
4177                     records, numrecords) == 0);
4178         }
4179         for (i = 0; i < numrecords; i++)
4180                 nvlist_free(records[i]);
4181         free(records);
4182
4183         return (err);
4184 }
4185
4186 void
4187 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
4188     char *pathname, size_t len)
4189 {
4190         zfs_cmd_t zc = { 0 };
4191         boolean_t mounted = B_FALSE;
4192         char *mntpnt = NULL;
4193         char dsname[ZFS_MAX_DATASET_NAME_LEN];
4194
4195         if (dsobj == 0) {
4196                 /* special case for the MOS */
4197                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>", obj);
4198                 return;
4199         }
4200
4201         /* get the dataset's name */
4202         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
4203         zc.zc_obj = dsobj;
4204         if (ioctl(zhp->zpool_hdl->libzfs_fd,
4205             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
4206                 /* just write out a path of two object numbers */
4207                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
4208                     dsobj, obj);
4209                 return;
4210         }
4211         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
4212
4213         /* find out if the dataset is mounted */
4214         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
4215
4216         /* get the corrupted object's path */
4217         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
4218         zc.zc_obj = obj;
4219         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
4220             &zc) == 0) {
4221                 if (mounted) {
4222                         (void) snprintf(pathname, len, "%s%s", mntpnt,
4223                             zc.zc_value);
4224                 } else {
4225                         (void) snprintf(pathname, len, "%s:%s",
4226                             dsname, zc.zc_value);
4227                 }
4228         } else {
4229                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname, obj);
4230         }
4231         free(mntpnt);
4232 }
4233
4234 #ifdef illumos
4235 /*
4236  * Read the EFI label from the config, if a label does not exist then
4237  * pass back the error to the caller. If the caller has passed a non-NULL
4238  * diskaddr argument then we set it to the starting address of the EFI
4239  * partition. If the caller has passed a non-NULL boolean argument, then
4240  * we set it to indicate if the disk does have efi system partition.
4241  */
4242 static int
4243 read_efi_label(nvlist_t *config, diskaddr_t *sb, boolean_t *system)
4244 {
4245         char *path;
4246         int fd;
4247         char diskname[MAXPATHLEN];
4248         boolean_t boot = B_FALSE;
4249         int err = -1;
4250         int slice;
4251
4252         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
4253                 return (err);
4254
4255         (void) snprintf(diskname, sizeof (diskname), "%s%s", ZFS_RDISK_ROOT,
4256             strrchr(path, '/'));
4257         if ((fd = open(diskname, O_RDONLY|O_NDELAY)) >= 0) {
4258                 struct dk_gpt *vtoc;
4259
4260                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
4261                         for (slice = 0; slice < vtoc->efi_nparts; slice++) {
4262                                 if (vtoc->efi_parts[slice].p_tag == V_SYSTEM)
4263                                         boot = B_TRUE;
4264                                 if (vtoc->efi_parts[slice].p_tag == V_USR)
4265                                         break;
4266                         }
4267                         if (sb != NULL && vtoc->efi_parts[slice].p_tag == V_USR)
4268                                 *sb = vtoc->efi_parts[slice].p_start;
4269                         if (system != NULL)
4270                                 *system = boot;
4271                         efi_free(vtoc);
4272                 }
4273                 (void) close(fd);
4274         }
4275         return (err);
4276 }
4277
4278 /*
4279  * determine where a partition starts on a disk in the current
4280  * configuration
4281  */
4282 static diskaddr_t
4283 find_start_block(nvlist_t *config)
4284 {
4285         nvlist_t **child;
4286         uint_t c, children;
4287         diskaddr_t sb = MAXOFFSET_T;
4288         uint64_t wholedisk;
4289
4290         if (nvlist_lookup_nvlist_array(config,
4291             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
4292                 if (nvlist_lookup_uint64(config,
4293                     ZPOOL_CONFIG_WHOLE_DISK,
4294                     &wholedisk) != 0 || !wholedisk) {
4295                         return (MAXOFFSET_T);
4296                 }
4297                 if (read_efi_label(config, &sb, NULL) < 0)
4298                         sb = MAXOFFSET_T;
4299                 return (sb);
4300         }
4301
4302         for (c = 0; c < children; c++) {
4303                 sb = find_start_block(child[c]);
4304                 if (sb != MAXOFFSET_T) {
4305                         return (sb);
4306                 }
4307         }
4308         return (MAXOFFSET_T);
4309 }
4310 #endif /* illumos */
4311
4312 /*
4313  * Label an individual disk.  The name provided is the short name,
4314  * stripped of any leading /dev path.
4315  */
4316 int
4317 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, const char *name,
4318     zpool_boot_label_t boot_type, uint64_t boot_size, int *slice)
4319 {
4320 #ifdef illumos
4321         char path[MAXPATHLEN];
4322         struct dk_gpt *vtoc;
4323         int fd;
4324         size_t resv = EFI_MIN_RESV_SIZE;
4325         uint64_t slice_size;
4326         diskaddr_t start_block;
4327         char errbuf[1024];
4328
4329         /* prepare an error message just in case */
4330         (void) snprintf(errbuf, sizeof (errbuf),
4331             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
4332
4333         if (zhp) {
4334                 nvlist_t *nvroot;
4335
4336                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
4337                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
4338
4339                 if (zhp->zpool_start_block == 0)
4340                         start_block = find_start_block(nvroot);
4341                 else
4342                         start_block = zhp->zpool_start_block;
4343                 zhp->zpool_start_block = start_block;
4344         } else {
4345                 /* new pool */
4346                 start_block = NEW_START_BLOCK;
4347         }
4348
4349         (void) snprintf(path, sizeof (path), "%s/%s%s", ZFS_RDISK_ROOT, name,
4350             BACKUP_SLICE);
4351
4352         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
4353                 /*
4354                  * This shouldn't happen.  We've long since verified that this
4355                  * is a valid device.
4356                  */
4357                 zfs_error_aux(hdl,
4358                     dgettext(TEXT_DOMAIN, "unable to open device"));
4359                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
4360         }
4361
4362         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
4363                 /*
4364                  * The only way this can fail is if we run out of memory, or we
4365                  * were unable to read the disk's capacity
4366                  */
4367                 if (errno == ENOMEM)
4368                         (void) no_memory(hdl);
4369
4370                 (void) close(fd);
4371                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4372                     "unable to read disk capacity"), name);
4373
4374                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
4375         }
4376
4377         /*
4378          * Why we use V_USR: V_BACKUP confuses users, and is considered
4379          * disposable by some EFI utilities (since EFI doesn't have a backup
4380          * slice).  V_UNASSIGNED is supposed to be used only for zero size
4381          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
4382          * etc. were all pretty specific.  V_USR is as close to reality as we
4383          * can get, in the absence of V_OTHER.
4384          */
4385         /* first fix the partition start block */
4386         if (start_block == MAXOFFSET_T)
4387                 start_block = NEW_START_BLOCK;
4388
4389         /*
4390          * EFI System partition is using slice 0.
4391          * ZFS is on slice 1 and slice 8 is reserved.
4392          * We assume the GPT partition table without system
4393          * partition has zfs p_start == NEW_START_BLOCK.
4394          * If start_block != NEW_START_BLOCK, it means we have
4395          * system partition. Correct solution would be to query/cache vtoc
4396          * from existing vdev member.
4397          */
4398         if (boot_type == ZPOOL_CREATE_BOOT_LABEL) {
4399                 if (boot_size % vtoc->efi_lbasize != 0) {
4400                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4401                             "boot partition size must be a multiple of %d"),
4402                             vtoc->efi_lbasize);
4403                         (void) close(fd);
4404                         efi_free(vtoc);
4405                         return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4406                 }
4407                 /*
4408                  * System partition size checks.
4409                  * Note the 1MB is quite arbitrary value, since we
4410                  * are creating dedicated pool, it should be enough
4411                  * to hold fat + efi bootloader. May need to be
4412                  * adjusted if the bootloader size will grow.
4413                  */
4414                 if (boot_size < 1024 * 1024) {
4415                         char buf[64];
4416                         zfs_nicenum(boot_size, buf, sizeof (buf));
4417                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4418                             "Specified size %s for EFI System partition is too "
4419                             "small, the minimum size is 1MB."), buf);
4420                         (void) close(fd);
4421                         efi_free(vtoc);
4422                         return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4423                 }
4424                 /* 33MB is tested with mkfs -F pcfs */
4425                 if (hdl->libzfs_printerr &&
4426                     ((vtoc->efi_lbasize == 512 &&
4427                     boot_size < 33 * 1024 * 1024) ||
4428                     (vtoc->efi_lbasize == 4096 &&
4429                     boot_size < 256 * 1024 * 1024)))  {
4430                         char buf[64];
4431                         zfs_nicenum(boot_size, buf, sizeof (buf));
4432                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN,
4433                             "Warning: EFI System partition size %s is "
4434                             "not allowing to create FAT32 file\nsystem, which "
4435                             "may result in unbootable system.\n"), buf);
4436                 }
4437                 /* Adjust zfs partition start by size of system partition. */
4438                 start_block += boot_size / vtoc->efi_lbasize;
4439         }
4440
4441         if (start_block == NEW_START_BLOCK) {
4442                 /*
4443                  * Use default layout.
4444                  * ZFS is on slice 0 and slice 8 is reserved.
4445                  */
4446                 slice_size = vtoc->efi_last_u_lba + 1;
4447                 slice_size -= EFI_MIN_RESV_SIZE;
4448                 slice_size -= start_block;
4449                 if (slice != NULL)
4450                         *slice = 0;
4451
4452                 vtoc->efi_parts[0].p_start = start_block;
4453                 vtoc->efi_parts[0].p_size = slice_size;
4454
4455                 vtoc->efi_parts[0].p_tag = V_USR;
4456                 (void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
4457
4458                 vtoc->efi_parts[8].p_start = slice_size + start_block;
4459                 vtoc->efi_parts[8].p_size = resv;
4460                 vtoc->efi_parts[8].p_tag = V_RESERVED;
4461         } else {
4462                 slice_size = start_block - NEW_START_BLOCK;
4463                 vtoc->efi_parts[0].p_start = NEW_START_BLOCK;
4464                 vtoc->efi_parts[0].p_size = slice_size;
4465                 vtoc->efi_parts[0].p_tag = V_SYSTEM;
4466                 (void) strcpy(vtoc->efi_parts[0].p_name, "loader");
4467                 if (slice != NULL)
4468                         *slice = 1;
4469                 /* prepare slice 1 */
4470                 slice_size = vtoc->efi_last_u_lba + 1 - slice_size;
4471                 slice_size -= resv;
4472                 slice_size -= NEW_START_BLOCK;
4473                 vtoc->efi_parts[1].p_start = start_block;
4474                 vtoc->efi_parts[1].p_size = slice_size;
4475                 vtoc->efi_parts[1].p_tag = V_USR;
4476                 (void) strcpy(vtoc->efi_parts[1].p_name, "zfs");
4477
4478                 vtoc->efi_parts[8].p_start = slice_size + start_block;
4479                 vtoc->efi_parts[8].p_size = resv;
4480                 vtoc->efi_parts[8].p_tag = V_RESERVED;
4481         }
4482
4483         if (efi_write(fd, vtoc) != 0) {
4484                 /*
4485                  * Some block drivers (like pcata) may not support EFI
4486                  * GPT labels.  Print out a helpful error message dir-
4487                  * ecting the user to manually label the disk and give
4488                  * a specific slice.
4489                  */
4490                 (void) close(fd);
4491                 efi_free(vtoc);
4492
4493                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4494                     "try using fdisk(1M) and then provide a specific slice"));
4495                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4496         }
4497
4498         (void) close(fd);
4499         efi_free(vtoc);
4500 #endif /* illumos */
4501         return (0);
4502 }
4503
4504 static boolean_t
4505 supported_dump_vdev_type(libzfs_handle_t *hdl, nvlist_t *config, char *errbuf)
4506 {
4507         char *type;
4508         nvlist_t **child;
4509         uint_t children, c;
4510
4511         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_TYPE, &type) == 0);
4512         if (strcmp(type, VDEV_TYPE_FILE) == 0 ||
4513             strcmp(type, VDEV_TYPE_HOLE) == 0 ||
4514             strcmp(type, VDEV_TYPE_MISSING) == 0) {
4515                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4516                     "vdev type '%s' is not supported"), type);
4517                 (void) zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf);
4518                 return (B_FALSE);
4519         }
4520         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
4521             &child, &children) == 0) {
4522                 for (c = 0; c < children; c++) {
4523                         if (!supported_dump_vdev_type(hdl, child[c], errbuf))
4524                                 return (B_FALSE);
4525                 }
4526         }
4527         return (B_TRUE);
4528 }
4529
4530 /*
4531  * Check if this zvol is allowable for use as a dump device; zero if
4532  * it is, > 0 if it isn't, < 0 if it isn't a zvol.
4533  *
4534  * Allowable storage configurations include mirrors, all raidz variants, and
4535  * pools with log, cache, and spare devices.  Pools which are backed by files or
4536  * have missing/hole vdevs are not suitable.
4537  */
4538 int
4539 zvol_check_dump_config(char *arg)
4540 {
4541         zpool_handle_t *zhp = NULL;
4542         nvlist_t *config, *nvroot;
4543         char *p, *volname;
4544         nvlist_t **top;
4545         uint_t toplevels;
4546         libzfs_handle_t *hdl;
4547         char errbuf[1024];
4548         char poolname[ZFS_MAX_DATASET_NAME_LEN];
4549         int pathlen = strlen(ZVOL_FULL_DEV_DIR);
4550         int ret = 1;
4551
4552         if (strncmp(arg, ZVOL_FULL_DEV_DIR, pathlen)) {
4553                 return (-1);
4554         }
4555
4556         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
4557             "dump is not supported on device '%s'"), arg);
4558
4559         if ((hdl = libzfs_init()) == NULL)
4560                 return (1);
4561         libzfs_print_on_error(hdl, B_TRUE);
4562
4563         volname = arg + pathlen;
4564
4565         /* check the configuration of the pool */
4566         if ((p = strchr(volname, '/')) == NULL) {
4567                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4568                     "malformed dataset name"));
4569                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4570                 return (1);
4571         } else if (p - volname >= ZFS_MAX_DATASET_NAME_LEN) {
4572                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4573                     "dataset name is too long"));
4574                 (void) zfs_error(hdl, EZFS_NAMETOOLONG, errbuf);
4575                 return (1);
4576         } else {
4577                 (void) strncpy(poolname, volname, p - volname);
4578                 poolname[p - volname] = '\0';
4579         }
4580
4581         if ((zhp = zpool_open(hdl, poolname)) == NULL) {
4582                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4583                     "could not open pool '%s'"), poolname);
4584                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
4585                 goto out;
4586         }
4587         config = zpool_get_config(zhp, NULL);
4588         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4589             &nvroot) != 0) {
4590                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4591                     "could not obtain vdev configuration for  '%s'"), poolname);
4592                 (void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
4593                 goto out;
4594         }
4595
4596         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4597             &top, &toplevels) == 0);
4598
4599         if (!supported_dump_vdev_type(hdl, top[0], errbuf)) {
4600                 goto out;
4601         }
4602         ret = 0;
4603
4604 out:
4605         if (zhp)
4606                 zpool_close(zhp);
4607         libzfs_fini(hdl);
4608         return (ret);
4609 }
4610
4611 int
4612 zpool_nextboot(libzfs_handle_t *hdl, uint64_t pool_guid, uint64_t dev_guid,
4613     const char *command)
4614 {
4615         zfs_cmd_t zc = { 0 };
4616         nvlist_t *args;
4617         char *packed;
4618         size_t size;
4619         int error;
4620
4621         args = fnvlist_alloc();
4622         fnvlist_add_uint64(args, ZPOOL_CONFIG_POOL_GUID, pool_guid);
4623         fnvlist_add_uint64(args, ZPOOL_CONFIG_GUID, dev_guid);
4624         fnvlist_add_string(args, "command", command);
4625         error = zcmd_write_src_nvlist(hdl, &zc, args);
4626         if (error == 0)
4627                 error = ioctl(hdl->libzfs_fd, ZFS_IOC_NEXTBOOT, &zc);
4628         zcmd_free_nvlists(&zc);
4629         nvlist_free(args);
4630         return (error);
4631 }