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