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