lib/libzfs/libzfs_pool.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
  24  * Use is subject to license terms.
  25  */
  26
  27 #include <alloca.h>
  28 #include <assert.h>
  29 #include <ctype.h>
  30 #include <errno.h>
  31 #include <devid.h>
  32 #include <dirent.h>
  33 #include <fcntl.h>
  34 #include <libintl.h>
  35 #include <stdio.h>
  36 #include <stdlib.h>
  37 #include <strings.h>
  38 #include <unistd.h>
  39 #include <zone.h>
  40 #include <sys/efi_partition.h>
  41 #include <sys/vtoc.h>
  42 #include <sys/zfs_ioctl.h>
  43 #include <sys/zio.h>
  44 #include <strings.h>
  45
  46 #include "zfs_namecheck.h"
  47 #include "zfs_prop.h"
  48 #include "libzfs_impl.h"
  49
  50 static int read_efi_label(nvlist_t *config, diskaddr_t *sb);
  51
  52 #if defined(__i386) || defined(__amd64)
  53 #define BOOTCMD "installgrub(1M)"
  54 #else
  55 #define BOOTCMD "installboot(1M)"
  56 #endif
  57
  58 /*
  59  * ====================================================================
  60  *   zpool property functions
  61  * ====================================================================
  62  */
  63
  64 static int
  65 zpool_get_all_props(zpool_handle_t *zhp)
  66 {
  67         zfs_cmd_t zc = { 0 };
  68         libzfs_handle_t *hdl = zhp->zpool_hdl;
  69
  70         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
  71
  72         if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
  73                 return (-1);
  74
  75         while (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
  76                 if (errno == ENOMEM) {
  77                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
  78                                 zcmd_free_nvlists(&zc);
  79                                 return (-1);
  80                         }
  81                 } else {
  82                         zcmd_free_nvlists(&zc);
  83                         return (-1);
  84                 }
  85         }
  86
  87         if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
  88                 zcmd_free_nvlists(&zc);
  89                 return (-1);
  90         }
  91
  92         zcmd_free_nvlists(&zc);
  93
  94         return (0);
  95 }
  96
  97 static int
  98 zpool_props_refresh(zpool_handle_t *zhp)
  99 {
 100         nvlist_t *old_props;
 101
 102         old_props = zhp->zpool_props;
 103
 104         if (zpool_get_all_props(zhp) != 0)
 105                 return (-1);
 106
 107         nvlist_free(old_props);
 108         return (0);
 109 }
 110
 111 static char *
 112 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
 113     zprop_source_t *src)
 114 {
 115         nvlist_t *nv, *nvl;
 116         uint64_t ival;
 117         char *value;
 118         zprop_source_t source;
 119
 120         nvl = zhp->zpool_props;
 121         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 122                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &ival) == 0);
 123                 source = ival;
 124                 verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
 125         } else {
 126                 source = ZPROP_SRC_DEFAULT;
 127                 if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
 128                         value = "-";
 129         }
 130
 131         if (src)
 132                 *src = source;
 133
 134         return (value);
 135 }
 136
 137 uint64_t
 138 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
 139 {
 140         nvlist_t *nv, *nvl;
 141         uint64_t value;
 142         zprop_source_t source;
 143
 144         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
 145                 /*
 146                  * zpool_get_all_props() has most likely failed because
 147                  * the pool is faulted, but if all we need is the top level
 148                  * vdev's guid then get it from the zhp config nvlist.
 149                  */
 150                 if ((prop == ZPOOL_PROP_GUID) &&
 151                     (nvlist_lookup_nvlist(zhp->zpool_config,
 152                     ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
 153                     (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
 154                     == 0)) {
 155                         return (value);
 156                 }
 157                 return (zpool_prop_default_numeric(prop));
 158         }
 159
 160         nvl = zhp->zpool_props;
 161         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 162                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &value) == 0);
 163                 source = value;
 164                 verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
 165         } else {
 166                 source = ZPROP_SRC_DEFAULT;
 167                 value = zpool_prop_default_numeric(prop);
 168         }
 169
 170         if (src)
 171                 *src = source;
 172
 173         return (value);
 174 }
 175
 176 /*
 177  * Map VDEV STATE to printed strings.
 178  */
 179 char *
 180 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
 181 {
 182         switch (state) {
 183         case VDEV_STATE_CLOSED:
 184         case VDEV_STATE_OFFLINE:
 185                 return (gettext("OFFLINE"));
 186         case VDEV_STATE_REMOVED:
 187                 return (gettext("REMOVED"));
 188         case VDEV_STATE_CANT_OPEN:
 189                 if (aux == VDEV_AUX_CORRUPT_DATA || aux == VDEV_AUX_BAD_LOG)
 190                         return (gettext("FAULTED"));
 191                 else
 192                         return (gettext("UNAVAIL"));
 193         case VDEV_STATE_FAULTED:
 194                 return (gettext("FAULTED"));
 195         case VDEV_STATE_DEGRADED:
 196                 return (gettext("DEGRADED"));
 197         case VDEV_STATE_HEALTHY:
 198                 return (gettext("ONLINE"));
 199         }
 200
 201         return (gettext("UNKNOWN"));
 202 }
 203
 204 /*
 205  * Get a zpool property value for 'prop' and return the value in
 206  * a pre-allocated buffer.
 207  */
 208 int
 209 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf, size_t len,
 210     zprop_source_t *srctype)
 211 {
 212         uint64_t intval;
 213         const char *strval;
 214         zprop_source_t src = ZPROP_SRC_NONE;
 215         nvlist_t *nvroot;
 216         vdev_stat_t *vs;
 217         uint_t vsc;
 218
 219         if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
 220                 switch (prop) {
 221                 case ZPOOL_PROP_NAME:
 222                         (void) strlcpy(buf, zpool_get_name(zhp), len);
 223                         break;
 224
 225                 case ZPOOL_PROP_HEALTH:
 226                         (void) strlcpy(buf, "FAULTED", len);
 227                         break;
 228
 229                 case ZPOOL_PROP_GUID:
 230                         intval = zpool_get_prop_int(zhp, prop, &src);
 231                         (void) snprintf(buf, len, "%llu", intval);
 232                         break;
 233
 234                 case ZPOOL_PROP_ALTROOT:
 235                 case ZPOOL_PROP_CACHEFILE:
 236                         if (zhp->zpool_props != NULL ||
 237                             zpool_get_all_props(zhp) == 0) {
 238                                 (void) strlcpy(buf,
 239                                     zpool_get_prop_string(zhp, prop, &src),
 240                                     len);
 241                                 if (srctype != NULL)
 242                                         *srctype = src;
 243                                 return (0);
 244                         }
 245                         /* FALLTHROUGH */
 246                 default:
 247                         (void) strlcpy(buf, "-", len);
 248                         break;
 249                 }
 250
 251                 if (srctype != NULL)
 252                         *srctype = src;
 253                 return (0);
 254         }
 255
 256         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
 257             prop != ZPOOL_PROP_NAME)
 258                 return (-1);
 259
 260         switch (zpool_prop_get_type(prop)) {
 261         case PROP_TYPE_STRING:
 262                 (void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
 263                     len);
 264                 break;
 265
 266         case PROP_TYPE_NUMBER:
 267                 intval = zpool_get_prop_int(zhp, prop, &src);
 268
 269                 switch (prop) {
 270                 case ZPOOL_PROP_SIZE:
 271                 case ZPOOL_PROP_USED:
 272                 case ZPOOL_PROP_AVAILABLE:
 273                         (void) zfs_nicenum(intval, buf, len);
 274                         break;
 275
 276                 case ZPOOL_PROP_CAPACITY:
 277                         (void) snprintf(buf, len, "%llu%%",
 278                             (u_longlong_t)intval);
 279                         break;
 280
 281                 case ZPOOL_PROP_HEALTH:
 282                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 283                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 284                         verify(nvlist_lookup_uint64_array(nvroot,
 285                             ZPOOL_CONFIG_STATS, (uint64_t **)&vs, &vsc) == 0);
 286
 287                         (void) strlcpy(buf, zpool_state_to_name(intval,
 288                             vs->vs_aux), len);
 289                         break;
 290                 default:
 291                         (void) snprintf(buf, len, "%llu", intval);
 292                 }
 293                 break;
 294
 295         case PROP_TYPE_INDEX:
 296                 intval = zpool_get_prop_int(zhp, prop, &src);
 297                 if (zpool_prop_index_to_string(prop, intval, &strval)
 298                     != 0)
 299                         return (-1);
 300                 (void) strlcpy(buf, strval, len);
 301                 break;
 302
 303         default:
 304                 abort();
 305         }
 306
 307         if (srctype)
 308                 *srctype = src;
 309
 310         return (0);
 311 }
 312
 313 /*
 314  * Check if the bootfs name has the same pool name as it is set to.
 315  * Assuming bootfs is a valid dataset name.
 316  */
 317 static boolean_t
 318 bootfs_name_valid(const char *pool, char *bootfs)
 319 {
 320         int len = strlen(pool);
 321
 322         if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
 323                 return (B_FALSE);
 324
 325         if (strncmp(pool, bootfs, len) == 0 &&
 326             (bootfs[len] == '/' || bootfs[len] == '\0'))
 327                 return (B_TRUE);
 328
 329         return (B_FALSE);
 330 }
 331
 332 /*
 333  * Inspect the configuration to determine if any of the devices contain
 334  * an EFI label.
 335  */
 336 static boolean_t
 337 pool_uses_efi(nvlist_t *config)
 338 {
 339         nvlist_t **child;
 340         uint_t c, children;
 341
 342         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
 343             &child, &children) != 0)
 344                 return (read_efi_label(config, NULL) >= 0);
 345
 346         for (c = 0; c < children; c++) {
 347                 if (pool_uses_efi(child[c]))
 348                         return (B_TRUE);
 349         }
 350         return (B_FALSE);
 351 }
 352
 353 static boolean_t
 354 pool_is_bootable(zpool_handle_t *zhp)
 355 {
 356         char bootfs[ZPOOL_MAXNAMELEN];
 357
 358         return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
 359             sizeof (bootfs), NULL) == 0 && strncmp(bootfs, "-",
 360             sizeof (bootfs)) != 0);
 361 }
 362
 363
 364 /*
 365  * Given an nvlist of zpool properties to be set, validate that they are
 366  * correct, and parse any numeric properties (index, boolean, etc) if they are
 367  * specified as strings.
 368  */
 369 static nvlist_t *
 370 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 371     nvlist_t *props, uint64_t version, boolean_t create_or_import, char *errbuf)
 372 {
 373         nvpair_t *elem;
 374         nvlist_t *retprops;
 375         zpool_prop_t prop;
 376         char *strval;
 377         uint64_t intval;
 378         char *slash;
 379         struct stat64 statbuf;
 380         zpool_handle_t *zhp;
 381         nvlist_t *nvroot;
 382
 383         if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
 384                 (void) no_memory(hdl);
 385                 return (NULL);
 386         }
 387
 388         elem = NULL;
 389         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 390                 const char *propname = nvpair_name(elem);
 391
 392                 /*
 393                  * Make sure this property is valid and applies to this type.
 394                  */
 395                 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL) {
 396                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 397                             "invalid property '%s'"), propname);
 398                         (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 399                         goto error;
 400                 }
 401
 402                 if (zpool_prop_readonly(prop)) {
 403                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 404                             "is readonly"), propname);
 405                         (void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
 406                         goto error;
 407                 }
 408
 409                 if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
 410                     &strval, &intval, errbuf) != 0)
 411                         goto error;
 412
 413                 /*
 414                  * Perform additional checking for specific properties.
 415                  */
 416                 switch (prop) {
 417                 case ZPOOL_PROP_VERSION:
 418                         if (intval < version || intval > SPA_VERSION) {
 419                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 420                                     "property '%s' number %d is invalid."),
 421                                     propname, intval);
 422                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 423                                 goto error;
 424                         }
 425                         break;
 426
 427                 case ZPOOL_PROP_BOOTFS:
 428                         if (create_or_import) {
 429                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 430                                     "property '%s' cannot be set at creation "
 431                                     "or import time"), propname);
 432                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 433                                 goto error;
 434                         }
 435
 436                         if (version < SPA_VERSION_BOOTFS) {
 437                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 438                                     "pool must be upgraded to support "
 439                                     "'%s' property"), propname);
 440                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 441                                 goto error;
 442                         }
 443
 444                         /*
 445                          * bootfs property value has to be a dataset name and
 446                          * the dataset has to be in the same pool as it sets to.
 447                          */
 448                         if (strval[0] != '\0' && !bootfs_name_valid(poolname,
 449                             strval)) {
 450                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 451                                     "is an invalid name"), strval);
 452                                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
 453                                 goto error;
 454                         }
 455
 456                         if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
 457                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 458                                     "could not open pool '%s'"), poolname);
 459                                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
 460                                 goto error;
 461                         }
 462                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 463                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 464
 465                         /*
 466                          * bootfs property cannot be set on a disk which has
 467                          * been EFI labeled.
 468                          */
 469                         if (pool_uses_efi(nvroot)) {
 470                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 471                                     "property '%s' not supported on "
 472                                     "EFI labeled devices"), propname);
 473                                 (void) zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf);
 474                                 zpool_close(zhp);
 475                                 goto error;
 476                         }
 477                         zpool_close(zhp);
 478                         break;
 479
 480                 case ZPOOL_PROP_ALTROOT:
 481                         if (!create_or_import) {
 482                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 483                                     "property '%s' can only be set during pool "
 484                                     "creation or import"), propname);
 485                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 486                                 goto error;
 487                         }
 488
 489                         if (strval[0] != '/') {
 490                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 491                                     "bad alternate root '%s'"), strval);
 492                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 493                                 goto error;
 494                         }
 495                         break;
 496
 497                 case ZPOOL_PROP_CACHEFILE:
 498                         if (strval[0] == '\0')
 499                                 break;
 500
 501                         if (strcmp(strval, "none") == 0)
 502                                 break;
 503
 504                         if (strval[0] != '/') {
 505                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 506                                     "property '%s' must be empty, an "
 507                                     "absolute path, or 'none'"), propname);
 508                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 509                                 goto error;
 510                         }
 511
 512                         slash = strrchr(strval, '/');
 513
 514                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 515                             strcmp(slash, "/..") == 0) {
 516                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 517                                     "'%s' is not a valid file"), strval);
 518                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 519                                 goto error;
 520                         }
 521
 522                         *slash = '\0';
 523
 524                         if (strval[0] != '\0' &&
 525                             (stat64(strval, &statbuf) != 0 ||
 526                             !S_ISDIR(statbuf.st_mode))) {
 527                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 528                                     "'%s' is not a valid directory"),
 529                                     strval);
 530                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 531                                 goto error;
 532                         }
 533
 534                         *slash = '/';
 535                         break;
 536                 }
 537         }
 538
 539         return (retprops);
 540 error:
 541         nvlist_free(retprops);
 542         return (NULL);
 543 }
 544
 545 /*
 546  * Set zpool property : propname=propval.
 547  */
 548 int
 549 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
 550 {
 551         zfs_cmd_t zc = { 0 };
 552         int ret = -1;
 553         char errbuf[1024];
 554         nvlist_t *nvl = NULL;
 555         nvlist_t *realprops;
 556         uint64_t version;
 557
 558         (void) snprintf(errbuf, sizeof (errbuf),
 559             dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
 560             zhp->zpool_name);
 561
 562         if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
 563                 return (no_memory(zhp->zpool_hdl));
 564
 565         if (nvlist_add_string(nvl, propname, propval) != 0) {
 566                 nvlist_free(nvl);
 567                 return (no_memory(zhp->zpool_hdl));
 568         }
 569
 570         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
 571         if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
 572             zhp->zpool_name, nvl, version, B_FALSE, errbuf)) == NULL) {
 573                 nvlist_free(nvl);
 574                 return (-1);
 575         }
 576
 577         nvlist_free(nvl);
 578         nvl = realprops;
 579
 580         /*
 581          * Execute the corresponding ioctl() to set this property.
 582          */
 583         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 584
 585         if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
 586                 nvlist_free(nvl);
 587                 return (-1);
 588         }
 589
 590         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
 591
 592         zcmd_free_nvlists(&zc);
 593         nvlist_free(nvl);
 594
 595         if (ret)
 596                 (void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
 597         else
 598                 (void) zpool_props_refresh(zhp);
 599
 600         return (ret);
 601 }
 602
 603 int
 604 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
 605 {
 606         libzfs_handle_t *hdl = zhp->zpool_hdl;
 607         zprop_list_t *entry;
 608         char buf[ZFS_MAXPROPLEN];
 609
 610         if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
 611                 return (-1);
 612
 613         for (entry = *plp; entry != NULL; entry = entry->pl_next) {
 614
 615                 if (entry->pl_fixed)
 616                         continue;
 617
 618                 if (entry->pl_prop != ZPROP_INVAL &&
 619                     zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
 620                     NULL) == 0) {
 621                         if (strlen(buf) > entry->pl_width)
 622                                 entry->pl_width = strlen(buf);
 623                 }
 624         }
 625
 626         return (0);
 627 }
 628
 629
 630 /*
 631  * Validate the given pool name, optionally putting an extended error message in
 632  * 'buf'.
 633  */
 634 boolean_t
 635 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
 636 {
 637         namecheck_err_t why;
 638         char what;
 639         int ret;
 640
 641         ret = pool_namecheck(pool, &why, &what);
 642
 643         /*
 644          * The rules for reserved pool names were extended at a later point.
 645          * But we need to support users with existing pools that may now be
 646          * invalid.  So we only check for this expanded set of names during a
 647          * create (or import), and only in userland.
 648          */
 649         if (ret == 0 && !isopen &&
 650             (strncmp(pool, "mirror", 6) == 0 ||
 651             strncmp(pool, "raidz", 5) == 0 ||
 652             strncmp(pool, "spare", 5) == 0 ||
 653             strcmp(pool, "log") == 0)) {
 654                 if (hdl != NULL)
 655                         zfs_error_aux(hdl,
 656                             dgettext(TEXT_DOMAIN, "name is reserved"));
 657                 return (B_FALSE);
 658         }
 659
 660
 661         if (ret != 0) {
 662                 if (hdl != NULL) {
 663                         switch (why) {
 664                         case NAME_ERR_TOOLONG:
 665                                 zfs_error_aux(hdl,
 666                                     dgettext(TEXT_DOMAIN, "name is too long"));
 667                                 break;
 668
 669                         case NAME_ERR_INVALCHAR:
 670                                 zfs_error_aux(hdl,
 671                                     dgettext(TEXT_DOMAIN, "invalid character "
 672                                     "'%c' in pool name"), what);
 673                                 break;
 674
 675                         case NAME_ERR_NOLETTER:
 676                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 677                                     "name must begin with a letter"));
 678                                 break;
 679
 680                         case NAME_ERR_RESERVED:
 681                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 682                                     "name is reserved"));
 683                                 break;
 684
 685                         case NAME_ERR_DISKLIKE:
 686                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 687                                     "pool name is reserved"));
 688                                 break;
 689
 690                         case NAME_ERR_LEADING_SLASH:
 691                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 692                                     "leading slash in name"));
 693                                 break;
 694
 695                         case NAME_ERR_EMPTY_COMPONENT:
 696                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 697                                     "empty component in name"));
 698                                 break;
 699
 700                         case NAME_ERR_TRAILING_SLASH:
 701                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 702                                     "trailing slash in name"));
 703                                 break;
 704
 705                         case NAME_ERR_MULTIPLE_AT:
 706                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 707                                     "multiple '@' delimiters in name"));
 708                                 break;
 709
 710                         }
 711                 }
 712                 return (B_FALSE);
 713         }
 714
 715         return (B_TRUE);
 716 }
 717
 718 /*
 719  * Open a handle to the given pool, even if the pool is currently in the FAULTED
 720  * state.
 721  */
 722 zpool_handle_t *
 723 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
 724 {
 725         zpool_handle_t *zhp;
 726         boolean_t missing;
 727
 728         /*
 729          * Make sure the pool name is valid.
 730          */
 731         if (!zpool_name_valid(hdl, B_TRUE, pool)) {
 732                 (void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
 733                     dgettext(TEXT_DOMAIN, "cannot open '%s'"),
 734                     pool);
 735                 return (NULL);
 736         }
 737
 738         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
 739                 return (NULL);
 740
 741         zhp->zpool_hdl = hdl;
 742         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
 743
 744         if (zpool_refresh_stats(zhp, &missing) != 0) {
 745                 zpool_close(zhp);
 746                 return (NULL);
 747         }
 748
 749         if (missing) {
 750                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
 751                 (void) zfs_error_fmt(hdl, EZFS_NOENT,
 752                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
 753                 zpool_close(zhp);
 754                 return (NULL);
 755         }
 756
 757         return (zhp);
 758 }
 759
 760 /*
 761  * Like the above, but silent on error.  Used when iterating over pools (because
 762  * the configuration cache may be out of date).
 763  */
 764 int
 765 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
 766 {
 767         zpool_handle_t *zhp;
 768         boolean_t missing;
 769
 770         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
 771                 return (-1);
 772
 773         zhp->zpool_hdl = hdl;
 774         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
 775
 776         if (zpool_refresh_stats(zhp, &missing) != 0) {
 777                 zpool_close(zhp);
 778                 return (-1);
 779         }
 780
 781         if (missing) {
 782                 zpool_close(zhp);
 783                 *ret = NULL;
 784                 return (0);
 785         }
 786
 787         *ret = zhp;
 788         return (0);
 789 }
 790
 791 /*
 792  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
 793  * state.
 794  */
 795 zpool_handle_t *
 796 zpool_open(libzfs_handle_t *hdl, const char *pool)
 797 {
 798         zpool_handle_t *zhp;
 799
 800         if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
 801                 return (NULL);
 802
 803         if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
 804                 (void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
 805                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
 806                 zpool_close(zhp);
 807                 return (NULL);
 808         }
 809
 810         return (zhp);
 811 }
 812
 813 /*
 814  * Close the handle.  Simply frees the memory associated with the handle.
 815  */
 816 void
 817 zpool_close(zpool_handle_t *zhp)
 818 {
 819         if (zhp->zpool_config)
 820                 nvlist_free(zhp->zpool_config);
 821         if (zhp->zpool_old_config)
 822                 nvlist_free(zhp->zpool_old_config);
 823         if (zhp->zpool_props)
 824                 nvlist_free(zhp->zpool_props);
 825         free(zhp);
 826 }
 827
 828 /*
 829  * Return the name of the pool.
 830  */
 831 const char *
 832 zpool_get_name(zpool_handle_t *zhp)
 833 {
 834         return (zhp->zpool_name);
 835 }
 836
 837
 838 /*
 839  * Return the state of the pool (ACTIVE or UNAVAILABLE)
 840  */
 841 int
 842 zpool_get_state(zpool_handle_t *zhp)
 843 {
 844         return (zhp->zpool_state);
 845 }
 846
 847 /*
 848  * Create the named pool, using the provided vdev list.  It is assumed
 849  * that the consumer has already validated the contents of the nvlist, so we
 850  * don't have to worry about error semantics.
 851  */
 852 int
 853 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
 854     nvlist_t *props, nvlist_t *fsprops)
 855 {
 856         zfs_cmd_t zc = { 0 };
 857         nvlist_t *zc_fsprops = NULL;
 858         nvlist_t *zc_props = NULL;
 859         char msg[1024];
 860         char *altroot;
 861         int ret = -1;
 862
 863         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
 864             "cannot create '%s'"), pool);
 865
 866         if (!zpool_name_valid(hdl, B_FALSE, pool))
 867                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
 868
 869         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
 870                 return (-1);
 871
 872         if (props) {
 873                 if ((zc_props = zpool_valid_proplist(hdl, pool, props,
 874                     SPA_VERSION_1, B_TRUE, msg)) == NULL) {
 875                         goto create_failed;
 876                 }
 877         }
 878
 879         if (fsprops) {
 880                 uint64_t zoned;
 881                 char *zonestr;
 882
 883                 zoned = ((nvlist_lookup_string(fsprops,
 884                     zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
 885                     strcmp(zonestr, "on") == 0);
 886
 887                 if ((zc_fsprops = zfs_valid_proplist(hdl,
 888                     ZFS_TYPE_FILESYSTEM, fsprops, zoned, NULL, msg)) == NULL) {
 889                         goto create_failed;
 890                 }
 891                 if (!zc_props &&
 892                     (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
 893                         goto create_failed;
 894                 }
 895                 if (nvlist_add_nvlist(zc_props,
 896                     ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
 897                         goto create_failed;
 898                 }
 899         }
 900
 901         if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
 902                 goto create_failed;
 903
 904         (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
 905
 906         if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
 907
 908                 zcmd_free_nvlists(&zc);
 909                 nvlist_free(zc_props);
 910                 nvlist_free(zc_fsprops);
 911
 912                 switch (errno) {
 913                 case EBUSY:
 914                         /*
 915                          * This can happen if the user has specified the same
 916                          * device multiple times.  We can't reliably detect this
 917                          * until we try to add it and see we already have a
 918                          * label.
 919                          */
 920                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 921                             "one or more vdevs refer to the same device"));
 922                         return (zfs_error(hdl, EZFS_BADDEV, msg));
 923
 924                 case EOVERFLOW:
 925                         /*
 926                          * This occurs when one of the devices is below
 927                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
 928                          * device was the problem device since there's no
 929                          * reliable way to determine device size from userland.
 930                          */
 931                         {
 932                                 char buf[64];
 933
 934                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
 935
 936                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 937                                     "one or more devices is less than the "
 938                                     "minimum size (%s)"), buf);
 939                         }
 940                         return (zfs_error(hdl, EZFS_BADDEV, msg));
 941
 942                 case ENOSPC:
 943                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 944                             "one or more devices is out of space"));
 945                         return (zfs_error(hdl, EZFS_BADDEV, msg));
 946
 947                 case ENOTBLK:
 948                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 949                             "cache device must be a disk or disk slice"));
 950                         return (zfs_error(hdl, EZFS_BADDEV, msg));
 951
 952                 default:
 953                         return (zpool_standard_error(hdl, errno, msg));
 954                 }
 955         }
 956
 957         /*
 958          * If this is an alternate root pool, then we automatically set the
 959          * mountpoint of the root dataset to be '/'.
 960          */
 961         if (nvlist_lookup_string(props, zpool_prop_to_name(ZPOOL_PROP_ALTROOT),
 962             &altroot) == 0) {
 963                 zfs_handle_t *zhp;
 964
 965                 verify((zhp = zfs_open(hdl, pool, ZFS_TYPE_DATASET)) != NULL);
 966                 verify(zfs_prop_set(zhp, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT),
 967                     "/") == 0);
 968
 969                 zfs_close(zhp);
 970         }
 971
 972 create_failed:
 973         zcmd_free_nvlists(&zc);
 974         nvlist_free(zc_props);
 975         nvlist_free(zc_fsprops);
 976         return (ret);
 977 }
 978
 979 /*
 980  * Destroy the given pool.  It is up to the caller to ensure that there are no
 981  * datasets left in the pool.
 982  */
 983 int
 984 zpool_destroy(zpool_handle_t *zhp)
 985 {
 986         zfs_cmd_t zc = { 0 };
 987         zfs_handle_t *zfp = NULL;
 988         libzfs_handle_t *hdl = zhp->zpool_hdl;
 989         char msg[1024];
 990
 991         if (zhp->zpool_state == POOL_STATE_ACTIVE &&
 992             (zfp = zfs_open(zhp->zpool_hdl, zhp->zpool_name,
 993             ZFS_TYPE_FILESYSTEM)) == NULL)
 994                 return (-1);
 995
 996         if (zpool_remove_zvol_links(zhp) != 0)
 997                 return (-1);
 998
 999         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1000
1001         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1002                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1003                     "cannot destroy '%s'"), zhp->zpool_name);
1004
1005                 if (errno == EROFS) {
1006                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1007                             "one or more devices is read only"));
1008                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1009                 } else {
1010                         (void) zpool_standard_error(hdl, errno, msg);
1011                 }
1012
1013                 if (zfp)
1014                         zfs_close(zfp);
1015                 return (-1);
1016         }
1017
1018         if (zfp) {
1019                 remove_mountpoint(zfp);
1020                 zfs_close(zfp);
1021         }
1022
1023         return (0);
1024 }
1025
1026 /*
1027  * Add the given vdevs to the pool.  The caller must have already performed the
1028  * necessary verification to ensure that the vdev specification is well-formed.
1029  */
1030 int
1031 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1032 {
1033         zfs_cmd_t zc = { 0 };
1034         int ret;
1035         libzfs_handle_t *hdl = zhp->zpool_hdl;
1036         char msg[1024];
1037         nvlist_t **spares, **l2cache;
1038         uint_t nspares, nl2cache;
1039
1040         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1041             "cannot add to '%s'"), zhp->zpool_name);
1042
1043         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1044             SPA_VERSION_SPARES &&
1045             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1046             &spares, &nspares) == 0) {
1047                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1048                     "upgraded to add hot spares"));
1049                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1050         }
1051
1052         if (pool_is_bootable(zhp) && nvlist_lookup_nvlist_array(nvroot,
1053             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) {
1054                 uint64_t s;
1055
1056                 for (s = 0; s < nspares; s++) {
1057                         char *path;
1058
1059                         if (nvlist_lookup_string(spares[s], ZPOOL_CONFIG_PATH,
1060                             &path) == 0 && pool_uses_efi(spares[s])) {
1061                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1062                                     "device '%s' contains an EFI label and "
1063                                     "cannot be used on root pools."),
1064                                     zpool_vdev_name(hdl, NULL, spares[s]));
1065                                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
1066                         }
1067                 }
1068         }
1069
1070         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1071             SPA_VERSION_L2CACHE &&
1072             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1073             &l2cache, &nl2cache) == 0) {
1074                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1075                     "upgraded to add cache devices"));
1076                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1077         }
1078
1079         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1080                 return (-1);
1081         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1082
1083         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1084                 switch (errno) {
1085                 case EBUSY:
1086                         /*
1087                          * This can happen if the user has specified the same
1088                          * device multiple times.  We can't reliably detect this
1089                          * until we try to add it and see we already have a
1090                          * label.
1091                          */
1092                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1093                             "one or more vdevs refer to the same device"));
1094                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1095                         break;
1096
1097                 case EOVERFLOW:
1098                         /*
1099                          * This occurrs when one of the devices is below
1100                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1101                          * device was the problem device since there's no
1102                          * reliable way to determine device size from userland.
1103                          */
1104                         {
1105                                 char buf[64];
1106
1107                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1108
1109                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1110                                     "device is less than the minimum "
1111                                     "size (%s)"), buf);
1112                         }
1113                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1114                         break;
1115
1116                 case ENOTSUP:
1117                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1118                             "pool must be upgraded to add these vdevs"));
1119                         (void) zfs_error(hdl, EZFS_BADVERSION, msg);
1120                         break;
1121
1122                 case EDOM:
1123                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1124                             "root pool can not have multiple vdevs"
1125                             " or separate logs"));
1126                         (void) zfs_error(hdl, EZFS_POOL_NOTSUP, msg);
1127                         break;
1128
1129                 case ENOTBLK:
1130                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1131                             "cache device must be a disk or disk slice"));
1132                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1133                         break;
1134
1135                 default:
1136                         (void) zpool_standard_error(hdl, errno, msg);
1137                 }
1138
1139                 ret = -1;
1140         } else {
1141                 ret = 0;
1142         }
1143
1144         zcmd_free_nvlists(&zc);
1145
1146         return (ret);
1147 }
1148
1149 /*
1150  * Exports the pool from the system.  The caller must ensure that there are no
1151  * mounted datasets in the pool.
1152  */
1153 int
1154 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce)
1155 {
1156         zfs_cmd_t zc = { 0 };
1157         char msg[1024];
1158
1159         if (zpool_remove_zvol_links(zhp) != 0)
1160                 return (-1);
1161
1162         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1163             "cannot export '%s'"), zhp->zpool_name);
1164
1165         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1166         zc.zc_cookie = force;
1167         zc.zc_guid = hardforce;
1168
1169         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1170                 switch (errno) {
1171                 case EXDEV:
1172                         zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1173                             "use '-f' to override the following errors:\n"
1174                             "'%s' has an active shared spare which could be"
1175                             " used by other pools once '%s' is exported."),
1176                             zhp->zpool_name, zhp->zpool_name);
1177                         return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1178                             msg));
1179                 default:
1180                         return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1181                             msg));
1182                 }
1183         }
1184
1185         return (0);
1186 }
1187
1188 int
1189 zpool_export(zpool_handle_t *zhp, boolean_t force)
1190 {
1191         return (zpool_export_common(zhp, force, B_FALSE));
1192 }
1193
1194 int
1195 zpool_export_force(zpool_handle_t *zhp)
1196 {
1197         return (zpool_export_common(zhp, B_TRUE, B_TRUE));
1198 }
1199
1200 /*
1201  * zpool_import() is a contracted interface. Should be kept the same
1202  * if possible.
1203  *
1204  * Applications should use zpool_import_props() to import a pool with
1205  * new properties value to be set.
1206  */
1207 int
1208 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1209     char *altroot)
1210 {
1211         nvlist_t *props = NULL;
1212         int ret;
1213
1214         if (altroot != NULL) {
1215                 if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1216                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1217                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1218                             newname));
1219                 }
1220
1221                 if (nvlist_add_string(props,
1222                     zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1223                     nvlist_add_string(props,
1224                     zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1225                         nvlist_free(props);
1226                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1227                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1228                             newname));
1229                 }
1230         }
1231
1232         ret = zpool_import_props(hdl, config, newname, props, B_FALSE);
1233         if (props)
1234                 nvlist_free(props);
1235         return (ret);
1236 }
1237
1238 /*
1239  * Import the given pool using the known configuration and a list of
1240  * properties to be set. The configuration should have come from
1241  * zpool_find_import(). The 'newname' parameters control whether the pool
1242  * is imported with a different name.
1243  */
1244 int
1245 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1246     nvlist_t *props, boolean_t importfaulted)
1247 {
1248         zfs_cmd_t zc = { 0 };
1249         char *thename;
1250         char *origname;
1251         int ret;
1252         char errbuf[1024];
1253
1254         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1255             &origname) == 0);
1256
1257         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1258             "cannot import pool '%s'"), origname);
1259
1260         if (newname != NULL) {
1261                 if (!zpool_name_valid(hdl, B_FALSE, newname))
1262                         return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1263                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1264                             newname));
1265                 thename = (char *)newname;
1266         } else {
1267                 thename = origname;
1268         }
1269
1270         if (props) {
1271                 uint64_t version;
1272
1273                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1274                     &version) == 0);
1275
1276                 if ((props = zpool_valid_proplist(hdl, origname,
1277                     props, version, B_TRUE, errbuf)) == NULL) {
1278                         return (-1);
1279                 } else if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1280                         nvlist_free(props);
1281                         return (-1);
1282                 }
1283         }
1284
1285         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1286
1287         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1288             &zc.zc_guid) == 0);
1289
1290         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1291                 nvlist_free(props);
1292                 return (-1);
1293         }
1294
1295         zc.zc_cookie = (uint64_t)importfaulted;
1296         ret = 0;
1297         if (zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc) != 0) {
1298                 char desc[1024];
1299                 if (newname == NULL)
1300                         (void) snprintf(desc, sizeof (desc),
1301                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1302                             thename);
1303                 else
1304                         (void) snprintf(desc, sizeof (desc),
1305                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1306                             origname, thename);
1307
1308                 switch (errno) {
1309                 case ENOTSUP:
1310                         /*
1311                          * Unsupported version.
1312                          */
1313                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1314                         break;
1315
1316                 case EINVAL:
1317                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1318                         break;
1319
1320                 default:
1321                         (void) zpool_standard_error(hdl, errno, desc);
1322                 }
1323
1324                 ret = -1;
1325         } else {
1326                 zpool_handle_t *zhp;
1327
1328                 /*
1329                  * This should never fail, but play it safe anyway.
1330                  */
1331                 if (zpool_open_silent(hdl, thename, &zhp) != 0) {
1332                         ret = -1;
1333                 } else if (zhp != NULL) {
1334                         ret = zpool_create_zvol_links(zhp);
1335                         zpool_close(zhp);
1336                 }
1337
1338         }
1339
1340         zcmd_free_nvlists(&zc);
1341         nvlist_free(props);
1342
1343         return (ret);
1344 }
1345
1346 /*
1347  * Scrub the pool.
1348  */
1349 int
1350 zpool_scrub(zpool_handle_t *zhp, pool_scrub_type_t type)
1351 {
1352         zfs_cmd_t zc = { 0 };
1353         char msg[1024];
1354         libzfs_handle_t *hdl = zhp->zpool_hdl;
1355
1356         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1357         zc.zc_cookie = type;
1358
1359         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SCRUB, &zc) == 0)
1360                 return (0);
1361
1362         (void) snprintf(msg, sizeof (msg),
1363             dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1364
1365         if (errno == EBUSY)
1366                 return (zfs_error(hdl, EZFS_RESILVERING, msg));
1367         else
1368                 return (zpool_standard_error(hdl, errno, msg));
1369 }
1370
1371 /*
1372  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1373  * spare; but FALSE if its an INUSE spare.
1374  */
1375 static nvlist_t *
1376 vdev_to_nvlist_iter(nvlist_t *nv, const char *search, uint64_t guid,
1377     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
1378 {
1379         uint_t c, children;
1380         nvlist_t **child;
1381         uint64_t theguid, present;
1382         char *path;
1383         uint64_t wholedisk = 0;
1384         nvlist_t *ret;
1385         uint64_t is_log;
1386
1387         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &theguid) == 0);
1388
1389         if (search == NULL &&
1390             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &present) == 0) {
1391                 /*
1392                  * If the device has never been present since import, the only
1393                  * reliable way to match the vdev is by GUID.
1394                  */
1395                 if (theguid == guid)
1396                         return (nv);
1397         } else if (search != NULL &&
1398             nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0) {
1399                 (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1400                     &wholedisk);
1401                 if (wholedisk) {
1402                         /*
1403                          * For whole disks, the internal path has 's0', but the
1404                          * path passed in by the user doesn't.
1405                          */
1406                         if (strlen(search) == strlen(path) - 2 &&
1407                             strncmp(search, path, strlen(search)) == 0)
1408                                 return (nv);
1409                 } else if (strcmp(search, path) == 0) {
1410                         return (nv);
1411                 }
1412         }
1413
1414         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1415             &child, &children) != 0)
1416                 return (NULL);
1417
1418         for (c = 0; c < children; c++) {
1419                 if ((ret = vdev_to_nvlist_iter(child[c], search, guid,
1420                     avail_spare, l2cache, NULL)) != NULL) {
1421                         /*
1422                          * The 'is_log' value is only set for the toplevel
1423                          * vdev, not the leaf vdevs.  So we always lookup the
1424                          * log device from the root of the vdev tree (where
1425                          * 'log' is non-NULL).
1426                          */
1427                         if (log != NULL &&
1428                             nvlist_lookup_uint64(child[c],
1429                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
1430                             is_log) {
1431                                 *log = B_TRUE;
1432                         }
1433                         return (ret);
1434                 }
1435         }
1436
1437         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1438             &child, &children) == 0) {
1439                 for (c = 0; c < children; c++) {
1440                         if ((ret = vdev_to_nvlist_iter(child[c], search, guid,
1441                             avail_spare, l2cache, NULL)) != NULL) {
1442                                 *avail_spare = B_TRUE;
1443                                 return (ret);
1444                         }
1445                 }
1446         }
1447
1448         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1449             &child, &children) == 0) {
1450                 for (c = 0; c < children; c++) {
1451                         if ((ret = vdev_to_nvlist_iter(child[c], search, guid,
1452                             avail_spare, l2cache, NULL)) != NULL) {
1453                                 *l2cache = B_TRUE;
1454                                 return (ret);
1455                         }
1456                 }
1457         }
1458
1459         return (NULL);
1460 }
1461
1462 nvlist_t *
1463 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
1464     boolean_t *l2cache, boolean_t *log)
1465 {
1466         char buf[MAXPATHLEN];
1467         const char *search;
1468         char *end;
1469         nvlist_t *nvroot;
1470         uint64_t guid;
1471
1472         guid = strtoull(path, &end, 10);
1473         if (guid != 0 && *end == '\0') {
1474                 search = NULL;
1475         } else if (path[0] != '/') {
1476                 (void) snprintf(buf, sizeof (buf), "%s%s", "/dev/dsk/", path);
1477                 search = buf;
1478         } else {
1479                 search = path;
1480         }
1481
1482         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1483             &nvroot) == 0);
1484
1485         *avail_spare = B_FALSE;
1486         *l2cache = B_FALSE;
1487         if (log != NULL)
1488                 *log = B_FALSE;
1489         return (vdev_to_nvlist_iter(nvroot, search, guid, avail_spare,
1490             l2cache, log));
1491 }
1492
1493 static int
1494 vdev_online(nvlist_t *nv)
1495 {
1496         uint64_t ival;
1497
1498         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
1499             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
1500             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
1501                 return (0);
1502
1503         return (1);
1504 }
1505
1506 /*
1507  * Get phys_path for a root pool
1508  * Return 0 on success; non-zeron on failure.
1509  */
1510 int
1511 zpool_get_physpath(zpool_handle_t *zhp, char *physpath)
1512 {
1513         nvlist_t *vdev_root;
1514         nvlist_t **child;
1515         uint_t count;
1516         int i;
1517
1518         /*
1519          * Make sure this is a root pool, as phys_path doesn't mean
1520          * anything to a non-root pool.
1521          */
1522         if (!pool_is_bootable(zhp))
1523                 return (-1);
1524
1525         verify(nvlist_lookup_nvlist(zhp->zpool_config,
1526             ZPOOL_CONFIG_VDEV_TREE, &vdev_root) == 0);
1527
1528         if (nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
1529             &child, &count) != 0)
1530                 return (-2);
1531
1532         for (i = 0; i < count; i++) {
1533                 nvlist_t **child2;
1534                 uint_t count2;
1535                 char *type;
1536                 char *tmppath;
1537                 int j;
1538
1539                 if (nvlist_lookup_string(child[i], ZPOOL_CONFIG_TYPE, &type)
1540                     != 0)
1541                         return (-3);
1542
1543                 if (strcmp(type, VDEV_TYPE_DISK) == 0) {
1544                         if (!vdev_online(child[i]))
1545                                 return (-8);
1546                         verify(nvlist_lookup_string(child[i],
1547                             ZPOOL_CONFIG_PHYS_PATH, &tmppath) == 0);
1548                         (void) strncpy(physpath, tmppath, strlen(tmppath));
1549                 } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0) {
1550                         if (nvlist_lookup_nvlist_array(child[i],
1551                             ZPOOL_CONFIG_CHILDREN, &child2, &count2) != 0)
1552                                 return (-4);
1553
1554                         for (j = 0; j < count2; j++) {
1555                                 if (!vdev_online(child2[j]))
1556                                         return (-8);
1557                                 if (nvlist_lookup_string(child2[j],
1558                                     ZPOOL_CONFIG_PHYS_PATH, &tmppath) != 0)
1559                                         return (-5);
1560
1561                                 if ((strlen(physpath) + strlen(tmppath)) >
1562                                     MAXNAMELEN)
1563                                         return (-6);
1564
1565                                 if (strlen(physpath) == 0) {
1566                                         (void) strncpy(physpath, tmppath,
1567                                             strlen(tmppath));
1568                                 } else {
1569                                         (void) strcat(physpath, " ");
1570                                         (void) strcat(physpath, tmppath);
1571                                 }
1572                         }
1573                 } else {
1574                         return (-7);
1575                 }
1576         }
1577
1578         return (0);
1579 }
1580
1581 /*
1582  * Returns TRUE if the given guid corresponds to the given type.
1583  * This is used to check for hot spares (INUSE or not), and level 2 cache
1584  * devices.
1585  */
1586 static boolean_t
1587 is_guid_type(zpool_handle_t *zhp, uint64_t guid, const char *type)
1588 {
1589         uint64_t target_guid;
1590         nvlist_t *nvroot;
1591         nvlist_t **list;
1592         uint_t count;
1593         int i;
1594
1595         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1596             &nvroot) == 0);
1597         if (nvlist_lookup_nvlist_array(nvroot, type, &list, &count) == 0) {
1598                 for (i = 0; i < count; i++) {
1599                         verify(nvlist_lookup_uint64(list[i], ZPOOL_CONFIG_GUID,
1600                             &target_guid) == 0);
1601                         if (guid == target_guid)
1602                                 return (B_TRUE);
1603                 }
1604         }
1605
1606         return (B_FALSE);
1607 }
1608
1609 /*
1610  * Bring the specified vdev online.   The 'flags' parameter is a set of the
1611  * ZFS_ONLINE_* flags.
1612  */
1613 int
1614 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
1615     vdev_state_t *newstate)
1616 {
1617         zfs_cmd_t zc = { 0 };
1618         char msg[1024];
1619         nvlist_t *tgt;
1620         boolean_t avail_spare, l2cache;
1621         libzfs_handle_t *hdl = zhp->zpool_hdl;
1622
1623         (void) snprintf(msg, sizeof (msg),
1624             dgettext(TEXT_DOMAIN, "cannot online %s"), path);
1625
1626         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1627         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
1628             NULL)) == NULL)
1629                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
1630
1631         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
1632
1633         if (avail_spare ||
1634             is_guid_type(zhp, zc.zc_guid, ZPOOL_CONFIG_SPARES) == B_TRUE)
1635                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
1636
1637         zc.zc_cookie = VDEV_STATE_ONLINE;
1638         zc.zc_obj = flags;
1639
1640         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0)
1641                 return (zpool_standard_error(hdl, errno, msg));
1642
1643         *newstate = zc.zc_cookie;
1644         return (0);
1645 }
1646
1647 /*
1648  * Take the specified vdev offline
1649  */
1650 int
1651 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
1652 {
1653         zfs_cmd_t zc = { 0 };
1654         char msg[1024];
1655         nvlist_t *tgt;
1656         boolean_t avail_spare, l2cache;
1657         libzfs_handle_t *hdl = zhp->zpool_hdl;
1658
1659         (void) snprintf(msg, sizeof (msg),
1660             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
1661
1662         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1663         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
1664             NULL)) == NULL)
1665                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
1666
1667         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
1668
1669         if (avail_spare ||
1670             is_guid_type(zhp, zc.zc_guid, ZPOOL_CONFIG_SPARES) == B_TRUE)
1671                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
1672
1673         zc.zc_cookie = VDEV_STATE_OFFLINE;
1674         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
1675
1676         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
1677                 return (0);
1678
1679         switch (errno) {
1680         case EBUSY:
1681
1682                 /*
1683                  * There are no other replicas of this device.
1684                  */
1685                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
1686
1687         default:
1688                 return (zpool_standard_error(hdl, errno, msg));
1689         }
1690 }
1691
1692 /*
1693  * Mark the given vdev faulted.
1694  */
1695 int
1696 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid)
1697 {
1698         zfs_cmd_t zc = { 0 };
1699         char msg[1024];
1700         libzfs_handle_t *hdl = zhp->zpool_hdl;
1701
1702         (void) snprintf(msg, sizeof (msg),
1703             dgettext(TEXT_DOMAIN, "cannot fault %llu"), guid);
1704
1705         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1706         zc.zc_guid = guid;
1707         zc.zc_cookie = VDEV_STATE_FAULTED;
1708
1709         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
1710                 return (0);
1711
1712         switch (errno) {
1713         case EBUSY:
1714
1715                 /*
1716                  * There are no other replicas of this device.
1717                  */
1718                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
1719
1720         default:
1721                 return (zpool_standard_error(hdl, errno, msg));
1722         }
1723
1724 }
1725
1726 /*
1727  * Mark the given vdev degraded.
1728  */
1729 int
1730 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid)
1731 {
1732         zfs_cmd_t zc = { 0 };
1733         char msg[1024];
1734         libzfs_handle_t *hdl = zhp->zpool_hdl;
1735
1736         (void) snprintf(msg, sizeof (msg),
1737             dgettext(TEXT_DOMAIN, "cannot degrade %llu"), guid);
1738
1739         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1740         zc.zc_guid = guid;
1741         zc.zc_cookie = VDEV_STATE_DEGRADED;
1742
1743         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
1744                 return (0);
1745
1746         return (zpool_standard_error(hdl, errno, msg));
1747 }
1748
1749 /*
1750  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
1751  * a hot spare.
1752  */
1753 static boolean_t
1754 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
1755 {
1756         nvlist_t **child;
1757         uint_t c, children;
1758         char *type;
1759
1760         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
1761             &children) == 0) {
1762                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
1763                     &type) == 0);
1764
1765                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
1766                     children == 2 && child[which] == tgt)
1767                         return (B_TRUE);
1768
1769                 for (c = 0; c < children; c++)
1770                         if (is_replacing_spare(child[c], tgt, which))
1771                                 return (B_TRUE);
1772         }
1773
1774         return (B_FALSE);
1775 }
1776
1777 /*
1778  * Attach new_disk (fully described by nvroot) to old_disk.
1779  * If 'replacing' is specified, the new disk will replace the old one.
1780  */
1781 int
1782 zpool_vdev_attach(zpool_handle_t *zhp,
1783     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
1784 {
1785         zfs_cmd_t zc = { 0 };
1786         char msg[1024];
1787         int ret;
1788         nvlist_t *tgt;
1789         boolean_t avail_spare, l2cache, islog;
1790         uint64_t val;
1791         char *path, *newname;
1792         nvlist_t **child;
1793         uint_t children;
1794         nvlist_t *config_root;
1795         libzfs_handle_t *hdl = zhp->zpool_hdl;
1796         boolean_t rootpool = pool_is_bootable(zhp);
1797
1798         if (replacing)
1799                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1800                     "cannot replace %s with %s"), old_disk, new_disk);
1801         else
1802                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1803                     "cannot attach %s to %s"), new_disk, old_disk);
1804
1805         /*
1806          * If this is a root pool, make sure that we're not attaching an
1807          * EFI labeled device.
1808          */
1809         if (rootpool && pool_uses_efi(nvroot)) {
1810                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1811                     "EFI labeled devices are not supported on root pools."));
1812                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
1813         }
1814
1815         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1816         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
1817             &islog)) == 0)
1818                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
1819
1820         if (avail_spare)
1821                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
1822
1823         if (l2cache)
1824                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
1825
1826         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
1827         zc.zc_cookie = replacing;
1828
1829         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
1830             &child, &children) != 0 || children != 1) {
1831                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1832                     "new device must be a single disk"));
1833                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
1834         }
1835
1836         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
1837             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
1838
1839         if ((newname = zpool_vdev_name(NULL, NULL, child[0])) == NULL)
1840                 return (-1);
1841
1842         /*
1843          * If the target is a hot spare that has been swapped in, we can only
1844          * replace it with another hot spare.
1845          */
1846         if (replacing &&
1847             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
1848             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
1849             NULL) == NULL || !avail_spare) &&
1850             is_replacing_spare(config_root, tgt, 1)) {
1851                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1852                     "can only be replaced by another hot spare"));
1853                 free(newname);
1854                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
1855         }
1856
1857         /*
1858          * If we are attempting to replace a spare, it canot be applied to an
1859          * already spared device.
1860          */
1861         if (replacing &&
1862             nvlist_lookup_string(child[0], ZPOOL_CONFIG_PATH, &path) == 0 &&
1863             zpool_find_vdev(zhp, newname, &avail_spare,
1864             &l2cache, NULL) != NULL && avail_spare &&
1865             is_replacing_spare(config_root, tgt, 0)) {
1866                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1867                     "device has already been replaced with a spare"));
1868                 free(newname);
1869                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
1870         }
1871
1872         free(newname);
1873
1874         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1875                 return (-1);
1876
1877         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_VDEV_ATTACH, &zc);
1878
1879         zcmd_free_nvlists(&zc);
1880
1881         if (ret == 0) {
1882                 if (rootpool) {
1883                         /*
1884                          * XXX - This should be removed once we can
1885                          * automatically install the bootblocks on the
1886                          * newly attached disk.
1887                          */
1888                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Please "
1889                             "be sure to invoke %s to make '%s' bootable.\n"),
1890                             BOOTCMD, new_disk);
1891                 }
1892                 return (0);
1893         }
1894
1895         switch (errno) {
1896         case ENOTSUP:
1897                 /*
1898                  * Can't attach to or replace this type of vdev.
1899                  */
1900                 if (replacing) {
1901                         if (islog)
1902                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1903                                     "cannot replace a log with a spare"));
1904                         else
1905                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1906                                     "cannot replace a replacing device"));
1907                 } else {
1908                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1909                             "can only attach to mirrors and top-level "
1910                             "disks"));
1911                 }
1912                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
1913                 break;
1914
1915         case EINVAL:
1916                 /*
1917                  * The new device must be a single disk.
1918                  */
1919                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1920                     "new device must be a single disk"));
1921                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
1922                 break;
1923
1924         case EBUSY:
1925                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
1926                     new_disk);
1927                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
1928                 break;
1929
1930         case EOVERFLOW:
1931                 /*
1932                  * The new device is too small.
1933                  */
1934                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1935                     "device is too small"));
1936                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
1937                 break;
1938
1939         case EDOM:
1940                 /*
1941                  * The new device has a different alignment requirement.
1942                  */
1943                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1944                     "devices have different sector alignment"));
1945                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
1946                 break;
1947
1948         case ENAMETOOLONG:
1949                 /*
1950                  * The resulting top-level vdev spec won't fit in the label.
1951                  */
1952                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
1953                 break;
1954
1955         default:
1956                 (void) zpool_standard_error(hdl, errno, msg);
1957         }
1958
1959         return (-1);
1960 }
1961
1962 /*
1963  * Detach the specified device.
1964  */
1965 int
1966 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
1967 {
1968         zfs_cmd_t zc = { 0 };
1969         char msg[1024];
1970         nvlist_t *tgt;
1971         boolean_t avail_spare, l2cache;
1972         libzfs_handle_t *hdl = zhp->zpool_hdl;
1973
1974         (void) snprintf(msg, sizeof (msg),
1975             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
1976
1977         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1978         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
1979             NULL)) == 0)
1980                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
1981
1982         if (avail_spare)
1983                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
1984
1985         if (l2cache)
1986                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
1987
1988         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
1989
1990         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
1991                 return (0);
1992
1993         switch (errno) {
1994
1995         case ENOTSUP:
1996                 /*
1997                  * Can't detach from this type of vdev.
1998                  */
1999                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2000                     "applicable to mirror and replacing vdevs"));
2001                 (void) zfs_error(zhp->zpool_hdl, EZFS_BADTARGET, msg);
2002                 break;
2003
2004         case EBUSY:
2005                 /*
2006                  * There are no other replicas of this device.
2007                  */
2008                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2009                 break;
2010
2011         default:
2012                 (void) zpool_standard_error(hdl, errno, msg);
2013         }
2014
2015         return (-1);
2016 }
2017
2018 /*
2019  * Remove the given device.  Currently, this is supported only for hot spares
2020  * and level 2 cache devices.
2021  */
2022 int
2023 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
2024 {
2025         zfs_cmd_t zc = { 0 };
2026         char msg[1024];
2027         nvlist_t *tgt;
2028         boolean_t avail_spare, l2cache;
2029         libzfs_handle_t *hdl = zhp->zpool_hdl;
2030
2031         (void) snprintf(msg, sizeof (msg),
2032             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
2033
2034         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2035         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2036             NULL)) == 0)
2037                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2038
2039         if (!avail_spare && !l2cache) {
2040                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2041                     "only inactive hot spares or cache devices "
2042                     "can be removed"));
2043                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2044         }
2045
2046         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2047
2048         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
2049                 return (0);
2050
2051         return (zpool_standard_error(hdl, errno, msg));
2052 }
2053
2054 /*
2055  * Clear the errors for the pool, or the particular device if specified.
2056  */
2057 int
2058 zpool_clear(zpool_handle_t *zhp, const char *path)
2059 {
2060         zfs_cmd_t zc = { 0 };
2061         char msg[1024];
2062         nvlist_t *tgt;
2063         boolean_t avail_spare, l2cache;
2064         libzfs_handle_t *hdl = zhp->zpool_hdl;
2065
2066         if (path)
2067                 (void) snprintf(msg, sizeof (msg),
2068                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
2069                     path);
2070         else
2071                 (void) snprintf(msg, sizeof (msg),
2072                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
2073                     zhp->zpool_name);
2074
2075         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2076         if (path) {
2077                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
2078                     &l2cache, NULL)) == 0)
2079                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
2080
2081                 /*
2082                  * Don't allow error clearing for hot spares.  Do allow
2083                  * error clearing for l2cache devices.
2084                  */
2085                 if (avail_spare)
2086                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
2087
2088                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
2089                     &zc.zc_guid) == 0);
2090         }
2091
2092         if (zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc) == 0)
2093                 return (0);
2094
2095         return (zpool_standard_error(hdl, errno, msg));
2096 }
2097
2098 /*
2099  * Similar to zpool_clear(), but takes a GUID (used by fmd).
2100  */
2101 int
2102 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
2103 {
2104         zfs_cmd_t zc = { 0 };
2105         char msg[1024];
2106         libzfs_handle_t *hdl = zhp->zpool_hdl;
2107
2108         (void) snprintf(msg, sizeof (msg),
2109             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
2110             guid);
2111
2112         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2113         zc.zc_guid = guid;
2114
2115         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
2116                 return (0);
2117
2118         return (zpool_standard_error(hdl, errno, msg));
2119 }
2120
2121 /*
2122  * Iterate over all zvols in a given pool by walking the /dev/zvol/dsk/<pool>
2123  * hierarchy.
2124  */
2125 int
2126 zpool_iter_zvol(zpool_handle_t *zhp, int (*cb)(const char *, void *),
2127     void *data)
2128 {
2129         libzfs_handle_t *hdl = zhp->zpool_hdl;
2130         char (*paths)[MAXPATHLEN];
2131         size_t size = 4;
2132         int curr, fd, base, ret = 0;
2133         DIR *dirp;
2134         struct dirent *dp;
2135         struct stat st;
2136
2137         if ((base = open("/dev/zvol/dsk", O_RDONLY)) < 0)
2138                 return (errno == ENOENT ? 0 : -1);
2139
2140         if (fstatat(base, zhp->zpool_name, &st, 0) != 0) {
2141                 int err = errno;
2142                 (void) close(base);
2143                 return (err == ENOENT ? 0 : -1);
2144         }
2145
2146         /*
2147          * Oddly this wasn't a directory -- ignore that failure since we
2148          * know there are no links lower in the (non-existant) hierarchy.
2149          */
2150         if (!S_ISDIR(st.st_mode)) {
2151                 (void) close(base);
2152                 return (0);
2153         }
2154
2155         if ((paths = zfs_alloc(hdl, size * sizeof (paths[0]))) == NULL) {
2156                 (void) close(base);
2157                 return (-1);
2158         }
2159
2160         (void) strlcpy(paths[0], zhp->zpool_name, sizeof (paths[0]));
2161         curr = 0;
2162
2163         while (curr >= 0) {
2164                 if (fstatat(base, paths[curr], &st, AT_SYMLINK_NOFOLLOW) != 0)
2165                         goto err;
2166
2167                 if (S_ISDIR(st.st_mode)) {
2168                         if ((fd = openat(base, paths[curr], O_RDONLY)) < 0)
2169                                 goto err;
2170
2171                         if ((dirp = fdopendir(fd)) == NULL) {
2172                                 (void) close(fd);
2173                                 goto err;
2174                         }
2175
2176                         while ((dp = readdir(dirp)) != NULL) {
2177                                 if (dp->d_name[0] == '.')
2178                                         continue;
2179
2180                                 if (curr + 1 == size) {
2181                                         paths = zfs_realloc(hdl, paths,
2182                                             size * sizeof (paths[0]),
2183                                             size * 2 * sizeof (paths[0]));
2184                                         if (paths == NULL) {
2185                                                 (void) closedir(dirp);
2186                                                 (void) close(fd);
2187                                                 goto err;
2188                                         }
2189
2190                                         size *= 2;
2191                                 }
2192
2193                                 (void) strlcpy(paths[curr + 1], paths[curr],
2194                                     sizeof (paths[curr + 1]));
2195                                 (void) strlcat(paths[curr], "/",
2196                                     sizeof (paths[curr]));
2197                                 (void) strlcat(paths[curr], dp->d_name,
2198                                     sizeof (paths[curr]));
2199                                 curr++;
2200                         }
2201
2202                         (void) closedir(dirp);
2203
2204                 } else {
2205                         if ((ret = cb(paths[curr], data)) != 0)
2206                                 break;
2207                 }
2208
2209                 curr--;
2210         }
2211
2212         free(paths);
2213         (void) close(base);
2214
2215         return (ret);
2216
2217 err:
2218         free(paths);
2219         (void) close(base);
2220         return (-1);
2221 }
2222
2223 typedef struct zvol_cb {
2224         zpool_handle_t *zcb_pool;
2225         boolean_t zcb_create;
2226 } zvol_cb_t;
2227
2228 /*ARGSUSED*/
2229 static int
2230 do_zvol_create(zfs_handle_t *zhp, void *data)
2231 {
2232         int ret = 0;
2233
2234         if (ZFS_IS_VOLUME(zhp)) {
2235                 (void) zvol_create_link(zhp->zfs_hdl, zhp->zfs_name);
2236                 ret = zfs_iter_snapshots(zhp, do_zvol_create, NULL);
2237         }
2238
2239         if (ret == 0)
2240                 ret = zfs_iter_filesystems(zhp, do_zvol_create, NULL);
2241
2242         zfs_close(zhp);
2243
2244         return (ret);
2245 }
2246
2247 /*
2248  * Iterate over all zvols in the pool and make any necessary minor nodes.
2249  */
2250 int
2251 zpool_create_zvol_links(zpool_handle_t *zhp)
2252 {
2253         zfs_handle_t *zfp;
2254         int ret;
2255
2256         /*
2257          * If the pool is unavailable, just return success.
2258          */
2259         if ((zfp = make_dataset_handle(zhp->zpool_hdl,
2260             zhp->zpool_name)) == NULL)
2261                 return (0);
2262
2263         ret = zfs_iter_filesystems(zfp, do_zvol_create, NULL);
2264
2265         zfs_close(zfp);
2266         return (ret);
2267 }
2268
2269 static int
2270 do_zvol_remove(const char *dataset, void *data)
2271 {
2272         zpool_handle_t *zhp = data;
2273
2274         return (zvol_remove_link(zhp->zpool_hdl, dataset));
2275 }
2276
2277 /*
2278  * Iterate over all zvols in the pool and remove any minor nodes.  We iterate
2279  * by examining the /dev links so that a corrupted pool doesn't impede this
2280  * operation.
2281  */
2282 int
2283 zpool_remove_zvol_links(zpool_handle_t *zhp)
2284 {
2285         return (zpool_iter_zvol(zhp, do_zvol_remove, zhp));
2286 }
2287
2288 /*
2289  * Convert from a devid string to a path.
2290  */
2291 static char *
2292 devid_to_path(char *devid_str)
2293 {
2294         ddi_devid_t devid;
2295         char *minor;
2296         char *path;
2297         devid_nmlist_t *list = NULL;
2298         int ret;
2299
2300         if (devid_str_decode(devid_str, &devid, &minor) != 0)
2301                 return (NULL);
2302
2303         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
2304
2305         devid_str_free(minor);
2306         devid_free(devid);
2307
2308         if (ret != 0)
2309                 return (NULL);
2310
2311         if ((path = strdup(list[0].devname)) == NULL)
2312                 return (NULL);
2313
2314         devid_free_nmlist(list);
2315
2316         return (path);
2317 }
2318
2319 /*
2320  * Convert from a path to a devid string.
2321  */
2322 static char *
2323 path_to_devid(const char *path)
2324 {
2325         int fd;
2326         ddi_devid_t devid;
2327         char *minor, *ret;
2328
2329         if ((fd = open(path, O_RDONLY)) < 0)
2330                 return (NULL);
2331
2332         minor = NULL;
2333         ret = NULL;
2334         if (devid_get(fd, &devid) == 0) {
2335                 if (devid_get_minor_name(fd, &minor) == 0)
2336                         ret = devid_str_encode(devid, minor);
2337                 if (minor != NULL)
2338                         devid_str_free(minor);
2339                 devid_free(devid);
2340         }
2341         (void) close(fd);
2342
2343         return (ret);
2344 }
2345
2346 /*
2347  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
2348  * ignore any failure here, since a common case is for an unprivileged user to
2349  * type 'zpool status', and we'll display the correct information anyway.
2350  */
2351 static void
2352 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
2353 {
2354         zfs_cmd_t zc = { 0 };
2355
2356         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2357         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
2358         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
2359             &zc.zc_guid) == 0);
2360
2361         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
2362 }
2363
2364 /*
2365  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
2366  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
2367  * We also check if this is a whole disk, in which case we strip off the
2368  * trailing 's0' slice name.
2369  *
2370  * This routine is also responsible for identifying when disks have been
2371  * reconfigured in a new location.  The kernel will have opened the device by
2372  * devid, but the path will still refer to the old location.  To catch this, we
2373  * first do a path -> devid translation (which is fast for the common case).  If
2374  * the devid matches, we're done.  If not, we do a reverse devid -> path
2375  * translation and issue the appropriate ioctl() to update the path of the vdev.
2376  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
2377  * of these checks.
2378  */
2379 char *
2380 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv)
2381 {
2382         char *path, *devid;
2383         uint64_t value;
2384         char buf[64];
2385         vdev_stat_t *vs;
2386         uint_t vsc;
2387
2388         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
2389             &value) == 0) {
2390                 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
2391                     &value) == 0);
2392                 (void) snprintf(buf, sizeof (buf), "%llu",
2393                     (u_longlong_t)value);
2394                 path = buf;
2395         } else if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0) {
2396
2397                 /*
2398                  * If the device is dead (faulted, offline, etc) then don't
2399                  * bother opening it.  Otherwise we may be forcing the user to
2400                  * open a misbehaving device, which can have undesirable
2401                  * effects.
2402                  */
2403                 if ((nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_STATS,
2404                     (uint64_t **)&vs, &vsc) != 0 ||
2405                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
2406                     zhp != NULL &&
2407                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
2408                         /*
2409                          * Determine if the current path is correct.
2410                          */
2411                         char *newdevid = path_to_devid(path);
2412
2413                         if (newdevid == NULL ||
2414                             strcmp(devid, newdevid) != 0) {
2415                                 char *newpath;
2416
2417                                 if ((newpath = devid_to_path(devid)) != NULL) {
2418                                         /*
2419                                          * Update the path appropriately.
2420                                          */
2421                                         set_path(zhp, nv, newpath);
2422                                         if (nvlist_add_string(nv,
2423                                             ZPOOL_CONFIG_PATH, newpath) == 0)
2424                                                 verify(nvlist_lookup_string(nv,
2425                                                     ZPOOL_CONFIG_PATH,
2426                                                     &path) == 0);
2427                                         free(newpath);
2428                                 }
2429                         }
2430
2431                         if (newdevid)
2432                                 devid_str_free(newdevid);
2433                 }
2434
2435                 if (strncmp(path, "/dev/dsk/", 9) == 0)
2436                         path += 9;
2437
2438                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
2439                     &value) == 0 && value) {
2440                         char *tmp = zfs_strdup(hdl, path);
2441                         if (tmp == NULL)
2442                                 return (NULL);
2443                         tmp[strlen(path) - 2] = '\0';
2444                         return (tmp);
2445                 }
2446         } else {
2447                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
2448
2449                 /*
2450                  * If it's a raidz device, we need to stick in the parity level.
2451                  */
2452                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
2453                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
2454                             &value) == 0);
2455                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
2456                             (u_longlong_t)value);
2457                         path = buf;
2458                 }
2459         }
2460
2461         return (zfs_strdup(hdl, path));
2462 }
2463
2464 static int
2465 zbookmark_compare(const void *a, const void *b)
2466 {
2467         return (memcmp(a, b, sizeof (zbookmark_t)));
2468 }
2469
2470 /*
2471  * Retrieve the persistent error log, uniquify the members, and return to the
2472  * caller.
2473  */
2474 int
2475 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
2476 {
2477         zfs_cmd_t zc = { 0 };
2478         uint64_t count;
2479         zbookmark_t *zb = NULL;
2480         int i;
2481
2482         /*
2483          * Retrieve the raw error list from the kernel.  If the number of errors
2484          * has increased, allocate more space and continue until we get the
2485          * entire list.
2486          */
2487         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
2488             &count) == 0);
2489         if (count == 0)
2490                 return (0);
2491         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
2492             count * sizeof (zbookmark_t))) == (uintptr_t)NULL)
2493                 return (-1);
2494         zc.zc_nvlist_dst_size = count;
2495         (void) strcpy(zc.zc_name, zhp->zpool_name);
2496         for (;;) {
2497                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
2498                     &zc) != 0) {
2499                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
2500                         if (errno == ENOMEM) {
2501                                 count = zc.zc_nvlist_dst_size;
2502                                 if ((zc.zc_nvlist_dst = (uintptr_t)
2503                                     zfs_alloc(zhp->zpool_hdl, count *
2504                                     sizeof (zbookmark_t))) == (uintptr_t)NULL)
2505                                         return (-1);
2506                         } else {
2507                                 return (-1);
2508                         }
2509                 } else {
2510                         break;
2511                 }
2512         }
2513
2514         /*
2515          * Sort the resulting bookmarks.  This is a little confusing due to the
2516          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
2517          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
2518          * _not_ copied as part of the process.  So we point the start of our
2519          * array appropriate and decrement the total number of elements.
2520          */
2521         zb = ((zbookmark_t *)(uintptr_t)zc.zc_nvlist_dst) +
2522             zc.zc_nvlist_dst_size;
2523         count -= zc.zc_nvlist_dst_size;
2524
2525         qsort(zb, count, sizeof (zbookmark_t), zbookmark_compare);
2526
2527         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
2528
2529         /*
2530          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
2531          */
2532         for (i = 0; i < count; i++) {
2533                 nvlist_t *nv;
2534
2535                 /* ignoring zb_blkid and zb_level for now */
2536                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
2537                     zb[i-1].zb_object == zb[i].zb_object)
2538                         continue;
2539
2540                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
2541                         goto nomem;
2542                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
2543                     zb[i].zb_objset) != 0) {
2544                         nvlist_free(nv);
2545                         goto nomem;
2546                 }
2547                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
2548                     zb[i].zb_object) != 0) {
2549                         nvlist_free(nv);
2550                         goto nomem;
2551                 }
2552                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
2553                         nvlist_free(nv);
2554                         goto nomem;
2555                 }
2556                 nvlist_free(nv);
2557         }
2558
2559         free((void *)(uintptr_t)zc.zc_nvlist_dst);
2560         return (0);
2561
2562 nomem:
2563         free((void *)(uintptr_t)zc.zc_nvlist_dst);
2564         return (no_memory(zhp->zpool_hdl));
2565 }
2566
2567 /*
2568  * Upgrade a ZFS pool to the latest on-disk version.
2569  */
2570 int
2571 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
2572 {
2573         zfs_cmd_t zc = { 0 };
2574         libzfs_handle_t *hdl = zhp->zpool_hdl;
2575
2576         (void) strcpy(zc.zc_name, zhp->zpool_name);
2577         zc.zc_cookie = new_version;
2578
2579         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
2580                 return (zpool_standard_error_fmt(hdl, errno,
2581                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
2582                     zhp->zpool_name));
2583         return (0);
2584 }
2585
2586 void
2587 zpool_set_history_str(const char *subcommand, int argc, char **argv,
2588     char *history_str)
2589 {
2590         int i;
2591
2592         (void) strlcpy(history_str, subcommand, HIS_MAX_RECORD_LEN);
2593         for (i = 1; i < argc; i++) {
2594                 if (strlen(history_str) + 1 + strlen(argv[i]) >
2595                     HIS_MAX_RECORD_LEN)
2596                         break;
2597                 (void) strlcat(history_str, " ", HIS_MAX_RECORD_LEN);
2598                 (void) strlcat(history_str, argv[i], HIS_MAX_RECORD_LEN);
2599         }
2600 }
2601
2602 /*
2603  * Stage command history for logging.
2604  */
2605 int
2606 zpool_stage_history(libzfs_handle_t *hdl, const char *history_str)
2607 {
2608         if (history_str == NULL)
2609                 return (EINVAL);
2610
2611         if (strlen(history_str) > HIS_MAX_RECORD_LEN)
2612                 return (EINVAL);
2613
2614         if (hdl->libzfs_log_str != NULL)
2615                 free(hdl->libzfs_log_str);
2616
2617         if ((hdl->libzfs_log_str = strdup(history_str)) == NULL)
2618                 return (no_memory(hdl));
2619
2620         return (0);
2621 }
2622
2623 /*
2624  * Perform ioctl to get some command history of a pool.
2625  *
2626  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
2627  * logical offset of the history buffer to start reading from.
2628  *
2629  * Upon return, 'off' is the next logical offset to read from and
2630  * 'len' is the actual amount of bytes read into 'buf'.
2631  */
2632 static int
2633 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
2634 {
2635         zfs_cmd_t zc = { 0 };
2636         libzfs_handle_t *hdl = zhp->zpool_hdl;
2637
2638         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2639
2640         zc.zc_history = (uint64_t)(uintptr_t)buf;
2641         zc.zc_history_len = *len;
2642         zc.zc_history_offset = *off;
2643
2644         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
2645                 switch (errno) {
2646                 case EPERM:
2647                         return (zfs_error_fmt(hdl, EZFS_PERM,
2648                             dgettext(TEXT_DOMAIN,
2649                             "cannot show history for pool '%s'"),
2650                             zhp->zpool_name));
2651                 case ENOENT:
2652                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
2653                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
2654                             "'%s'"), zhp->zpool_name));
2655                 case ENOTSUP:
2656                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
2657                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
2658                             "'%s', pool must be upgraded"), zhp->zpool_name));
2659                 default:
2660                         return (zpool_standard_error_fmt(hdl, errno,
2661                             dgettext(TEXT_DOMAIN,
2662                             "cannot get history for '%s'"), zhp->zpool_name));
2663                 }
2664         }
2665
2666         *len = zc.zc_history_len;
2667         *off = zc.zc_history_offset;
2668
2669         return (0);
2670 }
2671
2672 /*
2673  * Process the buffer of nvlists, unpacking and storing each nvlist record
2674  * into 'records'.  'leftover' is set to the number of bytes that weren't
2675  * processed as there wasn't a complete record.
2676  */
2677 static int
2678 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
2679     nvlist_t ***records, uint_t *numrecords)
2680 {
2681         uint64_t reclen;
2682         nvlist_t *nv;
2683         int i;
2684
2685         while (bytes_read > sizeof (reclen)) {
2686
2687                 /* get length of packed record (stored as little endian) */
2688                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
2689                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
2690
2691                 if (bytes_read < sizeof (reclen) + reclen)
2692                         break;
2693
2694                 /* unpack record */
2695                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
2696                         return (ENOMEM);
2697                 bytes_read -= sizeof (reclen) + reclen;
2698                 buf += sizeof (reclen) + reclen;
2699
2700                 /* add record to nvlist array */
2701                 (*numrecords)++;
2702                 if (ISP2(*numrecords + 1)) {
2703                         *records = realloc(*records,
2704                             *numrecords * 2 * sizeof (nvlist_t *));
2705                 }
2706                 (*records)[*numrecords - 1] = nv;
2707         }
2708
2709         *leftover = bytes_read;
2710         return (0);
2711 }
2712
2713 #define HIS_BUF_LEN     (128*1024)
2714
2715 /*
2716  * Retrieve the command history of a pool.
2717  */
2718 int
2719 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
2720 {
2721         char buf[HIS_BUF_LEN];
2722         uint64_t off = 0;
2723         nvlist_t **records = NULL;
2724         uint_t numrecords = 0;
2725         int err, i;
2726
2727         do {
2728                 uint64_t bytes_read = sizeof (buf);
2729                 uint64_t leftover;
2730
2731                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
2732                         break;
2733
2734                 /* if nothing else was read in, we're at EOF, just return */
2735                 if (!bytes_read)
2736                         break;
2737
2738                 if ((err = zpool_history_unpack(buf, bytes_read,
2739                     &leftover, &records, &numrecords)) != 0)
2740                         break;
2741                 off -= leftover;
2742
2743                 /* CONSTCOND */
2744         } while (1);
2745
2746         if (!err) {
2747                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
2748                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
2749                     records, numrecords) == 0);
2750         }
2751         for (i = 0; i < numrecords; i++)
2752                 nvlist_free(records[i]);
2753         free(records);
2754
2755         return (err);
2756 }
2757
2758 void
2759 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
2760     char *pathname, size_t len)
2761 {
2762         zfs_cmd_t zc = { 0 };
2763         boolean_t mounted = B_FALSE;
2764         char *mntpnt = NULL;
2765         char dsname[MAXNAMELEN];
2766
2767         if (dsobj == 0) {
2768                 /* special case for the MOS */
2769                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>", obj);
2770                 return;
2771         }
2772
2773         /* get the dataset's name */
2774         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2775         zc.zc_obj = dsobj;
2776         if (ioctl(zhp->zpool_hdl->libzfs_fd,
2777             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
2778                 /* just write out a path of two object numbers */
2779                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
2780                     dsobj, obj);
2781                 return;
2782         }
2783         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
2784
2785         /* find out if the dataset is mounted */
2786         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
2787
2788         /* get the corrupted object's path */
2789         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
2790         zc.zc_obj = obj;
2791         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
2792             &zc) == 0) {
2793                 if (mounted) {
2794                         (void) snprintf(pathname, len, "%s%s", mntpnt,
2795                             zc.zc_value);
2796                 } else {
2797                         (void) snprintf(pathname, len, "%s:%s",
2798                             dsname, zc.zc_value);
2799                 }
2800         } else {
2801                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname, obj);
2802         }
2803         free(mntpnt);
2804 }
2805
2806 #define RDISK_ROOT      "/dev/rdsk"
2807 #define BACKUP_SLICE    "s2"
2808 /*
2809  * Don't start the slice at the default block of 34; many storage
2810  * devices will use a stripe width of 128k, so start there instead.
2811  */
2812 #define NEW_START_BLOCK 256
2813
2814 /*
2815  * Read the EFI label from the config, if a label does not exist then
2816  * pass back the error to the caller. If the caller has passed a non-NULL
2817  * diskaddr argument then we set it to the starting address of the EFI
2818  * partition.
2819  */
2820 static int
2821 read_efi_label(nvlist_t *config, diskaddr_t *sb)
2822 {
2823         char *path;
2824         int fd;
2825         char diskname[MAXPATHLEN];
2826         int err = -1;
2827
2828         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
2829                 return (err);
2830
2831         (void) snprintf(diskname, sizeof (diskname), "%s%s", RDISK_ROOT,
2832             strrchr(path, '/'));
2833         if ((fd = open(diskname, O_RDONLY|O_NDELAY)) >= 0) {
2834                 struct dk_gpt *vtoc;
2835
2836                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
2837                         if (sb != NULL)
2838                                 *sb = vtoc->efi_parts[0].p_start;
2839                         efi_free(vtoc);
2840                 }
2841                 (void) close(fd);
2842         }
2843         return (err);
2844 }
2845
2846 /*
2847  * determine where a partition starts on a disk in the current
2848  * configuration
2849  */
2850 static diskaddr_t
2851 find_start_block(nvlist_t *config)
2852 {
2853         nvlist_t **child;
2854         uint_t c, children;
2855         diskaddr_t sb = MAXOFFSET_T;
2856         uint64_t wholedisk;
2857
2858         if (nvlist_lookup_nvlist_array(config,
2859             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
2860                 if (nvlist_lookup_uint64(config,
2861                     ZPOOL_CONFIG_WHOLE_DISK,
2862                     &wholedisk) != 0 || !wholedisk) {
2863                         return (MAXOFFSET_T);
2864                 }
2865                 if (read_efi_label(config, &sb) < 0)
2866                         sb = MAXOFFSET_T;
2867                 return (sb);
2868         }
2869
2870         for (c = 0; c < children; c++) {
2871                 sb = find_start_block(child[c]);
2872                 if (sb != MAXOFFSET_T) {
2873                         return (sb);
2874                 }
2875         }
2876         return (MAXOFFSET_T);
2877 }
2878
2879 /*
2880  * Label an individual disk.  The name provided is the short name,
2881  * stripped of any leading /dev path.
2882  */
2883 int
2884 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, char *name)
2885 {
2886         char path[MAXPATHLEN];
2887         struct dk_gpt *vtoc;
2888         int fd;
2889         size_t resv = EFI_MIN_RESV_SIZE;
2890         uint64_t slice_size;
2891         diskaddr_t start_block;
2892         char errbuf[1024];
2893
2894         /* prepare an error message just in case */
2895         (void) snprintf(errbuf, sizeof (errbuf),
2896             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
2897
2898         if (zhp) {
2899                 nvlist_t *nvroot;
2900
2901                 if (pool_is_bootable(zhp)) {
2902                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2903                             "EFI labeled devices are not supported on root "
2904                             "pools."));
2905                         return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
2906                 }
2907
2908                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
2909                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
2910
2911                 if (zhp->zpool_start_block == 0)
2912                         start_block = find_start_block(nvroot);
2913                 else
2914                         start_block = zhp->zpool_start_block;
2915                 zhp->zpool_start_block = start_block;
2916         } else {
2917                 /* new pool */
2918                 start_block = NEW_START_BLOCK;
2919         }
2920
2921         (void) snprintf(path, sizeof (path), "%s/%s%s", RDISK_ROOT, name,
2922             BACKUP_SLICE);
2923
2924         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
2925                 /*
2926                  * This shouldn't happen.  We've long since verified that this
2927                  * is a valid device.
2928                  */
2929                 zfs_error_aux(hdl,
2930                     dgettext(TEXT_DOMAIN, "unable to open device"));
2931                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
2932         }
2933
2934         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
2935                 /*
2936                  * The only way this can fail is if we run out of memory, or we
2937                  * were unable to read the disk's capacity
2938                  */
2939                 if (errno == ENOMEM)
2940                         (void) no_memory(hdl);
2941
2942                 (void) close(fd);
2943                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2944                     "unable to read disk capacity"), name);
2945
2946                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
2947         }
2948
2949         slice_size = vtoc->efi_last_u_lba + 1;
2950         slice_size -= EFI_MIN_RESV_SIZE;
2951         if (start_block == MAXOFFSET_T)
2952                 start_block = NEW_START_BLOCK;
2953         slice_size -= start_block;
2954
2955         vtoc->efi_parts[0].p_start = start_block;
2956         vtoc->efi_parts[0].p_size = slice_size;
2957
2958         /*
2959          * Why we use V_USR: V_BACKUP confuses users, and is considered
2960          * disposable by some EFI utilities (since EFI doesn't have a backup
2961          * slice).  V_UNASSIGNED is supposed to be used only for zero size
2962          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
2963          * etc. were all pretty specific.  V_USR is as close to reality as we
2964          * can get, in the absence of V_OTHER.
2965          */
2966         vtoc->efi_parts[0].p_tag = V_USR;
2967         (void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
2968
2969         vtoc->efi_parts[8].p_start = slice_size + start_block;
2970         vtoc->efi_parts[8].p_size = resv;
2971         vtoc->efi_parts[8].p_tag = V_RESERVED;
2972
2973         if (efi_write(fd, vtoc) != 0) {
2974                 /*
2975                  * Some block drivers (like pcata) may not support EFI
2976                  * GPT labels.  Print out a helpful error message dir-
2977                  * ecting the user to manually label the disk and give
2978                  * a specific slice.
2979                  */
2980                 (void) close(fd);
2981                 efi_free(vtoc);
2982
2983                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2984                     "try using fdisk(1M) and then provide a specific slice"));
2985                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
2986         }
2987
2988         (void) close(fd);
2989         efi_free(vtoc);
2990         return (0);
2991 }
2992
2993 static boolean_t
2994 supported_dump_vdev_type(libzfs_handle_t *hdl, nvlist_t *config, char *errbuf)
2995 {
2996         char *type;
2997         nvlist_t **child;
2998         uint_t children, c;
2999
3000         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_TYPE, &type) == 0);
3001         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0 ||
3002             strcmp(type, VDEV_TYPE_FILE) == 0 ||
3003             strcmp(type, VDEV_TYPE_LOG) == 0 ||
3004             strcmp(type, VDEV_TYPE_MISSING) == 0) {
3005                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3006                     "vdev type '%s' is not supported"), type);
3007                 (void) zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf);
3008                 return (B_FALSE);
3009         }
3010         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
3011             &child, &children) == 0) {
3012                 for (c = 0; c < children; c++) {
3013                         if (!supported_dump_vdev_type(hdl, child[c], errbuf))
3014                                 return (B_FALSE);
3015                 }
3016         }
3017         return (B_TRUE);
3018 }
3019
3020 /*
3021  * check if this zvol is allowable for use as a dump device; zero if
3022  * it is, > 0 if it isn't, < 0 if it isn't a zvol
3023  */
3024 int
3025 zvol_check_dump_config(char *arg)
3026 {
3027         zpool_handle_t *zhp = NULL;
3028         nvlist_t *config, *nvroot;
3029         char *p, *volname;
3030         nvlist_t **top;
3031         uint_t toplevels;
3032         libzfs_handle_t *hdl;
3033         char errbuf[1024];
3034         char poolname[ZPOOL_MAXNAMELEN];
3035         int pathlen = strlen(ZVOL_FULL_DEV_DIR);
3036         int ret = 1;
3037
3038         if (strncmp(arg, ZVOL_FULL_DEV_DIR, pathlen)) {
3039                 return (-1);
3040         }
3041
3042         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3043             "dump is not supported on device '%s'"), arg);
3044
3045         if ((hdl = libzfs_init()) == NULL)
3046                 return (1);
3047         libzfs_print_on_error(hdl, B_TRUE);
3048
3049         volname = arg + pathlen;
3050
3051         /* check the configuration of the pool */
3052         if ((p = strchr(volname, '/')) == NULL) {
3053                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3054                     "malformed dataset name"));
3055                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
3056                 return (1);
3057         } else if (p - volname >= ZFS_MAXNAMELEN) {
3058                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3059                     "dataset name is too long"));
3060                 (void) zfs_error(hdl, EZFS_NAMETOOLONG, errbuf);
3061                 return (1);
3062         } else {
3063                 (void) strncpy(poolname, volname, p - volname);
3064                 poolname[p - volname] = '\0';
3065         }
3066
3067         if ((zhp = zpool_open(hdl, poolname)) == NULL) {
3068                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3069                     "could not open pool '%s'"), poolname);
3070                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
3071                 goto out;
3072         }
3073         config = zpool_get_config(zhp, NULL);
3074         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3075             &nvroot) != 0) {
3076                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3077                     "could not obtain vdev configuration for  '%s'"), poolname);
3078                 (void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
3079                 goto out;
3080         }
3081
3082         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3083             &top, &toplevels) == 0);
3084         if (toplevels != 1) {
3085                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3086                     "'%s' has multiple top level vdevs"), poolname);
3087                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, errbuf);
3088                 goto out;
3089         }
3090
3091         if (!supported_dump_vdev_type(hdl, top[0], errbuf)) {
3092                 goto out;
3093         }
3094         ret = 0;
3095
3096 out:
3097         if (zhp)
3098                 zpool_close(zhp);
3099         libzfs_fini(hdl);
3100         return (ret);
3101 }