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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  25  * Copyright (c) 2012 by Delphix. All rights reserved.
  26  */
  27
  28 #include <ctype.h>
  29 #include <errno.h>
  30 #include <devid.h>
  31 #include <fcntl.h>
  32 #include <libintl.h>
  33 #include <stdio.h>
  34 #include <stdlib.h>
  35 #include <strings.h>
  36 #include <unistd.h>
  37 #include <zone.h>
  38 #include <sys/stat.h>
  39 #include <sys/efi_partition.h>
  40 #include <sys/vtoc.h>
  41 #include <sys/zfs_ioctl.h>
  42 #include <dlfcn.h>
  43
  44 #include "zfs_namecheck.h"
  45 #include "zfs_prop.h"
  46 #include "libzfs_impl.h"
  47 #include "zfs_comutil.h"
  48
  49 static int read_efi_label(nvlist_t *config, diskaddr_t *sb);
  50
  51 typedef struct prop_flags {
  52         int create:1;   /* Validate property on creation */
  53         int import:1;   /* Validate property on import */
  54 } prop_flags_t;
  55
  56 /*
  57  * ====================================================================
  58  *   zpool property functions
  59  * ====================================================================
  60  */
  61
  62 static int
  63 zpool_get_all_props(zpool_handle_t *zhp)
  64 {
  65         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
  66         libzfs_handle_t *hdl = zhp->zpool_hdl;
  67
  68         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
  69
  70         if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
  71                 return (-1);
  72
  73         while (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
  74                 if (errno == ENOMEM) {
  75                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
  76                                 zcmd_free_nvlists(&zc);
  77                                 return (-1);
  78                         }
  79                 } else {
  80                         zcmd_free_nvlists(&zc);
  81                         return (-1);
  82                 }
  83         }
  84
  85         if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
  86                 zcmd_free_nvlists(&zc);
  87                 return (-1);
  88         }
  89
  90         zcmd_free_nvlists(&zc);
  91
  92         return (0);
  93 }
  94
  95 static int
  96 zpool_props_refresh(zpool_handle_t *zhp)
  97 {
  98         nvlist_t *old_props;
  99
 100         old_props = zhp->zpool_props;
 101
 102         if (zpool_get_all_props(zhp) != 0)
 103                 return (-1);
 104
 105         nvlist_free(old_props);
 106         return (0);
 107 }
 108
 109 static char *
 110 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
 111     zprop_source_t *src)
 112 {
 113         nvlist_t *nv, *nvl;
 114         uint64_t ival;
 115         char *value;
 116         zprop_source_t source;
 117
 118         nvl = zhp->zpool_props;
 119         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 120                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &ival) == 0);
 121                 source = ival;
 122                 verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
 123         } else {
 124                 source = ZPROP_SRC_DEFAULT;
 125                 if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
 126                         value = "-";
 127         }
 128
 129         if (src)
 130                 *src = source;
 131
 132         return (value);
 133 }
 134
 135 uint64_t
 136 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
 137 {
 138         nvlist_t *nv, *nvl;
 139         uint64_t value;
 140         zprop_source_t source;
 141
 142         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
 143                 /*
 144                  * zpool_get_all_props() has most likely failed because
 145                  * the pool is faulted, but if all we need is the top level
 146                  * vdev's guid then get it from the zhp config nvlist.
 147                  */
 148                 if ((prop == ZPOOL_PROP_GUID) &&
 149                     (nvlist_lookup_nvlist(zhp->zpool_config,
 150                     ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
 151                     (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
 152                     == 0)) {
 153                         return (value);
 154                 }
 155                 return (zpool_prop_default_numeric(prop));
 156         }
 157
 158         nvl = zhp->zpool_props;
 159         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 160                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &value) == 0);
 161                 source = value;
 162                 verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
 163         } else {
 164                 source = ZPROP_SRC_DEFAULT;
 165                 value = zpool_prop_default_numeric(prop);
 166         }
 167
 168         if (src)
 169                 *src = source;
 170
 171         return (value);
 172 }
 173
 174 /*
 175  * Map VDEV STATE to printed strings.
 176  */
 177 char *
 178 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
 179 {
 180         switch (state) {
 181         default:
 182                 break;
 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 if (aux == VDEV_AUX_SPLIT_POOL)
 192                         return (gettext("SPLIT"));
 193                 else
 194                         return (gettext("UNAVAIL"));
 195         case VDEV_STATE_FAULTED:
 196                 return (gettext("FAULTED"));
 197         case VDEV_STATE_DEGRADED:
 198                 return (gettext("DEGRADED"));
 199         case VDEV_STATE_HEALTHY:
 200                 return (gettext("ONLINE"));
 201         }
 202
 203         return (gettext("UNKNOWN"));
 204 }
 205
 206 /*
 207  * Get a zpool property value for 'prop' and return the value in
 208  * a pre-allocated buffer.
 209  */
 210 int
 211 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf, size_t len,
 212     zprop_source_t *srctype)
 213 {
 214         uint64_t intval;
 215         const char *strval;
 216         zprop_source_t src = ZPROP_SRC_NONE;
 217         nvlist_t *nvroot;
 218         vdev_stat_t *vs;
 219         uint_t vsc;
 220
 221         if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
 222                 switch (prop) {
 223                 case ZPOOL_PROP_NAME:
 224                         (void) strlcpy(buf, zpool_get_name(zhp), len);
 225                         break;
 226
 227                 case ZPOOL_PROP_HEALTH:
 228                         (void) strlcpy(buf, "FAULTED", len);
 229                         break;
 230
 231                 case ZPOOL_PROP_GUID:
 232                         intval = zpool_get_prop_int(zhp, prop, &src);
 233                         (void) snprintf(buf, len, "%llu", (u_longlong_t)intval);
 234                         break;
 235
 236                 case ZPOOL_PROP_ALTROOT:
 237                 case ZPOOL_PROP_CACHEFILE:
 238                 case ZPOOL_PROP_COMMENT:
 239                         if (zhp->zpool_props != NULL ||
 240                             zpool_get_all_props(zhp) == 0) {
 241                                 (void) strlcpy(buf,
 242                                     zpool_get_prop_string(zhp, prop, &src),
 243                                     len);
 244                                 if (srctype != NULL)
 245                                         *srctype = src;
 246                                 return (0);
 247                         }
 248                         /* FALLTHROUGH */
 249                 default:
 250                         (void) strlcpy(buf, "-", len);
 251                         break;
 252                 }
 253
 254                 if (srctype != NULL)
 255                         *srctype = src;
 256                 return (0);
 257         }
 258
 259         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
 260             prop != ZPOOL_PROP_NAME)
 261                 return (-1);
 262
 263         switch (zpool_prop_get_type(prop)) {
 264         case PROP_TYPE_STRING:
 265                 (void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
 266                     len);
 267                 break;
 268
 269         case PROP_TYPE_NUMBER:
 270                 intval = zpool_get_prop_int(zhp, prop, &src);
 271
 272                 switch (prop) {
 273                 case ZPOOL_PROP_SIZE:
 274                 case ZPOOL_PROP_ALLOCATED:
 275                 case ZPOOL_PROP_FREE:
 276                 case ZPOOL_PROP_EXPANDSZ:
 277                 case ZPOOL_PROP_ASHIFT:
 278                         (void) zfs_nicenum(intval, buf, len);
 279                         break;
 280
 281                 case ZPOOL_PROP_CAPACITY:
 282                         (void) snprintf(buf, len, "%llu%%",
 283                             (u_longlong_t)intval);
 284                         break;
 285
 286                 case ZPOOL_PROP_DEDUPRATIO:
 287                         (void) snprintf(buf, len, "%llu.%02llux",
 288                             (u_longlong_t)(intval / 100),
 289                             (u_longlong_t)(intval % 100));
 290                         break;
 291
 292                 case ZPOOL_PROP_HEALTH:
 293                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 294                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 295                         verify(nvlist_lookup_uint64_array(nvroot,
 296                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
 297                             == 0);
 298
 299                         (void) strlcpy(buf, zpool_state_to_name(intval,
 300                             vs->vs_aux), len);
 301                         break;
 302                 default:
 303                         (void) snprintf(buf, len, "%llu", (u_longlong_t)intval);
 304                 }
 305                 break;
 306
 307         case PROP_TYPE_INDEX:
 308                 intval = zpool_get_prop_int(zhp, prop, &src);
 309                 if (zpool_prop_index_to_string(prop, intval, &strval)
 310                     != 0)
 311                         return (-1);
 312                 (void) strlcpy(buf, strval, len);
 313                 break;
 314
 315         default:
 316                 abort();
 317         }
 318
 319         if (srctype)
 320                 *srctype = src;
 321
 322         return (0);
 323 }
 324
 325 /*
 326  * Check if the bootfs name has the same pool name as it is set to.
 327  * Assuming bootfs is a valid dataset name.
 328  */
 329 static boolean_t
 330 bootfs_name_valid(const char *pool, char *bootfs)
 331 {
 332         int len = strlen(pool);
 333
 334         if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
 335                 return (B_FALSE);
 336
 337         if (strncmp(pool, bootfs, len) == 0 &&
 338             (bootfs[len] == '/' || bootfs[len] == '\0'))
 339                 return (B_TRUE);
 340
 341         return (B_FALSE);
 342 }
 343
 344 /*
 345  * Inspect the configuration to determine if any of the devices contain
 346  * an EFI label.
 347  */
 348 static boolean_t
 349 pool_uses_efi(nvlist_t *config)
 350 {
 351         nvlist_t **child;
 352         uint_t c, children;
 353
 354         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
 355             &child, &children) != 0)
 356                 return (read_efi_label(config, NULL) >= 0);
 357
 358         for (c = 0; c < children; c++) {
 359                 if (pool_uses_efi(child[c]))
 360                         return (B_TRUE);
 361         }
 362         return (B_FALSE);
 363 }
 364
 365 boolean_t
 366 zpool_is_bootable(zpool_handle_t *zhp)
 367 {
 368         char bootfs[ZPOOL_MAXNAMELEN];
 369
 370         return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
 371             sizeof (bootfs), NULL) == 0 && strncmp(bootfs, "-",
 372             sizeof (bootfs)) != 0);
 373 }
 374
 375
 376 /*
 377  * Given an nvlist of zpool properties to be set, validate that they are
 378  * correct, and parse any numeric properties (index, boolean, etc) if they are
 379  * specified as strings.
 380  */
 381 static nvlist_t *
 382 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 383     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
 384 {
 385         nvpair_t *elem;
 386         nvlist_t *retprops;
 387         zpool_prop_t prop;
 388         char *strval;
 389         uint64_t intval;
 390         char *slash, *check;
 391         struct stat64 statbuf;
 392         zpool_handle_t *zhp;
 393         nvlist_t *nvroot;
 394
 395         if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
 396                 (void) no_memory(hdl);
 397                 return (NULL);
 398         }
 399
 400         elem = NULL;
 401         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 402                 const char *propname = nvpair_name(elem);
 403
 404                 /*
 405                  * Make sure this property is valid and applies to this type.
 406                  */
 407                 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL) {
 408                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 409                             "invalid property '%s'"), propname);
 410                         (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 411                         goto error;
 412                 }
 413
 414                 if (zpool_prop_readonly(prop)) {
 415                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 416                             "is readonly"), propname);
 417                         (void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
 418                         goto error;
 419                 }
 420
 421                 if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
 422                     &strval, &intval, errbuf) != 0)
 423                         goto error;
 424
 425                 /*
 426                  * Perform additional checking for specific properties.
 427                  */
 428                 switch (prop) {
 429                 default:
 430                         break;
 431                 case ZPOOL_PROP_VERSION:
 432                         if (intval < version || intval > SPA_VERSION) {
 433                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 434                                     "property '%s' number %d is invalid."),
 435                                     propname, intval);
 436                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 437                                 goto error;
 438                         }
 439                         break;
 440
 441                 case ZPOOL_PROP_ASHIFT:
 442                         if (!flags.create) {
 443                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 444                                     "property '%s' can only be set at "
 445                                     "creation time"), propname);
 446                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 447                                 goto error;
 448                         }
 449
 450                         if (intval != 0 && (intval < 9 || intval > 13)) {
 451                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 452                                     "property '%s' number %d is invalid."),
 453                                     propname, intval);
 454                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 455                                 goto error;
 456                         }
 457                         break;
 458
 459                 case ZPOOL_PROP_BOOTFS:
 460                         if (flags.create || flags.import) {
 461                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 462                                     "property '%s' cannot be set at creation "
 463                                     "or import time"), propname);
 464                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 465                                 goto error;
 466                         }
 467
 468                         if (version < SPA_VERSION_BOOTFS) {
 469                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 470                                     "pool must be upgraded to support "
 471                                     "'%s' property"), propname);
 472                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 473                                 goto error;
 474                         }
 475
 476                         /*
 477                          * bootfs property value has to be a dataset name and
 478                          * the dataset has to be in the same pool as it sets to.
 479                          */
 480                         if (strval[0] != '\0' && !bootfs_name_valid(poolname,
 481                             strval)) {
 482                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 483                                     "is an invalid name"), strval);
 484                                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
 485                                 goto error;
 486                         }
 487
 488                         if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
 489                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 490                                     "could not open pool '%s'"), poolname);
 491                                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
 492                                 goto error;
 493                         }
 494                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 495                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 496
 497 #if defined(__sun__) || defined(__sun)
 498                         /*
 499                          * bootfs property cannot be set on a disk which has
 500                          * been EFI labeled.
 501                          */
 502                         if (pool_uses_efi(nvroot)) {
 503                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 504                                     "property '%s' not supported on "
 505                                     "EFI labeled devices"), propname);
 506                                 (void) zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf);
 507                                 zpool_close(zhp);
 508                                 goto error;
 509                         }
 510 #endif
 511                         zpool_close(zhp);
 512                         break;
 513
 514                 case ZPOOL_PROP_ALTROOT:
 515                         if (!flags.create && !flags.import) {
 516                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 517                                     "property '%s' can only be set during pool "
 518                                     "creation or import"), propname);
 519                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 520                                 goto error;
 521                         }
 522
 523                         if (strval[0] != '/') {
 524                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 525                                     "bad alternate root '%s'"), strval);
 526                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 527                                 goto error;
 528                         }
 529                         break;
 530
 531                 case ZPOOL_PROP_CACHEFILE:
 532                         if (strval[0] == '\0')
 533                                 break;
 534
 535                         if (strcmp(strval, "none") == 0)
 536                                 break;
 537
 538                         if (strval[0] != '/') {
 539                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 540                                     "property '%s' must be empty, an "
 541                                     "absolute path, or 'none'"), propname);
 542                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 543                                 goto error;
 544                         }
 545
 546                         slash = strrchr(strval, '/');
 547
 548                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 549                             strcmp(slash, "/..") == 0) {
 550                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 551                                     "'%s' is not a valid file"), strval);
 552                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 553                                 goto error;
 554                         }
 555
 556                         *slash = '\0';
 557
 558                         if (strval[0] != '\0' &&
 559                             (stat64(strval, &statbuf) != 0 ||
 560                             !S_ISDIR(statbuf.st_mode))) {
 561                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 562                                     "'%s' is not a valid directory"),
 563                                     strval);
 564                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 565                                 goto error;
 566                         }
 567
 568                         *slash = '/';
 569                         break;
 570
 571                 case ZPOOL_PROP_COMMENT:
 572                         for (check = strval; *check != '\0'; check++) {
 573                                 if (!isprint(*check)) {
 574                                         zfs_error_aux(hdl,
 575                                             dgettext(TEXT_DOMAIN,
 576                                             "comment may only have printable "
 577                                             "characters"));
 578                                         (void) zfs_error(hdl, EZFS_BADPROP,
 579                                             errbuf);
 580                                         goto error;
 581                                 }
 582                         }
 583                         if (strlen(strval) > ZPROP_MAX_COMMENT) {
 584                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 585                                     "comment must not exceed %d characters"),
 586                                     ZPROP_MAX_COMMENT);
 587                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 588                                 goto error;
 589                         }
 590                         break;
 591                 case ZPOOL_PROP_READONLY:
 592                         if (!flags.import) {
 593                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 594                                     "property '%s' can only be set at "
 595                                     "import time"), propname);
 596                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 597                                 goto error;
 598                         }
 599                         break;
 600                 }
 601         }
 602
 603         return (retprops);
 604 error:
 605         nvlist_free(retprops);
 606         return (NULL);
 607 }
 608
 609 /*
 610  * Set zpool property : propname=propval.
 611  */
 612 int
 613 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
 614 {
 615         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
 616         int ret = -1;
 617         char errbuf[1024];
 618         nvlist_t *nvl = NULL;
 619         nvlist_t *realprops;
 620         uint64_t version;
 621         prop_flags_t flags = { 0 };
 622
 623         (void) snprintf(errbuf, sizeof (errbuf),
 624             dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
 625             zhp->zpool_name);
 626
 627         if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
 628                 return (no_memory(zhp->zpool_hdl));
 629
 630         if (nvlist_add_string(nvl, propname, propval) != 0) {
 631                 nvlist_free(nvl);
 632                 return (no_memory(zhp->zpool_hdl));
 633         }
 634
 635         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
 636         if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
 637             zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
 638                 nvlist_free(nvl);
 639                 return (-1);
 640         }
 641
 642         nvlist_free(nvl);
 643         nvl = realprops;
 644
 645         /*
 646          * Execute the corresponding ioctl() to set this property.
 647          */
 648         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 649
 650         if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
 651                 nvlist_free(nvl);
 652                 return (-1);
 653         }
 654
 655         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
 656
 657         zcmd_free_nvlists(&zc);
 658         nvlist_free(nvl);
 659
 660         if (ret)
 661                 (void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
 662         else
 663                 (void) zpool_props_refresh(zhp);
 664
 665         return (ret);
 666 }
 667
 668 int
 669 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
 670 {
 671         libzfs_handle_t *hdl = zhp->zpool_hdl;
 672         zprop_list_t *entry;
 673         char buf[ZFS_MAXPROPLEN];
 674
 675         if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
 676                 return (-1);
 677
 678         for (entry = *plp; entry != NULL; entry = entry->pl_next) {
 679
 680                 if (entry->pl_fixed)
 681                         continue;
 682
 683                 if (entry->pl_prop != ZPROP_INVAL &&
 684                     zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
 685                     NULL) == 0) {
 686                         if (strlen(buf) > entry->pl_width)
 687                                 entry->pl_width = strlen(buf);
 688                 }
 689         }
 690
 691         return (0);
 692 }
 693
 694
 695 /*
 696  * Don't start the slice at the default block of 34; many storage
 697  * devices will use a stripe width of 128k, other vendors prefer a 1m
 698  * alignment.  It is best to play it safe and ensure a 1m alignment
 699  * given 512B blocks.  When the block size is larger by a power of 2
 700  * we will still be 1m aligned.  Some devices are sensitive to the
 701  * partition ending alignment as well.
 702  */
 703 #define NEW_START_BLOCK         2048
 704 #define PARTITION_END_ALIGNMENT 2048
 705
 706 /*
 707  * Validate the given pool name, optionally putting an extended error message in
 708  * 'buf'.
 709  */
 710 boolean_t
 711 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
 712 {
 713         namecheck_err_t why;
 714         char what;
 715         int ret;
 716
 717         ret = pool_namecheck(pool, &why, &what);
 718
 719         /*
 720          * The rules for reserved pool names were extended at a later point.
 721          * But we need to support users with existing pools that may now be
 722          * invalid.  So we only check for this expanded set of names during a
 723          * create (or import), and only in userland.
 724          */
 725         if (ret == 0 && !isopen &&
 726             (strncmp(pool, "mirror", 6) == 0 ||
 727             strncmp(pool, "raidz", 5) == 0 ||
 728             strncmp(pool, "spare", 5) == 0 ||
 729             strcmp(pool, "log") == 0)) {
 730                 if (hdl != NULL)
 731                         zfs_error_aux(hdl,
 732                             dgettext(TEXT_DOMAIN, "name is reserved"));
 733                 return (B_FALSE);
 734         }
 735
 736
 737         if (ret != 0) {
 738                 if (hdl != NULL) {
 739                         switch (why) {
 740                         case NAME_ERR_TOOLONG:
 741                                 zfs_error_aux(hdl,
 742                                     dgettext(TEXT_DOMAIN, "name is too long"));
 743                                 break;
 744
 745                         case NAME_ERR_INVALCHAR:
 746                                 zfs_error_aux(hdl,
 747                                     dgettext(TEXT_DOMAIN, "invalid character "
 748                                     "'%c' in pool name"), what);
 749                                 break;
 750
 751                         case NAME_ERR_NOLETTER:
 752                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 753                                     "name must begin with a letter"));
 754                                 break;
 755
 756                         case NAME_ERR_RESERVED:
 757                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 758                                     "name is reserved"));
 759                                 break;
 760
 761                         case NAME_ERR_DISKLIKE:
 762                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 763                                     "pool name is reserved"));
 764                                 break;
 765
 766                         case NAME_ERR_LEADING_SLASH:
 767                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 768                                     "leading slash in name"));
 769                                 break;
 770
 771                         case NAME_ERR_EMPTY_COMPONENT:
 772                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 773                                     "empty component in name"));
 774                                 break;
 775
 776                         case NAME_ERR_TRAILING_SLASH:
 777                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 778                                     "trailing slash in name"));
 779                                 break;
 780
 781                         case NAME_ERR_MULTIPLE_AT:
 782                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 783                                     "multiple '@' delimiters in name"));
 784                                 break;
 785                         case NAME_ERR_NO_AT:
 786                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 787                                     "permission set is missing '@'"));
 788                                 break;
 789                         }
 790                 }
 791                 return (B_FALSE);
 792         }
 793
 794         return (B_TRUE);
 795 }
 796
 797 /*
 798  * Open a handle to the given pool, even if the pool is currently in the FAULTED
 799  * state.
 800  */
 801 zpool_handle_t *
 802 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
 803 {
 804         zpool_handle_t *zhp;
 805         boolean_t missing;
 806
 807         /*
 808          * Make sure the pool name is valid.
 809          */
 810         if (!zpool_name_valid(hdl, B_TRUE, pool)) {
 811                 (void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
 812                     dgettext(TEXT_DOMAIN, "cannot open '%s'"),
 813                     pool);
 814                 return (NULL);
 815         }
 816
 817         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
 818                 return (NULL);
 819
 820         zhp->zpool_hdl = hdl;
 821         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
 822
 823         if (zpool_refresh_stats(zhp, &missing) != 0) {
 824                 zpool_close(zhp);
 825                 return (NULL);
 826         }
 827
 828         if (missing) {
 829                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
 830                 (void) zfs_error_fmt(hdl, EZFS_NOENT,
 831                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
 832                 zpool_close(zhp);
 833                 return (NULL);
 834         }
 835
 836         return (zhp);
 837 }
 838
 839 /*
 840  * Like the above, but silent on error.  Used when iterating over pools (because
 841  * the configuration cache may be out of date).
 842  */
 843 int
 844 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
 845 {
 846         zpool_handle_t *zhp;
 847         boolean_t missing;
 848
 849         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
 850                 return (-1);
 851
 852         zhp->zpool_hdl = hdl;
 853         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
 854
 855         if (zpool_refresh_stats(zhp, &missing) != 0) {
 856                 zpool_close(zhp);
 857                 return (-1);
 858         }
 859
 860         if (missing) {
 861                 zpool_close(zhp);
 862                 *ret = NULL;
 863                 return (0);
 864         }
 865
 866         *ret = zhp;
 867         return (0);
 868 }
 869
 870 /*
 871  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
 872  * state.
 873  */
 874 zpool_handle_t *
 875 zpool_open(libzfs_handle_t *hdl, const char *pool)
 876 {
 877         zpool_handle_t *zhp;
 878
 879         if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
 880                 return (NULL);
 881
 882         if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
 883                 (void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
 884                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
 885                 zpool_close(zhp);
 886                 return (NULL);
 887         }
 888
 889         return (zhp);
 890 }
 891
 892 /*
 893  * Close the handle.  Simply frees the memory associated with the handle.
 894  */
 895 void
 896 zpool_close(zpool_handle_t *zhp)
 897 {
 898         if (zhp->zpool_config)
 899                 nvlist_free(zhp->zpool_config);
 900         if (zhp->zpool_old_config)
 901                 nvlist_free(zhp->zpool_old_config);
 902         if (zhp->zpool_props)
 903                 nvlist_free(zhp->zpool_props);
 904         free(zhp);
 905 }
 906
 907 /*
 908  * Return the name of the pool.
 909  */
 910 const char *
 911 zpool_get_name(zpool_handle_t *zhp)
 912 {
 913         return (zhp->zpool_name);
 914 }
 915
 916
 917 /*
 918  * Return the state of the pool (ACTIVE or UNAVAILABLE)
 919  */
 920 int
 921 zpool_get_state(zpool_handle_t *zhp)
 922 {
 923         return (zhp->zpool_state);
 924 }
 925
 926 /*
 927  * Create the named pool, using the provided vdev list.  It is assumed
 928  * that the consumer has already validated the contents of the nvlist, so we
 929  * don't have to worry about error semantics.
 930  */
 931 int
 932 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
 933     nvlist_t *props, nvlist_t *fsprops)
 934 {
 935         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
 936         nvlist_t *zc_fsprops = NULL;
 937         nvlist_t *zc_props = NULL;
 938         char msg[1024];
 939         char *altroot;
 940         int ret = -1;
 941
 942         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
 943             "cannot create '%s'"), pool);
 944
 945         if (!zpool_name_valid(hdl, B_FALSE, pool))
 946                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
 947
 948         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
 949                 return (-1);
 950
 951         if (props) {
 952                 prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
 953
 954                 if ((zc_props = zpool_valid_proplist(hdl, pool, props,
 955                     SPA_VERSION_1, flags, msg)) == NULL) {
 956                         goto create_failed;
 957                 }
 958         }
 959
 960         if (fsprops) {
 961                 uint64_t zoned;
 962                 char *zonestr;
 963
 964                 zoned = ((nvlist_lookup_string(fsprops,
 965                     zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
 966                     strcmp(zonestr, "on") == 0);
 967
 968                 if ((zc_fsprops = zfs_valid_proplist(hdl,
 969                     ZFS_TYPE_FILESYSTEM, fsprops, zoned, NULL, msg)) == NULL) {
 970                         goto create_failed;
 971                 }
 972                 if (!zc_props &&
 973                     (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
 974                         goto create_failed;
 975                 }
 976                 if (nvlist_add_nvlist(zc_props,
 977                     ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
 978                         goto create_failed;
 979                 }
 980         }
 981
 982         if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
 983                 goto create_failed;
 984
 985         (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
 986
 987         if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
 988
 989                 zcmd_free_nvlists(&zc);
 990                 nvlist_free(zc_props);
 991                 nvlist_free(zc_fsprops);
 992
 993                 switch (errno) {
 994                 case EBUSY:
 995                         /*
 996                          * This can happen if the user has specified the same
 997                          * device multiple times.  We can't reliably detect this
 998                          * until we try to add it and see we already have a
 999                          * label.  This can also happen under if the device is
1000                          * part of an active md or lvm device.
1001                          */
1002                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1003                             "one or more vdevs refer to the same device, or one of\n"
1004                             "the devices is part of an active md or lvm device"));
1005                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1006
1007                 case EOVERFLOW:
1008                         /*
1009                          * This occurs when one of the devices is below
1010                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1011                          * device was the problem device since there's no
1012                          * reliable way to determine device size from userland.
1013                          */
1014                         {
1015                                 char buf[64];
1016
1017                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1018
1019                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1020                                     "one or more devices is less than the "
1021                                     "minimum size (%s)"), buf);
1022                         }
1023                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1024
1025                 case ENOSPC:
1026                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1027                             "one or more devices is out of space"));
1028                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1029
1030                 case ENOTBLK:
1031                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1032                             "cache device must be a disk or disk slice"));
1033                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1034
1035                 default:
1036                         return (zpool_standard_error(hdl, errno, msg));
1037                 }
1038         }
1039
1040         /*
1041          * If this is an alternate root pool, then we automatically set the
1042          * mountpoint of the root dataset to be '/'.
1043          */
1044         if (nvlist_lookup_string(props, zpool_prop_to_name(ZPOOL_PROP_ALTROOT),
1045             &altroot) == 0) {
1046                 zfs_handle_t *zhp;
1047
1048                 verify((zhp = zfs_open(hdl, pool, ZFS_TYPE_DATASET)) != NULL);
1049                 verify(zfs_prop_set(zhp, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT),
1050                     "/") == 0);
1051
1052                 zfs_close(zhp);
1053         }
1054
1055 create_failed:
1056         zcmd_free_nvlists(&zc);
1057         nvlist_free(zc_props);
1058         nvlist_free(zc_fsprops);
1059         return (ret);
1060 }
1061
1062 /*
1063  * Destroy the given pool.  It is up to the caller to ensure that there are no
1064  * datasets left in the pool.
1065  */
1066 int
1067 zpool_destroy(zpool_handle_t *zhp)
1068 {
1069         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1070         zfs_handle_t *zfp = NULL;
1071         libzfs_handle_t *hdl = zhp->zpool_hdl;
1072         char msg[1024];
1073
1074         if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1075             (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1076                 return (-1);
1077
1078         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1079
1080         if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1081                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1082                     "cannot destroy '%s'"), zhp->zpool_name);
1083
1084                 if (errno == EROFS) {
1085                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1086                             "one or more devices is read only"));
1087                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1088                 } else {
1089                         (void) zpool_standard_error(hdl, errno, msg);
1090                 }
1091
1092                 if (zfp)
1093                         zfs_close(zfp);
1094                 return (-1);
1095         }
1096
1097         if (zfp) {
1098                 remove_mountpoint(zfp);
1099                 zfs_close(zfp);
1100         }
1101
1102         return (0);
1103 }
1104
1105 /*
1106  * Add the given vdevs to the pool.  The caller must have already performed the
1107  * necessary verification to ensure that the vdev specification is well-formed.
1108  */
1109 int
1110 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1111 {
1112         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1113         int ret;
1114         libzfs_handle_t *hdl = zhp->zpool_hdl;
1115         char msg[1024];
1116         nvlist_t **spares, **l2cache;
1117         uint_t nspares, nl2cache;
1118
1119         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1120             "cannot add to '%s'"), zhp->zpool_name);
1121
1122         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1123             SPA_VERSION_SPARES &&
1124             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1125             &spares, &nspares) == 0) {
1126                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1127                     "upgraded to add hot spares"));
1128                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1129         }
1130
1131         if (zpool_is_bootable(zhp) && nvlist_lookup_nvlist_array(nvroot,
1132             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) {
1133                 uint64_t s;
1134
1135                 for (s = 0; s < nspares; s++) {
1136                         char *path;
1137
1138                         if (nvlist_lookup_string(spares[s], ZPOOL_CONFIG_PATH,
1139                             &path) == 0 && pool_uses_efi(spares[s])) {
1140                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1141                                     "device '%s' contains an EFI label and "
1142                                     "cannot be used on root pools."),
1143                                     zpool_vdev_name(hdl, NULL, spares[s],
1144                                     B_FALSE));
1145                                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
1146                         }
1147                 }
1148         }
1149
1150         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1151             SPA_VERSION_L2CACHE &&
1152             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1153             &l2cache, &nl2cache) == 0) {
1154                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1155                     "upgraded to add cache devices"));
1156                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1157         }
1158
1159         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1160                 return (-1);
1161         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1162
1163         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1164                 switch (errno) {
1165                 case EBUSY:
1166                         /*
1167                          * This can happen if the user has specified the same
1168                          * device multiple times.  We can't reliably detect this
1169                          * until we try to add it and see we already have a
1170                          * label.
1171                          */
1172                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1173                             "one or more vdevs refer to the same device"));
1174                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1175                         break;
1176
1177                 case EOVERFLOW:
1178                         /*
1179                          * This occurrs when one of the devices is below
1180                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1181                          * device was the problem device since there's no
1182                          * reliable way to determine device size from userland.
1183                          */
1184                         {
1185                                 char buf[64];
1186
1187                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1188
1189                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1190                                     "device is less than the minimum "
1191                                     "size (%s)"), buf);
1192                         }
1193                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1194                         break;
1195
1196                 case ENOTSUP:
1197                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1198                             "pool must be upgraded to add these vdevs"));
1199                         (void) zfs_error(hdl, EZFS_BADVERSION, msg);
1200                         break;
1201
1202                 case EDOM:
1203                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1204                             "root pool can not have multiple vdevs"
1205                             " or separate logs"));
1206                         (void) zfs_error(hdl, EZFS_POOL_NOTSUP, msg);
1207                         break;
1208
1209                 case ENOTBLK:
1210                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1211                             "cache device must be a disk or disk slice"));
1212                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1213                         break;
1214
1215                 default:
1216                         (void) zpool_standard_error(hdl, errno, msg);
1217                 }
1218
1219                 ret = -1;
1220         } else {
1221                 ret = 0;
1222         }
1223
1224         zcmd_free_nvlists(&zc);
1225
1226         return (ret);
1227 }
1228
1229 /*
1230  * Exports the pool from the system.  The caller must ensure that there are no
1231  * mounted datasets in the pool.
1232  */
1233 int
1234 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce)
1235 {
1236         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1237         char msg[1024];
1238
1239         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1240             "cannot export '%s'"), zhp->zpool_name);
1241
1242         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1243         zc.zc_cookie = force;
1244         zc.zc_guid = hardforce;
1245
1246         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1247                 switch (errno) {
1248                 case EXDEV:
1249                         zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1250                             "use '-f' to override the following errors:\n"
1251                             "'%s' has an active shared spare which could be"
1252                             " used by other pools once '%s' is exported."),
1253                             zhp->zpool_name, zhp->zpool_name);
1254                         return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1255                             msg));
1256                 default:
1257                         return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1258                             msg));
1259                 }
1260         }
1261
1262         return (0);
1263 }
1264
1265 int
1266 zpool_export(zpool_handle_t *zhp, boolean_t force)
1267 {
1268         return (zpool_export_common(zhp, force, B_FALSE));
1269 }
1270
1271 int
1272 zpool_export_force(zpool_handle_t *zhp)
1273 {
1274         return (zpool_export_common(zhp, B_TRUE, B_TRUE));
1275 }
1276
1277 static void
1278 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1279     nvlist_t *config)
1280 {
1281         nvlist_t *nv = NULL;
1282         uint64_t rewindto;
1283         int64_t loss = -1;
1284         struct tm t;
1285         char timestr[128];
1286
1287         if (!hdl->libzfs_printerr || config == NULL)
1288                 return;
1289
1290         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0)
1291                 return;
1292
1293         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1294                 return;
1295         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1296
1297         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1298             strftime(timestr, 128, "%c", &t) != 0) {
1299                 if (dryrun) {
1300                         (void) printf(dgettext(TEXT_DOMAIN,
1301                             "Would be able to return %s "
1302                             "to its state as of %s.\n"),
1303                             name, timestr);
1304                 } else {
1305                         (void) printf(dgettext(TEXT_DOMAIN,
1306                             "Pool %s returned to its state as of %s.\n"),
1307                             name, timestr);
1308                 }
1309                 if (loss > 120) {
1310                         (void) printf(dgettext(TEXT_DOMAIN,
1311                             "%s approximately %lld "),
1312                             dryrun ? "Would discard" : "Discarded",
1313                             ((longlong_t)loss + 30) / 60);
1314                         (void) printf(dgettext(TEXT_DOMAIN,
1315                             "minutes of transactions.\n"));
1316                 } else if (loss > 0) {
1317                         (void) printf(dgettext(TEXT_DOMAIN,
1318                             "%s approximately %lld "),
1319                             dryrun ? "Would discard" : "Discarded",
1320                             (longlong_t)loss);
1321                         (void) printf(dgettext(TEXT_DOMAIN,
1322                             "seconds of transactions.\n"));
1323                 }
1324         }
1325 }
1326
1327 void
1328 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1329     nvlist_t *config)
1330 {
1331         nvlist_t *nv = NULL;
1332         int64_t loss = -1;
1333         uint64_t edata = UINT64_MAX;
1334         uint64_t rewindto;
1335         struct tm t;
1336         char timestr[128];
1337
1338         if (!hdl->libzfs_printerr)
1339                 return;
1340
1341         if (reason >= 0)
1342                 (void) printf(dgettext(TEXT_DOMAIN, "action: "));
1343         else
1344                 (void) printf(dgettext(TEXT_DOMAIN, "\t"));
1345
1346         /* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1347         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1348             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1349                 goto no_info;
1350
1351         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1352         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1353             &edata);
1354
1355         (void) printf(dgettext(TEXT_DOMAIN,
1356             "Recovery is possible, but will result in some data loss.\n"));
1357
1358         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1359             strftime(timestr, 128, "%c", &t) != 0) {
1360                 (void) printf(dgettext(TEXT_DOMAIN,
1361                     "\tReturning the pool to its state as of %s\n"
1362                     "\tshould correct the problem.  "),
1363                     timestr);
1364         } else {
1365                 (void) printf(dgettext(TEXT_DOMAIN,
1366                     "\tReverting the pool to an earlier state "
1367                     "should correct the problem.\n\t"));
1368         }
1369
1370         if (loss > 120) {
1371                 (void) printf(dgettext(TEXT_DOMAIN,
1372                     "Approximately %lld minutes of data\n"
1373                     "\tmust be discarded, irreversibly.  "),
1374                     ((longlong_t)loss + 30) / 60);
1375         } else if (loss > 0) {
1376                 (void) printf(dgettext(TEXT_DOMAIN,
1377                     "Approximately %lld seconds of data\n"
1378                     "\tmust be discarded, irreversibly.  "),
1379                     (longlong_t)loss);
1380         }
1381         if (edata != 0 && edata != UINT64_MAX) {
1382                 if (edata == 1) {
1383                         (void) printf(dgettext(TEXT_DOMAIN,
1384                             "After rewind, at least\n"
1385                             "\tone persistent user-data error will remain.  "));
1386                 } else {
1387                         (void) printf(dgettext(TEXT_DOMAIN,
1388                             "After rewind, several\n"
1389                             "\tpersistent user-data errors will remain.  "));
1390                 }
1391         }
1392         (void) printf(dgettext(TEXT_DOMAIN,
1393             "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1394             reason >= 0 ? "clear" : "import", name);
1395
1396         (void) printf(dgettext(TEXT_DOMAIN,
1397             "A scrub of the pool\n"
1398             "\tis strongly recommended after recovery.\n"));
1399         return;
1400
1401 no_info:
1402         (void) printf(dgettext(TEXT_DOMAIN,
1403             "Destroy and re-create the pool from\n\ta backup source.\n"));
1404 }
1405
1406 /*
1407  * zpool_import() is a contracted interface. Should be kept the same
1408  * if possible.
1409  *
1410  * Applications should use zpool_import_props() to import a pool with
1411  * new properties value to be set.
1412  */
1413 int
1414 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1415     char *altroot)
1416 {
1417         nvlist_t *props = NULL;
1418         int ret;
1419
1420         if (altroot != NULL) {
1421                 if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1422                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1423                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1424                             newname));
1425                 }
1426
1427                 if (nvlist_add_string(props,
1428                     zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1429                     nvlist_add_string(props,
1430                     zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1431                         nvlist_free(props);
1432                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1433                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1434                             newname));
1435                 }
1436         }
1437
1438         ret = zpool_import_props(hdl, config, newname, props,
1439             ZFS_IMPORT_NORMAL);
1440         if (props)
1441                 nvlist_free(props);
1442         return (ret);
1443 }
1444
1445 static void
1446 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1447     int indent)
1448 {
1449         nvlist_t **child;
1450         uint_t c, children;
1451         char *vname;
1452         uint64_t is_log = 0;
1453
1454         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1455             &is_log);
1456
1457         if (name != NULL)
1458                 (void) printf("\t%*s%s%s\n", indent, "", name,
1459                     is_log ? " [log]" : "");
1460
1461         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1462             &child, &children) != 0)
1463                 return;
1464
1465         for (c = 0; c < children; c++) {
1466                 vname = zpool_vdev_name(hdl, NULL, child[c], B_TRUE);
1467                 print_vdev_tree(hdl, vname, child[c], indent + 2);
1468                 free(vname);
1469         }
1470 }
1471
1472 /*
1473  * Import the given pool using the known configuration and a list of
1474  * properties to be set. The configuration should have come from
1475  * zpool_find_import(). The 'newname' parameters control whether the pool
1476  * is imported with a different name.
1477  */
1478 int
1479 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1480     nvlist_t *props, int flags)
1481 {
1482         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1483         zpool_rewind_policy_t policy;
1484         nvlist_t *nv = NULL;
1485         nvlist_t *nvinfo = NULL;
1486         nvlist_t *missing = NULL;
1487         char *thename;
1488         char *origname;
1489         int ret;
1490         int error = 0;
1491         char errbuf[1024];
1492
1493         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1494             &origname) == 0);
1495
1496         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1497             "cannot import pool '%s'"), origname);
1498
1499         if (newname != NULL) {
1500                 if (!zpool_name_valid(hdl, B_FALSE, newname))
1501                         return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1502                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1503                             newname));
1504                 thename = (char *)newname;
1505         } else {
1506                 thename = origname;
1507         }
1508
1509         if (props) {
1510                 uint64_t version;
1511                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1512
1513                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1514                     &version) == 0);
1515
1516                 if ((props = zpool_valid_proplist(hdl, origname,
1517                     props, version, flags, errbuf)) == NULL) {
1518                         return (-1);
1519                 } else if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1520                         nvlist_free(props);
1521                         return (-1);
1522                 }
1523         }
1524
1525         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1526
1527         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1528             &zc.zc_guid) == 0);
1529
1530         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1531                 nvlist_free(props);
1532                 return (-1);
1533         }
1534         if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1535                 nvlist_free(props);
1536                 return (-1);
1537         }
1538
1539         zc.zc_cookie = flags;
1540         while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1541             errno == ENOMEM) {
1542                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1543                         zcmd_free_nvlists(&zc);
1544                         return (-1);
1545                 }
1546         }
1547         if (ret != 0)
1548                 error = errno;
1549
1550         (void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1551         zpool_get_rewind_policy(config, &policy);
1552
1553         if (error) {
1554                 char desc[1024];
1555
1556                 /*
1557                  * Dry-run failed, but we print out what success
1558                  * looks like if we found a best txg
1559                  */
1560                 if (policy.zrp_request & ZPOOL_TRY_REWIND) {
1561                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1562                             B_TRUE, nv);
1563                         nvlist_free(nv);
1564                         return (-1);
1565                 }
1566
1567                 if (newname == NULL)
1568                         (void) snprintf(desc, sizeof (desc),
1569                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1570                             thename);
1571                 else
1572                         (void) snprintf(desc, sizeof (desc),
1573                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1574                             origname, thename);
1575
1576                 switch (error) {
1577                 case ENOTSUP:
1578                         /*
1579                          * Unsupported version.
1580                          */
1581                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1582                         break;
1583
1584                 case EINVAL:
1585                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1586                         break;
1587
1588                 case EROFS:
1589                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1590                             "one or more devices is read only"));
1591                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1592                         break;
1593
1594                 case ENXIO:
1595                         if (nv && nvlist_lookup_nvlist(nv,
1596                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1597                             nvlist_lookup_nvlist(nvinfo,
1598                             ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1599                                 (void) printf(dgettext(TEXT_DOMAIN,
1600                                     "The devices below are missing, use "
1601                                     "'-m' to import the pool anyway:\n"));
1602                                 print_vdev_tree(hdl, NULL, missing, 2);
1603                                 (void) printf("\n");
1604                         }
1605                         (void) zpool_standard_error(hdl, error, desc);
1606                         break;
1607
1608                 case EEXIST:
1609                         (void) zpool_standard_error(hdl, error, desc);
1610                         break;
1611
1612                 case EBUSY:
1613                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1614                             "one or more devices are already in use\n"));
1615                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1616                         break;
1617
1618                 default:
1619                         (void) zpool_standard_error(hdl, error, desc);
1620                         zpool_explain_recover(hdl,
1621                             newname ? origname : thename, -error, nv);
1622                         break;
1623                 }
1624
1625                 nvlist_free(nv);
1626                 ret = -1;
1627         } else {
1628                 zpool_handle_t *zhp;
1629
1630                 /*
1631                  * This should never fail, but play it safe anyway.
1632                  */
1633                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1634                         ret = -1;
1635                 else if (zhp != NULL)
1636                         zpool_close(zhp);
1637                 if (policy.zrp_request &
1638                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1639                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1640                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1641                 }
1642                 nvlist_free(nv);
1643                 return (0);
1644         }
1645
1646         zcmd_free_nvlists(&zc);
1647         nvlist_free(props);
1648
1649         return (ret);
1650 }
1651
1652 /*
1653  * Scan the pool.
1654  */
1655 int
1656 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1657 {
1658         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
1659         char msg[1024];
1660         libzfs_handle_t *hdl = zhp->zpool_hdl;
1661
1662         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1663         zc.zc_cookie = func;
1664
1665         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1666             (errno == ENOENT && func != POOL_SCAN_NONE))
1667                 return (0);
1668
1669         if (func == POOL_SCAN_SCRUB) {
1670                 (void) snprintf(msg, sizeof (msg),
1671                     dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1672         } else if (func == POOL_SCAN_NONE) {
1673                 (void) snprintf(msg, sizeof (msg),
1674                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1675                     zc.zc_name);
1676         } else {
1677                 assert(!"unexpected result");
1678         }
1679
1680         if (errno == EBUSY) {
1681                 nvlist_t *nvroot;
1682                 pool_scan_stat_t *ps = NULL;
1683                 uint_t psc;
1684
1685                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
1686                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1687                 (void) nvlist_lookup_uint64_array(nvroot,
1688                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1689                 if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1690                         return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1691                 else
1692                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
1693         } else if (errno == ENOENT) {
1694                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1695         } else {
1696                 return (zpool_standard_error(hdl, errno, msg));
1697         }
1698 }
1699
1700 /*
1701  * Find a vdev that matches the search criteria specified. We use the
1702  * the nvpair name to determine how we should look for the device.
1703  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1704  * spare; but FALSE if its an INUSE spare.
1705  */
1706 static nvlist_t *
1707 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1708     boolean_t *l2cache, boolean_t *log)
1709 {
1710         uint_t c, children;
1711         nvlist_t **child;
1712         nvlist_t *ret;
1713         uint64_t is_log;
1714         char *srchkey;
1715         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
1716
1717         /* Nothing to look for */
1718         if (search == NULL || pair == NULL)
1719                 return (NULL);
1720
1721         /* Obtain the key we will use to search */
1722         srchkey = nvpair_name(pair);
1723
1724         switch (nvpair_type(pair)) {
1725         case DATA_TYPE_UINT64:
1726                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
1727                         uint64_t srchval, theguid;
1728
1729                         verify(nvpair_value_uint64(pair, &srchval) == 0);
1730                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
1731                             &theguid) == 0);
1732                         if (theguid == srchval)
1733                                 return (nv);
1734                 }
1735                 break;
1736
1737         case DATA_TYPE_STRING: {
1738                 char *srchval, *val;
1739
1740                 verify(nvpair_value_string(pair, &srchval) == 0);
1741                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
1742                         break;
1743
1744                 /*
1745                  * Search for the requested value. Special cases:
1746                  *
1747                  * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
1748                  *   "-part1", or "p1".  The suffix is hidden from the user,
1749                  *   but included in the string, so this matches around it.
1750                  * - ZPOOL_CONFIG_PATH for short names zfs_strcmp_shortname()
1751                  *   is used to check all possible expanded paths.
1752                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
1753                  *
1754                  * Otherwise, all other searches are simple string compares.
1755                  */
1756                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0) {
1757                         uint64_t wholedisk = 0;
1758
1759                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1760                             &wholedisk);
1761                         if (zfs_strcmp_pathname(srchval, val, wholedisk) == 0)
1762                                 return (nv);
1763
1764                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
1765                         char *type, *idx, *end, *p;
1766                         uint64_t id, vdev_id;
1767
1768                         /*
1769                          * Determine our vdev type, keeping in mind
1770                          * that the srchval is composed of a type and
1771                          * vdev id pair (i.e. mirror-4).
1772                          */
1773                         if ((type = strdup(srchval)) == NULL)
1774                                 return (NULL);
1775
1776                         if ((p = strrchr(type, '-')) == NULL) {
1777                                 free(type);
1778                                 break;
1779                         }
1780                         idx = p + 1;
1781                         *p = '\0';
1782
1783                         /*
1784                          * If the types don't match then keep looking.
1785                          */
1786                         if (strncmp(val, type, strlen(val)) != 0) {
1787                                 free(type);
1788                                 break;
1789                         }
1790
1791                         verify(strncmp(type, VDEV_TYPE_RAIDZ,
1792                             strlen(VDEV_TYPE_RAIDZ)) == 0 ||
1793                             strncmp(type, VDEV_TYPE_MIRROR,
1794                             strlen(VDEV_TYPE_MIRROR)) == 0);
1795                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
1796                             &id) == 0);
1797
1798                         errno = 0;
1799                         vdev_id = strtoull(idx, &end, 10);
1800
1801                         free(type);
1802                         if (errno != 0)
1803                                 return (NULL);
1804
1805                         /*
1806                          * Now verify that we have the correct vdev id.
1807                          */
1808                         if (vdev_id == id)
1809                                 return (nv);
1810                 }
1811
1812                 /*
1813                  * Common case
1814                  */
1815                 if (strcmp(srchval, val) == 0)
1816                         return (nv);
1817                 break;
1818         }
1819
1820         default:
1821                 break;
1822         }
1823
1824         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1825             &child, &children) != 0)
1826                 return (NULL);
1827
1828         for (c = 0; c < children; c++) {
1829                 if ((ret = vdev_to_nvlist_iter(child[c], search,
1830                     avail_spare, l2cache, NULL)) != NULL) {
1831                         /*
1832                          * The 'is_log' value is only set for the toplevel
1833                          * vdev, not the leaf vdevs.  So we always lookup the
1834                          * log device from the root of the vdev tree (where
1835                          * 'log' is non-NULL).
1836                          */
1837                         if (log != NULL &&
1838                             nvlist_lookup_uint64(child[c],
1839                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
1840                             is_log) {
1841                                 *log = B_TRUE;
1842                         }
1843                         return (ret);
1844                 }
1845         }
1846
1847         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1848             &child, &children) == 0) {
1849                 for (c = 0; c < children; c++) {
1850                         if ((ret = vdev_to_nvlist_iter(child[c], search,
1851                             avail_spare, l2cache, NULL)) != NULL) {
1852                                 *avail_spare = B_TRUE;
1853                                 return (ret);
1854                         }
1855                 }
1856         }
1857
1858         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1859             &child, &children) == 0) {
1860                 for (c = 0; c < children; c++) {
1861                         if ((ret = vdev_to_nvlist_iter(child[c], search,
1862                             avail_spare, l2cache, NULL)) != NULL) {
1863                                 *l2cache = B_TRUE;
1864                                 return (ret);
1865                         }
1866                 }
1867         }
1868
1869         return (NULL);
1870 }
1871
1872 /*
1873  * Given a physical path (minus the "/devices" prefix), find the
1874  * associated vdev.
1875  */
1876 nvlist_t *
1877 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
1878     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
1879 {
1880         nvlist_t *search, *nvroot, *ret;
1881
1882         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1883         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
1884
1885         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1886             &nvroot) == 0);
1887
1888         *avail_spare = B_FALSE;
1889         *l2cache = B_FALSE;
1890         if (log != NULL)
1891                 *log = B_FALSE;
1892         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
1893         nvlist_free(search);
1894
1895         return (ret);
1896 }
1897
1898 /*
1899  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
1900  */
1901 boolean_t
1902 zpool_vdev_is_interior(const char *name)
1903 {
1904         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
1905             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
1906                 return (B_TRUE);
1907         return (B_FALSE);
1908 }
1909
1910 nvlist_t *
1911 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
1912     boolean_t *l2cache, boolean_t *log)
1913 {
1914         char *end;
1915         nvlist_t *nvroot, *search, *ret;
1916         uint64_t guid;
1917
1918         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1919
1920         guid = strtoull(path, &end, 10);
1921         if (guid != 0 && *end == '\0') {
1922                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
1923         } else if (zpool_vdev_is_interior(path)) {
1924                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
1925         } else {
1926                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
1927         }
1928
1929         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1930             &nvroot) == 0);
1931
1932         *avail_spare = B_FALSE;
1933         *l2cache = B_FALSE;
1934         if (log != NULL)
1935                 *log = B_FALSE;
1936         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
1937         nvlist_free(search);
1938
1939         return (ret);
1940 }
1941
1942 static int
1943 vdev_online(nvlist_t *nv)
1944 {
1945         uint64_t ival;
1946
1947         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
1948             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
1949             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
1950                 return (0);
1951
1952         return (1);
1953 }
1954
1955 /*
1956  * Helper function for zpool_get_physpaths().
1957  */
1958 static int
1959 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
1960     size_t *bytes_written)
1961 {
1962         size_t bytes_left, pos, rsz;
1963         char *tmppath;
1964         const char *format;
1965
1966         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
1967             &tmppath) != 0)
1968                 return (EZFS_NODEVICE);
1969
1970         pos = *bytes_written;
1971         bytes_left = physpath_size - pos;
1972         format = (pos == 0) ? "%s" : " %s";
1973
1974         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
1975         *bytes_written += rsz;
1976
1977         if (rsz >= bytes_left) {
1978                 /* if physpath was not copied properly, clear it */
1979                 if (bytes_left != 0) {
1980                         physpath[pos] = 0;
1981                 }
1982                 return (EZFS_NOSPC);
1983         }
1984         return (0);
1985 }
1986
1987 static int
1988 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
1989     size_t *rsz, boolean_t is_spare)
1990 {
1991         char *type;
1992         int ret;
1993
1994         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
1995                 return (EZFS_INVALCONFIG);
1996
1997         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
1998                 /*
1999                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2000                  * For a spare vdev, we only want to boot from the active
2001                  * spare device.
2002                  */
2003                 if (is_spare) {
2004                         uint64_t spare = 0;
2005                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2006                             &spare);
2007                         if (!spare)
2008                                 return (EZFS_INVALCONFIG);
2009                 }
2010
2011                 if (vdev_online(nv)) {
2012                         if ((ret = vdev_get_one_physpath(nv, physpath,
2013                             phypath_size, rsz)) != 0)
2014                                 return (ret);
2015                 }
2016         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2017             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2018             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2019                 nvlist_t **child;
2020                 uint_t count;
2021                 int i, ret;
2022
2023                 if (nvlist_lookup_nvlist_array(nv,
2024                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2025                         return (EZFS_INVALCONFIG);
2026
2027                 for (i = 0; i < count; i++) {
2028                         ret = vdev_get_physpaths(child[i], physpath,
2029                             phypath_size, rsz, is_spare);
2030                         if (ret == EZFS_NOSPC)
2031                                 return (ret);
2032                 }
2033         }
2034
2035         return (EZFS_POOL_INVALARG);
2036 }
2037
2038 /*
2039  * Get phys_path for a root pool config.
2040  * Return 0 on success; non-zero on failure.
2041  */
2042 static int
2043 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2044 {
2045         size_t rsz;
2046         nvlist_t *vdev_root;
2047         nvlist_t **child;
2048         uint_t count;
2049         char *type;
2050
2051         rsz = 0;
2052
2053         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2054             &vdev_root) != 0)
2055                 return (EZFS_INVALCONFIG);
2056
2057         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2058             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2059             &child, &count) != 0)
2060                 return (EZFS_INVALCONFIG);
2061
2062         /*
2063          * root pool can not have EFI labeled disks and can only have
2064          * a single top-level vdev.
2065          */
2066         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1 ||
2067             pool_uses_efi(vdev_root))
2068                 return (EZFS_POOL_INVALARG);
2069
2070         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2071             B_FALSE);
2072
2073         /* No online devices */
2074         if (rsz == 0)
2075                 return (EZFS_NODEVICE);
2076
2077         return (0);
2078 }
2079
2080 /*
2081  * Get phys_path for a root pool
2082  * Return 0 on success; non-zero on failure.
2083  */
2084 int
2085 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2086 {
2087         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2088             phypath_size));
2089 }
2090
2091 /*
2092  * If the device has being dynamically expanded then we need to relabel
2093  * the disk to use the new unallocated space.
2094  */
2095 static int
2096 zpool_relabel_disk(libzfs_handle_t *hdl, const char *path, const char *msg)
2097 {
2098         int fd, error;
2099
2100         if ((fd = open(path, O_RDWR|O_DIRECT)) < 0) {
2101                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2102                     "relabel '%s': unable to open device: %d"), path, errno);
2103                 return (zfs_error(hdl, EZFS_OPENFAILED, msg));
2104         }
2105
2106         /*
2107          * It's possible that we might encounter an error if the device
2108          * does not have any unallocated space left. If so, we simply
2109          * ignore that error and continue on.
2110          *
2111          * Also, we don't call efi_rescan() - that would just return EBUSY.
2112          * The module will do it for us in vdev_disk_open().
2113          */
2114         error = efi_use_whole_disk(fd);
2115         (void) close(fd);
2116         if (error && error != VT_ENOSPC) {
2117                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2118                     "relabel '%s': unable to read disk capacity"), path);
2119                 return (zfs_error(hdl, EZFS_NOCAP, msg));
2120         }
2121         return (0);
2122 }
2123
2124 /*
2125  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2126  * ZFS_ONLINE_* flags.
2127  */
2128 int
2129 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2130     vdev_state_t *newstate)
2131 {
2132         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2133         char msg[1024];
2134         nvlist_t *tgt;
2135         boolean_t avail_spare, l2cache, islog;
2136         libzfs_handle_t *hdl = zhp->zpool_hdl;
2137         int error;
2138
2139         if (flags & ZFS_ONLINE_EXPAND) {
2140                 (void) snprintf(msg, sizeof (msg),
2141                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2142         } else {
2143                 (void) snprintf(msg, sizeof (msg),
2144                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2145         }
2146
2147         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2148         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2149             &islog)) == NULL)
2150                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2151
2152         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2153
2154         if (avail_spare)
2155                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2156
2157         if (flags & ZFS_ONLINE_EXPAND ||
2158             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2159                 uint64_t wholedisk = 0;
2160
2161                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2162                     &wholedisk);
2163
2164                 /*
2165                  * XXX - L2ARC 1.0 devices can't support expansion.
2166                  */
2167                 if (l2cache) {
2168                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2169                             "cannot expand cache devices"));
2170                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2171                 }
2172
2173                 if (wholedisk) {
2174                         const char *fullpath = path;
2175                         char buf[MAXPATHLEN];
2176
2177                         if (path[0] != '/') {
2178                                 error = zfs_resolve_shortname(path, buf,
2179                                     sizeof(buf));
2180                                 if (error != 0)
2181                                         return (zfs_error(hdl, EZFS_NODEVICE,
2182                                             msg));
2183
2184                                 fullpath = buf;
2185                         }
2186
2187                         error = zpool_relabel_disk(hdl, fullpath, msg);
2188                         if (error != 0)
2189                                 return (error);
2190                 }
2191         }
2192
2193         zc.zc_cookie = VDEV_STATE_ONLINE;
2194         zc.zc_obj = flags;
2195
2196         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2197                 if (errno == EINVAL) {
2198                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2199                             "from this pool into a new one.  Use '%s' "
2200                             "instead"), "zpool detach");
2201                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2202                 }
2203                 return (zpool_standard_error(hdl, errno, msg));
2204         }
2205
2206         *newstate = zc.zc_cookie;
2207         return (0);
2208 }
2209
2210 /*
2211  * Take the specified vdev offline
2212  */
2213 int
2214 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2215 {
2216         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2217         char msg[1024];
2218         nvlist_t *tgt;
2219         boolean_t avail_spare, l2cache;
2220         libzfs_handle_t *hdl = zhp->zpool_hdl;
2221
2222         (void) snprintf(msg, sizeof (msg),
2223             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2224
2225         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2226         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2227             NULL)) == NULL)
2228                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2229
2230         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2231
2232         if (avail_spare)
2233                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2234
2235         zc.zc_cookie = VDEV_STATE_OFFLINE;
2236         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2237
2238         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2239                 return (0);
2240
2241         switch (errno) {
2242         case EBUSY:
2243
2244                 /*
2245                  * There are no other replicas of this device.
2246                  */
2247                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2248
2249         case EEXIST:
2250                 /*
2251                  * The log device has unplayed logs
2252                  */
2253                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2254
2255         default:
2256                 return (zpool_standard_error(hdl, errno, msg));
2257         }
2258 }
2259
2260 /*
2261  * Mark the given vdev faulted.
2262  */
2263 int
2264 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2265 {
2266         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2267         char msg[1024];
2268         libzfs_handle_t *hdl = zhp->zpool_hdl;
2269
2270         (void) snprintf(msg, sizeof (msg),
2271            dgettext(TEXT_DOMAIN, "cannot fault %llu"), (u_longlong_t)guid);
2272
2273         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2274         zc.zc_guid = guid;
2275         zc.zc_cookie = VDEV_STATE_FAULTED;
2276         zc.zc_obj = aux;
2277
2278         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2279                 return (0);
2280
2281         switch (errno) {
2282         case EBUSY:
2283
2284                 /*
2285                  * There are no other replicas of this device.
2286                  */
2287                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2288
2289         default:
2290                 return (zpool_standard_error(hdl, errno, msg));
2291         }
2292
2293 }
2294
2295 /*
2296  * Mark the given vdev degraded.
2297  */
2298 int
2299 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2300 {
2301         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2302         char msg[1024];
2303         libzfs_handle_t *hdl = zhp->zpool_hdl;
2304
2305         (void) snprintf(msg, sizeof (msg),
2306            dgettext(TEXT_DOMAIN, "cannot degrade %llu"), (u_longlong_t)guid);
2307
2308         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2309         zc.zc_guid = guid;
2310         zc.zc_cookie = VDEV_STATE_DEGRADED;
2311         zc.zc_obj = aux;
2312
2313         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2314                 return (0);
2315
2316         return (zpool_standard_error(hdl, errno, msg));
2317 }
2318
2319 /*
2320  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2321  * a hot spare.
2322  */
2323 static boolean_t
2324 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2325 {
2326         nvlist_t **child;
2327         uint_t c, children;
2328         char *type;
2329
2330         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2331             &children) == 0) {
2332                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2333                     &type) == 0);
2334
2335                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2336                     children == 2 && child[which] == tgt)
2337                         return (B_TRUE);
2338
2339                 for (c = 0; c < children; c++)
2340                         if (is_replacing_spare(child[c], tgt, which))
2341                                 return (B_TRUE);
2342         }
2343
2344         return (B_FALSE);
2345 }
2346
2347 /*
2348  * Attach new_disk (fully described by nvroot) to old_disk.
2349  * If 'replacing' is specified, the new disk will replace the old one.
2350  */
2351 int
2352 zpool_vdev_attach(zpool_handle_t *zhp,
2353     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2354 {
2355         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2356         char msg[1024];
2357         int ret;
2358         nvlist_t *tgt;
2359         boolean_t avail_spare, l2cache, islog;
2360         uint64_t val;
2361         char *newname;
2362         nvlist_t **child;
2363         uint_t children;
2364         nvlist_t *config_root;
2365         libzfs_handle_t *hdl = zhp->zpool_hdl;
2366         boolean_t rootpool = zpool_is_bootable(zhp);
2367
2368         if (replacing)
2369                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2370                     "cannot replace %s with %s"), old_disk, new_disk);
2371         else
2372                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2373                     "cannot attach %s to %s"), new_disk, old_disk);
2374
2375         /*
2376          * If this is a root pool, make sure that we're not attaching an
2377          * EFI labeled device.
2378          */
2379         if (rootpool && pool_uses_efi(nvroot)) {
2380                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2381                     "EFI labeled devices are not supported on root pools."));
2382                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
2383         }
2384
2385         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2386         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2387             &islog)) == 0)
2388                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2389
2390         if (avail_spare)
2391                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2392
2393         if (l2cache)
2394                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2395
2396         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2397         zc.zc_cookie = replacing;
2398
2399         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2400             &child, &children) != 0 || children != 1) {
2401                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2402                     "new device must be a single disk"));
2403                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2404         }
2405
2406         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2407             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2408
2409         if ((newname = zpool_vdev_name(NULL, NULL, child[0], B_FALSE)) == NULL)
2410                 return (-1);
2411
2412         /*
2413          * If the target is a hot spare that has been swapped in, we can only
2414          * replace it with another hot spare.
2415          */
2416         if (replacing &&
2417             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2418             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2419             NULL) == NULL || !avail_spare) &&
2420             is_replacing_spare(config_root, tgt, 1)) {
2421                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2422                     "can only be replaced by another hot spare"));
2423                 free(newname);
2424                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2425         }
2426
2427         free(newname);
2428
2429         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2430                 return (-1);
2431
2432         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2433
2434         zcmd_free_nvlists(&zc);
2435
2436         if (ret == 0) {
2437                 if (rootpool) {
2438                         /*
2439                          * XXX need a better way to prevent user from
2440                          * booting up a half-baked vdev.
2441                          */
2442                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2443                             "sure to wait until resilver is done "
2444                             "before rebooting.\n"));
2445                 }
2446                 return (0);
2447         }
2448
2449         switch (errno) {
2450         case ENOTSUP:
2451                 /*
2452                  * Can't attach to or replace this type of vdev.
2453                  */
2454                 if (replacing) {
2455                         uint64_t version = zpool_get_prop_int(zhp,
2456                             ZPOOL_PROP_VERSION, NULL);
2457
2458                         if (islog)
2459                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2460                                     "cannot replace a log with a spare"));
2461                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2462                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2463                                     "already in replacing/spare config; wait "
2464                                     "for completion or use 'zpool detach'"));
2465                         else
2466                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2467                                     "cannot replace a replacing device"));
2468                 } else {
2469                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2470                             "can only attach to mirrors and top-level "
2471                             "disks"));
2472                 }
2473                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2474                 break;
2475
2476         case EINVAL:
2477                 /*
2478                  * The new device must be a single disk.
2479                  */
2480                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2481                     "new device must be a single disk"));
2482                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2483                 break;
2484
2485         case EBUSY:
2486                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2487                     new_disk);
2488                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2489                 break;
2490
2491         case EOVERFLOW:
2492                 /*
2493                  * The new device is too small.
2494                  */
2495                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2496                     "device is too small"));
2497                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2498                 break;
2499
2500         case EDOM:
2501                 /*
2502                  * The new device has a different alignment requirement.
2503                  */
2504                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2505                     "devices have different sector alignment"));
2506                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2507                 break;
2508
2509         case ENAMETOOLONG:
2510                 /*
2511                  * The resulting top-level vdev spec won't fit in the label.
2512                  */
2513                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2514                 break;
2515
2516         default:
2517                 (void) zpool_standard_error(hdl, errno, msg);
2518         }
2519
2520         return (-1);
2521 }
2522
2523 /*
2524  * Detach the specified device.
2525  */
2526 int
2527 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2528 {
2529         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2530         char msg[1024];
2531         nvlist_t *tgt;
2532         boolean_t avail_spare, l2cache;
2533         libzfs_handle_t *hdl = zhp->zpool_hdl;
2534
2535         (void) snprintf(msg, sizeof (msg),
2536             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2537
2538         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2539         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2540             NULL)) == 0)
2541                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2542
2543         if (avail_spare)
2544                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2545
2546         if (l2cache)
2547                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2548
2549         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2550
2551         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2552                 return (0);
2553
2554         switch (errno) {
2555
2556         case ENOTSUP:
2557                 /*
2558                  * Can't detach from this type of vdev.
2559                  */
2560                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2561                     "applicable to mirror and replacing vdevs"));
2562                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2563                 break;
2564
2565         case EBUSY:
2566                 /*
2567                  * There are no other replicas of this device.
2568                  */
2569                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2570                 break;
2571
2572         default:
2573                 (void) zpool_standard_error(hdl, errno, msg);
2574         }
2575
2576         return (-1);
2577 }
2578
2579 /*
2580  * Find a mirror vdev in the source nvlist.
2581  *
2582  * The mchild array contains a list of disks in one of the top-level mirrors
2583  * of the source pool.  The schild array contains a list of disks that the
2584  * user specified on the command line.  We loop over the mchild array to
2585  * see if any entry in the schild array matches.
2586  *
2587  * If a disk in the mchild array is found in the schild array, we return
2588  * the index of that entry.  Otherwise we return -1.
2589  */
2590 static int
2591 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2592     nvlist_t **schild, uint_t schildren)
2593 {
2594         uint_t mc;
2595
2596         for (mc = 0; mc < mchildren; mc++) {
2597                 uint_t sc;
2598                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2599                     mchild[mc], B_FALSE);
2600
2601                 for (sc = 0; sc < schildren; sc++) {
2602                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2603                             schild[sc], B_FALSE);
2604                         boolean_t result = (strcmp(mpath, spath) == 0);
2605
2606                         free(spath);
2607                         if (result) {
2608                                 free(mpath);
2609                                 return (mc);
2610                         }
2611                 }
2612
2613                 free(mpath);
2614         }
2615
2616         return (-1);
2617 }
2618
2619 /*
2620  * Split a mirror pool.  If newroot points to null, then a new nvlist
2621  * is generated and it is the responsibility of the caller to free it.
2622  */
2623 int
2624 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2625     nvlist_t *props, splitflags_t flags)
2626 {
2627         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2628         char msg[1024];
2629         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2630         nvlist_t **varray = NULL, *zc_props = NULL;
2631         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2632         libzfs_handle_t *hdl = zhp->zpool_hdl;
2633         uint64_t vers;
2634         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2635         int retval = 0;
2636
2637         (void) snprintf(msg, sizeof (msg),
2638             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2639
2640         if (!zpool_name_valid(hdl, B_FALSE, newname))
2641                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2642
2643         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2644                 (void) fprintf(stderr, gettext("Internal error: unable to "
2645                     "retrieve pool configuration\n"));
2646                 return (-1);
2647         }
2648
2649         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2650             == 0);
2651         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2652
2653         if (props) {
2654                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2655                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2656                     props, vers, flags, msg)) == NULL)
2657                         return (-1);
2658         }
2659
2660         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2661             &children) != 0) {
2662                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2663                     "Source pool is missing vdev tree"));
2664                 if (zc_props)
2665                         nvlist_free(zc_props);
2666                 return (-1);
2667         }
2668
2669         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2670         vcount = 0;
2671
2672         if (*newroot == NULL ||
2673             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2674             &newchild, &newchildren) != 0)
2675                 newchildren = 0;
2676
2677         for (c = 0; c < children; c++) {
2678                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2679                 char *type;
2680                 nvlist_t **mchild, *vdev;
2681                 uint_t mchildren;
2682                 int entry;
2683
2684                 /*
2685                  * Unlike cache & spares, slogs are stored in the
2686                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2687                  */
2688                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2689                     &is_log);
2690                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2691                     &is_hole);
2692                 if (is_log || is_hole) {
2693                         /*
2694                          * Create a hole vdev and put it in the config.
2695                          */
2696                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2697                                 goto out;
2698                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2699                             VDEV_TYPE_HOLE) != 0)
2700                                 goto out;
2701                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2702                             1) != 0)
2703                                 goto out;
2704                         if (lastlog == 0)
2705                                 lastlog = vcount;
2706                         varray[vcount++] = vdev;
2707                         continue;
2708                 }
2709                 lastlog = 0;
2710                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
2711                     == 0);
2712                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
2713                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2714                             "Source pool must be composed only of mirrors\n"));
2715                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2716                         goto out;
2717                 }
2718
2719                 verify(nvlist_lookup_nvlist_array(child[c],
2720                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2721
2722                 /* find or add an entry for this top-level vdev */
2723                 if (newchildren > 0 &&
2724                     (entry = find_vdev_entry(zhp, mchild, mchildren,
2725                     newchild, newchildren)) >= 0) {
2726                         /* We found a disk that the user specified. */
2727                         vdev = mchild[entry];
2728                         ++found;
2729                 } else {
2730                         /* User didn't specify a disk for this vdev. */
2731                         vdev = mchild[mchildren - 1];
2732                 }
2733
2734                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
2735                         goto out;
2736         }
2737
2738         /* did we find every disk the user specified? */
2739         if (found != newchildren) {
2740                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
2741                     "include at most one disk from each mirror"));
2742                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2743                 goto out;
2744         }
2745
2746         /* Prepare the nvlist for populating. */
2747         if (*newroot == NULL) {
2748                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
2749                         goto out;
2750                 freelist = B_TRUE;
2751                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
2752                     VDEV_TYPE_ROOT) != 0)
2753                         goto out;
2754         } else {
2755                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
2756         }
2757
2758         /* Add all the children we found */
2759         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
2760             lastlog == 0 ? vcount : lastlog) != 0)
2761                 goto out;
2762
2763         /*
2764          * If we're just doing a dry run, exit now with success.
2765          */
2766         if (flags.dryrun) {
2767                 memory_err = B_FALSE;
2768                 freelist = B_FALSE;
2769                 goto out;
2770         }
2771
2772         /* now build up the config list & call the ioctl */
2773         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
2774                 goto out;
2775
2776         if (nvlist_add_nvlist(newconfig,
2777             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
2778             nvlist_add_string(newconfig,
2779             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
2780             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
2781                 goto out;
2782
2783         /*
2784          * The new pool is automatically part of the namespace unless we
2785          * explicitly export it.
2786          */
2787         if (!flags.import)
2788                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
2789         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2790         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
2791         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
2792                 goto out;
2793         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
2794                 goto out;
2795
2796         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
2797                 retval = zpool_standard_error(hdl, errno, msg);
2798                 goto out;
2799         }
2800
2801         freelist = B_FALSE;
2802         memory_err = B_FALSE;
2803
2804 out:
2805         if (varray != NULL) {
2806                 int v;
2807
2808                 for (v = 0; v < vcount; v++)
2809                         nvlist_free(varray[v]);
2810                 free(varray);
2811         }
2812         zcmd_free_nvlists(&zc);
2813         if (zc_props)
2814                 nvlist_free(zc_props);
2815         if (newconfig)
2816                 nvlist_free(newconfig);
2817         if (freelist) {
2818                 nvlist_free(*newroot);
2819                 *newroot = NULL;
2820         }
2821
2822         if (retval != 0)
2823                 return (retval);
2824
2825         if (memory_err)
2826                 return (no_memory(hdl));
2827
2828         return (0);
2829 }
2830
2831 /*
2832  * Remove the given device.  Currently, this is supported only for hot spares
2833  * and level 2 cache devices.
2834  */
2835 int
2836 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
2837 {
2838         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2839         char msg[1024];
2840         nvlist_t *tgt;
2841         boolean_t avail_spare, l2cache, islog;
2842         libzfs_handle_t *hdl = zhp->zpool_hdl;
2843         uint64_t version;
2844
2845         (void) snprintf(msg, sizeof (msg),
2846             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
2847
2848         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2849         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2850             &islog)) == 0)
2851                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2852         /*
2853          * XXX - this should just go away.
2854          */
2855         if (!avail_spare && !l2cache && !islog) {
2856                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2857                     "only inactive hot spares, cache, top-level, "
2858                     "or log devices can be removed"));
2859                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2860         }
2861
2862         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
2863         if (islog && version < SPA_VERSION_HOLES) {
2864                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2865                     "pool must be upgrade to support log removal"));
2866                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
2867         }
2868
2869         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2870
2871         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
2872                 return (0);
2873
2874         return (zpool_standard_error(hdl, errno, msg));
2875 }
2876
2877 /*
2878  * Clear the errors for the pool, or the particular device if specified.
2879  */
2880 int
2881 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
2882 {
2883         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2884         char msg[1024];
2885         nvlist_t *tgt;
2886         zpool_rewind_policy_t policy;
2887         boolean_t avail_spare, l2cache;
2888         libzfs_handle_t *hdl = zhp->zpool_hdl;
2889         nvlist_t *nvi = NULL;
2890         int error;
2891
2892         if (path)
2893                 (void) snprintf(msg, sizeof (msg),
2894                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
2895                     path);
2896         else
2897                 (void) snprintf(msg, sizeof (msg),
2898                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
2899                     zhp->zpool_name);
2900
2901         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2902         if (path) {
2903                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
2904                     &l2cache, NULL)) == 0)
2905                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
2906
2907                 /*
2908                  * Don't allow error clearing for hot spares.  Do allow
2909                  * error clearing for l2cache devices.
2910                  */
2911                 if (avail_spare)
2912                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
2913
2914                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
2915                     &zc.zc_guid) == 0);
2916         }
2917
2918         zpool_get_rewind_policy(rewindnvl, &policy);
2919         zc.zc_cookie = policy.zrp_request;
2920
2921         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
2922                 return (-1);
2923
2924         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
2925                 return (-1);
2926
2927         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
2928             errno == ENOMEM) {
2929                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
2930                         zcmd_free_nvlists(&zc);
2931                         return (-1);
2932                 }
2933         }
2934
2935         if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
2936             errno != EPERM && errno != EACCES)) {
2937                 if (policy.zrp_request &
2938                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
2939                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
2940                         zpool_rewind_exclaim(hdl, zc.zc_name,
2941                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
2942                             nvi);
2943                         nvlist_free(nvi);
2944                 }
2945                 zcmd_free_nvlists(&zc);
2946                 return (0);
2947         }
2948
2949         zcmd_free_nvlists(&zc);
2950         return (zpool_standard_error(hdl, errno, msg));
2951 }
2952
2953 /*
2954  * Similar to zpool_clear(), but takes a GUID (used by fmd).
2955  */
2956 int
2957 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
2958 {
2959         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2960         char msg[1024];
2961         libzfs_handle_t *hdl = zhp->zpool_hdl;
2962
2963         (void) snprintf(msg, sizeof (msg),
2964             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
2965            (u_longlong_t)guid);
2966
2967         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2968         zc.zc_guid = guid;
2969         zc.zc_cookie = ZPOOL_NO_REWIND;
2970
2971         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
2972                 return (0);
2973
2974         return (zpool_standard_error(hdl, errno, msg));
2975 }
2976
2977 /*
2978  * Change the GUID for a pool.
2979  */
2980 int
2981 zpool_reguid(zpool_handle_t *zhp)
2982 {
2983         char msg[1024];
2984         libzfs_handle_t *hdl = zhp->zpool_hdl;
2985         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
2986
2987         (void) snprintf(msg, sizeof (msg),
2988             dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
2989
2990         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2991         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
2992                 return (0);
2993
2994         return (zpool_standard_error(hdl, errno, msg));
2995 }
2996
2997 /*
2998  * Reopen the pool.
2999  */
3000 int
3001 zpool_reopen(zpool_handle_t *zhp)
3002 {
3003         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3004         char msg[1024];
3005         libzfs_handle_t *hdl = zhp->zpool_hdl;
3006
3007         (void) snprintf(msg, sizeof (msg),
3008             dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3009             zhp->zpool_name);
3010
3011         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3012         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3013                 return (0);
3014         return (zpool_standard_error(hdl, errno, msg));
3015 }
3016
3017 /*
3018  * Convert from a devid string to a path.
3019  */
3020 static char *
3021 devid_to_path(char *devid_str)
3022 {
3023         ddi_devid_t devid;
3024         char *minor;
3025         char *path;
3026         devid_nmlist_t *list = NULL;
3027         int ret;
3028
3029         if (devid_str_decode(devid_str, &devid, &minor) != 0)
3030                 return (NULL);
3031
3032         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3033
3034         devid_str_free(minor);
3035         devid_free(devid);
3036
3037         if (ret != 0)
3038                 return (NULL);
3039
3040         if ((path = strdup(list[0].devname)) == NULL)
3041                 return (NULL);
3042
3043         devid_free_nmlist(list);
3044
3045         return (path);
3046 }
3047
3048 /*
3049  * Convert from a path to a devid string.
3050  */
3051 static char *
3052 path_to_devid(const char *path)
3053 {
3054         int fd;
3055         ddi_devid_t devid;
3056         char *minor, *ret;
3057
3058         if ((fd = open(path, O_RDONLY)) < 0)
3059                 return (NULL);
3060
3061         minor = NULL;
3062         ret = NULL;
3063         if (devid_get(fd, &devid) == 0) {
3064                 if (devid_get_minor_name(fd, &minor) == 0)
3065                         ret = devid_str_encode(devid, minor);
3066                 if (minor != NULL)
3067                         devid_str_free(minor);
3068                 devid_free(devid);
3069         }
3070         (void) close(fd);
3071
3072         return (ret);
3073 }
3074
3075 /*
3076  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3077  * ignore any failure here, since a common case is for an unprivileged user to
3078  * type 'zpool status', and we'll display the correct information anyway.
3079  */
3080 static void
3081 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3082 {
3083         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3084
3085         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3086         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3087         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3088             &zc.zc_guid) == 0);
3089
3090         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3091 }
3092
3093 /*
3094  * Remove partition suffix from a vdev path.  Partition suffixes may take three
3095  * forms: "-partX", "pX", or "X", where X is a string of digits.  The second
3096  * case only occurs when the suffix is preceded by a digit, i.e. "md0p0" The
3097  * third case only occurs when preceded by a string matching the regular
3098  * expression "^[hs]d[a-z]+", i.e. a scsi or ide disk.
3099  */
3100 static char *
3101 strip_partition(libzfs_handle_t *hdl, char *path)
3102 {
3103         char *tmp = zfs_strdup(hdl, path);
3104         char *part = NULL, *d = NULL;
3105
3106         if ((part = strstr(tmp, "-part")) && part != tmp) {
3107                 d = part + 5;
3108         } else if ((part = strrchr(tmp, 'p')) &&
3109             part > tmp + 1 && isdigit(*(part-1))) {
3110                 d = part + 1;
3111         } else if ((tmp[0] == 'h' || tmp[0] == 's') && tmp[1] == 'd') {
3112                 for (d = &tmp[2]; isalpha(*d); part = ++d);
3113         }
3114         if (part && d && *d != '\0') {
3115                 for (; isdigit(*d); d++);
3116                 if (*d == '\0')
3117                         *part = '\0';
3118         }
3119         return (tmp);
3120 }
3121
3122 #define PATH_BUF_LEN    64
3123
3124 /*
3125  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3126  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3127  * We also check if this is a whole disk, in which case we strip off the
3128  * trailing 's0' slice name.
3129  *
3130  * This routine is also responsible for identifying when disks have been
3131  * reconfigured in a new location.  The kernel will have opened the device by
3132  * devid, but the path will still refer to the old location.  To catch this, we
3133  * first do a path -> devid translation (which is fast for the common case).  If
3134  * the devid matches, we're done.  If not, we do a reverse devid -> path
3135  * translation and issue the appropriate ioctl() to update the path of the vdev.
3136  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3137  * of these checks.
3138  */
3139 char *
3140 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3141     boolean_t verbose)
3142 {
3143         char *path, *devid, *type;
3144         uint64_t value;
3145         char buf[PATH_BUF_LEN];
3146         char tmpbuf[PATH_BUF_LEN];
3147         vdev_stat_t *vs;
3148         uint_t vsc;
3149
3150         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
3151             &value) == 0) {
3152                 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3153                     &value) == 0);
3154                 (void) snprintf(buf, sizeof (buf), "%llu",
3155                     (u_longlong_t)value);
3156                 path = buf;
3157         } else if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0) {
3158                 /*
3159                  * If the device is dead (faulted, offline, etc) then don't
3160                  * bother opening it.  Otherwise we may be forcing the user to
3161                  * open a misbehaving device, which can have undesirable
3162                  * effects.
3163                  */
3164                 if ((nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3165                     (uint64_t **)&vs, &vsc) != 0 ||
3166                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3167                     zhp != NULL &&
3168                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3169                         /*
3170                          * Determine if the current path is correct.
3171                          */
3172                         char *newdevid = path_to_devid(path);
3173
3174                         if (newdevid == NULL ||
3175                             strcmp(devid, newdevid) != 0) {
3176                                 char *newpath;
3177
3178                                 if ((newpath = devid_to_path(devid)) != NULL) {
3179                                         /*
3180                                          * Update the path appropriately.
3181                                          */
3182                                         set_path(zhp, nv, newpath);
3183                                         if (nvlist_add_string(nv,
3184                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3185                                                 verify(nvlist_lookup_string(nv,
3186                                                     ZPOOL_CONFIG_PATH,
3187                                                     &path) == 0);
3188                                         free(newpath);
3189                                 }
3190                         }
3191
3192                         if (newdevid)
3193                                 devid_str_free(newdevid);
3194                 }
3195
3196                 /*
3197                  * For a block device only use the name.
3198                  */
3199                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) == 0);
3200                 if (strcmp(type, VDEV_TYPE_DISK) == 0) {
3201                         path = strrchr(path, '/');
3202                         path++;
3203                 }
3204
3205                 /*
3206                  * Remove the partition from the path it this is a whole disk.
3207                  */
3208                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
3209                     &value) == 0 && value) {
3210                         return strip_partition(hdl, path);
3211                 }
3212         } else {
3213                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3214
3215                 /*
3216                  * If it's a raidz device, we need to stick in the parity level.
3217                  */
3218                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3219
3220                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3221                             &value) == 0);
3222                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3223                             (u_longlong_t)value);
3224                         path = buf;
3225                 }
3226
3227                 /*
3228                  * We identify each top-level vdev by using a <type-id>
3229                  * naming convention.
3230                  */
3231                 if (verbose) {
3232                         uint64_t id;
3233
3234                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3235                             &id) == 0);
3236                         (void) snprintf(tmpbuf, sizeof (tmpbuf), "%s-%llu",
3237                             path, (u_longlong_t)id);
3238                         path = tmpbuf;
3239                 }
3240         }
3241
3242         return (zfs_strdup(hdl, path));
3243 }
3244
3245 static int
3246 zbookmark_compare(const void *a, const void *b)
3247 {
3248         return (memcmp(a, b, sizeof (zbookmark_t)));
3249 }
3250
3251 /*
3252  * Retrieve the persistent error log, uniquify the members, and return to the
3253  * caller.
3254  */
3255 int
3256 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3257 {
3258         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3259         uint64_t count;
3260         zbookmark_t *zb = NULL;
3261         int i;
3262
3263         /*
3264          * Retrieve the raw error list from the kernel.  If the number of errors
3265          * has increased, allocate more space and continue until we get the
3266          * entire list.
3267          */
3268         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3269             &count) == 0);
3270         if (count == 0)
3271                 return (0);
3272         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3273             count * sizeof (zbookmark_t))) == (uintptr_t)NULL)
3274                 return (-1);
3275         zc.zc_nvlist_dst_size = count;
3276         (void) strcpy(zc.zc_name, zhp->zpool_name);
3277         for (;;) {
3278                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3279                     &zc) != 0) {
3280                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3281                         if (errno == ENOMEM) {
3282                                 count = zc.zc_nvlist_dst_size;
3283                                 if ((zc.zc_nvlist_dst = (uintptr_t)
3284                                     zfs_alloc(zhp->zpool_hdl, count *
3285                                     sizeof (zbookmark_t))) == (uintptr_t)NULL)
3286                                         return (-1);
3287                         } else {
3288                                 return (-1);
3289                         }
3290                 } else {
3291                         break;
3292                 }
3293         }
3294
3295         /*
3296          * Sort the resulting bookmarks.  This is a little confusing due to the
3297          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3298          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3299          * _not_ copied as part of the process.  So we point the start of our
3300          * array appropriate and decrement the total number of elements.
3301          */
3302         zb = ((zbookmark_t *)(uintptr_t)zc.zc_nvlist_dst) +
3303             zc.zc_nvlist_dst_size;
3304         count -= zc.zc_nvlist_dst_size;
3305
3306         qsort(zb, count, sizeof (zbookmark_t), zbookmark_compare);
3307
3308         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3309
3310         /*
3311          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3312          */
3313         for (i = 0; i < count; i++) {
3314                 nvlist_t *nv;
3315
3316                 /* ignoring zb_blkid and zb_level for now */
3317                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3318                     zb[i-1].zb_object == zb[i].zb_object)
3319                         continue;
3320
3321                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3322                         goto nomem;
3323                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3324                     zb[i].zb_objset) != 0) {
3325                         nvlist_free(nv);
3326                         goto nomem;
3327                 }
3328                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3329                     zb[i].zb_object) != 0) {
3330                         nvlist_free(nv);
3331                         goto nomem;
3332                 }
3333                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3334                         nvlist_free(nv);
3335                         goto nomem;
3336                 }
3337                 nvlist_free(nv);
3338         }
3339
3340         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3341         return (0);
3342
3343 nomem:
3344         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3345         return (no_memory(zhp->zpool_hdl));
3346 }
3347
3348 /*
3349  * Upgrade a ZFS pool to the latest on-disk version.
3350  */
3351 int
3352 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3353 {
3354         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3355         libzfs_handle_t *hdl = zhp->zpool_hdl;
3356
3357         (void) strcpy(zc.zc_name, zhp->zpool_name);
3358         zc.zc_cookie = new_version;
3359
3360         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3361                 return (zpool_standard_error_fmt(hdl, errno,
3362                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3363                     zhp->zpool_name));
3364         return (0);
3365 }
3366
3367 void
3368 zpool_set_history_str(const char *subcommand, int argc, char **argv,
3369     char *history_str)
3370 {
3371         int i;
3372
3373         (void) strlcpy(history_str, subcommand, HIS_MAX_RECORD_LEN);
3374         for (i = 1; i < argc; i++) {
3375                 if (strlen(history_str) + 1 + strlen(argv[i]) >
3376                     HIS_MAX_RECORD_LEN)
3377                         break;
3378                 (void) strlcat(history_str, " ", HIS_MAX_RECORD_LEN);
3379                 (void) strlcat(history_str, argv[i], HIS_MAX_RECORD_LEN);
3380         }
3381 }
3382
3383 /*
3384  * Stage command history for logging.
3385  */
3386 int
3387 zpool_stage_history(libzfs_handle_t *hdl, const char *history_str)
3388 {
3389         if (history_str == NULL)
3390                 return (EINVAL);
3391
3392         if (strlen(history_str) > HIS_MAX_RECORD_LEN)
3393                 return (EINVAL);
3394
3395         if (hdl->libzfs_log_str != NULL)
3396                 free(hdl->libzfs_log_str);
3397
3398         if ((hdl->libzfs_log_str = strdup(history_str)) == NULL)
3399                 return (no_memory(hdl));
3400
3401         return (0);
3402 }
3403
3404 /*
3405  * Perform ioctl to get some command history of a pool.
3406  *
3407  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3408  * logical offset of the history buffer to start reading from.
3409  *
3410  * Upon return, 'off' is the next logical offset to read from and
3411  * 'len' is the actual amount of bytes read into 'buf'.
3412  */
3413 static int
3414 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3415 {
3416         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3417         libzfs_handle_t *hdl = zhp->zpool_hdl;
3418
3419         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3420
3421         zc.zc_history = (uint64_t)(uintptr_t)buf;
3422         zc.zc_history_len = *len;
3423         zc.zc_history_offset = *off;
3424
3425         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3426                 switch (errno) {
3427                 case EPERM:
3428                         return (zfs_error_fmt(hdl, EZFS_PERM,
3429                             dgettext(TEXT_DOMAIN,
3430                             "cannot show history for pool '%s'"),
3431                             zhp->zpool_name));
3432                 case ENOENT:
3433                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3434                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3435                             "'%s'"), zhp->zpool_name));
3436                 case ENOTSUP:
3437                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3438                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3439                             "'%s', pool must be upgraded"), zhp->zpool_name));
3440                 default:
3441                         return (zpool_standard_error_fmt(hdl, errno,
3442                             dgettext(TEXT_DOMAIN,
3443                             "cannot get history for '%s'"), zhp->zpool_name));
3444                 }
3445         }
3446
3447         *len = zc.zc_history_len;
3448         *off = zc.zc_history_offset;
3449
3450         return (0);
3451 }
3452
3453 /*
3454  * Process the buffer of nvlists, unpacking and storing each nvlist record
3455  * into 'records'.  'leftover' is set to the number of bytes that weren't
3456  * processed as there wasn't a complete record.
3457  */
3458 int
3459 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3460     nvlist_t ***records, uint_t *numrecords)
3461 {
3462         uint64_t reclen;
3463         nvlist_t *nv;
3464         int i;
3465
3466         while (bytes_read > sizeof (reclen)) {
3467
3468                 /* get length of packed record (stored as little endian) */
3469                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3470                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3471
3472                 if (bytes_read < sizeof (reclen) + reclen)
3473                         break;
3474
3475                 /* unpack record */
3476                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3477                         return (ENOMEM);
3478                 bytes_read -= sizeof (reclen) + reclen;
3479                 buf += sizeof (reclen) + reclen;
3480
3481                 /* add record to nvlist array */
3482                 (*numrecords)++;
3483                 if (ISP2(*numrecords + 1)) {
3484                         *records = realloc(*records,
3485                             *numrecords * 2 * sizeof (nvlist_t *));
3486                 }
3487                 (*records)[*numrecords - 1] = nv;
3488         }
3489
3490         *leftover = bytes_read;
3491         return (0);
3492 }
3493
3494 #define HIS_BUF_LEN     (128*1024)
3495
3496 /*
3497  * Retrieve the command history of a pool.
3498  */
3499 int
3500 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3501 {
3502         char buf[HIS_BUF_LEN];
3503         uint64_t off = 0;
3504         nvlist_t **records = NULL;
3505         uint_t numrecords = 0;
3506         int err, i;
3507
3508         do {
3509                 uint64_t bytes_read = sizeof (buf);
3510                 uint64_t leftover;
3511
3512                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3513                         break;
3514
3515                 /* if nothing else was read in, we're at EOF, just return */
3516                 if (!bytes_read)
3517                         break;
3518
3519                 if ((err = zpool_history_unpack(buf, bytes_read,
3520                     &leftover, &records, &numrecords)) != 0)
3521                         break;
3522                 off -= leftover;
3523
3524                 /* CONSTCOND */
3525         } while (1);
3526
3527         if (!err) {
3528                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3529                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3530                     records, numrecords) == 0);
3531         }
3532         for (i = 0; i < numrecords; i++)
3533                 nvlist_free(records[i]);
3534         free(records);
3535
3536         return (err);
3537 }
3538
3539 /*
3540  * Retrieve the next event.  If there is a new event available 'nvp' will
3541  * contain a newly allocated nvlist and 'dropped' will be set to the number
3542  * of missed events since the last call to this function.  When 'nvp' is
3543  * set to NULL it indicates no new events are available.  In either case
3544  * the function returns 0 and it is up to the caller to free 'nvp'.  In
3545  * the case of a fatal error the function will return a non-zero value.
3546  * When the function is called in blocking mode it will not return until
3547  * a new event is available.
3548  */
3549 int
3550 zpool_events_next(libzfs_handle_t *hdl, nvlist_t **nvp,
3551     int *dropped, int block, int cleanup_fd)
3552 {
3553         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3554         int error = 0;
3555
3556         *nvp = NULL;
3557         *dropped = 0;
3558         zc.zc_cleanup_fd = cleanup_fd;
3559
3560         if (!block)
3561                 zc.zc_guid = ZEVENT_NONBLOCK;
3562
3563         if (zcmd_alloc_dst_nvlist(hdl, &zc, ZEVENT_SIZE) != 0)
3564                 return (-1);
3565
3566 retry:
3567         if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_NEXT, &zc) != 0) {
3568                 switch (errno) {
3569                 case ESHUTDOWN:
3570                         error = zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
3571                             dgettext(TEXT_DOMAIN, "zfs shutdown"));
3572                         goto out;
3573                 case ENOENT:
3574                         /* Blocking error case should not occur */
3575                         if (block)
3576                                 error = zpool_standard_error_fmt(hdl, errno,
3577                                     dgettext(TEXT_DOMAIN, "cannot get event"));
3578
3579                         goto out;
3580                 case ENOMEM:
3581                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3582                                 error = zfs_error_fmt(hdl, EZFS_NOMEM,
3583                                     dgettext(TEXT_DOMAIN, "cannot get event"));
3584                                 goto out;
3585                         } else {
3586                                 goto retry;
3587                         }
3588                 default:
3589                         error = zpool_standard_error_fmt(hdl, errno,
3590                             dgettext(TEXT_DOMAIN, "cannot get event"));
3591                         goto out;
3592                 }
3593         }
3594
3595         error = zcmd_read_dst_nvlist(hdl, &zc, nvp);
3596         if (error != 0)
3597                 goto out;
3598
3599         *dropped = (int)zc.zc_cookie;
3600 out:
3601         zcmd_free_nvlists(&zc);
3602
3603         return (error);
3604 }
3605
3606 /*
3607  * Clear all events.
3608  */
3609 int
3610 zpool_events_clear(libzfs_handle_t *hdl, int *count)
3611 {
3612         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3613         char msg[1024];
3614
3615         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
3616             "cannot clear events"));
3617
3618         if (zfs_ioctl(hdl, ZFS_IOC_EVENTS_CLEAR, &zc) != 0)
3619                 return (zpool_standard_error_fmt(hdl, errno, msg));
3620
3621         if (count != NULL)
3622                 *count = (int)zc.zc_cookie; /* # of events cleared */
3623
3624         return (0);
3625 }
3626
3627 void
3628 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
3629     char *pathname, size_t len)
3630 {
3631         zfs_cmd_t zc = { "\0", "\0", "\0", "\0", 0 };
3632         boolean_t mounted = B_FALSE;
3633         char *mntpnt = NULL;
3634         char dsname[MAXNAMELEN];
3635
3636         if (dsobj == 0) {
3637                 /* special case for the MOS */
3638                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>", (longlong_t)obj);
3639                 return;
3640         }
3641
3642         /* get the dataset's name */
3643         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3644         zc.zc_obj = dsobj;
3645         if (ioctl(zhp->zpool_hdl->libzfs_fd,
3646             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
3647                 /* just write out a path of two object numbers */
3648                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
3649                     (longlong_t)dsobj, (longlong_t)obj);
3650                 return;
3651         }
3652         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
3653
3654         /* find out if the dataset is mounted */
3655         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
3656
3657         /* get the corrupted object's path */
3658         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
3659         zc.zc_obj = obj;
3660         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
3661             &zc) == 0) {
3662                 if (mounted) {
3663                         (void) snprintf(pathname, len, "%s%s", mntpnt,
3664                             zc.zc_value);
3665                 } else {
3666                         (void) snprintf(pathname, len, "%s:%s",
3667                             dsname, zc.zc_value);
3668                 }
3669         } else {
3670                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname, (longlong_t)obj);
3671         }
3672         free(mntpnt);
3673 }
3674
3675 /*
3676  * Read the EFI label from the config, if a label does not exist then
3677  * pass back the error to the caller. If the caller has passed a non-NULL
3678  * diskaddr argument then we set it to the starting address of the EFI
3679  * partition.
3680  */
3681 static int
3682 read_efi_label(nvlist_t *config, diskaddr_t *sb)
3683 {
3684         char *path;
3685         int fd;
3686         char diskname[MAXPATHLEN];
3687         int err = -1;
3688
3689         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
3690                 return (err);
3691
3692         (void) snprintf(diskname, sizeof (diskname), "%s%s", DISK_ROOT,
3693             strrchr(path, '/'));
3694         if ((fd = open(diskname, O_RDWR|O_DIRECT)) >= 0) {
3695                 struct dk_gpt *vtoc;
3696
3697                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
3698                         if (sb != NULL)
3699                                 *sb = vtoc->efi_parts[0].p_start;
3700                         efi_free(vtoc);
3701                 }
3702                 (void) close(fd);
3703         }
3704         return (err);
3705 }
3706
3707 /*
3708  * determine where a partition starts on a disk in the current
3709  * configuration
3710  */
3711 static diskaddr_t
3712 find_start_block(nvlist_t *config)
3713 {
3714         nvlist_t **child;
3715         uint_t c, children;
3716         diskaddr_t sb = MAXOFFSET_T;
3717         uint64_t wholedisk;
3718
3719         if (nvlist_lookup_nvlist_array(config,
3720             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
3721                 if (nvlist_lookup_uint64(config,
3722                     ZPOOL_CONFIG_WHOLE_DISK,
3723                     &wholedisk) != 0 || !wholedisk) {
3724                         return (MAXOFFSET_T);
3725                 }
3726                 if (read_efi_label(config, &sb) < 0)
3727                         sb = MAXOFFSET_T;
3728                 return (sb);
3729         }
3730
3731         for (c = 0; c < children; c++) {
3732                 sb = find_start_block(child[c]);
3733                 if (sb != MAXOFFSET_T) {
3734                         return (sb);
3735                 }
3736         }
3737         return (MAXOFFSET_T);
3738 }
3739
3740 int
3741 zpool_label_disk_wait(char *path, int timeout)
3742 {
3743         struct stat64 statbuf;
3744         int i;
3745
3746         /*
3747          * Wait timeout miliseconds for a newly created device to be available
3748          * from the given path.  There is a small window when a /dev/ device
3749          * will exist and the udev link will not, so we must wait for the
3750          * symlink.  Depending on the udev rules this may take a few seconds.
3751          */
3752         for (i = 0; i < timeout; i++) {
3753                 usleep(1000);
3754
3755                 errno = 0;
3756                 if ((stat64(path, &statbuf) == 0) && (errno == 0))
3757                         return (0);
3758         }
3759
3760         return (ENOENT);
3761 }
3762
3763 int
3764 zpool_label_disk_check(char *path)
3765 {
3766         struct dk_gpt *vtoc;
3767         int fd, err;
3768
3769         if ((fd = open(path, O_RDWR|O_DIRECT)) < 0)
3770                 return errno;
3771
3772         if ((err = efi_alloc_and_read(fd, &vtoc)) != 0) {
3773                 (void) close(fd);
3774                 return err;
3775         }
3776
3777         if (vtoc->efi_flags & EFI_GPT_PRIMARY_CORRUPT) {
3778                 efi_free(vtoc);
3779                 (void) close(fd);
3780                 return EIDRM;
3781         }
3782
3783         efi_free(vtoc);
3784         (void) close(fd);
3785         return 0;
3786 }
3787
3788 /*
3789  * Label an individual disk.  The name provided is the short name,
3790  * stripped of any leading /dev path.
3791  */
3792 int
3793 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, char *name)
3794 {
3795         char path[MAXPATHLEN];
3796         struct dk_gpt *vtoc;
3797         int rval, fd;
3798         size_t resv = EFI_MIN_RESV_SIZE;
3799         uint64_t slice_size;
3800         diskaddr_t start_block;
3801         char errbuf[1024];
3802
3803         /* prepare an error message just in case */
3804         (void) snprintf(errbuf, sizeof (errbuf),
3805             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
3806
3807         if (zhp) {
3808                 nvlist_t *nvroot;
3809
3810                 if (zpool_is_bootable(zhp)) {
3811                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3812                             "EFI labeled devices are not supported on root "
3813                             "pools."));
3814                         return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
3815                 }
3816
3817                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
3818                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3819
3820                 if (zhp->zpool_start_block == 0)
3821                         start_block = find_start_block(nvroot);
3822                 else
3823                         start_block = zhp->zpool_start_block;
3824                 zhp->zpool_start_block = start_block;
3825         } else {
3826                 /* new pool */
3827                 start_block = NEW_START_BLOCK;
3828         }
3829
3830         (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT, name);
3831
3832         if ((fd = open(path, O_RDWR|O_DIRECT)) < 0) {
3833                 /*
3834                  * This shouldn't happen.  We've long since verified that this
3835                  * is a valid device.
3836                  */
3837                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
3838                     "label '%s': unable to open device: %d"), path, errno);
3839                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
3840         }
3841
3842         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
3843                 /*
3844                  * The only way this can fail is if we run out of memory, or we
3845                  * were unable to read the disk's capacity
3846                  */
3847                 if (errno == ENOMEM)
3848                         (void) no_memory(hdl);
3849
3850                 (void) close(fd);
3851                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
3852                     "label '%s': unable to read disk capacity"), path);
3853
3854                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
3855         }
3856
3857         slice_size = vtoc->efi_last_u_lba + 1;
3858         slice_size -= EFI_MIN_RESV_SIZE;
3859         if (start_block == MAXOFFSET_T)
3860                 start_block = NEW_START_BLOCK;
3861         slice_size -= start_block;
3862         slice_size = P2ALIGN(slice_size, PARTITION_END_ALIGNMENT);
3863
3864         vtoc->efi_parts[0].p_start = start_block;
3865         vtoc->efi_parts[0].p_size = slice_size;
3866
3867         /*
3868          * Why we use V_USR: V_BACKUP confuses users, and is considered
3869          * disposable by some EFI utilities (since EFI doesn't have a backup
3870          * slice).  V_UNASSIGNED is supposed to be used only for zero size
3871          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
3872          * etc. were all pretty specific.  V_USR is as close to reality as we
3873          * can get, in the absence of V_OTHER.
3874          */
3875         vtoc->efi_parts[0].p_tag = V_USR;
3876         (void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
3877
3878         vtoc->efi_parts[8].p_start = slice_size + start_block;
3879         vtoc->efi_parts[8].p_size = resv;
3880         vtoc->efi_parts[8].p_tag = V_RESERVED;
3881
3882         if ((rval = efi_write(fd, vtoc)) != 0 || (rval = efi_rescan(fd)) != 0) {
3883                 /*
3884                  * Some block drivers (like pcata) may not support EFI
3885                  * GPT labels.  Print out a helpful error message dir-
3886                  * ecting the user to manually label the disk and give
3887                  * a specific slice.
3888                  */
3889                 (void) close(fd);
3890                 efi_free(vtoc);
3891
3892                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "try using "
3893                     "parted(8) and then provide a specific slice: %d"), rval);
3894                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
3895         }
3896
3897         (void) close(fd);
3898         efi_free(vtoc);
3899
3900         /* Wait for the first expected partition to appear. */
3901
3902         (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT, name);
3903         (void) zfs_append_partition(path, MAXPATHLEN);
3904
3905         rval = zpool_label_disk_wait(path, 3000);
3906         if (rval) {
3907                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "failed to "
3908                     "detect device partitions on '%s': %d"), path, rval);
3909                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
3910         }
3911
3912         /* We can't be to paranoid.  Read the label back and verify it. */
3913         (void) snprintf(path, sizeof (path), "%s/%s", DISK_ROOT, name);
3914         rval = zpool_label_disk_check(path);
3915         if (rval) {
3916                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "freshly written "
3917                     "EFI label on '%s' is damaged.  Ensure\nthis device "
3918                     "is not in in use, and is functioning properly: %d"),
3919                     path, rval);
3920                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
3921         }
3922
3923         return 0;
3924 }