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