cddl/contrib/opensolaris/lib/libzfs/common/libzfs_pool.c

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  25  * Copyright (c) 2012 by Delphix. All rights reserved.
  26  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  27  */
  28
  29 #include <sys/types.h>
  30 #include <sys/stat.h>
  31 #include <ctype.h>
  32 #include <errno.h>
  33 #include <devid.h>
  34 #include <fcntl.h>
  35 #include <libintl.h>
  36 #include <stdio.h>
  37 #include <stdlib.h>
  38 #include <strings.h>
  39 #include <unistd.h>
  40 #include <libgen.h>
  41 #include <sys/zfs_ioctl.h>
  42 #include <dlfcn.h>
  43
  44 #include "zfs_namecheck.h"
  45 #include "zfs_prop.h"
  46 #include "libzfs_impl.h"
  47 #include "zfs_comutil.h"
  48 #include "zfeature_common.h"
  49
  50 static int read_efi_label(nvlist_t *config, diskaddr_t *sb);
  51
  52 #define DISK_ROOT       "/dev/dsk"
  53 #define RDISK_ROOT      "/dev/rdsk"
  54 #define BACKUP_SLICE    "s2"
  55
  56 typedef struct prop_flags {
  57         int create:1;   /* Validate property on creation */
  58         int import:1;   /* Validate property on import */
  59 } prop_flags_t;
  60
  61 /*
  62  * ====================================================================
  63  *   zpool property functions
  64  * ====================================================================
  65  */
  66
  67 static int
  68 zpool_get_all_props(zpool_handle_t *zhp)
  69 {
  70         zfs_cmd_t zc = { 0 };
  71         libzfs_handle_t *hdl = zhp->zpool_hdl;
  72
  73         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
  74
  75         if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
  76                 return (-1);
  77
  78         while (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
  79                 if (errno == ENOMEM) {
  80                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
  81                                 zcmd_free_nvlists(&zc);
  82                                 return (-1);
  83                         }
  84                 } else {
  85                         zcmd_free_nvlists(&zc);
  86                         return (-1);
  87                 }
  88         }
  89
  90         if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
  91                 zcmd_free_nvlists(&zc);
  92                 return (-1);
  93         }
  94
  95         zcmd_free_nvlists(&zc);
  96
  97         return (0);
  98 }
  99
 100 static int
 101 zpool_props_refresh(zpool_handle_t *zhp)
 102 {
 103         nvlist_t *old_props;
 104
 105         old_props = zhp->zpool_props;
 106
 107         if (zpool_get_all_props(zhp) != 0)
 108                 return (-1);
 109
 110         nvlist_free(old_props);
 111         return (0);
 112 }
 113
 114 static char *
 115 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
 116     zprop_source_t *src)
 117 {
 118         nvlist_t *nv, *nvl;
 119         uint64_t ival;
 120         char *value;
 121         zprop_source_t source;
 122
 123         nvl = zhp->zpool_props;
 124         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 125                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &ival) == 0);
 126                 source = ival;
 127                 verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
 128         } else {
 129                 source = ZPROP_SRC_DEFAULT;
 130                 if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
 131                         value = "-";
 132         }
 133
 134         if (src)
 135                 *src = source;
 136
 137         return (value);
 138 }
 139
 140 uint64_t
 141 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
 142 {
 143         nvlist_t *nv, *nvl;
 144         uint64_t value;
 145         zprop_source_t source;
 146
 147         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
 148                 /*
 149                  * zpool_get_all_props() has most likely failed because
 150                  * the pool is faulted, but if all we need is the top level
 151                  * vdev's guid then get it from the zhp config nvlist.
 152                  */
 153                 if ((prop == ZPOOL_PROP_GUID) &&
 154                     (nvlist_lookup_nvlist(zhp->zpool_config,
 155                     ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
 156                     (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
 157                     == 0)) {
 158                         return (value);
 159                 }
 160                 return (zpool_prop_default_numeric(prop));
 161         }
 162
 163         nvl = zhp->zpool_props;
 164         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 165                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &value) == 0);
 166                 source = value;
 167                 verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
 168         } else {
 169                 source = ZPROP_SRC_DEFAULT;
 170                 value = zpool_prop_default_numeric(prop);
 171         }
 172
 173         if (src)
 174                 *src = source;
 175
 176         return (value);
 177 }
 178
 179 /*
 180  * Map VDEV STATE to printed strings.
 181  */
 182 const char *
 183 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
 184 {
 185         switch (state) {
 186         case VDEV_STATE_CLOSED:
 187         case VDEV_STATE_OFFLINE:
 188                 return (gettext("OFFLINE"));
 189         case VDEV_STATE_REMOVED:
 190                 return (gettext("REMOVED"));
 191         case VDEV_STATE_CANT_OPEN:
 192                 if (aux == VDEV_AUX_CORRUPT_DATA || aux == VDEV_AUX_BAD_LOG)
 193                         return (gettext("FAULTED"));
 194                 else if (aux == VDEV_AUX_SPLIT_POOL)
 195                         return (gettext("SPLIT"));
 196                 else
 197                         return (gettext("UNAVAIL"));
 198         case VDEV_STATE_FAULTED:
 199                 return (gettext("FAULTED"));
 200         case VDEV_STATE_DEGRADED:
 201                 return (gettext("DEGRADED"));
 202         case VDEV_STATE_HEALTHY:
 203                 return (gettext("ONLINE"));
 204         }
 205
 206         return (gettext("UNKNOWN"));
 207 }
 208
 209 /*
 210  * Map POOL STATE to printed strings.
 211  */
 212 const char *
 213 zpool_pool_state_to_name(pool_state_t state)
 214 {
 215         switch (state) {
 216         case POOL_STATE_ACTIVE:
 217                 return (gettext("ACTIVE"));
 218         case POOL_STATE_EXPORTED:
 219                 return (gettext("EXPORTED"));
 220         case POOL_STATE_DESTROYED:
 221                 return (gettext("DESTROYED"));
 222         case POOL_STATE_SPARE:
 223                 return (gettext("SPARE"));
 224         case POOL_STATE_L2CACHE:
 225                 return (gettext("L2CACHE"));
 226         case POOL_STATE_UNINITIALIZED:
 227                 return (gettext("UNINITIALIZED"));
 228         case POOL_STATE_UNAVAIL:
 229                 return (gettext("UNAVAIL"));
 230         case POOL_STATE_POTENTIALLY_ACTIVE:
 231                 return (gettext("POTENTIALLY_ACTIVE"));
 232         }
 233
 234         return (gettext("UNKNOWN"));
 235 }
 236
 237 /*
 238  * Get a zpool property value for 'prop' and return the value in
 239  * a pre-allocated buffer.
 240  */
 241 int
 242 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf, size_t len,
 243     zprop_source_t *srctype)
 244 {
 245         uint64_t intval;
 246         const char *strval;
 247         zprop_source_t src = ZPROP_SRC_NONE;
 248         nvlist_t *nvroot;
 249         vdev_stat_t *vs;
 250         uint_t vsc;
 251
 252         if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
 253                 switch (prop) {
 254                 case ZPOOL_PROP_NAME:
 255                         (void) strlcpy(buf, zpool_get_name(zhp), len);
 256                         break;
 257
 258                 case ZPOOL_PROP_HEALTH:
 259                         (void) strlcpy(buf, "FAULTED", len);
 260                         break;
 261
 262                 case ZPOOL_PROP_GUID:
 263                         intval = zpool_get_prop_int(zhp, prop, &src);
 264                         (void) snprintf(buf, len, "%llu", intval);
 265                         break;
 266
 267                 case ZPOOL_PROP_ALTROOT:
 268                 case ZPOOL_PROP_CACHEFILE:
 269                 case ZPOOL_PROP_COMMENT:
 270                         if (zhp->zpool_props != NULL ||
 271                             zpool_get_all_props(zhp) == 0) {
 272                                 (void) strlcpy(buf,
 273                                     zpool_get_prop_string(zhp, prop, &src),
 274                                     len);
 275                                 if (srctype != NULL)
 276                                         *srctype = src;
 277                                 return (0);
 278                         }
 279                         /* FALLTHROUGH */
 280                 default:
 281                         (void) strlcpy(buf, "-", len);
 282                         break;
 283                 }
 284
 285                 if (srctype != NULL)
 286                         *srctype = src;
 287                 return (0);
 288         }
 289
 290         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
 291             prop != ZPOOL_PROP_NAME)
 292                 return (-1);
 293
 294         switch (zpool_prop_get_type(prop)) {
 295         case PROP_TYPE_STRING:
 296                 (void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
 297                     len);
 298                 break;
 299
 300         case PROP_TYPE_NUMBER:
 301                 intval = zpool_get_prop_int(zhp, prop, &src);
 302
 303                 switch (prop) {
 304                 case ZPOOL_PROP_SIZE:
 305                 case ZPOOL_PROP_ALLOCATED:
 306                 case ZPOOL_PROP_FREE:
 307                 case ZPOOL_PROP_FREEING:
 308                 case ZPOOL_PROP_EXPANDSZ:
 309                         (void) zfs_nicenum(intval, buf, len);
 310                         break;
 311
 312                 case ZPOOL_PROP_CAPACITY:
 313                         (void) snprintf(buf, len, "%llu%%",
 314                             (u_longlong_t)intval);
 315                         break;
 316
 317                 case ZPOOL_PROP_DEDUPRATIO:
 318                         (void) snprintf(buf, len, "%llu.%02llux",
 319                             (u_longlong_t)(intval / 100),
 320                             (u_longlong_t)(intval % 100));
 321                         break;
 322
 323                 case ZPOOL_PROP_HEALTH:
 324                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 325                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 326                         verify(nvlist_lookup_uint64_array(nvroot,
 327                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
 328                             == 0);
 329
 330                         (void) strlcpy(buf, zpool_state_to_name(intval,
 331                             vs->vs_aux), len);
 332                         break;
 333                 case ZPOOL_PROP_VERSION:
 334                         if (intval >= SPA_VERSION_FEATURES) {
 335                                 (void) snprintf(buf, len, "-");
 336                                 break;
 337                         }
 338                         /* FALLTHROUGH */
 339                 default:
 340                         (void) snprintf(buf, len, "%llu", intval);
 341                 }
 342                 break;
 343
 344         case PROP_TYPE_INDEX:
 345                 intval = zpool_get_prop_int(zhp, prop, &src);
 346                 if (zpool_prop_index_to_string(prop, intval, &strval)
 347                     != 0)
 348                         return (-1);
 349                 (void) strlcpy(buf, strval, len);
 350                 break;
 351
 352         default:
 353                 abort();
 354         }
 355
 356         if (srctype)
 357                 *srctype = src;
 358
 359         return (0);
 360 }
 361
 362 /*
 363  * Check if the bootfs name has the same pool name as it is set to.
 364  * Assuming bootfs is a valid dataset name.
 365  */
 366 static boolean_t
 367 bootfs_name_valid(const char *pool, char *bootfs)
 368 {
 369         int len = strlen(pool);
 370
 371         if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
 372                 return (B_FALSE);
 373
 374         if (strncmp(pool, bootfs, len) == 0 &&
 375             (bootfs[len] == '/' || bootfs[len] == '\0'))
 376                 return (B_TRUE);
 377
 378         return (B_FALSE);
 379 }
 380
 381 /*
 382  * Inspect the configuration to determine if any of the devices contain
 383  * an EFI label.
 384  */
 385 static boolean_t
 386 pool_uses_efi(nvlist_t *config)
 387 {
 388 #ifdef sun
 389         nvlist_t **child;
 390         uint_t c, children;
 391
 392         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
 393             &child, &children) != 0)
 394                 return (read_efi_label(config, NULL) >= 0);
 395
 396         for (c = 0; c < children; c++) {
 397                 if (pool_uses_efi(child[c]))
 398                         return (B_TRUE);
 399         }
 400 #endif  /* sun */
 401         return (B_FALSE);
 402 }
 403
 404 boolean_t
 405 zpool_is_bootable(zpool_handle_t *zhp)
 406 {
 407         char bootfs[ZPOOL_MAXNAMELEN];
 408
 409         return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
 410             sizeof (bootfs), NULL) == 0 && strncmp(bootfs, "-",
 411             sizeof (bootfs)) != 0);
 412 }
 413
 414
 415 /*
 416  * Given an nvlist of zpool properties to be set, validate that they are
 417  * correct, and parse any numeric properties (index, boolean, etc) if they are
 418  * specified as strings.
 419  */
 420 static nvlist_t *
 421 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 422     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
 423 {
 424         nvpair_t *elem;
 425         nvlist_t *retprops;
 426         zpool_prop_t prop;
 427         char *strval;
 428         uint64_t intval;
 429         char *slash, *check;
 430         struct stat64 statbuf;
 431         zpool_handle_t *zhp;
 432         nvlist_t *nvroot;
 433
 434         if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
 435                 (void) no_memory(hdl);
 436                 return (NULL);
 437         }
 438
 439         elem = NULL;
 440         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 441                 const char *propname = nvpair_name(elem);
 442
 443                 prop = zpool_name_to_prop(propname);
 444                 if (prop == ZPROP_INVAL && zpool_prop_feature(propname)) {
 445                         int err;
 446                         zfeature_info_t *feature;
 447                         char *fname = strchr(propname, '@') + 1;
 448
 449                         err = zfeature_lookup_name(fname, &feature);
 450                         if (err != 0) {
 451                                 ASSERT3U(err, ==, ENOENT);
 452                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 453                                     "invalid feature '%s'"), fname);
 454                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 455                                 goto error;
 456                         }
 457
 458                         if (nvpair_type(elem) != DATA_TYPE_STRING) {
 459                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 460                                     "'%s' must be a string"), propname);
 461                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 462                                 goto error;
 463                         }
 464
 465                         (void) nvpair_value_string(elem, &strval);
 466                         if (strcmp(strval, ZFS_FEATURE_ENABLED) != 0) {
 467                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 468                                     "property '%s' can only be set to "
 469                                     "'enabled'"), propname);
 470                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 471                                 goto error;
 472                         }
 473
 474                         if (nvlist_add_uint64(retprops, propname, 0) != 0) {
 475                                 (void) no_memory(hdl);
 476                                 goto error;
 477                         }
 478                         continue;
 479                 }
 480
 481                 /*
 482                  * Make sure this property is valid and applies to this type.
 483                  */
 484                 if (prop == ZPROP_INVAL) {
 485                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 486                             "invalid property '%s'"), propname);
 487                         (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 488                         goto error;
 489                 }
 490
 491                 if (zpool_prop_readonly(prop)) {
 492                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 493                             "is readonly"), propname);
 494                         (void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
 495                         goto error;
 496                 }
 497
 498                 if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
 499                     &strval, &intval, errbuf) != 0)
 500                         goto error;
 501
 502                 /*
 503                  * Perform additional checking for specific properties.
 504                  */
 505                 switch (prop) {
 506                 case ZPOOL_PROP_VERSION:
 507                         if (intval < version ||
 508                             !SPA_VERSION_IS_SUPPORTED(intval)) {
 509                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 510                                     "property '%s' number %d is invalid."),
 511                                     propname, intval);
 512                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 513                                 goto error;
 514                         }
 515                         break;
 516
 517                 case ZPOOL_PROP_BOOTFS:
 518                         if (flags.create || flags.import) {
 519                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 520                                     "property '%s' cannot be set at creation "
 521                                     "or import time"), propname);
 522                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 523                                 goto error;
 524                         }
 525
 526                         if (version < SPA_VERSION_BOOTFS) {
 527                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 528                                     "pool must be upgraded to support "
 529                                     "'%s' property"), propname);
 530                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 531                                 goto error;
 532                         }
 533
 534                         /*
 535                          * bootfs property value has to be a dataset name and
 536                          * the dataset has to be in the same pool as it sets to.
 537                          */
 538                         if (strval[0] != '\0' && !bootfs_name_valid(poolname,
 539                             strval)) {
 540                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 541                                     "is an invalid name"), strval);
 542                                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
 543                                 goto error;
 544                         }
 545
 546                         if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
 547                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 548                                     "could not open pool '%s'"), poolname);
 549                                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
 550                                 goto error;
 551                         }
 552                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 553                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 554
 555 #ifdef sun
 556                         /*
 557                          * bootfs property cannot be set on a disk which has
 558                          * been EFI labeled.
 559                          */
 560                         if (pool_uses_efi(nvroot)) {
 561                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 562                                     "property '%s' not supported on "
 563                                     "EFI labeled devices"), propname);
 564                                 (void) zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf);
 565                                 zpool_close(zhp);
 566                                 goto error;
 567                         }
 568 #endif  /* sun */
 569                         zpool_close(zhp);
 570                         break;
 571
 572                 case ZPOOL_PROP_ALTROOT:
 573                         if (!flags.create && !flags.import) {
 574                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 575                                     "property '%s' can only be set during pool "
 576                                     "creation or import"), propname);
 577                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 578                                 goto error;
 579                         }
 580
 581                         if (strval[0] != '/') {
 582                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 583                                     "bad alternate root '%s'"), strval);
 584                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 585                                 goto error;
 586                         }
 587                         break;
 588
 589                 case ZPOOL_PROP_CACHEFILE:
 590                         if (strval[0] == '\0')
 591                                 break;
 592
 593                         if (strcmp(strval, "none") == 0)
 594                                 break;
 595
 596                         if (strval[0] != '/') {
 597                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 598                                     "property '%s' must be empty, an "
 599                                     "absolute path, or 'none'"), propname);
 600                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 601                                 goto error;
 602                         }
 603
 604                         slash = strrchr(strval, '/');
 605
 606                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 607                             strcmp(slash, "/..") == 0) {
 608                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 609                                     "'%s' is not a valid file"), strval);
 610                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 611                                 goto error;
 612                         }
 613
 614                         *slash = '\0';
 615
 616                         if (strval[0] != '\0' &&
 617                             (stat64(strval, &statbuf) != 0 ||
 618                             !S_ISDIR(statbuf.st_mode))) {
 619                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 620                                     "'%s' is not a valid directory"),
 621                                     strval);
 622                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 623                                 goto error;
 624                         }
 625
 626                         *slash = '/';
 627                         break;
 628
 629                 case ZPOOL_PROP_COMMENT:
 630                         for (check = strval; *check != '\0'; check++) {
 631                                 if (!isprint(*check)) {
 632                                         zfs_error_aux(hdl,
 633                                             dgettext(TEXT_DOMAIN,
 634                                             "comment may only have printable "
 635                                             "characters"));
 636                                         (void) zfs_error(hdl, EZFS_BADPROP,
 637                                             errbuf);
 638                                         goto error;
 639                                 }
 640                         }
 641                         if (strlen(strval) > ZPROP_MAX_COMMENT) {
 642                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 643                                     "comment must not exceed %d characters"),
 644                                     ZPROP_MAX_COMMENT);
 645                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 646                                 goto error;
 647                         }
 648                         break;
 649                 case ZPOOL_PROP_READONLY:
 650                         if (!flags.import) {
 651                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 652                                     "property '%s' can only be set at "
 653                                     "import time"), propname);
 654                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 655                                 goto error;
 656                         }
 657                         break;
 658                 }
 659         }
 660
 661         return (retprops);
 662 error:
 663         nvlist_free(retprops);
 664         return (NULL);
 665 }
 666
 667 /*
 668  * Set zpool property : propname=propval.
 669  */
 670 int
 671 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
 672 {
 673         zfs_cmd_t zc = { 0 };
 674         int ret = -1;
 675         char errbuf[1024];
 676         nvlist_t *nvl = NULL;
 677         nvlist_t *realprops;
 678         uint64_t version;
 679         prop_flags_t flags = { 0 };
 680
 681         (void) snprintf(errbuf, sizeof (errbuf),
 682             dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
 683             zhp->zpool_name);
 684
 685         if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
 686                 return (no_memory(zhp->zpool_hdl));
 687
 688         if (nvlist_add_string(nvl, propname, propval) != 0) {
 689                 nvlist_free(nvl);
 690                 return (no_memory(zhp->zpool_hdl));
 691         }
 692
 693         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
 694         if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
 695             zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
 696                 nvlist_free(nvl);
 697                 return (-1);
 698         }
 699
 700         nvlist_free(nvl);
 701         nvl = realprops;
 702
 703         /*
 704          * Execute the corresponding ioctl() to set this property.
 705          */
 706         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 707
 708         if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
 709                 nvlist_free(nvl);
 710                 return (-1);
 711         }
 712
 713         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
 714
 715         zcmd_free_nvlists(&zc);
 716         nvlist_free(nvl);
 717
 718         if (ret)
 719                 (void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
 720         else
 721                 (void) zpool_props_refresh(zhp);
 722
 723         return (ret);
 724 }
 725
 726 int
 727 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
 728 {
 729         libzfs_handle_t *hdl = zhp->zpool_hdl;
 730         zprop_list_t *entry;
 731         char buf[ZFS_MAXPROPLEN];
 732         nvlist_t *features = NULL;
 733         zprop_list_t **last;
 734         boolean_t firstexpand = (NULL == *plp);
 735
 736         if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
 737                 return (-1);
 738
 739         last = plp;
 740         while (*last != NULL)
 741                 last = &(*last)->pl_next;
 742
 743         if ((*plp)->pl_all)
 744                 features = zpool_get_features(zhp);
 745
 746         if ((*plp)->pl_all && firstexpand) {
 747                 for (int i = 0; i < SPA_FEATURES; i++) {
 748                         zprop_list_t *entry = zfs_alloc(hdl,
 749                             sizeof (zprop_list_t));
 750                         entry->pl_prop = ZPROP_INVAL;
 751                         entry->pl_user_prop = zfs_asprintf(hdl, "feature@%s",
 752                             spa_feature_table[i].fi_uname);
 753                         entry->pl_width = strlen(entry->pl_user_prop);
 754                         entry->pl_all = B_TRUE;
 755
 756                         *last = entry;
 757                         last = &entry->pl_next;
 758                 }
 759         }
 760
 761         /* add any unsupported features */
 762         for (nvpair_t *nvp = nvlist_next_nvpair(features, NULL);
 763             nvp != NULL; nvp = nvlist_next_nvpair(features, nvp)) {
 764                 char *propname;
 765                 boolean_t found;
 766                 zprop_list_t *entry;
 767
 768                 if (zfeature_is_supported(nvpair_name(nvp)))
 769                         continue;
 770
 771                 propname = zfs_asprintf(hdl, "unsupported@%s",
 772                     nvpair_name(nvp));
 773
 774                 /*
 775                  * Before adding the property to the list make sure that no
 776                  * other pool already added the same property.
 777                  */
 778                 found = B_FALSE;
 779                 entry = *plp;
 780                 while (entry != NULL) {
 781                         if (entry->pl_user_prop != NULL &&
 782                             strcmp(propname, entry->pl_user_prop) == 0) {
 783                                 found = B_TRUE;
 784                                 break;
 785                         }
 786                         entry = entry->pl_next;
 787                 }
 788                 if (found) {
 789                         free(propname);
 790                         continue;
 791                 }
 792
 793                 entry = zfs_alloc(hdl, sizeof (zprop_list_t));
 794                 entry->pl_prop = ZPROP_INVAL;
 795                 entry->pl_user_prop = propname;
 796                 entry->pl_width = strlen(entry->pl_user_prop);
 797                 entry->pl_all = B_TRUE;
 798
 799                 *last = entry;
 800                 last = &entry->pl_next;
 801         }
 802
 803         for (entry = *plp; entry != NULL; entry = entry->pl_next) {
 804
 805                 if (entry->pl_fixed)
 806                         continue;
 807
 808                 if (entry->pl_prop != ZPROP_INVAL &&
 809                     zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
 810                     NULL) == 0) {
 811                         if (strlen(buf) > entry->pl_width)
 812                                 entry->pl_width = strlen(buf);
 813                 }
 814         }
 815
 816         return (0);
 817 }
 818
 819 /*
 820  * Get the state for the given feature on the given ZFS pool.
 821  */
 822 int
 823 zpool_prop_get_feature(zpool_handle_t *zhp, const char *propname, char *buf,
 824     size_t len)
 825 {
 826         uint64_t refcount;
 827         boolean_t found = B_FALSE;
 828         nvlist_t *features = zpool_get_features(zhp);
 829         boolean_t supported;
 830         const char *feature = strchr(propname, '@') + 1;
 831
 832         supported = zpool_prop_feature(propname);
 833         ASSERT(supported || zpool_prop_unsupported(propname));
 834
 835         /*
 836          * Convert from feature name to feature guid. This conversion is
 837          * unecessary for unsupported@... properties because they already
 838          * use guids.
 839          */
 840         if (supported) {
 841                 int ret;
 842                 zfeature_info_t *fi;
 843
 844                 ret = zfeature_lookup_name(feature, &fi);
 845                 if (ret != 0) {
 846                         (void) strlcpy(buf, "-", len);
 847                         return (ENOTSUP);
 848                 }
 849                 feature = fi->fi_guid;
 850         }
 851
 852         if (nvlist_lookup_uint64(features, feature, &refcount) == 0)
 853                 found = B_TRUE;
 854
 855         if (supported) {
 856                 if (!found) {
 857                         (void) strlcpy(buf, ZFS_FEATURE_DISABLED, len);
 858                 } else  {
 859                         if (refcount == 0)
 860                                 (void) strlcpy(buf, ZFS_FEATURE_ENABLED, len);
 861                         else
 862                                 (void) strlcpy(buf, ZFS_FEATURE_ACTIVE, len);
 863                 }
 864         } else {
 865                 if (found) {
 866                         if (refcount == 0) {
 867                                 (void) strcpy(buf, ZFS_UNSUPPORTED_INACTIVE);
 868                         } else {
 869                                 (void) strcpy(buf, ZFS_UNSUPPORTED_READONLY);
 870                         }
 871                 } else {
 872                         (void) strlcpy(buf, "-", len);
 873                         return (ENOTSUP);
 874                 }
 875         }
 876
 877         return (0);
 878 }
 879
 880 /*
 881  * Don't start the slice at the default block of 34; many storage
 882  * devices will use a stripe width of 128k, so start there instead.
 883  */
 884 #define NEW_START_BLOCK 256
 885
 886 /*
 887  * Validate the given pool name, optionally putting an extended error message in
 888  * 'buf'.
 889  */
 890 boolean_t
 891 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
 892 {
 893         namecheck_err_t why;
 894         char what;
 895         int ret;
 896
 897         ret = pool_namecheck(pool, &why, &what);
 898
 899         /*
 900          * The rules for reserved pool names were extended at a later point.
 901          * But we need to support users with existing pools that may now be
 902          * invalid.  So we only check for this expanded set of names during a
 903          * create (or import), and only in userland.
 904          */
 905         if (ret == 0 && !isopen &&
 906             (strncmp(pool, "mirror", 6) == 0 ||
 907             strncmp(pool, "raidz", 5) == 0 ||
 908             strncmp(pool, "spare", 5) == 0 ||
 909             strcmp(pool, "log") == 0)) {
 910                 if (hdl != NULL)
 911                         zfs_error_aux(hdl,
 912                             dgettext(TEXT_DOMAIN, "name is reserved"));
 913                 return (B_FALSE);
 914         }
 915
 916
 917         if (ret != 0) {
 918                 if (hdl != NULL) {
 919                         switch (why) {
 920                         case NAME_ERR_TOOLONG:
 921                                 zfs_error_aux(hdl,
 922                                     dgettext(TEXT_DOMAIN, "name is too long"));
 923                                 break;
 924
 925                         case NAME_ERR_INVALCHAR:
 926                                 zfs_error_aux(hdl,
 927                                     dgettext(TEXT_DOMAIN, "invalid character "
 928                                     "'%c' in pool name"), what);
 929                                 break;
 930
 931                         case NAME_ERR_NOLETTER:
 932                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 933                                     "name must begin with a letter"));
 934                                 break;
 935
 936                         case NAME_ERR_RESERVED:
 937                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 938                                     "name is reserved"));
 939                                 break;
 940
 941                         case NAME_ERR_DISKLIKE:
 942                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 943                                     "pool name is reserved"));
 944                                 break;
 945
 946                         case NAME_ERR_LEADING_SLASH:
 947                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 948                                     "leading slash in name"));
 949                                 break;
 950
 951                         case NAME_ERR_EMPTY_COMPONENT:
 952                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 953                                     "empty component in name"));
 954                                 break;
 955
 956                         case NAME_ERR_TRAILING_SLASH:
 957                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 958                                     "trailing slash in name"));
 959                                 break;
 960
 961                         case NAME_ERR_MULTIPLE_AT:
 962                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 963                                     "multiple '@' delimiters in name"));
 964                                 break;
 965
 966                         }
 967                 }
 968                 return (B_FALSE);
 969         }
 970
 971         return (B_TRUE);
 972 }
 973
 974 /*
 975  * Open a handle to the given pool, even if the pool is currently in the FAULTED
 976  * state.
 977  */
 978 zpool_handle_t *
 979 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
 980 {
 981         zpool_handle_t *zhp;
 982         boolean_t missing;
 983
 984         /*
 985          * Make sure the pool name is valid.
 986          */
 987         if (!zpool_name_valid(hdl, B_TRUE, pool)) {
 988                 (void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
 989                     dgettext(TEXT_DOMAIN, "cannot open '%s'"),
 990                     pool);
 991                 return (NULL);
 992         }
 993
 994         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
 995                 return (NULL);
 996
 997         zhp->zpool_hdl = hdl;
 998         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
 999
1000         if (zpool_refresh_stats(zhp, &missing) != 0) {
1001                 zpool_close(zhp);
1002                 return (NULL);
1003         }
1004
1005         if (missing) {
1006                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
1007                 (void) zfs_error_fmt(hdl, EZFS_NOENT,
1008                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
1009                 zpool_close(zhp);
1010                 return (NULL);
1011         }
1012
1013         return (zhp);
1014 }
1015
1016 /*
1017  * Like the above, but silent on error.  Used when iterating over pools (because
1018  * the configuration cache may be out of date).
1019  */
1020 int
1021 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
1022 {
1023         zpool_handle_t *zhp;
1024         boolean_t missing;
1025
1026         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1027                 return (-1);
1028
1029         zhp->zpool_hdl = hdl;
1030         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1031
1032         if (zpool_refresh_stats(zhp, &missing) != 0) {
1033                 zpool_close(zhp);
1034                 return (-1);
1035         }
1036
1037         if (missing) {
1038                 zpool_close(zhp);
1039                 *ret = NULL;
1040                 return (0);
1041         }
1042
1043         *ret = zhp;
1044         return (0);
1045 }
1046
1047 /*
1048  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
1049  * state.
1050  */
1051 zpool_handle_t *
1052 zpool_open(libzfs_handle_t *hdl, const char *pool)
1053 {
1054         zpool_handle_t *zhp;
1055
1056         if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
1057                 return (NULL);
1058
1059         if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
1060                 (void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
1061                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
1062                 zpool_close(zhp);
1063                 return (NULL);
1064         }
1065
1066         return (zhp);
1067 }
1068
1069 /*
1070  * Close the handle.  Simply frees the memory associated with the handle.
1071  */
1072 void
1073 zpool_close(zpool_handle_t *zhp)
1074 {
1075         if (zhp->zpool_config)
1076                 nvlist_free(zhp->zpool_config);
1077         if (zhp->zpool_old_config)
1078                 nvlist_free(zhp->zpool_old_config);
1079         if (zhp->zpool_props)
1080                 nvlist_free(zhp->zpool_props);
1081         free(zhp);
1082 }
1083
1084 /*
1085  * Return the name of the pool.
1086  */
1087 const char *
1088 zpool_get_name(zpool_handle_t *zhp)
1089 {
1090         return (zhp->zpool_name);
1091 }
1092
1093
1094 /*
1095  * Return the state of the pool (ACTIVE or UNAVAILABLE)
1096  */
1097 int
1098 zpool_get_state(zpool_handle_t *zhp)
1099 {
1100         return (zhp->zpool_state);
1101 }
1102
1103 /*
1104  * Create the named pool, using the provided vdev list.  It is assumed
1105  * that the consumer has already validated the contents of the nvlist, so we
1106  * don't have to worry about error semantics.
1107  */
1108 int
1109 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
1110     nvlist_t *props, nvlist_t *fsprops)
1111 {
1112         zfs_cmd_t zc = { 0 };
1113         nvlist_t *zc_fsprops = NULL;
1114         nvlist_t *zc_props = NULL;
1115         char msg[1024];
1116         int ret = -1;
1117
1118         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1119             "cannot create '%s'"), pool);
1120
1121         if (!zpool_name_valid(hdl, B_FALSE, pool))
1122                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
1123
1124         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1125                 return (-1);
1126
1127         if (props) {
1128                 prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
1129
1130                 if ((zc_props = zpool_valid_proplist(hdl, pool, props,
1131                     SPA_VERSION_1, flags, msg)) == NULL) {
1132                         goto create_failed;
1133                 }
1134         }
1135
1136         if (fsprops) {
1137                 uint64_t zoned;
1138                 char *zonestr;
1139
1140                 zoned = ((nvlist_lookup_string(fsprops,
1141                     zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
1142                     strcmp(zonestr, "on") == 0);
1143
1144                 if ((zc_fsprops = zfs_valid_proplist(hdl,
1145                     ZFS_TYPE_FILESYSTEM, fsprops, zoned, NULL, msg)) == NULL) {
1146                         goto create_failed;
1147                 }
1148                 if (!zc_props &&
1149                     (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
1150                         goto create_failed;
1151                 }
1152                 if (nvlist_add_nvlist(zc_props,
1153                     ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
1154                         goto create_failed;
1155                 }
1156         }
1157
1158         if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
1159                 goto create_failed;
1160
1161         (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
1162
1163         if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
1164
1165                 zcmd_free_nvlists(&zc);
1166                 nvlist_free(zc_props);
1167                 nvlist_free(zc_fsprops);
1168
1169                 switch (errno) {
1170                 case EBUSY:
1171                         /*
1172                          * This can happen if the user has specified the same
1173                          * device multiple times.  We can't reliably detect this
1174                          * until we try to add it and see we already have a
1175                          * label.
1176                          */
1177                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1178                             "one or more vdevs refer to the same device"));
1179                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1180
1181                 case EOVERFLOW:
1182                         /*
1183                          * This occurs when one of the devices is below
1184                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1185                          * device was the problem device since there's no
1186                          * reliable way to determine device size from userland.
1187                          */
1188                         {
1189                                 char buf[64];
1190
1191                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1192
1193                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1194                                     "one or more devices is less than the "
1195                                     "minimum size (%s)"), buf);
1196                         }
1197                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1198
1199                 case ENOSPC:
1200                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1201                             "one or more devices is out of space"));
1202                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1203
1204                 case ENOTBLK:
1205                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1206                             "cache device must be a disk or disk slice"));
1207                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1208
1209                 default:
1210                         return (zpool_standard_error(hdl, errno, msg));
1211                 }
1212         }
1213
1214 create_failed:
1215         zcmd_free_nvlists(&zc);
1216         nvlist_free(zc_props);
1217         nvlist_free(zc_fsprops);
1218         return (ret);
1219 }
1220
1221 /*
1222  * Destroy the given pool.  It is up to the caller to ensure that there are no
1223  * datasets left in the pool.
1224  */
1225 int
1226 zpool_destroy(zpool_handle_t *zhp, const char *log_str)
1227 {
1228         zfs_cmd_t zc = { 0 };
1229         zfs_handle_t *zfp = NULL;
1230         libzfs_handle_t *hdl = zhp->zpool_hdl;
1231         char msg[1024];
1232
1233         if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1234             (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1235                 return (-1);
1236
1237         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1238         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1239
1240         if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1241                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1242                     "cannot destroy '%s'"), zhp->zpool_name);
1243
1244                 if (errno == EROFS) {
1245                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1246                             "one or more devices is read only"));
1247                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1248                 } else {
1249                         (void) zpool_standard_error(hdl, errno, msg);
1250                 }
1251
1252                 if (zfp)
1253                         zfs_close(zfp);
1254                 return (-1);
1255         }
1256
1257         if (zfp) {
1258                 remove_mountpoint(zfp);
1259                 zfs_close(zfp);
1260         }
1261
1262         return (0);
1263 }
1264
1265 /*
1266  * Add the given vdevs to the pool.  The caller must have already performed the
1267  * necessary verification to ensure that the vdev specification is well-formed.
1268  */
1269 int
1270 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1271 {
1272         zfs_cmd_t zc = { 0 };
1273         int ret;
1274         libzfs_handle_t *hdl = zhp->zpool_hdl;
1275         char msg[1024];
1276         nvlist_t **spares, **l2cache;
1277         uint_t nspares, nl2cache;
1278
1279         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1280             "cannot add to '%s'"), zhp->zpool_name);
1281
1282         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1283             SPA_VERSION_SPARES &&
1284             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1285             &spares, &nspares) == 0) {
1286                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1287                     "upgraded to add hot spares"));
1288                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1289         }
1290
1291         if (zpool_is_bootable(zhp) && nvlist_lookup_nvlist_array(nvroot,
1292             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) {
1293                 uint64_t s;
1294
1295                 for (s = 0; s < nspares; s++) {
1296                         char *path;
1297
1298                         if (nvlist_lookup_string(spares[s], ZPOOL_CONFIG_PATH,
1299                             &path) == 0 && pool_uses_efi(spares[s])) {
1300                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1301                                     "device '%s' contains an EFI label and "
1302                                     "cannot be used on root pools."),
1303                                     zpool_vdev_name(hdl, NULL, spares[s],
1304                                     B_FALSE));
1305                                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
1306                         }
1307                 }
1308         }
1309
1310         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1311             SPA_VERSION_L2CACHE &&
1312             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1313             &l2cache, &nl2cache) == 0) {
1314                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1315                     "upgraded to add cache devices"));
1316                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1317         }
1318
1319         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1320                 return (-1);
1321         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1322
1323         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1324                 switch (errno) {
1325                 case EBUSY:
1326                         /*
1327                          * This can happen if the user has specified the same
1328                          * device multiple times.  We can't reliably detect this
1329                          * until we try to add it and see we already have a
1330                          * label.
1331                          */
1332                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1333                             "one or more vdevs refer to the same device"));
1334                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1335                         break;
1336
1337                 case EOVERFLOW:
1338                         /*
1339                          * This occurrs when one of the devices is below
1340                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1341                          * device was the problem device since there's no
1342                          * reliable way to determine device size from userland.
1343                          */
1344                         {
1345                                 char buf[64];
1346
1347                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1348
1349                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1350                                     "device is less than the minimum "
1351                                     "size (%s)"), buf);
1352                         }
1353                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1354                         break;
1355
1356                 case ENOTSUP:
1357                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1358                             "pool must be upgraded to add these vdevs"));
1359                         (void) zfs_error(hdl, EZFS_BADVERSION, msg);
1360                         break;
1361
1362                 case EDOM:
1363                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1364                             "root pool can not have multiple vdevs"
1365                             " or separate logs"));
1366                         (void) zfs_error(hdl, EZFS_POOL_NOTSUP, msg);
1367                         break;
1368
1369                 case ENOTBLK:
1370                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1371                             "cache device must be a disk or disk slice"));
1372                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1373                         break;
1374
1375                 default:
1376                         (void) zpool_standard_error(hdl, errno, msg);
1377                 }
1378
1379                 ret = -1;
1380         } else {
1381                 ret = 0;
1382         }
1383
1384         zcmd_free_nvlists(&zc);
1385
1386         return (ret);
1387 }
1388
1389 /*
1390  * Exports the pool from the system.  The caller must ensure that there are no
1391  * mounted datasets in the pool.
1392  */
1393 static int
1394 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce,
1395     const char *log_str)
1396 {
1397         zfs_cmd_t zc = { 0 };
1398         char msg[1024];
1399
1400         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1401             "cannot export '%s'"), zhp->zpool_name);
1402
1403         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1404         zc.zc_cookie = force;
1405         zc.zc_guid = hardforce;
1406         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1407
1408         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1409                 switch (errno) {
1410                 case EXDEV:
1411                         zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1412                             "use '-f' to override the following errors:\n"
1413                             "'%s' has an active shared spare which could be"
1414                             " used by other pools once '%s' is exported."),
1415                             zhp->zpool_name, zhp->zpool_name);
1416                         return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1417                             msg));
1418                 default:
1419                         return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1420                             msg));
1421                 }
1422         }
1423
1424         return (0);
1425 }
1426
1427 int
1428 zpool_export(zpool_handle_t *zhp, boolean_t force, const char *log_str)
1429 {
1430         return (zpool_export_common(zhp, force, B_FALSE, log_str));
1431 }
1432
1433 int
1434 zpool_export_force(zpool_handle_t *zhp, const char *log_str)
1435 {
1436         return (zpool_export_common(zhp, B_TRUE, B_TRUE, log_str));
1437 }
1438
1439 static void
1440 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1441     nvlist_t *config)
1442 {
1443         nvlist_t *nv = NULL;
1444         uint64_t rewindto;
1445         int64_t loss = -1;
1446         struct tm t;
1447         char timestr[128];
1448
1449         if (!hdl->libzfs_printerr || config == NULL)
1450                 return;
1451
1452         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1453             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0) {
1454                 return;
1455         }
1456
1457         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1458                 return;
1459         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1460
1461         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1462             strftime(timestr, 128, 0, &t) != 0) {
1463                 if (dryrun) {
1464                         (void) printf(dgettext(TEXT_DOMAIN,
1465                             "Would be able to return %s "
1466                             "to its state as of %s.\n"),
1467                             name, timestr);
1468                 } else {
1469                         (void) printf(dgettext(TEXT_DOMAIN,
1470                             "Pool %s returned to its state as of %s.\n"),
1471                             name, timestr);
1472                 }
1473                 if (loss > 120) {
1474                         (void) printf(dgettext(TEXT_DOMAIN,
1475                             "%s approximately %lld "),
1476                             dryrun ? "Would discard" : "Discarded",
1477                             (loss + 30) / 60);
1478                         (void) printf(dgettext(TEXT_DOMAIN,
1479                             "minutes of transactions.\n"));
1480                 } else if (loss > 0) {
1481                         (void) printf(dgettext(TEXT_DOMAIN,
1482                             "%s approximately %lld "),
1483                             dryrun ? "Would discard" : "Discarded", loss);
1484                         (void) printf(dgettext(TEXT_DOMAIN,
1485                             "seconds of transactions.\n"));
1486                 }
1487         }
1488 }
1489
1490 void
1491 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1492     nvlist_t *config)
1493 {
1494         nvlist_t *nv = NULL;
1495         int64_t loss = -1;
1496         uint64_t edata = UINT64_MAX;
1497         uint64_t rewindto;
1498         struct tm t;
1499         char timestr[128];
1500
1501         if (!hdl->libzfs_printerr)
1502                 return;
1503
1504         if (reason >= 0)
1505                 (void) printf(dgettext(TEXT_DOMAIN, "action: "));
1506         else
1507                 (void) printf(dgettext(TEXT_DOMAIN, "\t"));
1508
1509         /* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1510         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1511             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0 ||
1512             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1513                 goto no_info;
1514
1515         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1516         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1517             &edata);
1518
1519         (void) printf(dgettext(TEXT_DOMAIN,
1520             "Recovery is possible, but will result in some data loss.\n"));
1521
1522         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1523             strftime(timestr, 128, 0, &t) != 0) {
1524                 (void) printf(dgettext(TEXT_DOMAIN,
1525                     "\tReturning the pool to its state as of %s\n"
1526                     "\tshould correct the problem.  "),
1527                     timestr);
1528         } else {
1529                 (void) printf(dgettext(TEXT_DOMAIN,
1530                     "\tReverting the pool to an earlier state "
1531                     "should correct the problem.\n\t"));
1532         }
1533
1534         if (loss > 120) {
1535                 (void) printf(dgettext(TEXT_DOMAIN,
1536                     "Approximately %lld minutes of data\n"
1537                     "\tmust be discarded, irreversibly.  "), (loss + 30) / 60);
1538         } else if (loss > 0) {
1539                 (void) printf(dgettext(TEXT_DOMAIN,
1540                     "Approximately %lld seconds of data\n"
1541                     "\tmust be discarded, irreversibly.  "), loss);
1542         }
1543         if (edata != 0 && edata != UINT64_MAX) {
1544                 if (edata == 1) {
1545                         (void) printf(dgettext(TEXT_DOMAIN,
1546                             "After rewind, at least\n"
1547                             "\tone persistent user-data error will remain.  "));
1548                 } else {
1549                         (void) printf(dgettext(TEXT_DOMAIN,
1550                             "After rewind, several\n"
1551                             "\tpersistent user-data errors will remain.  "));
1552                 }
1553         }
1554         (void) printf(dgettext(TEXT_DOMAIN,
1555             "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1556             reason >= 0 ? "clear" : "import", name);
1557
1558         (void) printf(dgettext(TEXT_DOMAIN,
1559             "A scrub of the pool\n"
1560             "\tis strongly recommended after recovery.\n"));
1561         return;
1562
1563 no_info:
1564         (void) printf(dgettext(TEXT_DOMAIN,
1565             "Destroy and re-create the pool from\n\ta backup source.\n"));
1566 }
1567
1568 /*
1569  * zpool_import() is a contracted interface. Should be kept the same
1570  * if possible.
1571  *
1572  * Applications should use zpool_import_props() to import a pool with
1573  * new properties value to be set.
1574  */
1575 int
1576 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1577     char *altroot)
1578 {
1579         nvlist_t *props = NULL;
1580         int ret;
1581
1582         if (altroot != NULL) {
1583                 if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1584                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1585                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1586                             newname));
1587                 }
1588
1589                 if (nvlist_add_string(props,
1590                     zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1591                     nvlist_add_string(props,
1592                     zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1593                         nvlist_free(props);
1594                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1595                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1596                             newname));
1597                 }
1598         }
1599
1600         ret = zpool_import_props(hdl, config, newname, props,
1601             ZFS_IMPORT_NORMAL);
1602         if (props)
1603                 nvlist_free(props);
1604         return (ret);
1605 }
1606
1607 static void
1608 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1609     int indent)
1610 {
1611         nvlist_t **child;
1612         uint_t c, children;
1613         char *vname;
1614         uint64_t is_log = 0;
1615
1616         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1617             &is_log);
1618
1619         if (name != NULL)
1620                 (void) printf("\t%*s%s%s\n", indent, "", name,
1621                     is_log ? " [log]" : "");
1622
1623         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1624             &child, &children) != 0)
1625                 return;
1626
1627         for (c = 0; c < children; c++) {
1628                 vname = zpool_vdev_name(hdl, NULL, child[c], B_TRUE);
1629                 print_vdev_tree(hdl, vname, child[c], indent + 2);
1630                 free(vname);
1631         }
1632 }
1633
1634 void
1635 zpool_print_unsup_feat(nvlist_t *config)
1636 {
1637         nvlist_t *nvinfo, *unsup_feat;
1638
1639         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nvinfo) ==
1640             0);
1641         verify(nvlist_lookup_nvlist(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT,
1642             &unsup_feat) == 0);
1643
1644         for (nvpair_t *nvp = nvlist_next_nvpair(unsup_feat, NULL); nvp != NULL;
1645             nvp = nvlist_next_nvpair(unsup_feat, nvp)) {
1646                 char *desc;
1647
1648                 verify(nvpair_type(nvp) == DATA_TYPE_STRING);
1649                 verify(nvpair_value_string(nvp, &desc) == 0);
1650
1651                 if (strlen(desc) > 0)
1652                         (void) printf("\t%s (%s)\n", nvpair_name(nvp), desc);
1653                 else
1654                         (void) printf("\t%s\n", nvpair_name(nvp));
1655         }
1656 }
1657
1658 /*
1659  * Import the given pool using the known configuration and a list of
1660  * properties to be set. The configuration should have come from
1661  * zpool_find_import(). The 'newname' parameters control whether the pool
1662  * is imported with a different name.
1663  */
1664 int
1665 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1666     nvlist_t *props, int flags)
1667 {
1668         zfs_cmd_t zc = { 0 };
1669         zpool_rewind_policy_t policy;
1670         nvlist_t *nv = NULL;
1671         nvlist_t *nvinfo = NULL;
1672         nvlist_t *missing = NULL;
1673         char *thename;
1674         char *origname;
1675         int ret;
1676         int error = 0;
1677         char errbuf[1024];
1678
1679         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1680             &origname) == 0);
1681
1682         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1683             "cannot import pool '%s'"), origname);
1684
1685         if (newname != NULL) {
1686                 if (!zpool_name_valid(hdl, B_FALSE, newname))
1687                         return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1688                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1689                             newname));
1690                 thename = (char *)newname;
1691         } else {
1692                 thename = origname;
1693         }
1694
1695         if (props) {
1696                 uint64_t version;
1697                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1698
1699                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1700                     &version) == 0);
1701
1702                 if ((props = zpool_valid_proplist(hdl, origname,
1703                     props, version, flags, errbuf)) == NULL) {
1704                         return (-1);
1705                 } else if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1706                         nvlist_free(props);
1707                         return (-1);
1708                 }
1709         }
1710
1711         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1712
1713         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1714             &zc.zc_guid) == 0);
1715
1716         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1717                 nvlist_free(props);
1718                 return (-1);
1719         }
1720         if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1721                 nvlist_free(props);
1722                 return (-1);
1723         }
1724
1725         zc.zc_cookie = flags;
1726         while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1727             errno == ENOMEM) {
1728                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1729                         zcmd_free_nvlists(&zc);
1730                         return (-1);
1731                 }
1732         }
1733         if (ret != 0)
1734                 error = errno;
1735
1736         (void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1737         zpool_get_rewind_policy(config, &policy);
1738
1739         if (error) {
1740                 char desc[1024];
1741
1742                 /*
1743                  * Dry-run failed, but we print out what success
1744                  * looks like if we found a best txg
1745                  */
1746                 if (policy.zrp_request & ZPOOL_TRY_REWIND) {
1747                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1748                             B_TRUE, nv);
1749                         nvlist_free(nv);
1750                         return (-1);
1751                 }
1752
1753                 if (newname == NULL)
1754                         (void) snprintf(desc, sizeof (desc),
1755                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1756                             thename);
1757                 else
1758                         (void) snprintf(desc, sizeof (desc),
1759                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1760                             origname, thename);
1761
1762                 switch (error) {
1763                 case ENOTSUP:
1764                         if (nv != NULL && nvlist_lookup_nvlist(nv,
1765                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1766                             nvlist_exists(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT)) {
1767                                 (void) printf(dgettext(TEXT_DOMAIN, "This "
1768                                     "pool uses the following feature(s) not "
1769                                     "supported by this system:\n"));
1770                                 zpool_print_unsup_feat(nv);
1771                                 if (nvlist_exists(nvinfo,
1772                                     ZPOOL_CONFIG_CAN_RDONLY)) {
1773                                         (void) printf(dgettext(TEXT_DOMAIN,
1774                                             "All unsupported features are only "
1775                                             "required for writing to the pool."
1776                                             "\nThe pool can be imported using "
1777                                             "'-o readonly=on'.\n"));
1778                                 }
1779                         }
1780                         /*
1781                          * Unsupported version.
1782                          */
1783                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1784                         break;
1785
1786                 case EINVAL:
1787                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1788                         break;
1789
1790                 case EROFS:
1791                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1792                             "one or more devices is read only"));
1793                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1794                         break;
1795
1796                 case ENXIO:
1797                         if (nv && nvlist_lookup_nvlist(nv,
1798                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1799                             nvlist_lookup_nvlist(nvinfo,
1800                             ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1801                                 (void) printf(dgettext(TEXT_DOMAIN,
1802                                     "The devices below are missing, use "
1803                                     "'-m' to import the pool anyway:\n"));
1804                                 print_vdev_tree(hdl, NULL, missing, 2);
1805                                 (void) printf("\n");
1806                         }
1807                         (void) zpool_standard_error(hdl, error, desc);
1808                         break;
1809
1810                 case EEXIST:
1811                         (void) zpool_standard_error(hdl, error, desc);
1812                         break;
1813
1814                 default:
1815                         (void) zpool_standard_error(hdl, error, desc);
1816                         zpool_explain_recover(hdl,
1817                             newname ? origname : thename, -error, nv);
1818                         break;
1819                 }
1820
1821                 nvlist_free(nv);
1822                 ret = -1;
1823         } else {
1824                 zpool_handle_t *zhp;
1825
1826                 /*
1827                  * This should never fail, but play it safe anyway.
1828                  */
1829                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1830                         ret = -1;
1831                 else if (zhp != NULL)
1832                         zpool_close(zhp);
1833                 if (policy.zrp_request &
1834                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1835                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1836                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1837                 }
1838                 nvlist_free(nv);
1839                 return (0);
1840         }
1841
1842         zcmd_free_nvlists(&zc);
1843         nvlist_free(props);
1844
1845         return (ret);
1846 }
1847
1848 /*
1849  * Scan the pool.
1850  */
1851 int
1852 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1853 {
1854         zfs_cmd_t zc = { 0 };
1855         char msg[1024];
1856         libzfs_handle_t *hdl = zhp->zpool_hdl;
1857
1858         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1859         zc.zc_cookie = func;
1860
1861         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1862             (errno == ENOENT && func != POOL_SCAN_NONE))
1863                 return (0);
1864
1865         if (func == POOL_SCAN_SCRUB) {
1866                 (void) snprintf(msg, sizeof (msg),
1867                     dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1868         } else if (func == POOL_SCAN_NONE) {
1869                 (void) snprintf(msg, sizeof (msg),
1870                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1871                     zc.zc_name);
1872         } else {
1873                 assert(!"unexpected result");
1874         }
1875
1876         if (errno == EBUSY) {
1877                 nvlist_t *nvroot;
1878                 pool_scan_stat_t *ps = NULL;
1879                 uint_t psc;
1880
1881                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
1882                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1883                 (void) nvlist_lookup_uint64_array(nvroot,
1884                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1885                 if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1886                         return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1887                 else
1888                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
1889         } else if (errno == ENOENT) {
1890                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1891         } else {
1892                 return (zpool_standard_error(hdl, errno, msg));
1893         }
1894 }
1895
1896 /*
1897  * This provides a very minimal check whether a given string is likely a
1898  * c#t#d# style string.  Users of this are expected to do their own
1899  * verification of the s# part.
1900  */
1901 #define CTD_CHECK(str)  (str && str[0] == 'c' && isdigit(str[1]))
1902
1903 /*
1904  * More elaborate version for ones which may start with "/dev/dsk/"
1905  * and the like.
1906  */
1907 static int
1908 ctd_check_path(char *str) {
1909         /*
1910          * If it starts with a slash, check the last component.
1911          */
1912         if (str && str[0] == '/') {
1913                 char *tmp = strrchr(str, '/');
1914
1915                 /*
1916                  * If it ends in "/old", check the second-to-last
1917                  * component of the string instead.
1918                  */
1919                 if (tmp != str && strcmp(tmp, "/old") == 0) {
1920                         for (tmp--; *tmp != '/'; tmp--)
1921                                 ;
1922                 }
1923                 str = tmp + 1;
1924         }
1925         return (CTD_CHECK(str));
1926 }
1927
1928 /*
1929  * Find a vdev that matches the search criteria specified. We use the
1930  * the nvpair name to determine how we should look for the device.
1931  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1932  * spare; but FALSE if its an INUSE spare.
1933  */
1934 static nvlist_t *
1935 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1936     boolean_t *l2cache, boolean_t *log)
1937 {
1938         uint_t c, children;
1939         nvlist_t **child;
1940         nvlist_t *ret;
1941         uint64_t is_log;
1942         char *srchkey;
1943         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
1944
1945         /* Nothing to look for */
1946         if (search == NULL || pair == NULL)
1947                 return (NULL);
1948
1949         /* Obtain the key we will use to search */
1950         srchkey = nvpair_name(pair);
1951
1952         switch (nvpair_type(pair)) {
1953         case DATA_TYPE_UINT64:
1954                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
1955                         uint64_t srchval, theguid;
1956
1957                         verify(nvpair_value_uint64(pair, &srchval) == 0);
1958                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
1959                             &theguid) == 0);
1960                         if (theguid == srchval)
1961                                 return (nv);
1962                 }
1963                 break;
1964
1965         case DATA_TYPE_STRING: {
1966                 char *srchval, *val;
1967
1968                 verify(nvpair_value_string(pair, &srchval) == 0);
1969                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
1970                         break;
1971
1972                 /*
1973                  * Search for the requested value. Special cases:
1974                  *
1975                  * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
1976                  *   "s0" or "s0/old".  The "s0" part is hidden from the user,
1977                  *   but included in the string, so this matches around it.
1978                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
1979                  *
1980                  * Otherwise, all other searches are simple string compares.
1981                  */
1982                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0 &&
1983                     ctd_check_path(val)) {
1984                         uint64_t wholedisk = 0;
1985
1986                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1987                             &wholedisk);
1988                         if (wholedisk) {
1989                                 int slen = strlen(srchval);
1990                                 int vlen = strlen(val);
1991
1992                                 if (slen != vlen - 2)
1993                                         break;
1994
1995                                 /*
1996                                  * make_leaf_vdev() should only set
1997                                  * wholedisk for ZPOOL_CONFIG_PATHs which
1998                                  * will include "/dev/dsk/", giving plenty of
1999                                  * room for the indices used next.
2000                                  */
2001                                 ASSERT(vlen >= 6);
2002
2003                                 /*
2004                                  * strings identical except trailing "s0"
2005                                  */
2006                                 if (strcmp(&val[vlen - 2], "s0") == 0 &&
2007                                     strncmp(srchval, val, slen) == 0)
2008                                         return (nv);
2009
2010                                 /*
2011                                  * strings identical except trailing "s0/old"
2012                                  */
2013                                 if (strcmp(&val[vlen - 6], "s0/old") == 0 &&
2014                                     strcmp(&srchval[slen - 4], "/old") == 0 &&
2015                                     strncmp(srchval, val, slen - 4) == 0)
2016                                         return (nv);
2017
2018                                 break;
2019                         }
2020                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2021                         char *type, *idx, *end, *p;
2022                         uint64_t id, vdev_id;
2023
2024                         /*
2025                          * Determine our vdev type, keeping in mind
2026                          * that the srchval is composed of a type and
2027                          * vdev id pair (i.e. mirror-4).
2028                          */
2029                         if ((type = strdup(srchval)) == NULL)
2030                                 return (NULL);
2031
2032                         if ((p = strrchr(type, '-')) == NULL) {
2033                                 free(type);
2034                                 break;
2035                         }
2036                         idx = p + 1;
2037                         *p = '\0';
2038
2039                         /*
2040                          * If the types don't match then keep looking.
2041                          */
2042                         if (strncmp(val, type, strlen(val)) != 0) {
2043                                 free(type);
2044                                 break;
2045                         }
2046
2047                         verify(strncmp(type, VDEV_TYPE_RAIDZ,
2048                             strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2049                             strncmp(type, VDEV_TYPE_MIRROR,
2050                             strlen(VDEV_TYPE_MIRROR)) == 0);
2051                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
2052                             &id) == 0);
2053
2054                         errno = 0;
2055                         vdev_id = strtoull(idx, &end, 10);
2056
2057                         free(type);
2058                         if (errno != 0)
2059                                 return (NULL);
2060
2061                         /*
2062                          * Now verify that we have the correct vdev id.
2063                          */
2064                         if (vdev_id == id)
2065                                 return (nv);
2066                 }
2067
2068                 /*
2069                  * Common case
2070                  */
2071                 if (strcmp(srchval, val) == 0)
2072                         return (nv);
2073                 break;
2074         }
2075
2076         default:
2077                 break;
2078         }
2079
2080         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2081             &child, &children) != 0)
2082                 return (NULL);
2083
2084         for (c = 0; c < children; c++) {
2085                 if ((ret = vdev_to_nvlist_iter(child[c], search,
2086                     avail_spare, l2cache, NULL)) != NULL) {
2087                         /*
2088                          * The 'is_log' value is only set for the toplevel
2089                          * vdev, not the leaf vdevs.  So we always lookup the
2090                          * log device from the root of the vdev tree (where
2091                          * 'log' is non-NULL).
2092                          */
2093                         if (log != NULL &&
2094                             nvlist_lookup_uint64(child[c],
2095                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2096                             is_log) {
2097                                 *log = B_TRUE;
2098                         }
2099                         return (ret);
2100                 }
2101         }
2102
2103         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2104             &child, &children) == 0) {
2105                 for (c = 0; c < children; c++) {
2106                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2107                             avail_spare, l2cache, NULL)) != NULL) {
2108                                 *avail_spare = B_TRUE;
2109                                 return (ret);
2110                         }
2111                 }
2112         }
2113
2114         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2115             &child, &children) == 0) {
2116                 for (c = 0; c < children; c++) {
2117                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2118                             avail_spare, l2cache, NULL)) != NULL) {
2119                                 *l2cache = B_TRUE;
2120                                 return (ret);
2121                         }
2122                 }
2123         }
2124
2125         return (NULL);
2126 }
2127
2128 /*
2129  * Given a physical path (minus the "/devices" prefix), find the
2130  * associated vdev.
2131  */
2132 nvlist_t *
2133 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2134     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2135 {
2136         nvlist_t *search, *nvroot, *ret;
2137
2138         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2139         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
2140
2141         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2142             &nvroot) == 0);
2143
2144         *avail_spare = B_FALSE;
2145         *l2cache = B_FALSE;
2146         if (log != NULL)
2147                 *log = B_FALSE;
2148         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2149         nvlist_free(search);
2150
2151         return (ret);
2152 }
2153
2154 /*
2155  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2156  */
2157 boolean_t
2158 zpool_vdev_is_interior(const char *name)
2159 {
2160         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2161             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2162                 return (B_TRUE);
2163         return (B_FALSE);
2164 }
2165
2166 nvlist_t *
2167 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2168     boolean_t *l2cache, boolean_t *log)
2169 {
2170         char buf[MAXPATHLEN];
2171         char *end;
2172         nvlist_t *nvroot, *search, *ret;
2173         uint64_t guid;
2174
2175         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2176
2177         guid = strtoull(path, &end, 10);
2178         if (guid != 0 && *end == '\0') {
2179                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
2180         } else if (zpool_vdev_is_interior(path)) {
2181                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
2182         } else if (path[0] != '/') {
2183                 (void) snprintf(buf, sizeof (buf), "%s%s", _PATH_DEV, path);
2184                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, buf) == 0);
2185         } else {
2186                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
2187         }
2188
2189         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2190             &nvroot) == 0);
2191
2192         *avail_spare = B_FALSE;
2193         *l2cache = B_FALSE;
2194         if (log != NULL)
2195                 *log = B_FALSE;
2196         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2197         nvlist_free(search);
2198
2199         return (ret);
2200 }
2201
2202 static int
2203 vdev_online(nvlist_t *nv)
2204 {
2205         uint64_t ival;
2206
2207         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2208             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2209             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2210                 return (0);
2211
2212         return (1);
2213 }
2214
2215 /*
2216  * Helper function for zpool_get_physpaths().
2217  */
2218 static int
2219 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2220     size_t *bytes_written)
2221 {
2222         size_t bytes_left, pos, rsz;
2223         char *tmppath;
2224         const char *format;
2225
2226         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2227             &tmppath) != 0)
2228                 return (EZFS_NODEVICE);
2229
2230         pos = *bytes_written;
2231         bytes_left = physpath_size - pos;
2232         format = (pos == 0) ? "%s" : " %s";
2233
2234         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2235         *bytes_written += rsz;
2236
2237         if (rsz >= bytes_left) {
2238                 /* if physpath was not copied properly, clear it */
2239                 if (bytes_left != 0) {
2240                         physpath[pos] = 0;
2241                 }
2242                 return (EZFS_NOSPC);
2243         }
2244         return (0);
2245 }
2246
2247 static int
2248 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2249     size_t *rsz, boolean_t is_spare)
2250 {
2251         char *type;
2252         int ret;
2253
2254         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2255                 return (EZFS_INVALCONFIG);
2256
2257         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2258                 /*
2259                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2260                  * For a spare vdev, we only want to boot from the active
2261                  * spare device.
2262                  */
2263                 if (is_spare) {
2264                         uint64_t spare = 0;
2265                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2266                             &spare);
2267                         if (!spare)
2268                                 return (EZFS_INVALCONFIG);
2269                 }
2270
2271                 if (vdev_online(nv)) {
2272                         if ((ret = vdev_get_one_physpath(nv, physpath,
2273                             phypath_size, rsz)) != 0)
2274                                 return (ret);
2275                 }
2276         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2277             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2278             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2279                 nvlist_t **child;
2280                 uint_t count;
2281                 int i, ret;
2282
2283                 if (nvlist_lookup_nvlist_array(nv,
2284                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2285                         return (EZFS_INVALCONFIG);
2286
2287                 for (i = 0; i < count; i++) {
2288                         ret = vdev_get_physpaths(child[i], physpath,
2289                             phypath_size, rsz, is_spare);
2290                         if (ret == EZFS_NOSPC)
2291                                 return (ret);
2292                 }
2293         }
2294
2295         return (EZFS_POOL_INVALARG);
2296 }
2297
2298 /*
2299  * Get phys_path for a root pool config.
2300  * Return 0 on success; non-zero on failure.
2301  */
2302 static int
2303 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2304 {
2305         size_t rsz;
2306         nvlist_t *vdev_root;
2307         nvlist_t **child;
2308         uint_t count;
2309         char *type;
2310
2311         rsz = 0;
2312
2313         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2314             &vdev_root) != 0)
2315                 return (EZFS_INVALCONFIG);
2316
2317         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2318             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2319             &child, &count) != 0)
2320                 return (EZFS_INVALCONFIG);
2321
2322         /*
2323          * root pool can not have EFI labeled disks and can only have
2324          * a single top-level vdev.
2325          */
2326         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1 ||
2327             pool_uses_efi(vdev_root))
2328                 return (EZFS_POOL_INVALARG);
2329
2330         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2331             B_FALSE);
2332
2333         /* No online devices */
2334         if (rsz == 0)
2335                 return (EZFS_NODEVICE);
2336
2337         return (0);
2338 }
2339
2340 /*
2341  * Get phys_path for a root pool
2342  * Return 0 on success; non-zero on failure.
2343  */
2344 int
2345 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2346 {
2347         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2348             phypath_size));
2349 }
2350
2351 /*
2352  * If the device has being dynamically expanded then we need to relabel
2353  * the disk to use the new unallocated space.
2354  */
2355 static int
2356 zpool_relabel_disk(libzfs_handle_t *hdl, const char *name)
2357 {
2358 #ifdef sun
2359         char path[MAXPATHLEN];
2360         char errbuf[1024];
2361         int fd, error;
2362         int (*_efi_use_whole_disk)(int);
2363
2364         if ((_efi_use_whole_disk = (int (*)(int))dlsym(RTLD_DEFAULT,
2365             "efi_use_whole_disk")) == NULL)
2366                 return (-1);
2367
2368         (void) snprintf(path, sizeof (path), "%s/%s", RDISK_ROOT, name);
2369
2370         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
2371                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2372                     "relabel '%s': unable to open device"), name);
2373                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
2374         }
2375
2376         /*
2377          * It's possible that we might encounter an error if the device
2378          * does not have any unallocated space left. If so, we simply
2379          * ignore that error and continue on.
2380          */
2381         error = _efi_use_whole_disk(fd);
2382         (void) close(fd);
2383         if (error && error != VT_ENOSPC) {
2384                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2385                     "relabel '%s': unable to read disk capacity"), name);
2386                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
2387         }
2388 #endif  /* sun */
2389         return (0);
2390 }
2391
2392 /*
2393  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2394  * ZFS_ONLINE_* flags.
2395  */
2396 int
2397 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2398     vdev_state_t *newstate)
2399 {
2400         zfs_cmd_t zc = { 0 };
2401         char msg[1024];
2402         nvlist_t *tgt;
2403         boolean_t avail_spare, l2cache, islog;
2404         libzfs_handle_t *hdl = zhp->zpool_hdl;
2405
2406         if (flags & ZFS_ONLINE_EXPAND) {
2407                 (void) snprintf(msg, sizeof (msg),
2408                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2409         } else {
2410                 (void) snprintf(msg, sizeof (msg),
2411                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2412         }
2413
2414         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2415         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2416             &islog)) == NULL)
2417                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2418
2419         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2420
2421         if (avail_spare)
2422                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2423
2424         if (flags & ZFS_ONLINE_EXPAND ||
2425             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2426                 char *pathname = NULL;
2427                 uint64_t wholedisk = 0;
2428
2429                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2430                     &wholedisk);
2431                 verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
2432                     &pathname) == 0);
2433
2434                 /*
2435                  * XXX - L2ARC 1.0 devices can't support expansion.
2436                  */
2437                 if (l2cache) {
2438                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2439                             "cannot expand cache devices"));
2440                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2441                 }
2442
2443                 if (wholedisk) {
2444                         pathname += strlen(DISK_ROOT) + 1;
2445                         (void) zpool_relabel_disk(hdl, pathname);
2446                 }
2447         }
2448
2449         zc.zc_cookie = VDEV_STATE_ONLINE;
2450         zc.zc_obj = flags;
2451
2452         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2453                 if (errno == EINVAL) {
2454                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2455                             "from this pool into a new one.  Use '%s' "
2456                             "instead"), "zpool detach");
2457                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2458                 }
2459                 return (zpool_standard_error(hdl, errno, msg));
2460         }
2461
2462         *newstate = zc.zc_cookie;
2463         return (0);
2464 }
2465
2466 /*
2467  * Take the specified vdev offline
2468  */
2469 int
2470 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2471 {
2472         zfs_cmd_t zc = { 0 };
2473         char msg[1024];
2474         nvlist_t *tgt;
2475         boolean_t avail_spare, l2cache;
2476         libzfs_handle_t *hdl = zhp->zpool_hdl;
2477
2478         (void) snprintf(msg, sizeof (msg),
2479             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2480
2481         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2482         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2483             NULL)) == NULL)
2484                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2485
2486         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2487
2488         if (avail_spare)
2489                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2490
2491         zc.zc_cookie = VDEV_STATE_OFFLINE;
2492         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2493
2494         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2495                 return (0);
2496
2497         switch (errno) {
2498         case EBUSY:
2499
2500                 /*
2501                  * There are no other replicas of this device.
2502                  */
2503                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2504
2505         case EEXIST:
2506                 /*
2507                  * The log device has unplayed logs
2508                  */
2509                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2510
2511         default:
2512                 return (zpool_standard_error(hdl, errno, msg));
2513         }
2514 }
2515
2516 /*
2517  * Mark the given vdev faulted.
2518  */
2519 int
2520 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2521 {
2522         zfs_cmd_t zc = { 0 };
2523         char msg[1024];
2524         libzfs_handle_t *hdl = zhp->zpool_hdl;
2525
2526         (void) snprintf(msg, sizeof (msg),
2527             dgettext(TEXT_DOMAIN, "cannot fault %llu"), guid);
2528
2529         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2530         zc.zc_guid = guid;
2531         zc.zc_cookie = VDEV_STATE_FAULTED;
2532         zc.zc_obj = aux;
2533
2534         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2535                 return (0);
2536
2537         switch (errno) {
2538         case EBUSY:
2539
2540                 /*
2541                  * There are no other replicas of this device.
2542                  */
2543                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2544
2545         default:
2546                 return (zpool_standard_error(hdl, errno, msg));
2547         }
2548
2549 }
2550
2551 /*
2552  * Mark the given vdev degraded.
2553  */
2554 int
2555 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2556 {
2557         zfs_cmd_t zc = { 0 };
2558         char msg[1024];
2559         libzfs_handle_t *hdl = zhp->zpool_hdl;
2560
2561         (void) snprintf(msg, sizeof (msg),
2562             dgettext(TEXT_DOMAIN, "cannot degrade %llu"), guid);
2563
2564         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2565         zc.zc_guid = guid;
2566         zc.zc_cookie = VDEV_STATE_DEGRADED;
2567         zc.zc_obj = aux;
2568
2569         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2570                 return (0);
2571
2572         return (zpool_standard_error(hdl, errno, msg));
2573 }
2574
2575 /*
2576  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2577  * a hot spare.
2578  */
2579 static boolean_t
2580 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2581 {
2582         nvlist_t **child;
2583         uint_t c, children;
2584         char *type;
2585
2586         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2587             &children) == 0) {
2588                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2589                     &type) == 0);
2590
2591                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2592                     children == 2 && child[which] == tgt)
2593                         return (B_TRUE);
2594
2595                 for (c = 0; c < children; c++)
2596                         if (is_replacing_spare(child[c], tgt, which))
2597                                 return (B_TRUE);
2598         }
2599
2600         return (B_FALSE);
2601 }
2602
2603 /*
2604  * Attach new_disk (fully described by nvroot) to old_disk.
2605  * If 'replacing' is specified, the new disk will replace the old one.
2606  */
2607 int
2608 zpool_vdev_attach(zpool_handle_t *zhp,
2609     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2610 {
2611         zfs_cmd_t zc = { 0 };
2612         char msg[1024];
2613         int ret;
2614         nvlist_t *tgt;
2615         boolean_t avail_spare, l2cache, islog;
2616         uint64_t val;
2617         char *newname;
2618         nvlist_t **child;
2619         uint_t children;
2620         nvlist_t *config_root;
2621         libzfs_handle_t *hdl = zhp->zpool_hdl;
2622         boolean_t rootpool = zpool_is_bootable(zhp);
2623
2624         if (replacing)
2625                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2626                     "cannot replace %s with %s"), old_disk, new_disk);
2627         else
2628                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2629                     "cannot attach %s to %s"), new_disk, old_disk);
2630
2631         /*
2632          * If this is a root pool, make sure that we're not attaching an
2633          * EFI labeled device.
2634          */
2635         if (rootpool && pool_uses_efi(nvroot)) {
2636                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2637                     "EFI labeled devices are not supported on root pools."));
2638                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
2639         }
2640
2641         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2642         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2643             &islog)) == 0)
2644                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2645
2646         if (avail_spare)
2647                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2648
2649         if (l2cache)
2650                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2651
2652         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2653         zc.zc_cookie = replacing;
2654
2655         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2656             &child, &children) != 0 || children != 1) {
2657                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2658                     "new device must be a single disk"));
2659                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2660         }
2661
2662         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2663             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2664
2665         if ((newname = zpool_vdev_name(NULL, NULL, child[0], B_FALSE)) == NULL)
2666                 return (-1);
2667
2668         /*
2669          * If the target is a hot spare that has been swapped in, we can only
2670          * replace it with another hot spare.
2671          */
2672         if (replacing &&
2673             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2674             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2675             NULL) == NULL || !avail_spare) &&
2676             is_replacing_spare(config_root, tgt, 1)) {
2677                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2678                     "can only be replaced by another hot spare"));
2679                 free(newname);
2680                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2681         }
2682
2683         free(newname);
2684
2685         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2686                 return (-1);
2687
2688         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2689
2690         zcmd_free_nvlists(&zc);
2691
2692         if (ret == 0) {
2693                 if (rootpool) {
2694                         /*
2695                          * XXX need a better way to prevent user from
2696                          * booting up a half-baked vdev.
2697                          */
2698                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2699                             "sure to wait until resilver is done "
2700                             "before rebooting.\n"));
2701                         (void) fprintf(stderr, "\n");
2702                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "If "
2703                             "you boot from pool '%s', you may need to update\n"
2704                             "boot code on newly attached disk '%s'.\n\n"
2705                             "Assuming you use GPT partitioning and 'da0' is "
2706                             "your new boot disk\n"
2707                             "you may use the following command:\n\n"
2708                             "\tgpart bootcode -b /boot/pmbr -p "
2709                             "/boot/gptzfsboot -i 1 da0\n\n"),
2710                             zhp->zpool_name, new_disk);
2711                 }
2712                 return (0);
2713         }
2714
2715         switch (errno) {
2716         case ENOTSUP:
2717                 /*
2718                  * Can't attach to or replace this type of vdev.
2719                  */
2720                 if (replacing) {
2721                         uint64_t version = zpool_get_prop_int(zhp,
2722                             ZPOOL_PROP_VERSION, NULL);
2723
2724                         if (islog)
2725                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2726                                     "cannot replace a log with a spare"));
2727                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2728                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2729                                     "already in replacing/spare config; wait "
2730                                     "for completion or use 'zpool detach'"));
2731                         else
2732                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2733                                     "cannot replace a replacing device"));
2734                 } else {
2735                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2736                             "can only attach to mirrors and top-level "
2737                             "disks"));
2738                 }
2739                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2740                 break;
2741
2742         case EINVAL:
2743                 /*
2744                  * The new device must be a single disk.
2745                  */
2746                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2747                     "new device must be a single disk"));
2748                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2749                 break;
2750
2751         case EBUSY:
2752                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2753                     new_disk);
2754                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2755                 break;
2756
2757         case EOVERFLOW:
2758                 /*
2759                  * The new device is too small.
2760                  */
2761                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2762                     "device is too small"));
2763                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2764                 break;
2765
2766         case EDOM:
2767                 /*
2768                  * The new device has a different alignment requirement.
2769                  */
2770                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2771                     "devices have different sector alignment"));
2772                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2773                 break;
2774
2775         case ENAMETOOLONG:
2776                 /*
2777                  * The resulting top-level vdev spec won't fit in the label.
2778                  */
2779                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2780                 break;
2781
2782         default:
2783                 (void) zpool_standard_error(hdl, errno, msg);
2784         }
2785
2786         return (-1);
2787 }
2788
2789 /*
2790  * Detach the specified device.
2791  */
2792 int
2793 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2794 {
2795         zfs_cmd_t zc = { 0 };
2796         char msg[1024];
2797         nvlist_t *tgt;
2798         boolean_t avail_spare, l2cache;
2799         libzfs_handle_t *hdl = zhp->zpool_hdl;
2800
2801         (void) snprintf(msg, sizeof (msg),
2802             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2803
2804         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2805         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2806             NULL)) == 0)
2807                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2808
2809         if (avail_spare)
2810                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2811
2812         if (l2cache)
2813                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2814
2815         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2816
2817         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2818                 return (0);
2819
2820         switch (errno) {
2821
2822         case ENOTSUP:
2823                 /*
2824                  * Can't detach from this type of vdev.
2825                  */
2826                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2827                     "applicable to mirror and replacing vdevs"));
2828                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2829                 break;
2830
2831         case EBUSY:
2832                 /*
2833                  * There are no other replicas of this device.
2834                  */
2835                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2836                 break;
2837
2838         default:
2839                 (void) zpool_standard_error(hdl, errno, msg);
2840         }
2841
2842         return (-1);
2843 }
2844
2845 /*
2846  * Find a mirror vdev in the source nvlist.
2847  *
2848  * The mchild array contains a list of disks in one of the top-level mirrors
2849  * of the source pool.  The schild array contains a list of disks that the
2850  * user specified on the command line.  We loop over the mchild array to
2851  * see if any entry in the schild array matches.
2852  *
2853  * If a disk in the mchild array is found in the schild array, we return
2854  * the index of that entry.  Otherwise we return -1.
2855  */
2856 static int
2857 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2858     nvlist_t **schild, uint_t schildren)
2859 {
2860         uint_t mc;
2861
2862         for (mc = 0; mc < mchildren; mc++) {
2863                 uint_t sc;
2864                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2865                     mchild[mc], B_FALSE);
2866
2867                 for (sc = 0; sc < schildren; sc++) {
2868                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2869                             schild[sc], B_FALSE);
2870                         boolean_t result = (strcmp(mpath, spath) == 0);
2871
2872                         free(spath);
2873                         if (result) {
2874                                 free(mpath);
2875                                 return (mc);
2876                         }
2877                 }
2878
2879                 free(mpath);
2880         }
2881
2882         return (-1);
2883 }
2884
2885 /*
2886  * Split a mirror pool.  If newroot points to null, then a new nvlist
2887  * is generated and it is the responsibility of the caller to free it.
2888  */
2889 int
2890 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2891     nvlist_t *props, splitflags_t flags)
2892 {
2893         zfs_cmd_t zc = { 0 };
2894         char msg[1024];
2895         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2896         nvlist_t **varray = NULL, *zc_props = NULL;
2897         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2898         libzfs_handle_t *hdl = zhp->zpool_hdl;
2899         uint64_t vers;
2900         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2901         int retval = 0;
2902
2903         (void) snprintf(msg, sizeof (msg),
2904             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2905
2906         if (!zpool_name_valid(hdl, B_FALSE, newname))
2907                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2908
2909         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2910                 (void) fprintf(stderr, gettext("Internal error: unable to "
2911                     "retrieve pool configuration\n"));
2912                 return (-1);
2913         }
2914
2915         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2916             == 0);
2917         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2918
2919         if (props) {
2920                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2921                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2922                     props, vers, flags, msg)) == NULL)
2923                         return (-1);
2924         }
2925
2926         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2927             &children) != 0) {
2928                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2929                     "Source pool is missing vdev tree"));
2930                 if (zc_props)
2931                         nvlist_free(zc_props);
2932                 return (-1);
2933         }
2934
2935         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2936         vcount = 0;
2937
2938         if (*newroot == NULL ||
2939             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2940             &newchild, &newchildren) != 0)
2941                 newchildren = 0;
2942
2943         for (c = 0; c < children; c++) {
2944                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2945                 char *type;
2946                 nvlist_t **mchild, *vdev;
2947                 uint_t mchildren;
2948                 int entry;
2949
2950                 /*
2951                  * Unlike cache & spares, slogs are stored in the
2952                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2953                  */
2954                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2955                     &is_log);
2956                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2957                     &is_hole);
2958                 if (is_log || is_hole) {
2959                         /*
2960                          * Create a hole vdev and put it in the config.
2961                          */
2962                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2963                                 goto out;
2964                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2965                             VDEV_TYPE_HOLE) != 0)
2966                                 goto out;
2967                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2968                             1) != 0)
2969                                 goto out;
2970                         if (lastlog == 0)
2971                                 lastlog = vcount;
2972                         varray[vcount++] = vdev;
2973                         continue;
2974                 }
2975                 lastlog = 0;
2976                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
2977                     == 0);
2978                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
2979                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2980                             "Source pool must be composed only of mirrors\n"));
2981                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2982                         goto out;
2983                 }
2984
2985                 verify(nvlist_lookup_nvlist_array(child[c],
2986                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2987
2988                 /* find or add an entry for this top-level vdev */
2989                 if (newchildren > 0 &&
2990                     (entry = find_vdev_entry(zhp, mchild, mchildren,
2991                     newchild, newchildren)) >= 0) {
2992                         /* We found a disk that the user specified. */
2993                         vdev = mchild[entry];
2994                         ++found;
2995                 } else {
2996                         /* User didn't specify a disk for this vdev. */
2997                         vdev = mchild[mchildren - 1];
2998                 }
2999
3000                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
3001                         goto out;
3002         }
3003
3004         /* did we find every disk the user specified? */
3005         if (found != newchildren) {
3006                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
3007                     "include at most one disk from each mirror"));
3008                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3009                 goto out;
3010         }
3011
3012         /* Prepare the nvlist for populating. */
3013         if (*newroot == NULL) {
3014                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
3015                         goto out;
3016                 freelist = B_TRUE;
3017                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
3018                     VDEV_TYPE_ROOT) != 0)
3019                         goto out;
3020         } else {
3021                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
3022         }
3023
3024         /* Add all the children we found */
3025         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
3026             lastlog == 0 ? vcount : lastlog) != 0)
3027                 goto out;
3028
3029         /*
3030          * If we're just doing a dry run, exit now with success.
3031          */
3032         if (flags.dryrun) {
3033                 memory_err = B_FALSE;
3034                 freelist = B_FALSE;
3035                 goto out;
3036         }
3037
3038         /* now build up the config list & call the ioctl */
3039         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3040                 goto out;
3041
3042         if (nvlist_add_nvlist(newconfig,
3043             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3044             nvlist_add_string(newconfig,
3045             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3046             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3047                 goto out;
3048
3049         /*
3050          * The new pool is automatically part of the namespace unless we
3051          * explicitly export it.
3052          */
3053         if (!flags.import)
3054                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3055         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3056         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3057         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
3058                 goto out;
3059         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
3060                 goto out;
3061
3062         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3063                 retval = zpool_standard_error(hdl, errno, msg);
3064                 goto out;
3065         }
3066
3067         freelist = B_FALSE;
3068         memory_err = B_FALSE;
3069
3070 out:
3071         if (varray != NULL) {
3072                 int v;
3073
3074                 for (v = 0; v < vcount; v++)
3075                         nvlist_free(varray[v]);
3076                 free(varray);
3077         }
3078         zcmd_free_nvlists(&zc);
3079         if (zc_props)
3080                 nvlist_free(zc_props);
3081         if (newconfig)
3082                 nvlist_free(newconfig);
3083         if (freelist) {
3084                 nvlist_free(*newroot);
3085                 *newroot = NULL;
3086         }
3087
3088         if (retval != 0)
3089                 return (retval);
3090
3091         if (memory_err)
3092                 return (no_memory(hdl));
3093
3094         return (0);
3095 }
3096
3097 /*
3098  * Remove the given device.  Currently, this is supported only for hot spares
3099  * and level 2 cache devices.
3100  */
3101 int
3102 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3103 {
3104         zfs_cmd_t zc = { 0 };
3105         char msg[1024];
3106         nvlist_t *tgt;
3107         boolean_t avail_spare, l2cache, islog;
3108         libzfs_handle_t *hdl = zhp->zpool_hdl;
3109         uint64_t version;
3110
3111         (void) snprintf(msg, sizeof (msg),
3112             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3113
3114         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3115         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3116             &islog)) == 0)
3117                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3118         /*
3119          * XXX - this should just go away.
3120          */
3121         if (!avail_spare && !l2cache && !islog) {
3122                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3123                     "only inactive hot spares, cache, top-level, "
3124                     "or log devices can be removed"));
3125                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3126         }
3127
3128         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3129         if (islog && version < SPA_VERSION_HOLES) {
3130                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3131                     "pool must be upgrade to support log removal"));
3132                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
3133         }
3134
3135         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
3136
3137         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3138                 return (0);
3139
3140         return (zpool_standard_error(hdl, errno, msg));
3141 }
3142
3143 /*
3144  * Clear the errors for the pool, or the particular device if specified.
3145  */
3146 int
3147 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3148 {
3149         zfs_cmd_t zc = { 0 };
3150         char msg[1024];
3151         nvlist_t *tgt;
3152         zpool_rewind_policy_t policy;
3153         boolean_t avail_spare, l2cache;
3154         libzfs_handle_t *hdl = zhp->zpool_hdl;
3155         nvlist_t *nvi = NULL;
3156         int error;
3157
3158         if (path)
3159                 (void) snprintf(msg, sizeof (msg),
3160                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3161                     path);
3162         else
3163                 (void) snprintf(msg, sizeof (msg),
3164                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3165                     zhp->zpool_name);
3166
3167         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3168         if (path) {
3169                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3170                     &l2cache, NULL)) == 0)
3171                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
3172
3173                 /*
3174                  * Don't allow error clearing for hot spares.  Do allow
3175                  * error clearing for l2cache devices.
3176                  */
3177                 if (avail_spare)
3178                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
3179
3180                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
3181                     &zc.zc_guid) == 0);
3182         }
3183
3184         zpool_get_rewind_policy(rewindnvl, &policy);
3185         zc.zc_cookie = policy.zrp_request;
3186
3187         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
3188                 return (-1);
3189
3190         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
3191                 return (-1);
3192
3193         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3194             errno == ENOMEM) {
3195                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3196                         zcmd_free_nvlists(&zc);
3197                         return (-1);
3198                 }
3199         }
3200
3201         if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
3202             errno != EPERM && errno != EACCES)) {
3203                 if (policy.zrp_request &
3204                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
3205                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3206                         zpool_rewind_exclaim(hdl, zc.zc_name,
3207                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
3208                             nvi);
3209                         nvlist_free(nvi);
3210                 }
3211                 zcmd_free_nvlists(&zc);
3212                 return (0);
3213         }
3214
3215         zcmd_free_nvlists(&zc);
3216         return (zpool_standard_error(hdl, errno, msg));
3217 }
3218
3219 /*
3220  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3221  */
3222 int
3223 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3224 {
3225         zfs_cmd_t zc = { 0 };
3226         char msg[1024];
3227         libzfs_handle_t *hdl = zhp->zpool_hdl;
3228
3229         (void) snprintf(msg, sizeof (msg),
3230             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3231             guid);
3232
3233         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3234         zc.zc_guid = guid;
3235         zc.zc_cookie = ZPOOL_NO_REWIND;
3236
3237         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
3238                 return (0);
3239
3240         return (zpool_standard_error(hdl, errno, msg));
3241 }
3242
3243 /*
3244  * Change the GUID for a pool.
3245  */
3246 int
3247 zpool_reguid(zpool_handle_t *zhp)
3248 {
3249         char msg[1024];
3250         libzfs_handle_t *hdl = zhp->zpool_hdl;
3251         zfs_cmd_t zc = { 0 };
3252
3253         (void) snprintf(msg, sizeof (msg),
3254             dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3255
3256         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3257         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3258                 return (0);
3259
3260         return (zpool_standard_error(hdl, errno, msg));
3261 }
3262
3263 /*
3264  * Reopen the pool.
3265  */
3266 int
3267 zpool_reopen(zpool_handle_t *zhp)
3268 {
3269         zfs_cmd_t zc = { 0 };
3270         char msg[1024];
3271         libzfs_handle_t *hdl = zhp->zpool_hdl;
3272
3273         (void) snprintf(msg, sizeof (msg),
3274             dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3275             zhp->zpool_name);
3276
3277         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3278         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3279                 return (0);
3280         return (zpool_standard_error(hdl, errno, msg));
3281 }
3282
3283 /*
3284  * Convert from a devid string to a path.
3285  */
3286 static char *
3287 devid_to_path(char *devid_str)
3288 {
3289         ddi_devid_t devid;
3290         char *minor;
3291         char *path;
3292         devid_nmlist_t *list = NULL;
3293         int ret;
3294
3295         if (devid_str_decode(devid_str, &devid, &minor) != 0)
3296                 return (NULL);
3297
3298         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3299
3300         devid_str_free(minor);
3301         devid_free(devid);
3302
3303         if (ret != 0)
3304                 return (NULL);
3305
3306         if ((path = strdup(list[0].devname)) == NULL)
3307                 return (NULL);
3308
3309         devid_free_nmlist(list);
3310
3311         return (path);
3312 }
3313
3314 /*
3315  * Convert from a path to a devid string.
3316  */
3317 static char *
3318 path_to_devid(const char *path)
3319 {
3320         int fd;
3321         ddi_devid_t devid;
3322         char *minor, *ret;
3323
3324         if ((fd = open(path, O_RDONLY)) < 0)
3325                 return (NULL);
3326
3327         minor = NULL;
3328         ret = NULL;
3329         if (devid_get(fd, &devid) == 0) {
3330                 if (devid_get_minor_name(fd, &minor) == 0)
3331                         ret = devid_str_encode(devid, minor);
3332                 if (minor != NULL)
3333                         devid_str_free(minor);
3334                 devid_free(devid);
3335         }
3336         (void) close(fd);
3337
3338         return (ret);
3339 }
3340
3341 /*
3342  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3343  * ignore any failure here, since a common case is for an unprivileged user to
3344  * type 'zpool status', and we'll display the correct information anyway.
3345  */
3346 static void
3347 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3348 {
3349         zfs_cmd_t zc = { 0 };
3350
3351         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3352         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3353         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3354             &zc.zc_guid) == 0);
3355
3356         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3357 }
3358
3359 /*
3360  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3361  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3362  * We also check if this is a whole disk, in which case we strip off the
3363  * trailing 's0' slice name.
3364  *
3365  * This routine is also responsible for identifying when disks have been
3366  * reconfigured in a new location.  The kernel will have opened the device by
3367  * devid, but the path will still refer to the old location.  To catch this, we
3368  * first do a path -> devid translation (which is fast for the common case).  If
3369  * the devid matches, we're done.  If not, we do a reverse devid -> path
3370  * translation and issue the appropriate ioctl() to update the path of the vdev.
3371  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3372  * of these checks.
3373  */
3374 char *
3375 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3376     boolean_t verbose)
3377 {
3378         char *path, *devid;
3379         uint64_t value;
3380         char buf[64];
3381         vdev_stat_t *vs;
3382         uint_t vsc;
3383         int have_stats;
3384         int have_path;
3385
3386         have_stats = nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3387             (uint64_t **)&vs, &vsc) == 0;
3388         have_path = nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0;
3389
3390         /*
3391          * If the device is not currently present, assume it will not
3392          * come back at the same device path.  Display the device by GUID.
3393          */
3394         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &value) == 0 ||
3395             have_path && have_stats && vs->vs_state <= VDEV_STATE_CANT_OPEN) {
3396                 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3397                     &value) == 0);
3398                 (void) snprintf(buf, sizeof (buf), "%llu",
3399                     (u_longlong_t)value);
3400                 path = buf;
3401         } else if (have_path) {
3402
3403                 /*
3404                  * If the device is dead (faulted, offline, etc) then don't
3405                  * bother opening it.  Otherwise we may be forcing the user to
3406                  * open a misbehaving device, which can have undesirable
3407                  * effects.
3408                  */
3409                 if ((have_stats == 0 ||
3410                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3411                     zhp != NULL &&
3412                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3413                         /*
3414                          * Determine if the current path is correct.
3415                          */
3416                         char *newdevid = path_to_devid(path);
3417
3418                         if (newdevid == NULL ||
3419                             strcmp(devid, newdevid) != 0) {
3420                                 char *newpath;
3421
3422                                 if ((newpath = devid_to_path(devid)) != NULL) {
3423                                         /*
3424                                          * Update the path appropriately.
3425                                          */
3426                                         set_path(zhp, nv, newpath);
3427                                         if (nvlist_add_string(nv,
3428                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3429                                                 verify(nvlist_lookup_string(nv,
3430                                                     ZPOOL_CONFIG_PATH,
3431                                                     &path) == 0);
3432                                         free(newpath);
3433                                 }
3434                         }
3435
3436                         if (newdevid)
3437                                 devid_str_free(newdevid);
3438                 }
3439
3440 #ifdef sun
3441                 if (strncmp(path, "/dev/dsk/", 9) == 0)
3442                         path += 9;
3443
3444                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
3445                     &value) == 0 && value) {
3446                         int pathlen = strlen(path);
3447                         char *tmp = zfs_strdup(hdl, path);
3448
3449                         /*
3450                          * If it starts with c#, and ends with "s0", chop
3451                          * the "s0" off, or if it ends with "s0/old", remove
3452                          * the "s0" from the middle.
3453                          */
3454                         if (CTD_CHECK(tmp)) {
3455                                 if (strcmp(&tmp[pathlen - 2], "s0") == 0) {
3456                                         tmp[pathlen - 2] = '\0';
3457                                 } else if (pathlen > 6 &&
3458                                     strcmp(&tmp[pathlen - 6], "s0/old") == 0) {
3459                                         (void) strcpy(&tmp[pathlen - 6],
3460                                             "/old");
3461                                 }
3462                         }
3463                         return (tmp);
3464                 }
3465 #else   /* !sun */
3466                 if (strncmp(path, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
3467                         path += sizeof(_PATH_DEV) - 1;
3468 #endif  /* !sun */
3469         } else {
3470                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3471
3472                 /*
3473                  * If it's a raidz device, we need to stick in the parity level.
3474                  */
3475                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3476                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3477                             &value) == 0);
3478                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3479                             (u_longlong_t)value);
3480                         path = buf;
3481                 }
3482
3483                 /*
3484                  * We identify each top-level vdev by using a <type-id>
3485                  * naming convention.
3486                  */
3487                 if (verbose) {
3488                         uint64_t id;
3489
3490                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3491                             &id) == 0);
3492                         (void) snprintf(buf, sizeof (buf), "%s-%llu", path,
3493                             (u_longlong_t)id);
3494                         path = buf;
3495                 }
3496         }
3497
3498         return (zfs_strdup(hdl, path));
3499 }
3500
3501 static int
3502 zbookmark_compare(const void *a, const void *b)
3503 {
3504         return (memcmp(a, b, sizeof (zbookmark_t)));
3505 }
3506
3507 /*
3508  * Retrieve the persistent error log, uniquify the members, and return to the
3509  * caller.
3510  */
3511 int
3512 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3513 {
3514         zfs_cmd_t zc = { 0 };
3515         uint64_t count;
3516         zbookmark_t *zb = NULL;
3517         int i;
3518
3519         /*
3520          * Retrieve the raw error list from the kernel.  If the number of errors
3521          * has increased, allocate more space and continue until we get the
3522          * entire list.
3523          */
3524         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3525             &count) == 0);
3526         if (count == 0)
3527                 return (0);
3528         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3529             count * sizeof (zbookmark_t))) == (uintptr_t)NULL)
3530                 return (-1);
3531         zc.zc_nvlist_dst_size = count;
3532         (void) strcpy(zc.zc_name, zhp->zpool_name);
3533         for (;;) {
3534                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3535                     &zc) != 0) {
3536                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3537                         if (errno == ENOMEM) {
3538                                 count = zc.zc_nvlist_dst_size;
3539                                 if ((zc.zc_nvlist_dst = (uintptr_t)
3540                                     zfs_alloc(zhp->zpool_hdl, count *
3541                                     sizeof (zbookmark_t))) == (uintptr_t)NULL)
3542                                         return (-1);
3543                         } else {
3544                                 return (-1);
3545                         }
3546                 } else {
3547                         break;
3548                 }
3549         }
3550
3551         /*
3552          * Sort the resulting bookmarks.  This is a little confusing due to the
3553          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3554          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3555          * _not_ copied as part of the process.  So we point the start of our
3556          * array appropriate and decrement the total number of elements.
3557          */
3558         zb = ((zbookmark_t *)(uintptr_t)zc.zc_nvlist_dst) +
3559             zc.zc_nvlist_dst_size;
3560         count -= zc.zc_nvlist_dst_size;
3561
3562         qsort(zb, count, sizeof (zbookmark_t), zbookmark_compare);
3563
3564         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3565
3566         /*
3567          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3568          */
3569         for (i = 0; i < count; i++) {
3570                 nvlist_t *nv;
3571
3572                 /* ignoring zb_blkid and zb_level for now */
3573                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3574                     zb[i-1].zb_object == zb[i].zb_object)
3575                         continue;
3576
3577                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3578                         goto nomem;
3579                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3580                     zb[i].zb_objset) != 0) {
3581                         nvlist_free(nv);
3582                         goto nomem;
3583                 }
3584                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3585                     zb[i].zb_object) != 0) {
3586                         nvlist_free(nv);
3587                         goto nomem;
3588                 }
3589                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3590                         nvlist_free(nv);
3591                         goto nomem;
3592                 }
3593                 nvlist_free(nv);
3594         }
3595
3596         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3597         return (0);
3598
3599 nomem:
3600         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3601         return (no_memory(zhp->zpool_hdl));
3602 }
3603
3604 /*
3605  * Upgrade a ZFS pool to the latest on-disk version.
3606  */
3607 int
3608 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3609 {
3610         zfs_cmd_t zc = { 0 };
3611         libzfs_handle_t *hdl = zhp->zpool_hdl;
3612
3613         (void) strcpy(zc.zc_name, zhp->zpool_name);
3614         zc.zc_cookie = new_version;
3615
3616         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3617                 return (zpool_standard_error_fmt(hdl, errno,
3618                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3619                     zhp->zpool_name));
3620         return (0);
3621 }
3622
3623 void
3624 zfs_save_arguments(int argc, char **argv, char *string, int len)
3625 {
3626         (void) strlcpy(string, basename(argv[0]), len);
3627         for (int i = 1; i < argc; i++) {
3628                 (void) strlcat(string, " ", len);
3629                 (void) strlcat(string, argv[i], len);
3630         }
3631 }
3632
3633 int
3634 zpool_log_history(libzfs_handle_t *hdl, const char *message)
3635 {
3636         zfs_cmd_t zc = { 0 };
3637         nvlist_t *args;
3638         int err;
3639
3640         args = fnvlist_alloc();
3641         fnvlist_add_string(args, "message", message);
3642         err = zcmd_write_src_nvlist(hdl, &zc, args);
3643         if (err == 0)
3644                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_LOG_HISTORY, &zc);
3645         nvlist_free(args);
3646         zcmd_free_nvlists(&zc);
3647         return (err);
3648 }
3649
3650 /*
3651  * Perform ioctl to get some command history of a pool.
3652  *
3653  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3654  * logical offset of the history buffer to start reading from.
3655  *
3656  * Upon return, 'off' is the next logical offset to read from and
3657  * 'len' is the actual amount of bytes read into 'buf'.
3658  */
3659 static int
3660 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3661 {
3662         zfs_cmd_t zc = { 0 };
3663         libzfs_handle_t *hdl = zhp->zpool_hdl;
3664
3665         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3666
3667         zc.zc_history = (uint64_t)(uintptr_t)buf;
3668         zc.zc_history_len = *len;
3669         zc.zc_history_offset = *off;
3670
3671         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3672                 switch (errno) {
3673                 case EPERM:
3674                         return (zfs_error_fmt(hdl, EZFS_PERM,
3675                             dgettext(TEXT_DOMAIN,
3676                             "cannot show history for pool '%s'"),
3677                             zhp->zpool_name));
3678                 case ENOENT:
3679                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3680                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3681                             "'%s'"), zhp->zpool_name));
3682                 case ENOTSUP:
3683                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3684                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3685                             "'%s', pool must be upgraded"), zhp->zpool_name));
3686                 default:
3687                         return (zpool_standard_error_fmt(hdl, errno,
3688                             dgettext(TEXT_DOMAIN,
3689                             "cannot get history for '%s'"), zhp->zpool_name));
3690                 }
3691         }
3692
3693         *len = zc.zc_history_len;
3694         *off = zc.zc_history_offset;
3695
3696         return (0);
3697 }
3698
3699 /*
3700  * Process the buffer of nvlists, unpacking and storing each nvlist record
3701  * into 'records'.  'leftover' is set to the number of bytes that weren't
3702  * processed as there wasn't a complete record.
3703  */
3704 int
3705 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3706     nvlist_t ***records, uint_t *numrecords)
3707 {
3708         uint64_t reclen;
3709         nvlist_t *nv;
3710         int i;
3711
3712         while (bytes_read > sizeof (reclen)) {
3713
3714                 /* get length of packed record (stored as little endian) */
3715                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3716                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3717
3718                 if (bytes_read < sizeof (reclen) + reclen)
3719                         break;
3720
3721                 /* unpack record */
3722                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3723                         return (ENOMEM);
3724                 bytes_read -= sizeof (reclen) + reclen;
3725                 buf += sizeof (reclen) + reclen;
3726
3727                 /* add record to nvlist array */
3728                 (*numrecords)++;
3729                 if (ISP2(*numrecords + 1)) {
3730                         *records = realloc(*records,
3731                             *numrecords * 2 * sizeof (nvlist_t *));
3732                 }
3733                 (*records)[*numrecords - 1] = nv;
3734         }
3735
3736         *leftover = bytes_read;
3737         return (0);
3738 }
3739
3740 #define HIS_BUF_LEN     (128*1024)
3741
3742 /*
3743  * Retrieve the command history of a pool.
3744  */
3745 int
3746 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3747 {
3748         char buf[HIS_BUF_LEN];
3749         uint64_t off = 0;
3750         nvlist_t **records = NULL;
3751         uint_t numrecords = 0;
3752         int err, i;
3753
3754         do {
3755                 uint64_t bytes_read = sizeof (buf);
3756                 uint64_t leftover;
3757
3758                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3759                         break;
3760
3761                 /* if nothing else was read in, we're at EOF, just return */
3762                 if (!bytes_read)
3763                         break;
3764
3765                 if ((err = zpool_history_unpack(buf, bytes_read,
3766                     &leftover, &records, &numrecords)) != 0)
3767                         break;
3768                 off -= leftover;
3769
3770                 /* CONSTCOND */
3771         } while (1);
3772
3773         if (!err) {
3774                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3775                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3776                     records, numrecords) == 0);
3777         }
3778         for (i = 0; i < numrecords; i++)
3779                 nvlist_free(records[i]);
3780         free(records);
3781
3782         return (err);
3783 }
3784
3785 void
3786 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
3787     char *pathname, size_t len)
3788 {
3789         zfs_cmd_t zc = { 0 };
3790         boolean_t mounted = B_FALSE;
3791         char *mntpnt = NULL;
3792         char dsname[MAXNAMELEN];
3793
3794         if (dsobj == 0) {
3795                 /* special case for the MOS */
3796                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>", obj);
3797                 return;
3798         }
3799
3800         /* get the dataset's name */
3801         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3802         zc.zc_obj = dsobj;
3803         if (ioctl(zhp->zpool_hdl->libzfs_fd,
3804             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
3805                 /* just write out a path of two object numbers */
3806                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
3807                     dsobj, obj);
3808                 return;
3809         }
3810         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
3811
3812         /* find out if the dataset is mounted */
3813         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
3814
3815         /* get the corrupted object's path */
3816         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
3817         zc.zc_obj = obj;
3818         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
3819             &zc) == 0) {
3820                 if (mounted) {
3821                         (void) snprintf(pathname, len, "%s%s", mntpnt,
3822                             zc.zc_value);
3823                 } else {
3824                         (void) snprintf(pathname, len, "%s:%s",
3825                             dsname, zc.zc_value);
3826                 }
3827         } else {
3828                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname, obj);
3829         }
3830         free(mntpnt);
3831 }
3832
3833 #ifdef sun
3834 /*
3835  * Read the EFI label from the config, if a label does not exist then
3836  * pass back the error to the caller. If the caller has passed a non-NULL
3837  * diskaddr argument then we set it to the starting address of the EFI
3838  * partition.
3839  */
3840 static int
3841 read_efi_label(nvlist_t *config, diskaddr_t *sb)
3842 {
3843         char *path;
3844         int fd;
3845         char diskname[MAXPATHLEN];
3846         int err = -1;
3847
3848         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
3849                 return (err);
3850
3851         (void) snprintf(diskname, sizeof (diskname), "%s%s", RDISK_ROOT,
3852             strrchr(path, '/'));
3853         if ((fd = open(diskname, O_RDONLY|O_NDELAY)) >= 0) {
3854                 struct dk_gpt *vtoc;
3855
3856                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
3857                         if (sb != NULL)
3858                                 *sb = vtoc->efi_parts[0].p_start;
3859                         efi_free(vtoc);
3860                 }
3861                 (void) close(fd);
3862         }
3863         return (err);
3864 }
3865
3866 /*
3867  * determine where a partition starts on a disk in the current
3868  * configuration
3869  */
3870 static diskaddr_t
3871 find_start_block(nvlist_t *config)
3872 {
3873         nvlist_t **child;
3874         uint_t c, children;
3875         diskaddr_t sb = MAXOFFSET_T;
3876         uint64_t wholedisk;
3877
3878         if (nvlist_lookup_nvlist_array(config,
3879             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
3880                 if (nvlist_lookup_uint64(config,
3881                     ZPOOL_CONFIG_WHOLE_DISK,
3882                     &wholedisk) != 0 || !wholedisk) {
3883                         return (MAXOFFSET_T);
3884                 }
3885                 if (read_efi_label(config, &sb) < 0)
3886                         sb = MAXOFFSET_T;
3887                 return (sb);
3888         }
3889
3890         for (c = 0; c < children; c++) {
3891                 sb = find_start_block(child[c]);
3892                 if (sb != MAXOFFSET_T) {
3893                         return (sb);
3894                 }
3895         }
3896         return (MAXOFFSET_T);
3897 }
3898 #endif /* sun */
3899
3900 /*
3901  * Label an individual disk.  The name provided is the short name,
3902  * stripped of any leading /dev path.
3903  */
3904 int
3905 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, const char *name)
3906 {
3907 #ifdef sun
3908         char path[MAXPATHLEN];
3909         struct dk_gpt *vtoc;
3910         int fd;
3911         size_t resv = EFI_MIN_RESV_SIZE;
3912         uint64_t slice_size;
3913         diskaddr_t start_block;
3914         char errbuf[1024];
3915
3916         /* prepare an error message just in case */
3917         (void) snprintf(errbuf, sizeof (errbuf),
3918             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
3919
3920         if (zhp) {
3921                 nvlist_t *nvroot;
3922
3923                 if (zpool_is_bootable(zhp)) {
3924                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3925                             "EFI labeled devices are not supported on root "
3926                             "pools."));
3927                         return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
3928                 }
3929
3930                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
3931                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3932
3933                 if (zhp->zpool_start_block == 0)
3934                         start_block = find_start_block(nvroot);
3935                 else
3936                         start_block = zhp->zpool_start_block;
3937                 zhp->zpool_start_block = start_block;
3938         } else {
3939                 /* new pool */
3940                 start_block = NEW_START_BLOCK;
3941         }
3942
3943         (void) snprintf(path, sizeof (path), "%s/%s%s", RDISK_ROOT, name,
3944             BACKUP_SLICE);
3945
3946         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
3947                 /*
3948                  * This shouldn't happen.  We've long since verified that this
3949                  * is a valid device.
3950                  */
3951                 zfs_error_aux(hdl,
3952                     dgettext(TEXT_DOMAIN, "unable to open device"));
3953                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
3954         }
3955
3956         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
3957                 /*
3958                  * The only way this can fail is if we run out of memory, or we
3959                  * were unable to read the disk's capacity
3960                  */
3961                 if (errno == ENOMEM)
3962                         (void) no_memory(hdl);
3963
3964                 (void) close(fd);
3965                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3966                     "unable to read disk capacity"), name);
3967
3968                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
3969         }
3970
3971         slice_size = vtoc->efi_last_u_lba + 1;
3972         slice_size -= EFI_MIN_RESV_SIZE;
3973         if (start_block == MAXOFFSET_T)
3974                 start_block = NEW_START_BLOCK;
3975         slice_size -= start_block;
3976
3977         vtoc->efi_parts[0].p_start = start_block;
3978         vtoc->efi_parts[0].p_size = slice_size;
3979
3980         /*
3981          * Why we use V_USR: V_BACKUP confuses users, and is considered
3982          * disposable by some EFI utilities (since EFI doesn't have a backup
3983          * slice).  V_UNASSIGNED is supposed to be used only for zero size
3984          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
3985          * etc. were all pretty specific.  V_USR is as close to reality as we
3986          * can get, in the absence of V_OTHER.
3987          */
3988         vtoc->efi_parts[0].p_tag = V_USR;
3989         (void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
3990
3991         vtoc->efi_parts[8].p_start = slice_size + start_block;
3992         vtoc->efi_parts[8].p_size = resv;
3993         vtoc->efi_parts[8].p_tag = V_RESERVED;
3994
3995         if (efi_write(fd, vtoc) != 0) {
3996                 /*
3997                  * Some block drivers (like pcata) may not support EFI
3998                  * GPT labels.  Print out a helpful error message dir-
3999                  * ecting the user to manually label the disk and give
4000                  * a specific slice.
4001                  */
4002                 (void) close(fd);
4003                 efi_free(vtoc);
4004
4005                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4006                     "try using fdisk(1M) and then provide a specific slice"));
4007                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4008         }
4009
4010         (void) close(fd);
4011         efi_free(vtoc);
4012 #endif /* sun */
4013         return (0);
4014 }
4015
4016 static boolean_t
4017 supported_dump_vdev_type(libzfs_handle_t *hdl, nvlist_t *config, char *errbuf)
4018 {
4019         char *type;
4020         nvlist_t **child;
4021         uint_t children, c;
4022
4023         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_TYPE, &type) == 0);
4024         if (strcmp(type, VDEV_TYPE_FILE) == 0 ||
4025             strcmp(type, VDEV_TYPE_HOLE) == 0 ||
4026             strcmp(type, VDEV_TYPE_MISSING) == 0) {
4027                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4028                     "vdev type '%s' is not supported"), type);
4029                 (void) zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf);
4030                 return (B_FALSE);
4031         }
4032         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
4033             &child, &children) == 0) {
4034                 for (c = 0; c < children; c++) {
4035                         if (!supported_dump_vdev_type(hdl, child[c], errbuf))
4036                                 return (B_FALSE);
4037                 }
4038         }
4039         return (B_TRUE);
4040 }
4041
4042 /*
4043  * Check if this zvol is allowable for use as a dump device; zero if
4044  * it is, > 0 if it isn't, < 0 if it isn't a zvol.
4045  *
4046  * Allowable storage configurations include mirrors, all raidz variants, and
4047  * pools with log, cache, and spare devices.  Pools which are backed by files or
4048  * have missing/hole vdevs are not suitable.
4049  */
4050 int
4051 zvol_check_dump_config(char *arg)
4052 {
4053         zpool_handle_t *zhp = NULL;
4054         nvlist_t *config, *nvroot;
4055         char *p, *volname;
4056         nvlist_t **top;
4057         uint_t toplevels;
4058         libzfs_handle_t *hdl;
4059         char errbuf[1024];
4060         char poolname[ZPOOL_MAXNAMELEN];
4061         int pathlen = strlen(ZVOL_FULL_DEV_DIR);
4062         int ret = 1;
4063
4064         if (strncmp(arg, ZVOL_FULL_DEV_DIR, pathlen)) {
4065                 return (-1);
4066         }
4067
4068         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
4069             "dump is not supported on device '%s'"), arg);
4070
4071         if ((hdl = libzfs_init()) == NULL)
4072                 return (1);
4073         libzfs_print_on_error(hdl, B_TRUE);
4074
4075         volname = arg + pathlen;
4076
4077         /* check the configuration of the pool */
4078         if ((p = strchr(volname, '/')) == NULL) {
4079                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4080                     "malformed dataset name"));
4081                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4082                 return (1);
4083         } else if (p - volname >= ZFS_MAXNAMELEN) {
4084                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4085                     "dataset name is too long"));
4086                 (void) zfs_error(hdl, EZFS_NAMETOOLONG, errbuf);
4087                 return (1);
4088         } else {
4089                 (void) strncpy(poolname, volname, p - volname);
4090                 poolname[p - volname] = '\0';
4091         }
4092
4093         if ((zhp = zpool_open(hdl, poolname)) == NULL) {
4094                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4095                     "could not open pool '%s'"), poolname);
4096                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
4097                 goto out;
4098         }
4099         config = zpool_get_config(zhp, NULL);
4100         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4101             &nvroot) != 0) {
4102                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4103                     "could not obtain vdev configuration for  '%s'"), poolname);
4104                 (void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
4105                 goto out;
4106         }
4107
4108         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4109             &top, &toplevels) == 0);
4110
4111         if (!supported_dump_vdev_type(hdl, top[0], errbuf)) {
4112                 goto out;
4113         }
4114         ret = 0;
4115
4116 out:
4117         if (zhp)
4118                 zpool_close(zhp);
4119         libzfs_fini(hdl);
4120         return (ret);
4121 }