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) 2011, 2014 by Delphix. All rights reserved.
  26  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  27  */
  28
  29 #include <sys/types.h>
  30 #include <sys/stat.h>
  31 #include <ctype.h>
  32 #include <errno.h>
  33 #include <devid.h>
  34 #include <fcntl.h>
  35 #include <libintl.h>
  36 #include <stdio.h>
  37 #include <stdlib.h>
  38 #include <strings.h>
  39 #include <unistd.h>
  40 #include <libgen.h>
  41 #include <sys/zfs_ioctl.h>
  42 #include <dlfcn.h>
  43
  44 #include "zfs_namecheck.h"
  45 #include "zfs_prop.h"
  46 #include "libzfs_impl.h"
  47 #include "zfs_comutil.h"
  48 #include "zfeature_common.h"
  49
  50 static int read_efi_label(nvlist_t *config, diskaddr_t *sb);
  51
  52 #define DISK_ROOT       "/dev/dsk"
  53 #define RDISK_ROOT      "/dev/rdsk"
  54 #define BACKUP_SLICE    "s2"
  55
  56 typedef struct prop_flags {
  57         int create:1;   /* Validate property on creation */
  58         int import:1;   /* Validate property on import */
  59 } prop_flags_t;
  60
  61 /*
  62  * ====================================================================
  63  *   zpool property functions
  64  * ====================================================================
  65  */
  66
  67 static int
  68 zpool_get_all_props(zpool_handle_t *zhp)
  69 {
  70         zfs_cmd_t zc = { 0 };
  71         libzfs_handle_t *hdl = zhp->zpool_hdl;
  72
  73         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
  74
  75         if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
  76                 return (-1);
  77
  78         while (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
  79                 if (errno == ENOMEM) {
  80                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
  81                                 zcmd_free_nvlists(&zc);
  82                                 return (-1);
  83                         }
  84                 } else {
  85                         zcmd_free_nvlists(&zc);
  86                         return (-1);
  87                 }
  88         }
  89
  90         if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
  91                 zcmd_free_nvlists(&zc);
  92                 return (-1);
  93         }
  94
  95         zcmd_free_nvlists(&zc);
  96
  97         return (0);
  98 }
  99
 100 static int
 101 zpool_props_refresh(zpool_handle_t *zhp)
 102 {
 103         nvlist_t *old_props;
 104
 105         old_props = zhp->zpool_props;
 106
 107         if (zpool_get_all_props(zhp) != 0)
 108                 return (-1);
 109
 110         nvlist_free(old_props);
 111         return (0);
 112 }
 113
 114 static char *
 115 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
 116     zprop_source_t *src)
 117 {
 118         nvlist_t *nv, *nvl;
 119         uint64_t ival;
 120         char *value;
 121         zprop_source_t source;
 122
 123         nvl = zhp->zpool_props;
 124         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 125                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &ival) == 0);
 126                 source = ival;
 127                 verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
 128         } else {
 129                 source = ZPROP_SRC_DEFAULT;
 130                 if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
 131                         value = "-";
 132         }
 133
 134         if (src)
 135                 *src = source;
 136
 137         return (value);
 138 }
 139
 140 uint64_t
 141 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
 142 {
 143         nvlist_t *nv, *nvl;
 144         uint64_t value;
 145         zprop_source_t source;
 146
 147         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
 148                 /*
 149                  * zpool_get_all_props() has most likely failed because
 150                  * the pool is faulted, but if all we need is the top level
 151                  * vdev's guid then get it from the zhp config nvlist.
 152                  */
 153                 if ((prop == ZPOOL_PROP_GUID) &&
 154                     (nvlist_lookup_nvlist(zhp->zpool_config,
 155                     ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
 156                     (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
 157                     == 0)) {
 158                         return (value);
 159                 }
 160                 return (zpool_prop_default_numeric(prop));
 161         }
 162
 163         nvl = zhp->zpool_props;
 164         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 165                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &value) == 0);
 166                 source = value;
 167                 verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
 168         } else {
 169                 source = ZPROP_SRC_DEFAULT;
 170                 value = zpool_prop_default_numeric(prop);
 171         }
 172
 173         if (src)
 174                 *src = source;
 175
 176         return (value);
 177 }
 178
 179 /*
 180  * Map VDEV STATE to printed strings.
 181  */
 182 const char *
 183 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
 184 {
 185         switch (state) {
 186         case VDEV_STATE_CLOSED:
 187         case VDEV_STATE_OFFLINE:
 188                 return (gettext("OFFLINE"));
 189         case VDEV_STATE_REMOVED:
 190                 return (gettext("REMOVED"));
 191         case VDEV_STATE_CANT_OPEN:
 192                 if (aux == VDEV_AUX_CORRUPT_DATA || aux == VDEV_AUX_BAD_LOG)
 193                         return (gettext("FAULTED"));
 194                 else if (aux == VDEV_AUX_SPLIT_POOL)
 195                         return (gettext("SPLIT"));
 196                 else
 197                         return (gettext("UNAVAIL"));
 198         case VDEV_STATE_FAULTED:
 199                 return (gettext("FAULTED"));
 200         case VDEV_STATE_DEGRADED:
 201                 return (gettext("DEGRADED"));
 202         case VDEV_STATE_HEALTHY:
 203                 return (gettext("ONLINE"));
 204         }
 205
 206         return (gettext("UNKNOWN"));
 207 }
 208
 209 /*
 210  * Map POOL STATE to printed strings.
 211  */
 212 const char *
 213 zpool_pool_state_to_name(pool_state_t state)
 214 {
 215         switch (state) {
 216         case POOL_STATE_ACTIVE:
 217                 return (gettext("ACTIVE"));
 218         case POOL_STATE_EXPORTED:
 219                 return (gettext("EXPORTED"));
 220         case POOL_STATE_DESTROYED:
 221                 return (gettext("DESTROYED"));
 222         case POOL_STATE_SPARE:
 223                 return (gettext("SPARE"));
 224         case POOL_STATE_L2CACHE:
 225                 return (gettext("L2CACHE"));
 226         case POOL_STATE_UNINITIALIZED:
 227                 return (gettext("UNINITIALIZED"));
 228         case POOL_STATE_UNAVAIL:
 229                 return (gettext("UNAVAIL"));
 230         case POOL_STATE_POTENTIALLY_ACTIVE:
 231                 return (gettext("POTENTIALLY_ACTIVE"));
 232         }
 233
 234         return (gettext("UNKNOWN"));
 235 }
 236
 237 /*
 238  * Get a zpool property value for 'prop' and return the value in
 239  * a pre-allocated buffer.
 240  */
 241 int
 242 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf, size_t len,
 243     zprop_source_t *srctype, boolean_t literal)
 244 {
 245         uint64_t intval;
 246         const char *strval;
 247         zprop_source_t src = ZPROP_SRC_NONE;
 248         nvlist_t *nvroot;
 249         vdev_stat_t *vs;
 250         uint_t vsc;
 251
 252         if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
 253                 switch (prop) {
 254                 case ZPOOL_PROP_NAME:
 255                         (void) strlcpy(buf, zpool_get_name(zhp), len);
 256                         break;
 257
 258                 case ZPOOL_PROP_HEALTH:
 259                         (void) strlcpy(buf,
 260                             zpool_pool_state_to_name(POOL_STATE_UNAVAIL), len);
 261                         break;
 262
 263                 case ZPOOL_PROP_GUID:
 264                         intval = zpool_get_prop_int(zhp, prop, &src);
 265                         (void) snprintf(buf, len, "%llu", intval);
 266                         break;
 267
 268                 case ZPOOL_PROP_ALTROOT:
 269                 case ZPOOL_PROP_CACHEFILE:
 270                 case ZPOOL_PROP_COMMENT:
 271                         if (zhp->zpool_props != NULL ||
 272                             zpool_get_all_props(zhp) == 0) {
 273                                 (void) strlcpy(buf,
 274                                     zpool_get_prop_string(zhp, prop, &src),
 275                                     len);
 276                                 break;
 277                         }
 278                         /* FALLTHROUGH */
 279                 default:
 280                         (void) strlcpy(buf, "-", len);
 281                         break;
 282                 }
 283
 284                 if (srctype != NULL)
 285                         *srctype = src;
 286                 return (0);
 287         }
 288
 289         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
 290             prop != ZPOOL_PROP_NAME)
 291                 return (-1);
 292
 293         switch (zpool_prop_get_type(prop)) {
 294         case PROP_TYPE_STRING:
 295                 (void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
 296                     len);
 297                 break;
 298
 299         case PROP_TYPE_NUMBER:
 300                 intval = zpool_get_prop_int(zhp, prop, &src);
 301
 302                 switch (prop) {
 303                 case ZPOOL_PROP_SIZE:
 304                 case ZPOOL_PROP_ALLOCATED:
 305                 case ZPOOL_PROP_FREE:
 306                 case ZPOOL_PROP_FREEING:
 307                 case ZPOOL_PROP_LEAKED:
 308                         if (literal) {
 309                                 (void) snprintf(buf, len, "%llu",
 310                                     (u_longlong_t)intval);
 311                         } else {
 312                                 (void) zfs_nicenum(intval, buf, len);
 313                         }
 314                         break;
 315                 case ZPOOL_PROP_EXPANDSZ:
 316                         if (intval == 0) {
 317                                 (void) strlcpy(buf, "-", len);
 318                         } else if (literal) {
 319                                 (void) snprintf(buf, len, "%llu",
 320                                     (u_longlong_t)intval);
 321                         } else {
 322                                 (void) zfs_nicenum(intval, buf, len);
 323                         }
 324                         break;
 325                 case ZPOOL_PROP_CAPACITY:
 326                         if (literal) {
 327                                 (void) snprintf(buf, len, "%llu",
 328                                     (u_longlong_t)intval);
 329                         } else {
 330                                 (void) snprintf(buf, len, "%llu%%",
 331                                     (u_longlong_t)intval);
 332                         }
 333                         break;
 334                 case ZPOOL_PROP_FRAGMENTATION:
 335                         if (intval == UINT64_MAX) {
 336                                 (void) strlcpy(buf, "-", len);
 337                         } else {
 338                                 (void) snprintf(buf, len, "%llu%%",
 339                                     (u_longlong_t)intval);
 340                         }
 341                         break;
 342                 case ZPOOL_PROP_DEDUPRATIO:
 343                         (void) snprintf(buf, len, "%llu.%02llux",
 344                             (u_longlong_t)(intval / 100),
 345                             (u_longlong_t)(intval % 100));
 346                         break;
 347                 case ZPOOL_PROP_HEALTH:
 348                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 349                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 350                         verify(nvlist_lookup_uint64_array(nvroot,
 351                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
 352                             == 0);
 353
 354                         (void) strlcpy(buf, zpool_state_to_name(intval,
 355                             vs->vs_aux), len);
 356                         break;
 357                 case ZPOOL_PROP_VERSION:
 358                         if (intval >= SPA_VERSION_FEATURES) {
 359                                 (void) snprintf(buf, len, "-");
 360                                 break;
 361                         }
 362                         /* FALLTHROUGH */
 363                 default:
 364                         (void) snprintf(buf, len, "%llu", intval);
 365                 }
 366                 break;
 367
 368         case PROP_TYPE_INDEX:
 369                 intval = zpool_get_prop_int(zhp, prop, &src);
 370                 if (zpool_prop_index_to_string(prop, intval, &strval)
 371                     != 0)
 372                         return (-1);
 373                 (void) strlcpy(buf, strval, len);
 374                 break;
 375
 376         default:
 377                 abort();
 378         }
 379
 380         if (srctype)
 381                 *srctype = src;
 382
 383         return (0);
 384 }
 385
 386 /*
 387  * Check if the bootfs name has the same pool name as it is set to.
 388  * Assuming bootfs is a valid dataset name.
 389  */
 390 static boolean_t
 391 bootfs_name_valid(const char *pool, char *bootfs)
 392 {
 393         int len = strlen(pool);
 394
 395         if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
 396                 return (B_FALSE);
 397
 398         if (strncmp(pool, bootfs, len) == 0 &&
 399             (bootfs[len] == '/' || bootfs[len] == '\0'))
 400                 return (B_TRUE);
 401
 402         return (B_FALSE);
 403 }
 404
 405 /*
 406  * Inspect the configuration to determine if any of the devices contain
 407  * an EFI label.
 408  */
 409 static boolean_t
 410 pool_uses_efi(nvlist_t *config)
 411 {
 412 #ifdef sun
 413         nvlist_t **child;
 414         uint_t c, children;
 415
 416         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
 417             &child, &children) != 0)
 418                 return (read_efi_label(config, NULL) >= 0);
 419
 420         for (c = 0; c < children; c++) {
 421                 if (pool_uses_efi(child[c]))
 422                         return (B_TRUE);
 423         }
 424 #endif  /* sun */
 425         return (B_FALSE);
 426 }
 427
 428 boolean_t
 429 zpool_is_bootable(zpool_handle_t *zhp)
 430 {
 431         char bootfs[ZPOOL_MAXNAMELEN];
 432
 433         return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
 434             sizeof (bootfs), NULL, B_FALSE) == 0 && strncmp(bootfs, "-",
 435             sizeof (bootfs)) != 0);
 436 }
 437
 438
 439 /*
 440  * Given an nvlist of zpool properties to be set, validate that they are
 441  * correct, and parse any numeric properties (index, boolean, etc) if they are
 442  * specified as strings.
 443  */
 444 static nvlist_t *
 445 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 446     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
 447 {
 448         nvpair_t *elem;
 449         nvlist_t *retprops;
 450         zpool_prop_t prop;
 451         char *strval;
 452         uint64_t intval;
 453         char *slash, *check;
 454         struct stat64 statbuf;
 455         zpool_handle_t *zhp;
 456         nvlist_t *nvroot;
 457
 458         if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
 459                 (void) no_memory(hdl);
 460                 return (NULL);
 461         }
 462
 463         elem = NULL;
 464         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 465                 const char *propname = nvpair_name(elem);
 466
 467                 prop = zpool_name_to_prop(propname);
 468                 if (prop == ZPROP_INVAL && zpool_prop_feature(propname)) {
 469                         int err;
 470                         char *fname = strchr(propname, '@') + 1;
 471
 472                         err = zfeature_lookup_name(fname, NULL);
 473                         if (err != 0) {
 474                                 ASSERT3U(err, ==, ENOENT);
 475                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 476                                     "invalid feature '%s'"), fname);
 477                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 478                                 goto error;
 479                         }
 480
 481                         if (nvpair_type(elem) != DATA_TYPE_STRING) {
 482                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 483                                     "'%s' must be a string"), propname);
 484                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 485                                 goto error;
 486                         }
 487
 488                         (void) nvpair_value_string(elem, &strval);
 489                         if (strcmp(strval, ZFS_FEATURE_ENABLED) != 0) {
 490                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 491                                     "property '%s' can only be set to "
 492                                     "'enabled'"), propname);
 493                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 494                                 goto error;
 495                         }
 496
 497                         if (nvlist_add_uint64(retprops, propname, 0) != 0) {
 498                                 (void) no_memory(hdl);
 499                                 goto error;
 500                         }
 501                         continue;
 502                 }
 503
 504                 /*
 505                  * Make sure this property is valid and applies to this type.
 506                  */
 507                 if (prop == ZPROP_INVAL) {
 508                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 509                             "invalid property '%s'"), propname);
 510                         (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 511                         goto error;
 512                 }
 513
 514                 if (zpool_prop_readonly(prop)) {
 515                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 516                             "is readonly"), propname);
 517                         (void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
 518                         goto error;
 519                 }
 520
 521                 if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
 522                     &strval, &intval, errbuf) != 0)
 523                         goto error;
 524
 525                 /*
 526                  * Perform additional checking for specific properties.
 527                  */
 528                 switch (prop) {
 529                 case ZPOOL_PROP_VERSION:
 530                         if (intval < version ||
 531                             !SPA_VERSION_IS_SUPPORTED(intval)) {
 532                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 533                                     "property '%s' number %d is invalid."),
 534                                     propname, intval);
 535                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 536                                 goto error;
 537                         }
 538                         break;
 539
 540                 case ZPOOL_PROP_BOOTFS:
 541                         if (flags.create || flags.import) {
 542                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 543                                     "property '%s' cannot be set at creation "
 544                                     "or import time"), propname);
 545                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 546                                 goto error;
 547                         }
 548
 549                         if (version < SPA_VERSION_BOOTFS) {
 550                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 551                                     "pool must be upgraded to support "
 552                                     "'%s' property"), propname);
 553                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 554                                 goto error;
 555                         }
 556
 557                         /*
 558                          * bootfs property value has to be a dataset name and
 559                          * the dataset has to be in the same pool as it sets to.
 560                          */
 561                         if (strval[0] != '\0' && !bootfs_name_valid(poolname,
 562                             strval)) {
 563                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 564                                     "is an invalid name"), strval);
 565                                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
 566                                 goto error;
 567                         }
 568
 569                         if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
 570                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 571                                     "could not open pool '%s'"), poolname);
 572                                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
 573                                 goto error;
 574                         }
 575                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 576                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 577
 578 #ifdef sun
 579                         /*
 580                          * bootfs property cannot be set on a disk which has
 581                          * been EFI labeled.
 582                          */
 583                         if (pool_uses_efi(nvroot)) {
 584                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 585                                     "property '%s' not supported on "
 586                                     "EFI labeled devices"), propname);
 587                                 (void) zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf);
 588                                 zpool_close(zhp);
 589                                 goto error;
 590                         }
 591 #endif  /* sun */
 592                         zpool_close(zhp);
 593                         break;
 594
 595                 case ZPOOL_PROP_ALTROOT:
 596                         if (!flags.create && !flags.import) {
 597                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 598                                     "property '%s' can only be set during pool "
 599                                     "creation or import"), propname);
 600                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 601                                 goto error;
 602                         }
 603
 604                         if (strval[0] != '/') {
 605                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 606                                     "bad alternate root '%s'"), strval);
 607                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 608                                 goto error;
 609                         }
 610                         break;
 611
 612                 case ZPOOL_PROP_CACHEFILE:
 613                         if (strval[0] == '\0')
 614                                 break;
 615
 616                         if (strcmp(strval, "none") == 0)
 617                                 break;
 618
 619                         if (strval[0] != '/') {
 620                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 621                                     "property '%s' must be empty, an "
 622                                     "absolute path, or 'none'"), propname);
 623                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 624                                 goto error;
 625                         }
 626
 627                         slash = strrchr(strval, '/');
 628
 629                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 630                             strcmp(slash, "/..") == 0) {
 631                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 632                                     "'%s' is not a valid file"), strval);
 633                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 634                                 goto error;
 635                         }
 636
 637                         *slash = '\0';
 638
 639                         if (strval[0] != '\0' &&
 640                             (stat64(strval, &statbuf) != 0 ||
 641                             !S_ISDIR(statbuf.st_mode))) {
 642                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 643                                     "'%s' is not a valid directory"),
 644                                     strval);
 645                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 646                                 goto error;
 647                         }
 648
 649                         *slash = '/';
 650                         break;
 651
 652                 case ZPOOL_PROP_COMMENT:
 653                         for (check = strval; *check != '\0'; check++) {
 654                                 if (!isprint(*check)) {
 655                                         zfs_error_aux(hdl,
 656                                             dgettext(TEXT_DOMAIN,
 657                                             "comment may only have printable "
 658                                             "characters"));
 659                                         (void) zfs_error(hdl, EZFS_BADPROP,
 660                                             errbuf);
 661                                         goto error;
 662                                 }
 663                         }
 664                         if (strlen(strval) > ZPROP_MAX_COMMENT) {
 665                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 666                                     "comment must not exceed %d characters"),
 667                                     ZPROP_MAX_COMMENT);
 668                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 669                                 goto error;
 670                         }
 671                         break;
 672                 case ZPOOL_PROP_READONLY:
 673                         if (!flags.import) {
 674                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 675                                     "property '%s' can only be set at "
 676                                     "import time"), propname);
 677                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 678                                 goto error;
 679                         }
 680                         break;
 681                 }
 682         }
 683
 684         return (retprops);
 685 error:
 686         nvlist_free(retprops);
 687         return (NULL);
 688 }
 689
 690 /*
 691  * Set zpool property : propname=propval.
 692  */
 693 int
 694 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
 695 {
 696         zfs_cmd_t zc = { 0 };
 697         int ret = -1;
 698         char errbuf[1024];
 699         nvlist_t *nvl = NULL;
 700         nvlist_t *realprops;
 701         uint64_t version;
 702         prop_flags_t flags = { 0 };
 703
 704         (void) snprintf(errbuf, sizeof (errbuf),
 705             dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
 706             zhp->zpool_name);
 707
 708         if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
 709                 return (no_memory(zhp->zpool_hdl));
 710
 711         if (nvlist_add_string(nvl, propname, propval) != 0) {
 712                 nvlist_free(nvl);
 713                 return (no_memory(zhp->zpool_hdl));
 714         }
 715
 716         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
 717         if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
 718             zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
 719                 nvlist_free(nvl);
 720                 return (-1);
 721         }
 722
 723         nvlist_free(nvl);
 724         nvl = realprops;
 725
 726         /*
 727          * Execute the corresponding ioctl() to set this property.
 728          */
 729         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 730
 731         if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
 732                 nvlist_free(nvl);
 733                 return (-1);
 734         }
 735
 736         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
 737
 738         zcmd_free_nvlists(&zc);
 739         nvlist_free(nvl);
 740
 741         if (ret)
 742                 (void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
 743         else
 744                 (void) zpool_props_refresh(zhp);
 745
 746         return (ret);
 747 }
 748
 749 int
 750 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
 751 {
 752         libzfs_handle_t *hdl = zhp->zpool_hdl;
 753         zprop_list_t *entry;
 754         char buf[ZFS_MAXPROPLEN];
 755         nvlist_t *features = NULL;
 756         zprop_list_t **last;
 757         boolean_t firstexpand = (NULL == *plp);
 758
 759         if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
 760                 return (-1);
 761
 762         last = plp;
 763         while (*last != NULL)
 764                 last = &(*last)->pl_next;
 765
 766         if ((*plp)->pl_all)
 767                 features = zpool_get_features(zhp);
 768
 769         if ((*plp)->pl_all && firstexpand) {
 770                 for (int i = 0; i < SPA_FEATURES; i++) {
 771                         zprop_list_t *entry = zfs_alloc(hdl,
 772                             sizeof (zprop_list_t));
 773                         entry->pl_prop = ZPROP_INVAL;
 774                         entry->pl_user_prop = zfs_asprintf(hdl, "feature@%s",
 775                             spa_feature_table[i].fi_uname);
 776                         entry->pl_width = strlen(entry->pl_user_prop);
 777                         entry->pl_all = B_TRUE;
 778
 779                         *last = entry;
 780                         last = &entry->pl_next;
 781                 }
 782         }
 783
 784         /* add any unsupported features */
 785         for (nvpair_t *nvp = nvlist_next_nvpair(features, NULL);
 786             nvp != NULL; nvp = nvlist_next_nvpair(features, nvp)) {
 787                 char *propname;
 788                 boolean_t found;
 789                 zprop_list_t *entry;
 790
 791                 if (zfeature_is_supported(nvpair_name(nvp)))
 792                         continue;
 793
 794                 propname = zfs_asprintf(hdl, "unsupported@%s",
 795                     nvpair_name(nvp));
 796
 797                 /*
 798                  * Before adding the property to the list make sure that no
 799                  * other pool already added the same property.
 800                  */
 801                 found = B_FALSE;
 802                 entry = *plp;
 803                 while (entry != NULL) {
 804                         if (entry->pl_user_prop != NULL &&
 805                             strcmp(propname, entry->pl_user_prop) == 0) {
 806                                 found = B_TRUE;
 807                                 break;
 808                         }
 809                         entry = entry->pl_next;
 810                 }
 811                 if (found) {
 812                         free(propname);
 813                         continue;
 814                 }
 815
 816                 entry = zfs_alloc(hdl, sizeof (zprop_list_t));
 817                 entry->pl_prop = ZPROP_INVAL;
 818                 entry->pl_user_prop = propname;
 819                 entry->pl_width = strlen(entry->pl_user_prop);
 820                 entry->pl_all = B_TRUE;
 821
 822                 *last = entry;
 823                 last = &entry->pl_next;
 824         }
 825
 826         for (entry = *plp; entry != NULL; entry = entry->pl_next) {
 827
 828                 if (entry->pl_fixed)
 829                         continue;
 830
 831                 if (entry->pl_prop != ZPROP_INVAL &&
 832                     zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
 833                     NULL, B_FALSE) == 0) {
 834                         if (strlen(buf) > entry->pl_width)
 835                                 entry->pl_width = strlen(buf);
 836                 }
 837         }
 838
 839         return (0);
 840 }
 841
 842 /*
 843  * Get the state for the given feature on the given ZFS pool.
 844  */
 845 int
 846 zpool_prop_get_feature(zpool_handle_t *zhp, const char *propname, char *buf,
 847     size_t len)
 848 {
 849         uint64_t refcount;
 850         boolean_t found = B_FALSE;
 851         nvlist_t *features = zpool_get_features(zhp);
 852         boolean_t supported;
 853         const char *feature = strchr(propname, '@') + 1;
 854
 855         supported = zpool_prop_feature(propname);
 856         ASSERT(supported || zpool_prop_unsupported(propname));
 857
 858         /*
 859          * Convert from feature name to feature guid. This conversion is
 860          * unecessary for unsupported@... properties because they already
 861          * use guids.
 862          */
 863         if (supported) {
 864                 int ret;
 865                 spa_feature_t fid;
 866
 867                 ret = zfeature_lookup_name(feature, &fid);
 868                 if (ret != 0) {
 869                         (void) strlcpy(buf, "-", len);
 870                         return (ENOTSUP);
 871                 }
 872                 feature = spa_feature_table[fid].fi_guid;
 873         }
 874
 875         if (nvlist_lookup_uint64(features, feature, &refcount) == 0)
 876                 found = B_TRUE;
 877
 878         if (supported) {
 879                 if (!found) {
 880                         (void) strlcpy(buf, ZFS_FEATURE_DISABLED, len);
 881                 } else  {
 882                         if (refcount == 0)
 883                                 (void) strlcpy(buf, ZFS_FEATURE_ENABLED, len);
 884                         else
 885                                 (void) strlcpy(buf, ZFS_FEATURE_ACTIVE, len);
 886                 }
 887         } else {
 888                 if (found) {
 889                         if (refcount == 0) {
 890                                 (void) strcpy(buf, ZFS_UNSUPPORTED_INACTIVE);
 891                         } else {
 892                                 (void) strcpy(buf, ZFS_UNSUPPORTED_READONLY);
 893                         }
 894                 } else {
 895                         (void) strlcpy(buf, "-", len);
 896                         return (ENOTSUP);
 897                 }
 898         }
 899
 900         return (0);
 901 }
 902
 903 /*
 904  * Don't start the slice at the default block of 34; many storage
 905  * devices will use a stripe width of 128k, so start there instead.
 906  */
 907 #define NEW_START_BLOCK 256
 908
 909 /*
 910  * Validate the given pool name, optionally putting an extended error message in
 911  * 'buf'.
 912  */
 913 boolean_t
 914 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
 915 {
 916         namecheck_err_t why;
 917         char what;
 918         int ret;
 919
 920         ret = pool_namecheck(pool, &why, &what);
 921
 922         /*
 923          * The rules for reserved pool names were extended at a later point.
 924          * But we need to support users with existing pools that may now be
 925          * invalid.  So we only check for this expanded set of names during a
 926          * create (or import), and only in userland.
 927          */
 928         if (ret == 0 && !isopen &&
 929             (strncmp(pool, "mirror", 6) == 0 ||
 930             strncmp(pool, "raidz", 5) == 0 ||
 931             strncmp(pool, "spare", 5) == 0 ||
 932             strcmp(pool, "log") == 0)) {
 933                 if (hdl != NULL)
 934                         zfs_error_aux(hdl,
 935                             dgettext(TEXT_DOMAIN, "name is reserved"));
 936                 return (B_FALSE);
 937         }
 938
 939
 940         if (ret != 0) {
 941                 if (hdl != NULL) {
 942                         switch (why) {
 943                         case NAME_ERR_TOOLONG:
 944                                 zfs_error_aux(hdl,
 945                                     dgettext(TEXT_DOMAIN, "name is too long"));
 946                                 break;
 947
 948                         case NAME_ERR_INVALCHAR:
 949                                 zfs_error_aux(hdl,
 950                                     dgettext(TEXT_DOMAIN, "invalid character "
 951                                     "'%c' in pool name"), what);
 952                                 break;
 953
 954                         case NAME_ERR_NOLETTER:
 955                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 956                                     "name must begin with a letter"));
 957                                 break;
 958
 959                         case NAME_ERR_RESERVED:
 960                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 961                                     "name is reserved"));
 962                                 break;
 963
 964                         case NAME_ERR_DISKLIKE:
 965                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 966                                     "pool name is reserved"));
 967                                 break;
 968
 969                         case NAME_ERR_LEADING_SLASH:
 970                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 971                                     "leading slash in name"));
 972                                 break;
 973
 974                         case NAME_ERR_EMPTY_COMPONENT:
 975                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 976                                     "empty component in name"));
 977                                 break;
 978
 979                         case NAME_ERR_TRAILING_SLASH:
 980                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 981                                     "trailing slash in name"));
 982                                 break;
 983
 984                         case NAME_ERR_MULTIPLE_AT:
 985                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 986                                     "multiple '@' delimiters in name"));
 987                                 break;
 988
 989                         }
 990                 }
 991                 return (B_FALSE);
 992         }
 993
 994         return (B_TRUE);
 995 }
 996
 997 /*
 998  * Open a handle to the given pool, even if the pool is currently in the FAULTED
 999  * state.
1000  */
1001 zpool_handle_t *
1002 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
1003 {
1004         zpool_handle_t *zhp;
1005         boolean_t missing;
1006
1007         /*
1008          * Make sure the pool name is valid.
1009          */
1010         if (!zpool_name_valid(hdl, B_TRUE, pool)) {
1011                 (void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1012                     dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1013                     pool);
1014                 return (NULL);
1015         }
1016
1017         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1018                 return (NULL);
1019
1020         zhp->zpool_hdl = hdl;
1021         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1022
1023         if (zpool_refresh_stats(zhp, &missing) != 0) {
1024                 zpool_close(zhp);
1025                 return (NULL);
1026         }
1027
1028         if (missing) {
1029                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
1030                 (void) zfs_error_fmt(hdl, EZFS_NOENT,
1031                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
1032                 zpool_close(zhp);
1033                 return (NULL);
1034         }
1035
1036         return (zhp);
1037 }
1038
1039 /*
1040  * Like the above, but silent on error.  Used when iterating over pools (because
1041  * the configuration cache may be out of date).
1042  */
1043 int
1044 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
1045 {
1046         zpool_handle_t *zhp;
1047         boolean_t missing;
1048
1049         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1050                 return (-1);
1051
1052         zhp->zpool_hdl = hdl;
1053         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1054
1055         if (zpool_refresh_stats(zhp, &missing) != 0) {
1056                 zpool_close(zhp);
1057                 return (-1);
1058         }
1059
1060         if (missing) {
1061                 zpool_close(zhp);
1062                 *ret = NULL;
1063                 return (0);
1064         }
1065
1066         *ret = zhp;
1067         return (0);
1068 }
1069
1070 /*
1071  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
1072  * state.
1073  */
1074 zpool_handle_t *
1075 zpool_open(libzfs_handle_t *hdl, const char *pool)
1076 {
1077         zpool_handle_t *zhp;
1078
1079         if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
1080                 return (NULL);
1081
1082         if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
1083                 (void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
1084                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
1085                 zpool_close(zhp);
1086                 return (NULL);
1087         }
1088
1089         return (zhp);
1090 }
1091
1092 /*
1093  * Close the handle.  Simply frees the memory associated with the handle.
1094  */
1095 void
1096 zpool_close(zpool_handle_t *zhp)
1097 {
1098         if (zhp->zpool_config)
1099                 nvlist_free(zhp->zpool_config);
1100         if (zhp->zpool_old_config)
1101                 nvlist_free(zhp->zpool_old_config);
1102         if (zhp->zpool_props)
1103                 nvlist_free(zhp->zpool_props);
1104         free(zhp);
1105 }
1106
1107 /*
1108  * Return the name of the pool.
1109  */
1110 const char *
1111 zpool_get_name(zpool_handle_t *zhp)
1112 {
1113         return (zhp->zpool_name);
1114 }
1115
1116
1117 /*
1118  * Return the state of the pool (ACTIVE or UNAVAILABLE)
1119  */
1120 int
1121 zpool_get_state(zpool_handle_t *zhp)
1122 {
1123         return (zhp->zpool_state);
1124 }
1125
1126 /*
1127  * Create the named pool, using the provided vdev list.  It is assumed
1128  * that the consumer has already validated the contents of the nvlist, so we
1129  * don't have to worry about error semantics.
1130  */
1131 int
1132 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
1133     nvlist_t *props, nvlist_t *fsprops)
1134 {
1135         zfs_cmd_t zc = { 0 };
1136         nvlist_t *zc_fsprops = NULL;
1137         nvlist_t *zc_props = NULL;
1138         char msg[1024];
1139         int ret = -1;
1140
1141         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1142             "cannot create '%s'"), pool);
1143
1144         if (!zpool_name_valid(hdl, B_FALSE, pool))
1145                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
1146
1147         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1148                 return (-1);
1149
1150         if (props) {
1151                 prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
1152
1153                 if ((zc_props = zpool_valid_proplist(hdl, pool, props,
1154                     SPA_VERSION_1, flags, msg)) == NULL) {
1155                         goto create_failed;
1156                 }
1157         }
1158
1159         if (fsprops) {
1160                 uint64_t zoned;
1161                 char *zonestr;
1162
1163                 zoned = ((nvlist_lookup_string(fsprops,
1164                     zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
1165                     strcmp(zonestr, "on") == 0);
1166
1167                 if ((zc_fsprops = zfs_valid_proplist(hdl,
1168                     ZFS_TYPE_FILESYSTEM, fsprops, zoned, NULL, msg)) == NULL) {
1169                         goto create_failed;
1170                 }
1171                 if (!zc_props &&
1172                     (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
1173                         goto create_failed;
1174                 }
1175                 if (nvlist_add_nvlist(zc_props,
1176                     ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
1177                         goto create_failed;
1178                 }
1179         }
1180
1181         if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
1182                 goto create_failed;
1183
1184         (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
1185
1186         if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
1187
1188                 zcmd_free_nvlists(&zc);
1189                 nvlist_free(zc_props);
1190                 nvlist_free(zc_fsprops);
1191
1192                 switch (errno) {
1193                 case EBUSY:
1194                         /*
1195                          * This can happen if the user has specified the same
1196                          * device multiple times.  We can't reliably detect this
1197                          * until we try to add it and see we already have a
1198                          * label.
1199                          */
1200                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1201                             "one or more vdevs refer to the same device"));
1202                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1203
1204                 case EOVERFLOW:
1205                         /*
1206                          * This occurs when one of the devices is below
1207                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1208                          * device was the problem device since there's no
1209                          * reliable way to determine device size from userland.
1210                          */
1211                         {
1212                                 char buf[64];
1213
1214                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1215
1216                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1217                                     "one or more devices is less than the "
1218                                     "minimum size (%s)"), buf);
1219                         }
1220                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1221
1222                 case ENOSPC:
1223                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1224                             "one or more devices is out of space"));
1225                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1226
1227                 case ENOTBLK:
1228                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1229                             "cache device must be a disk or disk slice"));
1230                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1231
1232                 default:
1233                         return (zpool_standard_error(hdl, errno, msg));
1234                 }
1235         }
1236
1237 create_failed:
1238         zcmd_free_nvlists(&zc);
1239         nvlist_free(zc_props);
1240         nvlist_free(zc_fsprops);
1241         return (ret);
1242 }
1243
1244 /*
1245  * Destroy the given pool.  It is up to the caller to ensure that there are no
1246  * datasets left in the pool.
1247  */
1248 int
1249 zpool_destroy(zpool_handle_t *zhp, const char *log_str)
1250 {
1251         zfs_cmd_t zc = { 0 };
1252         zfs_handle_t *zfp = NULL;
1253         libzfs_handle_t *hdl = zhp->zpool_hdl;
1254         char msg[1024];
1255
1256         if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1257             (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1258                 return (-1);
1259
1260         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1261         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1262
1263         if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1264                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1265                     "cannot destroy '%s'"), zhp->zpool_name);
1266
1267                 if (errno == EROFS) {
1268                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1269                             "one or more devices is read only"));
1270                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1271                 } else {
1272                         (void) zpool_standard_error(hdl, errno, msg);
1273                 }
1274
1275                 if (zfp)
1276                         zfs_close(zfp);
1277                 return (-1);
1278         }
1279
1280         if (zfp) {
1281                 remove_mountpoint(zfp);
1282                 zfs_close(zfp);
1283         }
1284
1285         return (0);
1286 }
1287
1288 /*
1289  * Add the given vdevs to the pool.  The caller must have already performed the
1290  * necessary verification to ensure that the vdev specification is well-formed.
1291  */
1292 int
1293 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1294 {
1295         zfs_cmd_t zc = { 0 };
1296         int ret;
1297         libzfs_handle_t *hdl = zhp->zpool_hdl;
1298         char msg[1024];
1299         nvlist_t **spares, **l2cache;
1300         uint_t nspares, nl2cache;
1301
1302         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1303             "cannot add to '%s'"), zhp->zpool_name);
1304
1305         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1306             SPA_VERSION_SPARES &&
1307             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1308             &spares, &nspares) == 0) {
1309                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1310                     "upgraded to add hot spares"));
1311                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1312         }
1313
1314         if (zpool_is_bootable(zhp) && nvlist_lookup_nvlist_array(nvroot,
1315             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) {
1316                 uint64_t s;
1317
1318                 for (s = 0; s < nspares; s++) {
1319                         char *path;
1320
1321                         if (nvlist_lookup_string(spares[s], ZPOOL_CONFIG_PATH,
1322                             &path) == 0 && pool_uses_efi(spares[s])) {
1323                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1324                                     "device '%s' contains an EFI label and "
1325                                     "cannot be used on root pools."),
1326                                     zpool_vdev_name(hdl, NULL, spares[s],
1327                                     B_FALSE));
1328                                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
1329                         }
1330                 }
1331         }
1332
1333         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1334             SPA_VERSION_L2CACHE &&
1335             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1336             &l2cache, &nl2cache) == 0) {
1337                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1338                     "upgraded to add cache devices"));
1339                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1340         }
1341
1342         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1343                 return (-1);
1344         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1345
1346         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1347                 switch (errno) {
1348                 case EBUSY:
1349                         /*
1350                          * This can happen if the user has specified the same
1351                          * device multiple times.  We can't reliably detect this
1352                          * until we try to add it and see we already have a
1353                          * label.
1354                          */
1355                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1356                             "one or more vdevs refer to the same device"));
1357                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1358                         break;
1359
1360                 case EOVERFLOW:
1361                         /*
1362                          * This occurrs when one of the devices is below
1363                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1364                          * device was the problem device since there's no
1365                          * reliable way to determine device size from userland.
1366                          */
1367                         {
1368                                 char buf[64];
1369
1370                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1371
1372                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1373                                     "device is less than the minimum "
1374                                     "size (%s)"), buf);
1375                         }
1376                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1377                         break;
1378
1379                 case ENOTSUP:
1380                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1381                             "pool must be upgraded to add these vdevs"));
1382                         (void) zfs_error(hdl, EZFS_BADVERSION, msg);
1383                         break;
1384
1385                 case EDOM:
1386                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1387                             "root pool can not have multiple vdevs"
1388                             " or separate logs"));
1389                         (void) zfs_error(hdl, EZFS_POOL_NOTSUP, msg);
1390                         break;
1391
1392                 case ENOTBLK:
1393                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1394                             "cache device must be a disk or disk slice"));
1395                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1396                         break;
1397
1398                 default:
1399                         (void) zpool_standard_error(hdl, errno, msg);
1400                 }
1401
1402                 ret = -1;
1403         } else {
1404                 ret = 0;
1405         }
1406
1407         zcmd_free_nvlists(&zc);
1408
1409         return (ret);
1410 }
1411
1412 /*
1413  * Exports the pool from the system.  The caller must ensure that there are no
1414  * mounted datasets in the pool.
1415  */
1416 static int
1417 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce,
1418     const char *log_str)
1419 {
1420         zfs_cmd_t zc = { 0 };
1421         char msg[1024];
1422
1423         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1424             "cannot export '%s'"), zhp->zpool_name);
1425
1426         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1427         zc.zc_cookie = force;
1428         zc.zc_guid = hardforce;
1429         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1430
1431         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1432                 switch (errno) {
1433                 case EXDEV:
1434                         zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1435                             "use '-f' to override the following errors:\n"
1436                             "'%s' has an active shared spare which could be"
1437                             " used by other pools once '%s' is exported."),
1438                             zhp->zpool_name, zhp->zpool_name);
1439                         return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1440                             msg));
1441                 default:
1442                         return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1443                             msg));
1444                 }
1445         }
1446
1447         return (0);
1448 }
1449
1450 int
1451 zpool_export(zpool_handle_t *zhp, boolean_t force, const char *log_str)
1452 {
1453         return (zpool_export_common(zhp, force, B_FALSE, log_str));
1454 }
1455
1456 int
1457 zpool_export_force(zpool_handle_t *zhp, const char *log_str)
1458 {
1459         return (zpool_export_common(zhp, B_TRUE, B_TRUE, log_str));
1460 }
1461
1462 static void
1463 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1464     nvlist_t *config)
1465 {
1466         nvlist_t *nv = NULL;
1467         uint64_t rewindto;
1468         int64_t loss = -1;
1469         struct tm t;
1470         char timestr[128];
1471
1472         if (!hdl->libzfs_printerr || config == NULL)
1473                 return;
1474
1475         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1476             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0) {
1477                 return;
1478         }
1479
1480         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1481                 return;
1482         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1483
1484         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1485             strftime(timestr, 128, 0, &t) != 0) {
1486                 if (dryrun) {
1487                         (void) printf(dgettext(TEXT_DOMAIN,
1488                             "Would be able to return %s "
1489                             "to its state as of %s.\n"),
1490                             name, timestr);
1491                 } else {
1492                         (void) printf(dgettext(TEXT_DOMAIN,
1493                             "Pool %s returned to its state as of %s.\n"),
1494                             name, timestr);
1495                 }
1496                 if (loss > 120) {
1497                         (void) printf(dgettext(TEXT_DOMAIN,
1498                             "%s approximately %lld "),
1499                             dryrun ? "Would discard" : "Discarded",
1500                             (loss + 30) / 60);
1501                         (void) printf(dgettext(TEXT_DOMAIN,
1502                             "minutes of transactions.\n"));
1503                 } else if (loss > 0) {
1504                         (void) printf(dgettext(TEXT_DOMAIN,
1505                             "%s approximately %lld "),
1506                             dryrun ? "Would discard" : "Discarded", loss);
1507                         (void) printf(dgettext(TEXT_DOMAIN,
1508                             "seconds of transactions.\n"));
1509                 }
1510         }
1511 }
1512
1513 void
1514 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1515     nvlist_t *config)
1516 {
1517         nvlist_t *nv = NULL;
1518         int64_t loss = -1;
1519         uint64_t edata = UINT64_MAX;
1520         uint64_t rewindto;
1521         struct tm t;
1522         char timestr[128];
1523
1524         if (!hdl->libzfs_printerr)
1525                 return;
1526
1527         if (reason >= 0)
1528                 (void) printf(dgettext(TEXT_DOMAIN, "action: "));
1529         else
1530                 (void) printf(dgettext(TEXT_DOMAIN, "\t"));
1531
1532         /* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1533         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1534             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0 ||
1535             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1536                 goto no_info;
1537
1538         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1539         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1540             &edata);
1541
1542         (void) printf(dgettext(TEXT_DOMAIN,
1543             "Recovery is possible, but will result in some data loss.\n"));
1544
1545         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1546             strftime(timestr, 128, 0, &t) != 0) {
1547                 (void) printf(dgettext(TEXT_DOMAIN,
1548                     "\tReturning the pool to its state as of %s\n"
1549                     "\tshould correct the problem.  "),
1550                     timestr);
1551         } else {
1552                 (void) printf(dgettext(TEXT_DOMAIN,
1553                     "\tReverting the pool to an earlier state "
1554                     "should correct the problem.\n\t"));
1555         }
1556
1557         if (loss > 120) {
1558                 (void) printf(dgettext(TEXT_DOMAIN,
1559                     "Approximately %lld minutes of data\n"
1560                     "\tmust be discarded, irreversibly.  "), (loss + 30) / 60);
1561         } else if (loss > 0) {
1562                 (void) printf(dgettext(TEXT_DOMAIN,
1563                     "Approximately %lld seconds of data\n"
1564                     "\tmust be discarded, irreversibly.  "), loss);
1565         }
1566         if (edata != 0 && edata != UINT64_MAX) {
1567                 if (edata == 1) {
1568                         (void) printf(dgettext(TEXT_DOMAIN,
1569                             "After rewind, at least\n"
1570                             "\tone persistent user-data error will remain.  "));
1571                 } else {
1572                         (void) printf(dgettext(TEXT_DOMAIN,
1573                             "After rewind, several\n"
1574                             "\tpersistent user-data errors will remain.  "));
1575                 }
1576         }
1577         (void) printf(dgettext(TEXT_DOMAIN,
1578             "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1579             reason >= 0 ? "clear" : "import", name);
1580
1581         (void) printf(dgettext(TEXT_DOMAIN,
1582             "A scrub of the pool\n"
1583             "\tis strongly recommended after recovery.\n"));
1584         return;
1585
1586 no_info:
1587         (void) printf(dgettext(TEXT_DOMAIN,
1588             "Destroy and re-create the pool from\n\ta backup source.\n"));
1589 }
1590
1591 /*
1592  * zpool_import() is a contracted interface. Should be kept the same
1593  * if possible.
1594  *
1595  * Applications should use zpool_import_props() to import a pool with
1596  * new properties value to be set.
1597  */
1598 int
1599 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1600     char *altroot)
1601 {
1602         nvlist_t *props = NULL;
1603         int ret;
1604
1605         if (altroot != NULL) {
1606                 if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1607                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1608                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1609                             newname));
1610                 }
1611
1612                 if (nvlist_add_string(props,
1613                     zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1614                     nvlist_add_string(props,
1615                     zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1616                         nvlist_free(props);
1617                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1618                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1619                             newname));
1620                 }
1621         }
1622
1623         ret = zpool_import_props(hdl, config, newname, props,
1624             ZFS_IMPORT_NORMAL);
1625         if (props)
1626                 nvlist_free(props);
1627         return (ret);
1628 }
1629
1630 static void
1631 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1632     int indent)
1633 {
1634         nvlist_t **child;
1635         uint_t c, children;
1636         char *vname;
1637         uint64_t is_log = 0;
1638
1639         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1640             &is_log);
1641
1642         if (name != NULL)
1643                 (void) printf("\t%*s%s%s\n", indent, "", name,
1644                     is_log ? " [log]" : "");
1645
1646         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1647             &child, &children) != 0)
1648                 return;
1649
1650         for (c = 0; c < children; c++) {
1651                 vname = zpool_vdev_name(hdl, NULL, child[c], B_TRUE);
1652                 print_vdev_tree(hdl, vname, child[c], indent + 2);
1653                 free(vname);
1654         }
1655 }
1656
1657 void
1658 zpool_print_unsup_feat(nvlist_t *config)
1659 {
1660         nvlist_t *nvinfo, *unsup_feat;
1661
1662         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nvinfo) ==
1663             0);
1664         verify(nvlist_lookup_nvlist(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT,
1665             &unsup_feat) == 0);
1666
1667         for (nvpair_t *nvp = nvlist_next_nvpair(unsup_feat, NULL); nvp != NULL;
1668             nvp = nvlist_next_nvpair(unsup_feat, nvp)) {
1669                 char *desc;
1670
1671                 verify(nvpair_type(nvp) == DATA_TYPE_STRING);
1672                 verify(nvpair_value_string(nvp, &desc) == 0);
1673
1674                 if (strlen(desc) > 0)
1675                         (void) printf("\t%s (%s)\n", nvpair_name(nvp), desc);
1676                 else
1677                         (void) printf("\t%s\n", nvpair_name(nvp));
1678         }
1679 }
1680
1681 /*
1682  * Import the given pool using the known configuration and a list of
1683  * properties to be set. The configuration should have come from
1684  * zpool_find_import(). The 'newname' parameters control whether the pool
1685  * is imported with a different name.
1686  */
1687 int
1688 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1689     nvlist_t *props, int flags)
1690 {
1691         zfs_cmd_t zc = { 0 };
1692         zpool_rewind_policy_t policy;
1693         nvlist_t *nv = NULL;
1694         nvlist_t *nvinfo = NULL;
1695         nvlist_t *missing = NULL;
1696         char *thename;
1697         char *origname;
1698         int ret;
1699         int error = 0;
1700         char errbuf[1024];
1701
1702         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1703             &origname) == 0);
1704
1705         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1706             "cannot import pool '%s'"), origname);
1707
1708         if (newname != NULL) {
1709                 if (!zpool_name_valid(hdl, B_FALSE, newname))
1710                         return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1711                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1712                             newname));
1713                 thename = (char *)newname;
1714         } else {
1715                 thename = origname;
1716         }
1717
1718         if (props) {
1719                 uint64_t version;
1720                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1721
1722                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1723                     &version) == 0);
1724
1725                 if ((props = zpool_valid_proplist(hdl, origname,
1726                     props, version, flags, errbuf)) == NULL) {
1727                         return (-1);
1728                 } else if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1729                         nvlist_free(props);
1730                         return (-1);
1731                 }
1732         }
1733
1734         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1735
1736         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1737             &zc.zc_guid) == 0);
1738
1739         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1740                 nvlist_free(props);
1741                 return (-1);
1742         }
1743         if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1744                 nvlist_free(props);
1745                 return (-1);
1746         }
1747
1748         zc.zc_cookie = flags;
1749         while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1750             errno == ENOMEM) {
1751                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1752                         zcmd_free_nvlists(&zc);
1753                         return (-1);
1754                 }
1755         }
1756         if (ret != 0)
1757                 error = errno;
1758
1759         (void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1760         zpool_get_rewind_policy(config, &policy);
1761
1762         if (error) {
1763                 char desc[1024];
1764
1765                 /*
1766                  * Dry-run failed, but we print out what success
1767                  * looks like if we found a best txg
1768                  */
1769                 if (policy.zrp_request & ZPOOL_TRY_REWIND) {
1770                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1771                             B_TRUE, nv);
1772                         nvlist_free(nv);
1773                         return (-1);
1774                 }
1775
1776                 if (newname == NULL)
1777                         (void) snprintf(desc, sizeof (desc),
1778                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1779                             thename);
1780                 else
1781                         (void) snprintf(desc, sizeof (desc),
1782                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1783                             origname, thename);
1784
1785                 switch (error) {
1786                 case ENOTSUP:
1787                         if (nv != NULL && nvlist_lookup_nvlist(nv,
1788                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1789                             nvlist_exists(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT)) {
1790                                 (void) printf(dgettext(TEXT_DOMAIN, "This "
1791                                     "pool uses the following feature(s) not "
1792                                     "supported by this system:\n"));
1793                                 zpool_print_unsup_feat(nv);
1794                                 if (nvlist_exists(nvinfo,
1795                                     ZPOOL_CONFIG_CAN_RDONLY)) {
1796                                         (void) printf(dgettext(TEXT_DOMAIN,
1797                                             "All unsupported features are only "
1798                                             "required for writing to the pool."
1799                                             "\nThe pool can be imported using "
1800                                             "'-o readonly=on'.\n"));
1801                                 }
1802                         }
1803                         /*
1804                          * Unsupported version.
1805                          */
1806                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1807                         break;
1808
1809                 case EINVAL:
1810                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1811                         break;
1812
1813                 case EROFS:
1814                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1815                             "one or more devices is read only"));
1816                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1817                         break;
1818
1819                 case ENXIO:
1820                         if (nv && nvlist_lookup_nvlist(nv,
1821                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1822                             nvlist_lookup_nvlist(nvinfo,
1823                             ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1824                                 (void) printf(dgettext(TEXT_DOMAIN,
1825                                     "The devices below are missing, use "
1826                                     "'-m' to import the pool anyway:\n"));
1827                                 print_vdev_tree(hdl, NULL, missing, 2);
1828                                 (void) printf("\n");
1829                         }
1830                         (void) zpool_standard_error(hdl, error, desc);
1831                         break;
1832
1833                 case EEXIST:
1834                         (void) zpool_standard_error(hdl, error, desc);
1835                         break;
1836
1837                 default:
1838                         (void) zpool_standard_error(hdl, error, desc);
1839                         zpool_explain_recover(hdl,
1840                             newname ? origname : thename, -error, nv);
1841                         break;
1842                 }
1843
1844                 nvlist_free(nv);
1845                 ret = -1;
1846         } else {
1847                 zpool_handle_t *zhp;
1848
1849                 /*
1850                  * This should never fail, but play it safe anyway.
1851                  */
1852                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1853                         ret = -1;
1854                 else if (zhp != NULL)
1855                         zpool_close(zhp);
1856                 if (policy.zrp_request &
1857                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1858                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1859                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1860                 }
1861                 nvlist_free(nv);
1862                 return (0);
1863         }
1864
1865         zcmd_free_nvlists(&zc);
1866         nvlist_free(props);
1867
1868         return (ret);
1869 }
1870
1871 /*
1872  * Scan the pool.
1873  */
1874 int
1875 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1876 {
1877         zfs_cmd_t zc = { 0 };
1878         char msg[1024];
1879         libzfs_handle_t *hdl = zhp->zpool_hdl;
1880
1881         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1882         zc.zc_cookie = func;
1883
1884         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1885             (errno == ENOENT && func != POOL_SCAN_NONE))
1886                 return (0);
1887
1888         if (func == POOL_SCAN_SCRUB) {
1889                 (void) snprintf(msg, sizeof (msg),
1890                     dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1891         } else if (func == POOL_SCAN_NONE) {
1892                 (void) snprintf(msg, sizeof (msg),
1893                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1894                     zc.zc_name);
1895         } else {
1896                 assert(!"unexpected result");
1897         }
1898
1899         if (errno == EBUSY) {
1900                 nvlist_t *nvroot;
1901                 pool_scan_stat_t *ps = NULL;
1902                 uint_t psc;
1903
1904                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
1905                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1906                 (void) nvlist_lookup_uint64_array(nvroot,
1907                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1908                 if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1909                         return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1910                 else
1911                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
1912         } else if (errno == ENOENT) {
1913                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1914         } else {
1915                 return (zpool_standard_error(hdl, errno, msg));
1916         }
1917 }
1918
1919 /*
1920  * This provides a very minimal check whether a given string is likely a
1921  * c#t#d# style string.  Users of this are expected to do their own
1922  * verification of the s# part.
1923  */
1924 #define CTD_CHECK(str)  (str && str[0] == 'c' && isdigit(str[1]))
1925
1926 /*
1927  * More elaborate version for ones which may start with "/dev/dsk/"
1928  * and the like.
1929  */
1930 static int
1931 ctd_check_path(char *str) {
1932         /*
1933          * If it starts with a slash, check the last component.
1934          */
1935         if (str && str[0] == '/') {
1936                 char *tmp = strrchr(str, '/');
1937
1938                 /*
1939                  * If it ends in "/old", check the second-to-last
1940                  * component of the string instead.
1941                  */
1942                 if (tmp != str && strcmp(tmp, "/old") == 0) {
1943                         for (tmp--; *tmp != '/'; tmp--)
1944                                 ;
1945                 }
1946                 str = tmp + 1;
1947         }
1948         return (CTD_CHECK(str));
1949 }
1950
1951 /*
1952  * Find a vdev that matches the search criteria specified. We use the
1953  * the nvpair name to determine how we should look for the device.
1954  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1955  * spare; but FALSE if its an INUSE spare.
1956  */
1957 static nvlist_t *
1958 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1959     boolean_t *l2cache, boolean_t *log)
1960 {
1961         uint_t c, children;
1962         nvlist_t **child;
1963         nvlist_t *ret;
1964         uint64_t is_log;
1965         char *srchkey;
1966         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
1967
1968         /* Nothing to look for */
1969         if (search == NULL || pair == NULL)
1970                 return (NULL);
1971
1972         /* Obtain the key we will use to search */
1973         srchkey = nvpair_name(pair);
1974
1975         switch (nvpair_type(pair)) {
1976         case DATA_TYPE_UINT64:
1977                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
1978                         uint64_t srchval, theguid;
1979
1980                         verify(nvpair_value_uint64(pair, &srchval) == 0);
1981                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
1982                             &theguid) == 0);
1983                         if (theguid == srchval)
1984                                 return (nv);
1985                 }
1986                 break;
1987
1988         case DATA_TYPE_STRING: {
1989                 char *srchval, *val;
1990
1991                 verify(nvpair_value_string(pair, &srchval) == 0);
1992                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
1993                         break;
1994
1995                 /*
1996                  * Search for the requested value. Special cases:
1997                  *
1998                  * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
1999                  *   "s0" or "s0/old".  The "s0" part is hidden from the user,
2000                  *   but included in the string, so this matches around it.
2001                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
2002                  *
2003                  * Otherwise, all other searches are simple string compares.
2004                  */
2005                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0 &&
2006                     ctd_check_path(val)) {
2007                         uint64_t wholedisk = 0;
2008
2009                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
2010                             &wholedisk);
2011                         if (wholedisk) {
2012                                 int slen = strlen(srchval);
2013                                 int vlen = strlen(val);
2014
2015                                 if (slen != vlen - 2)
2016                                         break;
2017
2018                                 /*
2019                                  * make_leaf_vdev() should only set
2020                                  * wholedisk for ZPOOL_CONFIG_PATHs which
2021                                  * will include "/dev/dsk/", giving plenty of
2022                                  * room for the indices used next.
2023                                  */
2024                                 ASSERT(vlen >= 6);
2025
2026                                 /*
2027                                  * strings identical except trailing "s0"
2028                                  */
2029                                 if (strcmp(&val[vlen - 2], "s0") == 0 &&
2030                                     strncmp(srchval, val, slen) == 0)
2031                                         return (nv);
2032
2033                                 /*
2034                                  * strings identical except trailing "s0/old"
2035                                  */
2036                                 if (strcmp(&val[vlen - 6], "s0/old") == 0 &&
2037                                     strcmp(&srchval[slen - 4], "/old") == 0 &&
2038                                     strncmp(srchval, val, slen - 4) == 0)
2039                                         return (nv);
2040
2041                                 break;
2042                         }
2043                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2044                         char *type, *idx, *end, *p;
2045                         uint64_t id, vdev_id;
2046
2047                         /*
2048                          * Determine our vdev type, keeping in mind
2049                          * that the srchval is composed of a type and
2050                          * vdev id pair (i.e. mirror-4).
2051                          */
2052                         if ((type = strdup(srchval)) == NULL)
2053                                 return (NULL);
2054
2055                         if ((p = strrchr(type, '-')) == NULL) {
2056                                 free(type);
2057                                 break;
2058                         }
2059                         idx = p + 1;
2060                         *p = '\0';
2061
2062                         /*
2063                          * If the types don't match then keep looking.
2064                          */
2065                         if (strncmp(val, type, strlen(val)) != 0) {
2066                                 free(type);
2067                                 break;
2068                         }
2069
2070                         verify(strncmp(type, VDEV_TYPE_RAIDZ,
2071                             strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2072                             strncmp(type, VDEV_TYPE_MIRROR,
2073                             strlen(VDEV_TYPE_MIRROR)) == 0);
2074                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
2075                             &id) == 0);
2076
2077                         errno = 0;
2078                         vdev_id = strtoull(idx, &end, 10);
2079
2080                         free(type);
2081                         if (errno != 0)
2082                                 return (NULL);
2083
2084                         /*
2085                          * Now verify that we have the correct vdev id.
2086                          */
2087                         if (vdev_id == id)
2088                                 return (nv);
2089                 }
2090
2091                 /*
2092                  * Common case
2093                  */
2094                 if (strcmp(srchval, val) == 0)
2095                         return (nv);
2096                 break;
2097         }
2098
2099         default:
2100                 break;
2101         }
2102
2103         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2104             &child, &children) != 0)
2105                 return (NULL);
2106
2107         for (c = 0; c < children; c++) {
2108                 if ((ret = vdev_to_nvlist_iter(child[c], search,
2109                     avail_spare, l2cache, NULL)) != NULL) {
2110                         /*
2111                          * The 'is_log' value is only set for the toplevel
2112                          * vdev, not the leaf vdevs.  So we always lookup the
2113                          * log device from the root of the vdev tree (where
2114                          * 'log' is non-NULL).
2115                          */
2116                         if (log != NULL &&
2117                             nvlist_lookup_uint64(child[c],
2118                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2119                             is_log) {
2120                                 *log = B_TRUE;
2121                         }
2122                         return (ret);
2123                 }
2124         }
2125
2126         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2127             &child, &children) == 0) {
2128                 for (c = 0; c < children; c++) {
2129                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2130                             avail_spare, l2cache, NULL)) != NULL) {
2131                                 *avail_spare = B_TRUE;
2132                                 return (ret);
2133                         }
2134                 }
2135         }
2136
2137         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2138             &child, &children) == 0) {
2139                 for (c = 0; c < children; c++) {
2140                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2141                             avail_spare, l2cache, NULL)) != NULL) {
2142                                 *l2cache = B_TRUE;
2143                                 return (ret);
2144                         }
2145                 }
2146         }
2147
2148         return (NULL);
2149 }
2150
2151 /*
2152  * Given a physical path (minus the "/devices" prefix), find the
2153  * associated vdev.
2154  */
2155 nvlist_t *
2156 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2157     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2158 {
2159         nvlist_t *search, *nvroot, *ret;
2160
2161         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2162         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
2163
2164         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2165             &nvroot) == 0);
2166
2167         *avail_spare = B_FALSE;
2168         *l2cache = B_FALSE;
2169         if (log != NULL)
2170                 *log = B_FALSE;
2171         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2172         nvlist_free(search);
2173
2174         return (ret);
2175 }
2176
2177 /*
2178  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2179  */
2180 boolean_t
2181 zpool_vdev_is_interior(const char *name)
2182 {
2183         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2184             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2185                 return (B_TRUE);
2186         return (B_FALSE);
2187 }
2188
2189 nvlist_t *
2190 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2191     boolean_t *l2cache, boolean_t *log)
2192 {
2193         char buf[MAXPATHLEN];
2194         char *end;
2195         nvlist_t *nvroot, *search, *ret;
2196         uint64_t guid;
2197
2198         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2199
2200         guid = strtoull(path, &end, 10);
2201         if (guid != 0 && *end == '\0') {
2202                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
2203         } else if (zpool_vdev_is_interior(path)) {
2204                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
2205         } else if (path[0] != '/') {
2206                 (void) snprintf(buf, sizeof (buf), "%s%s", _PATH_DEV, path);
2207                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, buf) == 0);
2208         } else {
2209                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
2210         }
2211
2212         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2213             &nvroot) == 0);
2214
2215         *avail_spare = B_FALSE;
2216         *l2cache = B_FALSE;
2217         if (log != NULL)
2218                 *log = B_FALSE;
2219         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2220         nvlist_free(search);
2221
2222         return (ret);
2223 }
2224
2225 static int
2226 vdev_online(nvlist_t *nv)
2227 {
2228         uint64_t ival;
2229
2230         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2231             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2232             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2233                 return (0);
2234
2235         return (1);
2236 }
2237
2238 /*
2239  * Helper function for zpool_get_physpaths().
2240  */
2241 static int
2242 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2243     size_t *bytes_written)
2244 {
2245         size_t bytes_left, pos, rsz;
2246         char *tmppath;
2247         const char *format;
2248
2249         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2250             &tmppath) != 0)
2251                 return (EZFS_NODEVICE);
2252
2253         pos = *bytes_written;
2254         bytes_left = physpath_size - pos;
2255         format = (pos == 0) ? "%s" : " %s";
2256
2257         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2258         *bytes_written += rsz;
2259
2260         if (rsz >= bytes_left) {
2261                 /* if physpath was not copied properly, clear it */
2262                 if (bytes_left != 0) {
2263                         physpath[pos] = 0;
2264                 }
2265                 return (EZFS_NOSPC);
2266         }
2267         return (0);
2268 }
2269
2270 static int
2271 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2272     size_t *rsz, boolean_t is_spare)
2273 {
2274         char *type;
2275         int ret;
2276
2277         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2278                 return (EZFS_INVALCONFIG);
2279
2280         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2281                 /*
2282                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2283                  * For a spare vdev, we only want to boot from the active
2284                  * spare device.
2285                  */
2286                 if (is_spare) {
2287                         uint64_t spare = 0;
2288                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2289                             &spare);
2290                         if (!spare)
2291                                 return (EZFS_INVALCONFIG);
2292                 }
2293
2294                 if (vdev_online(nv)) {
2295                         if ((ret = vdev_get_one_physpath(nv, physpath,
2296                             phypath_size, rsz)) != 0)
2297                                 return (ret);
2298                 }
2299         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2300             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2301             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2302                 nvlist_t **child;
2303                 uint_t count;
2304                 int i, ret;
2305
2306                 if (nvlist_lookup_nvlist_array(nv,
2307                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2308                         return (EZFS_INVALCONFIG);
2309
2310                 for (i = 0; i < count; i++) {
2311                         ret = vdev_get_physpaths(child[i], physpath,
2312                             phypath_size, rsz, is_spare);
2313                         if (ret == EZFS_NOSPC)
2314                                 return (ret);
2315                 }
2316         }
2317
2318         return (EZFS_POOL_INVALARG);
2319 }
2320
2321 /*
2322  * Get phys_path for a root pool config.
2323  * Return 0 on success; non-zero on failure.
2324  */
2325 static int
2326 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2327 {
2328         size_t rsz;
2329         nvlist_t *vdev_root;
2330         nvlist_t **child;
2331         uint_t count;
2332         char *type;
2333
2334         rsz = 0;
2335
2336         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2337             &vdev_root) != 0)
2338                 return (EZFS_INVALCONFIG);
2339
2340         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2341             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2342             &child, &count) != 0)
2343                 return (EZFS_INVALCONFIG);
2344
2345         /*
2346          * root pool can not have EFI labeled disks and can only have
2347          * a single top-level vdev.
2348          */
2349         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1 ||
2350             pool_uses_efi(vdev_root))
2351                 return (EZFS_POOL_INVALARG);
2352
2353         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2354             B_FALSE);
2355
2356         /* No online devices */
2357         if (rsz == 0)
2358                 return (EZFS_NODEVICE);
2359
2360         return (0);
2361 }
2362
2363 /*
2364  * Get phys_path for a root pool
2365  * Return 0 on success; non-zero on failure.
2366  */
2367 int
2368 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2369 {
2370         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2371             phypath_size));
2372 }
2373
2374 /*
2375  * If the device has being dynamically expanded then we need to relabel
2376  * the disk to use the new unallocated space.
2377  */
2378 static int
2379 zpool_relabel_disk(libzfs_handle_t *hdl, const char *name)
2380 {
2381 #ifdef sun
2382         char path[MAXPATHLEN];
2383         char errbuf[1024];
2384         int fd, error;
2385         int (*_efi_use_whole_disk)(int);
2386
2387         if ((_efi_use_whole_disk = (int (*)(int))dlsym(RTLD_DEFAULT,
2388             "efi_use_whole_disk")) == NULL)
2389                 return (-1);
2390
2391         (void) snprintf(path, sizeof (path), "%s/%s", RDISK_ROOT, name);
2392
2393         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
2394                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2395                     "relabel '%s': unable to open device"), name);
2396                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
2397         }
2398
2399         /*
2400          * It's possible that we might encounter an error if the device
2401          * does not have any unallocated space left. If so, we simply
2402          * ignore that error and continue on.
2403          */
2404         error = _efi_use_whole_disk(fd);
2405         (void) close(fd);
2406         if (error && error != VT_ENOSPC) {
2407                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2408                     "relabel '%s': unable to read disk capacity"), name);
2409                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
2410         }
2411 #endif  /* sun */
2412         return (0);
2413 }
2414
2415 /*
2416  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2417  * ZFS_ONLINE_* flags.
2418  */
2419 int
2420 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2421     vdev_state_t *newstate)
2422 {
2423         zfs_cmd_t zc = { 0 };
2424         char msg[1024];
2425         nvlist_t *tgt;
2426         boolean_t avail_spare, l2cache, islog;
2427         libzfs_handle_t *hdl = zhp->zpool_hdl;
2428
2429         if (flags & ZFS_ONLINE_EXPAND) {
2430                 (void) snprintf(msg, sizeof (msg),
2431                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2432         } else {
2433                 (void) snprintf(msg, sizeof (msg),
2434                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2435         }
2436
2437         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2438         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2439             &islog)) == NULL)
2440                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2441
2442         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2443
2444         if (avail_spare)
2445                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2446
2447         if (flags & ZFS_ONLINE_EXPAND ||
2448             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2449                 char *pathname = NULL;
2450                 uint64_t wholedisk = 0;
2451
2452                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2453                     &wholedisk);
2454                 verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
2455                     &pathname) == 0);
2456
2457                 /*
2458                  * XXX - L2ARC 1.0 devices can't support expansion.
2459                  */
2460                 if (l2cache) {
2461                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2462                             "cannot expand cache devices"));
2463                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2464                 }
2465
2466                 if (wholedisk) {
2467                         pathname += strlen(DISK_ROOT) + 1;
2468                         (void) zpool_relabel_disk(hdl, pathname);
2469                 }
2470         }
2471
2472         zc.zc_cookie = VDEV_STATE_ONLINE;
2473         zc.zc_obj = flags;
2474
2475         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2476                 if (errno == EINVAL) {
2477                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2478                             "from this pool into a new one.  Use '%s' "
2479                             "instead"), "zpool detach");
2480                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2481                 }
2482                 return (zpool_standard_error(hdl, errno, msg));
2483         }
2484
2485         *newstate = zc.zc_cookie;
2486         return (0);
2487 }
2488
2489 /*
2490  * Take the specified vdev offline
2491  */
2492 int
2493 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2494 {
2495         zfs_cmd_t zc = { 0 };
2496         char msg[1024];
2497         nvlist_t *tgt;
2498         boolean_t avail_spare, l2cache;
2499         libzfs_handle_t *hdl = zhp->zpool_hdl;
2500
2501         (void) snprintf(msg, sizeof (msg),
2502             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2503
2504         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2505         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2506             NULL)) == NULL)
2507                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2508
2509         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2510
2511         if (avail_spare)
2512                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2513
2514         zc.zc_cookie = VDEV_STATE_OFFLINE;
2515         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2516
2517         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2518                 return (0);
2519
2520         switch (errno) {
2521         case EBUSY:
2522
2523                 /*
2524                  * There are no other replicas of this device.
2525                  */
2526                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2527
2528         case EEXIST:
2529                 /*
2530                  * The log device has unplayed logs
2531                  */
2532                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2533
2534         default:
2535                 return (zpool_standard_error(hdl, errno, msg));
2536         }
2537 }
2538
2539 /*
2540  * Mark the given vdev faulted.
2541  */
2542 int
2543 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2544 {
2545         zfs_cmd_t zc = { 0 };
2546         char msg[1024];
2547         libzfs_handle_t *hdl = zhp->zpool_hdl;
2548
2549         (void) snprintf(msg, sizeof (msg),
2550             dgettext(TEXT_DOMAIN, "cannot fault %llu"), guid);
2551
2552         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2553         zc.zc_guid = guid;
2554         zc.zc_cookie = VDEV_STATE_FAULTED;
2555         zc.zc_obj = aux;
2556
2557         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2558                 return (0);
2559
2560         switch (errno) {
2561         case EBUSY:
2562
2563                 /*
2564                  * There are no other replicas of this device.
2565                  */
2566                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2567
2568         default:
2569                 return (zpool_standard_error(hdl, errno, msg));
2570         }
2571
2572 }
2573
2574 /*
2575  * Mark the given vdev degraded.
2576  */
2577 int
2578 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2579 {
2580         zfs_cmd_t zc = { 0 };
2581         char msg[1024];
2582         libzfs_handle_t *hdl = zhp->zpool_hdl;
2583
2584         (void) snprintf(msg, sizeof (msg),
2585             dgettext(TEXT_DOMAIN, "cannot degrade %llu"), guid);
2586
2587         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2588         zc.zc_guid = guid;
2589         zc.zc_cookie = VDEV_STATE_DEGRADED;
2590         zc.zc_obj = aux;
2591
2592         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2593                 return (0);
2594
2595         return (zpool_standard_error(hdl, errno, msg));
2596 }
2597
2598 /*
2599  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2600  * a hot spare.
2601  */
2602 static boolean_t
2603 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2604 {
2605         nvlist_t **child;
2606         uint_t c, children;
2607         char *type;
2608
2609         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2610             &children) == 0) {
2611                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2612                     &type) == 0);
2613
2614                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2615                     children == 2 && child[which] == tgt)
2616                         return (B_TRUE);
2617
2618                 for (c = 0; c < children; c++)
2619                         if (is_replacing_spare(child[c], tgt, which))
2620                                 return (B_TRUE);
2621         }
2622
2623         return (B_FALSE);
2624 }
2625
2626 /*
2627  * Attach new_disk (fully described by nvroot) to old_disk.
2628  * If 'replacing' is specified, the new disk will replace the old one.
2629  */
2630 int
2631 zpool_vdev_attach(zpool_handle_t *zhp,
2632     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2633 {
2634         zfs_cmd_t zc = { 0 };
2635         char msg[1024];
2636         int ret;
2637         nvlist_t *tgt;
2638         boolean_t avail_spare, l2cache, islog;
2639         uint64_t val;
2640         char *newname;
2641         nvlist_t **child;
2642         uint_t children;
2643         nvlist_t *config_root;
2644         libzfs_handle_t *hdl = zhp->zpool_hdl;
2645         boolean_t rootpool = zpool_is_bootable(zhp);
2646
2647         if (replacing)
2648                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2649                     "cannot replace %s with %s"), old_disk, new_disk);
2650         else
2651                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2652                     "cannot attach %s to %s"), new_disk, old_disk);
2653
2654         /*
2655          * If this is a root pool, make sure that we're not attaching an
2656          * EFI labeled device.
2657          */
2658         if (rootpool && pool_uses_efi(nvroot)) {
2659                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2660                     "EFI labeled devices are not supported on root pools."));
2661                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
2662         }
2663
2664         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2665         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2666             &islog)) == 0)
2667                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2668
2669         if (avail_spare)
2670                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2671
2672         if (l2cache)
2673                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2674
2675         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2676         zc.zc_cookie = replacing;
2677
2678         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2679             &child, &children) != 0 || children != 1) {
2680                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2681                     "new device must be a single disk"));
2682                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2683         }
2684
2685         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2686             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2687
2688         if ((newname = zpool_vdev_name(NULL, NULL, child[0], B_FALSE)) == NULL)
2689                 return (-1);
2690
2691         /*
2692          * If the target is a hot spare that has been swapped in, we can only
2693          * replace it with another hot spare.
2694          */
2695         if (replacing &&
2696             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2697             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2698             NULL) == NULL || !avail_spare) &&
2699             is_replacing_spare(config_root, tgt, 1)) {
2700                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2701                     "can only be replaced by another hot spare"));
2702                 free(newname);
2703                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2704         }
2705
2706         free(newname);
2707
2708         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2709                 return (-1);
2710
2711         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2712
2713         zcmd_free_nvlists(&zc);
2714
2715         if (ret == 0) {
2716                 if (rootpool) {
2717                         /*
2718                          * XXX need a better way to prevent user from
2719                          * booting up a half-baked vdev.
2720                          */
2721                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2722                             "sure to wait until resilver is done "
2723                             "before rebooting.\n"));
2724                         (void) fprintf(stderr, "\n");
2725                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "If "
2726                             "you boot from pool '%s', you may need to update\n"
2727                             "boot code on newly attached disk '%s'.\n\n"
2728                             "Assuming you use GPT partitioning and 'da0' is "
2729                             "your new boot disk\n"
2730                             "you may use the following command:\n\n"
2731                             "\tgpart bootcode -b /boot/pmbr -p "
2732                             "/boot/gptzfsboot -i 1 da0\n\n"),
2733                             zhp->zpool_name, new_disk);
2734                 }
2735                 return (0);
2736         }
2737
2738         switch (errno) {
2739         case ENOTSUP:
2740                 /*
2741                  * Can't attach to or replace this type of vdev.
2742                  */
2743                 if (replacing) {
2744                         uint64_t version = zpool_get_prop_int(zhp,
2745                             ZPOOL_PROP_VERSION, NULL);
2746
2747                         if (islog)
2748                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2749                                     "cannot replace a log with a spare"));
2750                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2751                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2752                                     "already in replacing/spare config; wait "
2753                                     "for completion or use 'zpool detach'"));
2754                         else
2755                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2756                                     "cannot replace a replacing device"));
2757                 } else {
2758                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2759                             "can only attach to mirrors and top-level "
2760                             "disks"));
2761                 }
2762                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2763                 break;
2764
2765         case EINVAL:
2766                 /*
2767                  * The new device must be a single disk.
2768                  */
2769                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2770                     "new device must be a single disk"));
2771                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2772                 break;
2773
2774         case EBUSY:
2775                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2776                     new_disk);
2777                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2778                 break;
2779
2780         case EOVERFLOW:
2781                 /*
2782                  * The new device is too small.
2783                  */
2784                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2785                     "device is too small"));
2786                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2787                 break;
2788
2789         case EDOM:
2790                 /*
2791                  * The new device has a different alignment requirement.
2792                  */
2793                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2794                     "devices have different sector alignment"));
2795                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2796                 break;
2797
2798         case ENAMETOOLONG:
2799                 /*
2800                  * The resulting top-level vdev spec won't fit in the label.
2801                  */
2802                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2803                 break;
2804
2805         default:
2806                 (void) zpool_standard_error(hdl, errno, msg);
2807         }
2808
2809         return (-1);
2810 }
2811
2812 /*
2813  * Detach the specified device.
2814  */
2815 int
2816 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2817 {
2818         zfs_cmd_t zc = { 0 };
2819         char msg[1024];
2820         nvlist_t *tgt;
2821         boolean_t avail_spare, l2cache;
2822         libzfs_handle_t *hdl = zhp->zpool_hdl;
2823
2824         (void) snprintf(msg, sizeof (msg),
2825             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2826
2827         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2828         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2829             NULL)) == 0)
2830                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2831
2832         if (avail_spare)
2833                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2834
2835         if (l2cache)
2836                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2837
2838         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2839
2840         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2841                 return (0);
2842
2843         switch (errno) {
2844
2845         case ENOTSUP:
2846                 /*
2847                  * Can't detach from this type of vdev.
2848                  */
2849                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2850                     "applicable to mirror and replacing vdevs"));
2851                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2852                 break;
2853
2854         case EBUSY:
2855                 /*
2856                  * There are no other replicas of this device.
2857                  */
2858                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2859                 break;
2860
2861         default:
2862                 (void) zpool_standard_error(hdl, errno, msg);
2863         }
2864
2865         return (-1);
2866 }
2867
2868 /*
2869  * Find a mirror vdev in the source nvlist.
2870  *
2871  * The mchild array contains a list of disks in one of the top-level mirrors
2872  * of the source pool.  The schild array contains a list of disks that the
2873  * user specified on the command line.  We loop over the mchild array to
2874  * see if any entry in the schild array matches.
2875  *
2876  * If a disk in the mchild array is found in the schild array, we return
2877  * the index of that entry.  Otherwise we return -1.
2878  */
2879 static int
2880 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2881     nvlist_t **schild, uint_t schildren)
2882 {
2883         uint_t mc;
2884
2885         for (mc = 0; mc < mchildren; mc++) {
2886                 uint_t sc;
2887                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2888                     mchild[mc], B_FALSE);
2889
2890                 for (sc = 0; sc < schildren; sc++) {
2891                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2892                             schild[sc], B_FALSE);
2893                         boolean_t result = (strcmp(mpath, spath) == 0);
2894
2895                         free(spath);
2896                         if (result) {
2897                                 free(mpath);
2898                                 return (mc);
2899                         }
2900                 }
2901
2902                 free(mpath);
2903         }
2904
2905         return (-1);
2906 }
2907
2908 /*
2909  * Split a mirror pool.  If newroot points to null, then a new nvlist
2910  * is generated and it is the responsibility of the caller to free it.
2911  */
2912 int
2913 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2914     nvlist_t *props, splitflags_t flags)
2915 {
2916         zfs_cmd_t zc = { 0 };
2917         char msg[1024];
2918         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2919         nvlist_t **varray = NULL, *zc_props = NULL;
2920         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2921         libzfs_handle_t *hdl = zhp->zpool_hdl;
2922         uint64_t vers;
2923         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2924         int retval = 0;
2925
2926         (void) snprintf(msg, sizeof (msg),
2927             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2928
2929         if (!zpool_name_valid(hdl, B_FALSE, newname))
2930                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2931
2932         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2933                 (void) fprintf(stderr, gettext("Internal error: unable to "
2934                     "retrieve pool configuration\n"));
2935                 return (-1);
2936         }
2937
2938         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2939             == 0);
2940         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2941
2942         if (props) {
2943                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2944                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2945                     props, vers, flags, msg)) == NULL)
2946                         return (-1);
2947         }
2948
2949         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2950             &children) != 0) {
2951                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2952                     "Source pool is missing vdev tree"));
2953                 if (zc_props)
2954                         nvlist_free(zc_props);
2955                 return (-1);
2956         }
2957
2958         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2959         vcount = 0;
2960
2961         if (*newroot == NULL ||
2962             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2963             &newchild, &newchildren) != 0)
2964                 newchildren = 0;
2965
2966         for (c = 0; c < children; c++) {
2967                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2968                 char *type;
2969                 nvlist_t **mchild, *vdev;
2970                 uint_t mchildren;
2971                 int entry;
2972
2973                 /*
2974                  * Unlike cache & spares, slogs are stored in the
2975                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2976                  */
2977                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2978                     &is_log);
2979                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2980                     &is_hole);
2981                 if (is_log || is_hole) {
2982                         /*
2983                          * Create a hole vdev and put it in the config.
2984                          */
2985                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2986                                 goto out;
2987                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2988                             VDEV_TYPE_HOLE) != 0)
2989                                 goto out;
2990                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2991                             1) != 0)
2992                                 goto out;
2993                         if (lastlog == 0)
2994                                 lastlog = vcount;
2995                         varray[vcount++] = vdev;
2996                         continue;
2997                 }
2998                 lastlog = 0;
2999                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
3000                     == 0);
3001                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
3002                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3003                             "Source pool must be composed only of mirrors\n"));
3004                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3005                         goto out;
3006                 }
3007
3008                 verify(nvlist_lookup_nvlist_array(child[c],
3009                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
3010
3011                 /* find or add an entry for this top-level vdev */
3012                 if (newchildren > 0 &&
3013                     (entry = find_vdev_entry(zhp, mchild, mchildren,
3014                     newchild, newchildren)) >= 0) {
3015                         /* We found a disk that the user specified. */
3016                         vdev = mchild[entry];
3017                         ++found;
3018                 } else {
3019                         /* User didn't specify a disk for this vdev. */
3020                         vdev = mchild[mchildren - 1];
3021                 }
3022
3023                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
3024                         goto out;
3025         }
3026
3027         /* did we find every disk the user specified? */
3028         if (found != newchildren) {
3029                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
3030                     "include at most one disk from each mirror"));
3031                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3032                 goto out;
3033         }
3034
3035         /* Prepare the nvlist for populating. */
3036         if (*newroot == NULL) {
3037                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
3038                         goto out;
3039                 freelist = B_TRUE;
3040                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
3041                     VDEV_TYPE_ROOT) != 0)
3042                         goto out;
3043         } else {
3044                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
3045         }
3046
3047         /* Add all the children we found */
3048         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
3049             lastlog == 0 ? vcount : lastlog) != 0)
3050                 goto out;
3051
3052         /*
3053          * If we're just doing a dry run, exit now with success.
3054          */
3055         if (flags.dryrun) {
3056                 memory_err = B_FALSE;
3057                 freelist = B_FALSE;
3058                 goto out;
3059         }
3060
3061         /* now build up the config list & call the ioctl */
3062         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3063                 goto out;
3064
3065         if (nvlist_add_nvlist(newconfig,
3066             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3067             nvlist_add_string(newconfig,
3068             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3069             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3070                 goto out;
3071
3072         /*
3073          * The new pool is automatically part of the namespace unless we
3074          * explicitly export it.
3075          */
3076         if (!flags.import)
3077                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3078         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3079         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3080         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
3081                 goto out;
3082         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
3083                 goto out;
3084
3085         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3086                 retval = zpool_standard_error(hdl, errno, msg);
3087                 goto out;
3088         }
3089
3090         freelist = B_FALSE;
3091         memory_err = B_FALSE;
3092
3093 out:
3094         if (varray != NULL) {
3095                 int v;
3096
3097                 for (v = 0; v < vcount; v++)
3098                         nvlist_free(varray[v]);
3099                 free(varray);
3100         }
3101         zcmd_free_nvlists(&zc);
3102         if (zc_props)
3103                 nvlist_free(zc_props);
3104         if (newconfig)
3105                 nvlist_free(newconfig);
3106         if (freelist) {
3107                 nvlist_free(*newroot);
3108                 *newroot = NULL;
3109         }
3110
3111         if (retval != 0)
3112                 return (retval);
3113
3114         if (memory_err)
3115                 return (no_memory(hdl));
3116
3117         return (0);
3118 }
3119
3120 /*
3121  * Remove the given device.  Currently, this is supported only for hot spares
3122  * and level 2 cache devices.
3123  */
3124 int
3125 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3126 {
3127         zfs_cmd_t zc = { 0 };
3128         char msg[1024];
3129         nvlist_t *tgt;
3130         boolean_t avail_spare, l2cache, islog;
3131         libzfs_handle_t *hdl = zhp->zpool_hdl;
3132         uint64_t version;
3133
3134         (void) snprintf(msg, sizeof (msg),
3135             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3136
3137         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3138         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3139             &islog)) == 0)
3140                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3141         /*
3142          * XXX - this should just go away.
3143          */
3144         if (!avail_spare && !l2cache && !islog) {
3145                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3146                     "only inactive hot spares, cache, top-level, "
3147                     "or log devices can be removed"));
3148                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3149         }
3150
3151         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3152         if (islog && version < SPA_VERSION_HOLES) {
3153                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3154                     "pool must be upgrade to support log removal"));
3155                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
3156         }
3157
3158         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
3159
3160         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3161                 return (0);
3162
3163         return (zpool_standard_error(hdl, errno, msg));
3164 }
3165
3166 /*
3167  * Clear the errors for the pool, or the particular device if specified.
3168  */
3169 int
3170 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3171 {
3172         zfs_cmd_t zc = { 0 };
3173         char msg[1024];
3174         nvlist_t *tgt;
3175         zpool_rewind_policy_t policy;
3176         boolean_t avail_spare, l2cache;
3177         libzfs_handle_t *hdl = zhp->zpool_hdl;
3178         nvlist_t *nvi = NULL;
3179         int error;
3180
3181         if (path)
3182                 (void) snprintf(msg, sizeof (msg),
3183                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3184                     path);
3185         else
3186                 (void) snprintf(msg, sizeof (msg),
3187                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3188                     zhp->zpool_name);
3189
3190         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3191         if (path) {
3192                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3193                     &l2cache, NULL)) == 0)
3194                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
3195
3196                 /*
3197                  * Don't allow error clearing for hot spares.  Do allow
3198                  * error clearing for l2cache devices.
3199                  */
3200                 if (avail_spare)
3201                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
3202
3203                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
3204                     &zc.zc_guid) == 0);
3205         }
3206
3207         zpool_get_rewind_policy(rewindnvl, &policy);
3208         zc.zc_cookie = policy.zrp_request;
3209
3210         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
3211                 return (-1);
3212
3213         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
3214                 return (-1);
3215
3216         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3217             errno == ENOMEM) {
3218                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3219                         zcmd_free_nvlists(&zc);
3220                         return (-1);
3221                 }
3222         }
3223
3224         if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
3225             errno != EPERM && errno != EACCES)) {
3226                 if (policy.zrp_request &
3227                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
3228                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3229                         zpool_rewind_exclaim(hdl, zc.zc_name,
3230                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
3231                             nvi);
3232                         nvlist_free(nvi);
3233                 }
3234                 zcmd_free_nvlists(&zc);
3235                 return (0);
3236         }
3237
3238         zcmd_free_nvlists(&zc);
3239         return (zpool_standard_error(hdl, errno, msg));
3240 }
3241
3242 /*
3243  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3244  */
3245 int
3246 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3247 {
3248         zfs_cmd_t zc = { 0 };
3249         char msg[1024];
3250         libzfs_handle_t *hdl = zhp->zpool_hdl;
3251
3252         (void) snprintf(msg, sizeof (msg),
3253             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3254             guid);
3255
3256         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3257         zc.zc_guid = guid;
3258         zc.zc_cookie = ZPOOL_NO_REWIND;
3259
3260         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
3261                 return (0);
3262
3263         return (zpool_standard_error(hdl, errno, msg));
3264 }
3265
3266 /*
3267  * Change the GUID for a pool.
3268  */
3269 int
3270 zpool_reguid(zpool_handle_t *zhp)
3271 {
3272         char msg[1024];
3273         libzfs_handle_t *hdl = zhp->zpool_hdl;
3274         zfs_cmd_t zc = { 0 };
3275
3276         (void) snprintf(msg, sizeof (msg),
3277             dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3278
3279         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3280         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3281                 return (0);
3282
3283         return (zpool_standard_error(hdl, errno, msg));
3284 }
3285
3286 /*
3287  * Reopen the pool.
3288  */
3289 int
3290 zpool_reopen(zpool_handle_t *zhp)
3291 {
3292         zfs_cmd_t zc = { 0 };
3293         char msg[1024];
3294         libzfs_handle_t *hdl = zhp->zpool_hdl;
3295
3296         (void) snprintf(msg, sizeof (msg),
3297             dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3298             zhp->zpool_name);
3299
3300         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3301         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3302                 return (0);
3303         return (zpool_standard_error(hdl, errno, msg));
3304 }
3305
3306 /*
3307  * Convert from a devid string to a path.
3308  */
3309 static char *
3310 devid_to_path(char *devid_str)
3311 {
3312         ddi_devid_t devid;
3313         char *minor;
3314         char *path;
3315         devid_nmlist_t *list = NULL;
3316         int ret;
3317
3318         if (devid_str_decode(devid_str, &devid, &minor) != 0)
3319                 return (NULL);
3320
3321         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3322
3323         devid_str_free(minor);
3324         devid_free(devid);
3325
3326         if (ret != 0)
3327                 return (NULL);
3328
3329         if ((path = strdup(list[0].devname)) == NULL)
3330                 return (NULL);
3331
3332         devid_free_nmlist(list);
3333
3334         return (path);
3335 }
3336
3337 /*
3338  * Convert from a path to a devid string.
3339  */
3340 static char *
3341 path_to_devid(const char *path)
3342 {
3343 #ifdef have_devid
3344         int fd;
3345         ddi_devid_t devid;
3346         char *minor, *ret;
3347
3348         if ((fd = open(path, O_RDONLY)) < 0)
3349                 return (NULL);
3350
3351         minor = NULL;
3352         ret = NULL;
3353         if (devid_get(fd, &devid) == 0) {
3354                 if (devid_get_minor_name(fd, &minor) == 0)
3355                         ret = devid_str_encode(devid, minor);
3356                 if (minor != NULL)
3357                         devid_str_free(minor);
3358                 devid_free(devid);
3359         }
3360         (void) close(fd);
3361
3362         return (ret);
3363 #else
3364         return (NULL);
3365 #endif
3366 }
3367
3368 /*
3369  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3370  * ignore any failure here, since a common case is for an unprivileged user to
3371  * type 'zpool status', and we'll display the correct information anyway.
3372  */
3373 static void
3374 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3375 {
3376         zfs_cmd_t zc = { 0 };
3377
3378         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3379         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3380         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3381             &zc.zc_guid) == 0);
3382
3383         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3384 }
3385
3386 /*
3387  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3388  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3389  * We also check if this is a whole disk, in which case we strip off the
3390  * trailing 's0' slice name.
3391  *
3392  * This routine is also responsible for identifying when disks have been
3393  * reconfigured in a new location.  The kernel will have opened the device by
3394  * devid, but the path will still refer to the old location.  To catch this, we
3395  * first do a path -> devid translation (which is fast for the common case).  If
3396  * the devid matches, we're done.  If not, we do a reverse devid -> path
3397  * translation and issue the appropriate ioctl() to update the path of the vdev.
3398  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3399  * of these checks.
3400  */
3401 char *
3402 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3403     boolean_t verbose)
3404 {
3405         char *path, *devid;
3406         uint64_t value;
3407         char buf[64];
3408         vdev_stat_t *vs;
3409         uint_t vsc;
3410         int have_stats;
3411         int have_path;
3412
3413         have_stats = nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3414             (uint64_t **)&vs, &vsc) == 0;
3415         have_path = nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0;
3416
3417         /*
3418          * If the device is not currently present, assume it will not
3419          * come back at the same device path.  Display the device by GUID.
3420          */
3421         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &value) == 0 ||
3422             have_path && have_stats && vs->vs_state <= VDEV_STATE_CANT_OPEN) {
3423                 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3424                     &value) == 0);
3425                 (void) snprintf(buf, sizeof (buf), "%llu",
3426                     (u_longlong_t)value);
3427                 path = buf;
3428         } else if (have_path) {
3429
3430                 /*
3431                  * If the device is dead (faulted, offline, etc) then don't
3432                  * bother opening it.  Otherwise we may be forcing the user to
3433                  * open a misbehaving device, which can have undesirable
3434                  * effects.
3435                  */
3436                 if ((have_stats == 0 ||
3437                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3438                     zhp != NULL &&
3439                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3440                         /*
3441                          * Determine if the current path is correct.
3442                          */
3443                         char *newdevid = path_to_devid(path);
3444
3445                         if (newdevid == NULL ||
3446                             strcmp(devid, newdevid) != 0) {
3447                                 char *newpath;
3448
3449                                 if ((newpath = devid_to_path(devid)) != NULL) {
3450                                         /*
3451                                          * Update the path appropriately.
3452                                          */
3453                                         set_path(zhp, nv, newpath);
3454                                         if (nvlist_add_string(nv,
3455                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3456                                                 verify(nvlist_lookup_string(nv,
3457                                                     ZPOOL_CONFIG_PATH,
3458                                                     &path) == 0);
3459                                         free(newpath);
3460                                 }
3461                         }
3462
3463                         if (newdevid)
3464                                 devid_str_free(newdevid);
3465                 }
3466
3467 #ifdef sun
3468                 if (strncmp(path, "/dev/dsk/", 9) == 0)
3469                         path += 9;
3470
3471                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
3472                     &value) == 0 && value) {
3473                         int pathlen = strlen(path);
3474                         char *tmp = zfs_strdup(hdl, path);
3475
3476                         /*
3477                          * If it starts with c#, and ends with "s0", chop
3478                          * the "s0" off, or if it ends with "s0/old", remove
3479                          * the "s0" from the middle.
3480                          */
3481                         if (CTD_CHECK(tmp)) {
3482                                 if (strcmp(&tmp[pathlen - 2], "s0") == 0) {
3483                                         tmp[pathlen - 2] = '\0';
3484                                 } else if (pathlen > 6 &&
3485                                     strcmp(&tmp[pathlen - 6], "s0/old") == 0) {
3486                                         (void) strcpy(&tmp[pathlen - 6],
3487                                             "/old");
3488                                 }
3489                         }
3490                         return (tmp);
3491                 }
3492 #else   /* !sun */
3493                 if (strncmp(path, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
3494                         path += sizeof(_PATH_DEV) - 1;
3495 #endif  /* !sun */
3496         } else {
3497                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3498
3499                 /*
3500                  * If it's a raidz device, we need to stick in the parity level.
3501                  */
3502                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3503                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3504                             &value) == 0);
3505                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3506                             (u_longlong_t)value);
3507                         path = buf;
3508                 }
3509
3510                 /*
3511                  * We identify each top-level vdev by using a <type-id>
3512                  * naming convention.
3513                  */
3514                 if (verbose) {
3515                         uint64_t id;
3516
3517                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3518                             &id) == 0);
3519                         (void) snprintf(buf, sizeof (buf), "%s-%llu", path,
3520                             (u_longlong_t)id);
3521                         path = buf;
3522                 }
3523         }
3524
3525         return (zfs_strdup(hdl, path));
3526 }
3527
3528 static int
3529 zbookmark_compare(const void *a, const void *b)
3530 {
3531         return (memcmp(a, b, sizeof (zbookmark_phys_t)));
3532 }
3533
3534 /*
3535  * Retrieve the persistent error log, uniquify the members, and return to the
3536  * caller.
3537  */
3538 int
3539 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3540 {
3541         zfs_cmd_t zc = { 0 };
3542         uint64_t count;
3543         zbookmark_phys_t *zb = NULL;
3544         int i;
3545
3546         /*
3547          * Retrieve the raw error list from the kernel.  If the number of errors
3548          * has increased, allocate more space and continue until we get the
3549          * entire list.
3550          */
3551         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3552             &count) == 0);
3553         if (count == 0)
3554                 return (0);
3555         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3556             count * sizeof (zbookmark_phys_t))) == (uintptr_t)NULL)
3557                 return (-1);
3558         zc.zc_nvlist_dst_size = count;
3559         (void) strcpy(zc.zc_name, zhp->zpool_name);
3560         for (;;) {
3561                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3562                     &zc) != 0) {
3563                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3564                         if (errno == ENOMEM) {
3565                                 void *dst;
3566
3567                                 count = zc.zc_nvlist_dst_size;
3568                                 dst = zfs_alloc(zhp->zpool_hdl, count *
3569                                     sizeof (zbookmark_phys_t));
3570                                 if (dst == NULL)
3571                                         return (-1);
3572                                 zc.zc_nvlist_dst = (uintptr_t)dst;
3573                         } else {
3574                                 return (-1);
3575                         }
3576                 } else {
3577                         break;
3578                 }
3579         }
3580
3581         /*
3582          * Sort the resulting bookmarks.  This is a little confusing due to the
3583          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3584          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3585          * _not_ copied as part of the process.  So we point the start of our
3586          * array appropriate and decrement the total number of elements.
3587          */
3588         zb = ((zbookmark_phys_t *)(uintptr_t)zc.zc_nvlist_dst) +
3589             zc.zc_nvlist_dst_size;
3590         count -= zc.zc_nvlist_dst_size;
3591
3592         qsort(zb, count, sizeof (zbookmark_phys_t), zbookmark_compare);
3593
3594         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3595
3596         /*
3597          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3598          */
3599         for (i = 0; i < count; i++) {
3600                 nvlist_t *nv;
3601
3602                 /* ignoring zb_blkid and zb_level for now */
3603                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3604                     zb[i-1].zb_object == zb[i].zb_object)
3605                         continue;
3606
3607                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3608                         goto nomem;
3609                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3610                     zb[i].zb_objset) != 0) {
3611                         nvlist_free(nv);
3612                         goto nomem;
3613                 }
3614                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3615                     zb[i].zb_object) != 0) {
3616                         nvlist_free(nv);
3617                         goto nomem;
3618                 }
3619                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3620                         nvlist_free(nv);
3621                         goto nomem;
3622                 }
3623                 nvlist_free(nv);
3624         }
3625
3626         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3627         return (0);
3628
3629 nomem:
3630         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3631         return (no_memory(zhp->zpool_hdl));
3632 }
3633
3634 /*
3635  * Upgrade a ZFS pool to the latest on-disk version.
3636  */
3637 int
3638 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3639 {
3640         zfs_cmd_t zc = { 0 };
3641         libzfs_handle_t *hdl = zhp->zpool_hdl;
3642
3643         (void) strcpy(zc.zc_name, zhp->zpool_name);
3644         zc.zc_cookie = new_version;
3645
3646         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3647                 return (zpool_standard_error_fmt(hdl, errno,
3648                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3649                     zhp->zpool_name));
3650         return (0);
3651 }
3652
3653 void
3654 zfs_save_arguments(int argc, char **argv, char *string, int len)
3655 {
3656         (void) strlcpy(string, basename(argv[0]), len);
3657         for (int i = 1; i < argc; i++) {
3658                 (void) strlcat(string, " ", len);
3659                 (void) strlcat(string, argv[i], len);
3660         }
3661 }
3662
3663 int
3664 zpool_log_history(libzfs_handle_t *hdl, const char *message)
3665 {
3666         zfs_cmd_t zc = { 0 };
3667         nvlist_t *args;
3668         int err;
3669
3670         args = fnvlist_alloc();
3671         fnvlist_add_string(args, "message", message);
3672         err = zcmd_write_src_nvlist(hdl, &zc, args);
3673         if (err == 0)
3674                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_LOG_HISTORY, &zc);
3675         nvlist_free(args);
3676         zcmd_free_nvlists(&zc);
3677         return (err);
3678 }
3679
3680 /*
3681  * Perform ioctl to get some command history of a pool.
3682  *
3683  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3684  * logical offset of the history buffer to start reading from.
3685  *
3686  * Upon return, 'off' is the next logical offset to read from and
3687  * 'len' is the actual amount of bytes read into 'buf'.
3688  */
3689 static int
3690 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3691 {
3692         zfs_cmd_t zc = { 0 };
3693         libzfs_handle_t *hdl = zhp->zpool_hdl;
3694
3695         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3696
3697         zc.zc_history = (uint64_t)(uintptr_t)buf;
3698         zc.zc_history_len = *len;
3699         zc.zc_history_offset = *off;
3700
3701         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3702                 switch (errno) {
3703                 case EPERM:
3704                         return (zfs_error_fmt(hdl, EZFS_PERM,
3705                             dgettext(TEXT_DOMAIN,
3706                             "cannot show history for pool '%s'"),
3707                             zhp->zpool_name));
3708                 case ENOENT:
3709                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3710                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3711                             "'%s'"), zhp->zpool_name));
3712                 case ENOTSUP:
3713                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3714                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3715                             "'%s', pool must be upgraded"), zhp->zpool_name));
3716                 default:
3717                         return (zpool_standard_error_fmt(hdl, errno,
3718                             dgettext(TEXT_DOMAIN,
3719                             "cannot get history for '%s'"), zhp->zpool_name));
3720                 }
3721         }
3722
3723         *len = zc.zc_history_len;
3724         *off = zc.zc_history_offset;
3725
3726         return (0);
3727 }
3728
3729 /*
3730  * Process the buffer of nvlists, unpacking and storing each nvlist record
3731  * into 'records'.  'leftover' is set to the number of bytes that weren't
3732  * processed as there wasn't a complete record.
3733  */
3734 int
3735 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3736     nvlist_t ***records, uint_t *numrecords)
3737 {
3738         uint64_t reclen;
3739         nvlist_t *nv;
3740         int i;
3741
3742         while (bytes_read > sizeof (reclen)) {
3743
3744                 /* get length of packed record (stored as little endian) */
3745                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3746                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3747
3748                 if (bytes_read < sizeof (reclen) + reclen)
3749                         break;
3750
3751                 /* unpack record */
3752                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3753                         return (ENOMEM);
3754                 bytes_read -= sizeof (reclen) + reclen;
3755                 buf += sizeof (reclen) + reclen;
3756
3757                 /* add record to nvlist array */
3758                 (*numrecords)++;
3759                 if (ISP2(*numrecords + 1)) {
3760                         *records = realloc(*records,
3761                             *numrecords * 2 * sizeof (nvlist_t *));
3762                 }
3763                 (*records)[*numrecords - 1] = nv;
3764         }
3765
3766         *leftover = bytes_read;
3767         return (0);
3768 }
3769
3770 /* from spa_history.c: spa_history_create_obj() */
3771 #define HIS_BUF_LEN_DEF (128 << 10)
3772 #define HIS_BUF_LEN_MAX (1 << 30)
3773
3774 /*
3775  * Retrieve the command history of a pool.
3776  */
3777 int
3778 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3779 {
3780         char *buf = NULL;
3781         uint64_t bufsize = HIS_BUF_LEN_DEF;
3782         uint64_t off = 0;
3783         nvlist_t **records = NULL;
3784         uint_t numrecords = 0;
3785         int err, i;
3786
3787         if ((buf = malloc(bufsize)) == NULL)
3788                 return (ENOMEM);
3789         do {
3790                 uint64_t bytes_read = bufsize;
3791                 uint64_t leftover;
3792
3793                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3794                         break;
3795
3796                 /* if nothing else was read in, we're at EOF, just return */
3797                 if (bytes_read == 0)
3798                         break;
3799
3800                 if ((err = zpool_history_unpack(buf, bytes_read,
3801                     &leftover, &records, &numrecords)) != 0)
3802                         break;
3803                 off -= leftover;
3804
3805                 /*
3806                  * If the history block is too big, double the buffer
3807                  * size and try again.
3808                  */
3809                 if (leftover == bytes_read) {
3810                         free(buf);
3811                         buf = NULL;
3812
3813                         bufsize <<= 1;
3814                         if ((bufsize >= HIS_BUF_LEN_MAX) ||
3815                             ((buf = malloc(bufsize)) == NULL)) {
3816                                 err = ENOMEM;
3817                                 break;
3818                         }
3819                 }
3820
3821                 /* CONSTCOND */
3822         } while (1);
3823         free(buf);
3824
3825         if (!err) {
3826                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3827                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3828                     records, numrecords) == 0);
3829         }
3830         for (i = 0; i < numrecords; i++)
3831                 nvlist_free(records[i]);
3832         free(records);
3833
3834         return (err);
3835 }
3836
3837 void
3838 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
3839     char *pathname, size_t len)
3840 {
3841         zfs_cmd_t zc = { 0 };
3842         boolean_t mounted = B_FALSE;
3843         char *mntpnt = NULL;
3844         char dsname[MAXNAMELEN];
3845
3846         if (dsobj == 0) {
3847                 /* special case for the MOS */
3848                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>", obj);
3849                 return;
3850         }
3851
3852         /* get the dataset's name */
3853         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3854         zc.zc_obj = dsobj;
3855         if (ioctl(zhp->zpool_hdl->libzfs_fd,
3856             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
3857                 /* just write out a path of two object numbers */
3858                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
3859                     dsobj, obj);
3860                 return;
3861         }
3862         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
3863
3864         /* find out if the dataset is mounted */
3865         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
3866
3867         /* get the corrupted object's path */
3868         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
3869         zc.zc_obj = obj;
3870         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
3871             &zc) == 0) {
3872                 if (mounted) {
3873                         (void) snprintf(pathname, len, "%s%s", mntpnt,
3874                             zc.zc_value);
3875                 } else {
3876                         (void) snprintf(pathname, len, "%s:%s",
3877                             dsname, zc.zc_value);
3878                 }
3879         } else {
3880                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname, obj);
3881         }
3882         free(mntpnt);
3883 }
3884
3885 #ifdef sun
3886 /*
3887  * Read the EFI label from the config, if a label does not exist then
3888  * pass back the error to the caller. If the caller has passed a non-NULL
3889  * diskaddr argument then we set it to the starting address of the EFI
3890  * partition.
3891  */
3892 static int
3893 read_efi_label(nvlist_t *config, diskaddr_t *sb)
3894 {
3895         char *path;
3896         int fd;
3897         char diskname[MAXPATHLEN];
3898         int err = -1;
3899
3900         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
3901                 return (err);
3902
3903         (void) snprintf(diskname, sizeof (diskname), "%s%s", RDISK_ROOT,
3904             strrchr(path, '/'));
3905         if ((fd = open(diskname, O_RDONLY|O_NDELAY)) >= 0) {
3906                 struct dk_gpt *vtoc;
3907
3908                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
3909                         if (sb != NULL)
3910                                 *sb = vtoc->efi_parts[0].p_start;
3911                         efi_free(vtoc);
3912                 }
3913                 (void) close(fd);
3914         }
3915         return (err);
3916 }
3917
3918 /*
3919  * determine where a partition starts on a disk in the current
3920  * configuration
3921  */
3922 static diskaddr_t
3923 find_start_block(nvlist_t *config)
3924 {
3925         nvlist_t **child;
3926         uint_t c, children;
3927         diskaddr_t sb = MAXOFFSET_T;
3928         uint64_t wholedisk;
3929
3930         if (nvlist_lookup_nvlist_array(config,
3931             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
3932                 if (nvlist_lookup_uint64(config,
3933                     ZPOOL_CONFIG_WHOLE_DISK,
3934                     &wholedisk) != 0 || !wholedisk) {
3935                         return (MAXOFFSET_T);
3936                 }
3937                 if (read_efi_label(config, &sb) < 0)
3938                         sb = MAXOFFSET_T;
3939                 return (sb);
3940         }
3941
3942         for (c = 0; c < children; c++) {
3943                 sb = find_start_block(child[c]);
3944                 if (sb != MAXOFFSET_T) {
3945                         return (sb);
3946                 }
3947         }
3948         return (MAXOFFSET_T);
3949 }
3950 #endif /* sun */
3951
3952 /*
3953  * Label an individual disk.  The name provided is the short name,
3954  * stripped of any leading /dev path.
3955  */
3956 int
3957 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, const char *name)
3958 {
3959 #ifdef sun
3960         char path[MAXPATHLEN];
3961         struct dk_gpt *vtoc;
3962         int fd;
3963         size_t resv = EFI_MIN_RESV_SIZE;
3964         uint64_t slice_size;
3965         diskaddr_t start_block;
3966         char errbuf[1024];
3967
3968         /* prepare an error message just in case */
3969         (void) snprintf(errbuf, sizeof (errbuf),
3970             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
3971
3972         if (zhp) {
3973                 nvlist_t *nvroot;
3974
3975                 if (zpool_is_bootable(zhp)) {
3976                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3977                             "EFI labeled devices are not supported on root "
3978                             "pools."));
3979                         return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
3980                 }
3981
3982                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
3983                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3984
3985                 if (zhp->zpool_start_block == 0)
3986                         start_block = find_start_block(nvroot);
3987                 else
3988                         start_block = zhp->zpool_start_block;
3989                 zhp->zpool_start_block = start_block;
3990         } else {
3991                 /* new pool */
3992                 start_block = NEW_START_BLOCK;
3993         }
3994
3995         (void) snprintf(path, sizeof (path), "%s/%s%s", RDISK_ROOT, name,
3996             BACKUP_SLICE);
3997
3998         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
3999                 /*
4000                  * This shouldn't happen.  We've long since verified that this
4001                  * is a valid device.
4002                  */
4003                 zfs_error_aux(hdl,
4004                     dgettext(TEXT_DOMAIN, "unable to open device"));
4005                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
4006         }
4007
4008         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
4009                 /*
4010                  * The only way this can fail is if we run out of memory, or we
4011                  * were unable to read the disk's capacity
4012                  */
4013                 if (errno == ENOMEM)
4014                         (void) no_memory(hdl);
4015
4016                 (void) close(fd);
4017                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4018                     "unable to read disk capacity"), name);
4019
4020                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
4021         }
4022
4023         slice_size = vtoc->efi_last_u_lba + 1;
4024         slice_size -= EFI_MIN_RESV_SIZE;
4025         if (start_block == MAXOFFSET_T)
4026                 start_block = NEW_START_BLOCK;
4027         slice_size -= start_block;
4028
4029         vtoc->efi_parts[0].p_start = start_block;
4030         vtoc->efi_parts[0].p_size = slice_size;
4031
4032         /*
4033          * Why we use V_USR: V_BACKUP confuses users, and is considered
4034          * disposable by some EFI utilities (since EFI doesn't have a backup
4035          * slice).  V_UNASSIGNED is supposed to be used only for zero size
4036          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
4037          * etc. were all pretty specific.  V_USR is as close to reality as we
4038          * can get, in the absence of V_OTHER.
4039          */
4040         vtoc->efi_parts[0].p_tag = V_USR;
4041         (void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
4042
4043         vtoc->efi_parts[8].p_start = slice_size + start_block;
4044         vtoc->efi_parts[8].p_size = resv;
4045         vtoc->efi_parts[8].p_tag = V_RESERVED;
4046
4047         if (efi_write(fd, vtoc) != 0) {
4048                 /*
4049                  * Some block drivers (like pcata) may not support EFI
4050                  * GPT labels.  Print out a helpful error message dir-
4051                  * ecting the user to manually label the disk and give
4052                  * a specific slice.
4053                  */
4054                 (void) close(fd);
4055                 efi_free(vtoc);
4056
4057                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4058                     "try using fdisk(1M) and then provide a specific slice"));
4059                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4060         }
4061
4062         (void) close(fd);
4063         efi_free(vtoc);
4064 #endif /* sun */
4065         return (0);
4066 }
4067
4068 static boolean_t
4069 supported_dump_vdev_type(libzfs_handle_t *hdl, nvlist_t *config, char *errbuf)
4070 {
4071         char *type;
4072         nvlist_t **child;
4073         uint_t children, c;
4074
4075         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_TYPE, &type) == 0);
4076         if (strcmp(type, VDEV_TYPE_FILE) == 0 ||
4077             strcmp(type, VDEV_TYPE_HOLE) == 0 ||
4078             strcmp(type, VDEV_TYPE_MISSING) == 0) {
4079                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4080                     "vdev type '%s' is not supported"), type);
4081                 (void) zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf);
4082                 return (B_FALSE);
4083         }
4084         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
4085             &child, &children) == 0) {
4086                 for (c = 0; c < children; c++) {
4087                         if (!supported_dump_vdev_type(hdl, child[c], errbuf))
4088                                 return (B_FALSE);
4089                 }
4090         }
4091         return (B_TRUE);
4092 }
4093
4094 /*
4095  * Check if this zvol is allowable for use as a dump device; zero if
4096  * it is, > 0 if it isn't, < 0 if it isn't a zvol.
4097  *
4098  * Allowable storage configurations include mirrors, all raidz variants, and
4099  * pools with log, cache, and spare devices.  Pools which are backed by files or
4100  * have missing/hole vdevs are not suitable.
4101  */
4102 int
4103 zvol_check_dump_config(char *arg)
4104 {
4105         zpool_handle_t *zhp = NULL;
4106         nvlist_t *config, *nvroot;
4107         char *p, *volname;
4108         nvlist_t **top;
4109         uint_t toplevels;
4110         libzfs_handle_t *hdl;
4111         char errbuf[1024];
4112         char poolname[ZPOOL_MAXNAMELEN];
4113         int pathlen = strlen(ZVOL_FULL_DEV_DIR);
4114         int ret = 1;
4115
4116         if (strncmp(arg, ZVOL_FULL_DEV_DIR, pathlen)) {
4117                 return (-1);
4118         }
4119
4120         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
4121             "dump is not supported on device '%s'"), arg);
4122
4123         if ((hdl = libzfs_init()) == NULL)
4124                 return (1);
4125         libzfs_print_on_error(hdl, B_TRUE);
4126
4127         volname = arg + pathlen;
4128
4129         /* check the configuration of the pool */
4130         if ((p = strchr(volname, '/')) == NULL) {
4131                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4132                     "malformed dataset name"));
4133                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4134                 return (1);
4135         } else if (p - volname >= ZFS_MAXNAMELEN) {
4136                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4137                     "dataset name is too long"));
4138                 (void) zfs_error(hdl, EZFS_NAMETOOLONG, errbuf);
4139                 return (1);
4140         } else {
4141                 (void) strncpy(poolname, volname, p - volname);
4142                 poolname[p - volname] = '\0';
4143         }
4144
4145         if ((zhp = zpool_open(hdl, poolname)) == NULL) {
4146                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4147                     "could not open pool '%s'"), poolname);
4148                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
4149                 goto out;
4150         }
4151         config = zpool_get_config(zhp, NULL);
4152         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4153             &nvroot) != 0) {
4154                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4155                     "could not obtain vdev configuration for  '%s'"), poolname);
4156                 (void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
4157                 goto out;
4158         }
4159
4160         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4161             &top, &toplevels) == 0);
4162
4163         if (!supported_dump_vdev_type(hdl, top[0], errbuf)) {
4164                 goto out;
4165         }
4166         ret = 0;
4167
4168 out:
4169         if (zhp)
4170                 zpool_close(zhp);
4171         libzfs_fini(hdl);
4172         return (ret);
4173 }