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 #ifdef illumos
1920 /*
1921  * This provides a very minimal check whether a given string is likely a
1922  * c#t#d# style string.  Users of this are expected to do their own
1923  * verification of the s# part.
1924  */
1925 #define CTD_CHECK(str)  (str && str[0] == 'c' && isdigit(str[1]))
1926
1927 /*
1928  * More elaborate version for ones which may start with "/dev/dsk/"
1929  * and the like.
1930  */
1931 static int
1932 ctd_check_path(char *str) {
1933         /*
1934          * If it starts with a slash, check the last component.
1935          */
1936         if (str && str[0] == '/') {
1937                 char *tmp = strrchr(str, '/');
1938
1939                 /*
1940                  * If it ends in "/old", check the second-to-last
1941                  * component of the string instead.
1942                  */
1943                 if (tmp != str && strcmp(tmp, "/old") == 0) {
1944                         for (tmp--; *tmp != '/'; tmp--)
1945                                 ;
1946                 }
1947                 str = tmp + 1;
1948         }
1949         return (CTD_CHECK(str));
1950 }
1951 #endif
1952
1953 /*
1954  * Find a vdev that matches the search criteria specified. We use the
1955  * the nvpair name to determine how we should look for the device.
1956  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1957  * spare; but FALSE if its an INUSE spare.
1958  */
1959 static nvlist_t *
1960 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1961     boolean_t *l2cache, boolean_t *log)
1962 {
1963         uint_t c, children;
1964         nvlist_t **child;
1965         nvlist_t *ret;
1966         uint64_t is_log;
1967         char *srchkey;
1968         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
1969
1970         /* Nothing to look for */
1971         if (search == NULL || pair == NULL)
1972                 return (NULL);
1973
1974         /* Obtain the key we will use to search */
1975         srchkey = nvpair_name(pair);
1976
1977         switch (nvpair_type(pair)) {
1978         case DATA_TYPE_UINT64:
1979                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
1980                         uint64_t srchval, theguid;
1981
1982                         verify(nvpair_value_uint64(pair, &srchval) == 0);
1983                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
1984                             &theguid) == 0);
1985                         if (theguid == srchval)
1986                                 return (nv);
1987                 }
1988                 break;
1989
1990         case DATA_TYPE_STRING: {
1991                 char *srchval, *val;
1992
1993                 verify(nvpair_value_string(pair, &srchval) == 0);
1994                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
1995                         break;
1996
1997                 /*
1998                  * Search for the requested value. Special cases:
1999                  *
2000                  * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
2001                  *   "s0" or "s0/old".  The "s0" part is hidden from the user,
2002                  *   but included in the string, so this matches around it.
2003                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
2004                  *
2005                  * Otherwise, all other searches are simple string compares.
2006                  */
2007 #ifdef illumos
2008                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0 &&
2009                     ctd_check_path(val)) {
2010                         uint64_t wholedisk = 0;
2011
2012                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
2013                             &wholedisk);
2014                         if (wholedisk) {
2015                                 int slen = strlen(srchval);
2016                                 int vlen = strlen(val);
2017
2018                                 if (slen != vlen - 2)
2019                                         break;
2020
2021                                 /*
2022                                  * make_leaf_vdev() should only set
2023                                  * wholedisk for ZPOOL_CONFIG_PATHs which
2024                                  * will include "/dev/dsk/", giving plenty of
2025                                  * room for the indices used next.
2026                                  */
2027                                 ASSERT(vlen >= 6);
2028
2029                                 /*
2030                                  * strings identical except trailing "s0"
2031                                  */
2032                                 if (strcmp(&val[vlen - 2], "s0") == 0 &&
2033                                     strncmp(srchval, val, slen) == 0)
2034                                         return (nv);
2035
2036                                 /*
2037                                  * strings identical except trailing "s0/old"
2038                                  */
2039                                 if (strcmp(&val[vlen - 6], "s0/old") == 0 &&
2040                                     strcmp(&srchval[slen - 4], "/old") == 0 &&
2041                                     strncmp(srchval, val, slen - 4) == 0)
2042                                         return (nv);
2043
2044                                 break;
2045                         }
2046                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2047 #else
2048                 if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2049 #endif
2050                         char *type, *idx, *end, *p;
2051                         uint64_t id, vdev_id;
2052
2053                         /*
2054                          * Determine our vdev type, keeping in mind
2055                          * that the srchval is composed of a type and
2056                          * vdev id pair (i.e. mirror-4).
2057                          */
2058                         if ((type = strdup(srchval)) == NULL)
2059                                 return (NULL);
2060
2061                         if ((p = strrchr(type, '-')) == NULL) {
2062                                 free(type);
2063                                 break;
2064                         }
2065                         idx = p + 1;
2066                         *p = '\0';
2067
2068                         /*
2069                          * If the types don't match then keep looking.
2070                          */
2071                         if (strncmp(val, type, strlen(val)) != 0) {
2072                                 free(type);
2073                                 break;
2074                         }
2075
2076                         verify(strncmp(type, VDEV_TYPE_RAIDZ,
2077                             strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2078                             strncmp(type, VDEV_TYPE_MIRROR,
2079                             strlen(VDEV_TYPE_MIRROR)) == 0);
2080                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
2081                             &id) == 0);
2082
2083                         errno = 0;
2084                         vdev_id = strtoull(idx, &end, 10);
2085
2086                         free(type);
2087                         if (errno != 0)
2088                                 return (NULL);
2089
2090                         /*
2091                          * Now verify that we have the correct vdev id.
2092                          */
2093                         if (vdev_id == id)
2094                                 return (nv);
2095                 }
2096
2097                 /*
2098                  * Common case
2099                  */
2100                 if (strcmp(srchval, val) == 0)
2101                         return (nv);
2102                 break;
2103         }
2104
2105         default:
2106                 break;
2107         }
2108
2109         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2110             &child, &children) != 0)
2111                 return (NULL);
2112
2113         for (c = 0; c < children; c++) {
2114                 if ((ret = vdev_to_nvlist_iter(child[c], search,
2115                     avail_spare, l2cache, NULL)) != NULL) {
2116                         /*
2117                          * The 'is_log' value is only set for the toplevel
2118                          * vdev, not the leaf vdevs.  So we always lookup the
2119                          * log device from the root of the vdev tree (where
2120                          * 'log' is non-NULL).
2121                          */
2122                         if (log != NULL &&
2123                             nvlist_lookup_uint64(child[c],
2124                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2125                             is_log) {
2126                                 *log = B_TRUE;
2127                         }
2128                         return (ret);
2129                 }
2130         }
2131
2132         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2133             &child, &children) == 0) {
2134                 for (c = 0; c < children; c++) {
2135                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2136                             avail_spare, l2cache, NULL)) != NULL) {
2137                                 *avail_spare = B_TRUE;
2138                                 return (ret);
2139                         }
2140                 }
2141         }
2142
2143         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2144             &child, &children) == 0) {
2145                 for (c = 0; c < children; c++) {
2146                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2147                             avail_spare, l2cache, NULL)) != NULL) {
2148                                 *l2cache = B_TRUE;
2149                                 return (ret);
2150                         }
2151                 }
2152         }
2153
2154         return (NULL);
2155 }
2156
2157 /*
2158  * Given a physical path (minus the "/devices" prefix), find the
2159  * associated vdev.
2160  */
2161 nvlist_t *
2162 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2163     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2164 {
2165         nvlist_t *search, *nvroot, *ret;
2166
2167         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2168         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
2169
2170         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2171             &nvroot) == 0);
2172
2173         *avail_spare = B_FALSE;
2174         *l2cache = B_FALSE;
2175         if (log != NULL)
2176                 *log = B_FALSE;
2177         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2178         nvlist_free(search);
2179
2180         return (ret);
2181 }
2182
2183 /*
2184  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2185  */
2186 boolean_t
2187 zpool_vdev_is_interior(const char *name)
2188 {
2189         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2190             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2191                 return (B_TRUE);
2192         return (B_FALSE);
2193 }
2194
2195 nvlist_t *
2196 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2197     boolean_t *l2cache, boolean_t *log)
2198 {
2199         char buf[MAXPATHLEN];
2200         char *end;
2201         nvlist_t *nvroot, *search, *ret;
2202         uint64_t guid;
2203
2204         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2205
2206         guid = strtoull(path, &end, 10);
2207         if (guid != 0 && *end == '\0') {
2208                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
2209         } else if (zpool_vdev_is_interior(path)) {
2210                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
2211         } else if (path[0] != '/') {
2212                 (void) snprintf(buf, sizeof (buf), "%s%s", _PATH_DEV, path);
2213                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, buf) == 0);
2214         } else {
2215                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
2216         }
2217
2218         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2219             &nvroot) == 0);
2220
2221         *avail_spare = B_FALSE;
2222         *l2cache = B_FALSE;
2223         if (log != NULL)
2224                 *log = B_FALSE;
2225         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2226         nvlist_free(search);
2227
2228         return (ret);
2229 }
2230
2231 static int
2232 vdev_online(nvlist_t *nv)
2233 {
2234         uint64_t ival;
2235
2236         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2237             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2238             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2239                 return (0);
2240
2241         return (1);
2242 }
2243
2244 /*
2245  * Helper function for zpool_get_physpaths().
2246  */
2247 static int
2248 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2249     size_t *bytes_written)
2250 {
2251         size_t bytes_left, pos, rsz;
2252         char *tmppath;
2253         const char *format;
2254
2255         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2256             &tmppath) != 0)
2257                 return (EZFS_NODEVICE);
2258
2259         pos = *bytes_written;
2260         bytes_left = physpath_size - pos;
2261         format = (pos == 0) ? "%s" : " %s";
2262
2263         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2264         *bytes_written += rsz;
2265
2266         if (rsz >= bytes_left) {
2267                 /* if physpath was not copied properly, clear it */
2268                 if (bytes_left != 0) {
2269                         physpath[pos] = 0;
2270                 }
2271                 return (EZFS_NOSPC);
2272         }
2273         return (0);
2274 }
2275
2276 static int
2277 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2278     size_t *rsz, boolean_t is_spare)
2279 {
2280         char *type;
2281         int ret;
2282
2283         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2284                 return (EZFS_INVALCONFIG);
2285
2286         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2287                 /*
2288                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2289                  * For a spare vdev, we only want to boot from the active
2290                  * spare device.
2291                  */
2292                 if (is_spare) {
2293                         uint64_t spare = 0;
2294                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2295                             &spare);
2296                         if (!spare)
2297                                 return (EZFS_INVALCONFIG);
2298                 }
2299
2300                 if (vdev_online(nv)) {
2301                         if ((ret = vdev_get_one_physpath(nv, physpath,
2302                             phypath_size, rsz)) != 0)
2303                                 return (ret);
2304                 }
2305         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2306             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2307             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2308                 nvlist_t **child;
2309                 uint_t count;
2310                 int i, ret;
2311
2312                 if (nvlist_lookup_nvlist_array(nv,
2313                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2314                         return (EZFS_INVALCONFIG);
2315
2316                 for (i = 0; i < count; i++) {
2317                         ret = vdev_get_physpaths(child[i], physpath,
2318                             phypath_size, rsz, is_spare);
2319                         if (ret == EZFS_NOSPC)
2320                                 return (ret);
2321                 }
2322         }
2323
2324         return (EZFS_POOL_INVALARG);
2325 }
2326
2327 /*
2328  * Get phys_path for a root pool config.
2329  * Return 0 on success; non-zero on failure.
2330  */
2331 static int
2332 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2333 {
2334         size_t rsz;
2335         nvlist_t *vdev_root;
2336         nvlist_t **child;
2337         uint_t count;
2338         char *type;
2339
2340         rsz = 0;
2341
2342         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2343             &vdev_root) != 0)
2344                 return (EZFS_INVALCONFIG);
2345
2346         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2347             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2348             &child, &count) != 0)
2349                 return (EZFS_INVALCONFIG);
2350
2351         /*
2352          * root pool can not have EFI labeled disks and can only have
2353          * a single top-level vdev.
2354          */
2355         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1 ||
2356             pool_uses_efi(vdev_root))
2357                 return (EZFS_POOL_INVALARG);
2358
2359         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2360             B_FALSE);
2361
2362         /* No online devices */
2363         if (rsz == 0)
2364                 return (EZFS_NODEVICE);
2365
2366         return (0);
2367 }
2368
2369 /*
2370  * Get phys_path for a root pool
2371  * Return 0 on success; non-zero on failure.
2372  */
2373 int
2374 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2375 {
2376         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2377             phypath_size));
2378 }
2379
2380 /*
2381  * If the device has being dynamically expanded then we need to relabel
2382  * the disk to use the new unallocated space.
2383  */
2384 static int
2385 zpool_relabel_disk(libzfs_handle_t *hdl, const char *name)
2386 {
2387 #ifdef sun
2388         char path[MAXPATHLEN];
2389         char errbuf[1024];
2390         int fd, error;
2391         int (*_efi_use_whole_disk)(int);
2392
2393         if ((_efi_use_whole_disk = (int (*)(int))dlsym(RTLD_DEFAULT,
2394             "efi_use_whole_disk")) == NULL)
2395                 return (-1);
2396
2397         (void) snprintf(path, sizeof (path), "%s/%s", RDISK_ROOT, name);
2398
2399         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
2400                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2401                     "relabel '%s': unable to open device"), name);
2402                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
2403         }
2404
2405         /*
2406          * It's possible that we might encounter an error if the device
2407          * does not have any unallocated space left. If so, we simply
2408          * ignore that error and continue on.
2409          */
2410         error = _efi_use_whole_disk(fd);
2411         (void) close(fd);
2412         if (error && error != VT_ENOSPC) {
2413                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2414                     "relabel '%s': unable to read disk capacity"), name);
2415                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
2416         }
2417 #endif  /* sun */
2418         return (0);
2419 }
2420
2421 /*
2422  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2423  * ZFS_ONLINE_* flags.
2424  */
2425 int
2426 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2427     vdev_state_t *newstate)
2428 {
2429         zfs_cmd_t zc = { 0 };
2430         char msg[1024];
2431         nvlist_t *tgt;
2432         boolean_t avail_spare, l2cache, islog;
2433         libzfs_handle_t *hdl = zhp->zpool_hdl;
2434
2435         if (flags & ZFS_ONLINE_EXPAND) {
2436                 (void) snprintf(msg, sizeof (msg),
2437                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2438         } else {
2439                 (void) snprintf(msg, sizeof (msg),
2440                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2441         }
2442
2443         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2444         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2445             &islog)) == NULL)
2446                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2447
2448         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2449
2450         if (avail_spare)
2451                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2452
2453         if (flags & ZFS_ONLINE_EXPAND ||
2454             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2455                 char *pathname = NULL;
2456                 uint64_t wholedisk = 0;
2457
2458                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2459                     &wholedisk);
2460                 verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
2461                     &pathname) == 0);
2462
2463                 /*
2464                  * XXX - L2ARC 1.0 devices can't support expansion.
2465                  */
2466                 if (l2cache) {
2467                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2468                             "cannot expand cache devices"));
2469                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2470                 }
2471
2472                 if (wholedisk) {
2473                         pathname += strlen(DISK_ROOT) + 1;
2474                         (void) zpool_relabel_disk(hdl, pathname);
2475                 }
2476         }
2477
2478         zc.zc_cookie = VDEV_STATE_ONLINE;
2479         zc.zc_obj = flags;
2480
2481         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2482                 if (errno == EINVAL) {
2483                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2484                             "from this pool into a new one.  Use '%s' "
2485                             "instead"), "zpool detach");
2486                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2487                 }
2488                 return (zpool_standard_error(hdl, errno, msg));
2489         }
2490
2491         *newstate = zc.zc_cookie;
2492         return (0);
2493 }
2494
2495 /*
2496  * Take the specified vdev offline
2497  */
2498 int
2499 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2500 {
2501         zfs_cmd_t zc = { 0 };
2502         char msg[1024];
2503         nvlist_t *tgt;
2504         boolean_t avail_spare, l2cache;
2505         libzfs_handle_t *hdl = zhp->zpool_hdl;
2506
2507         (void) snprintf(msg, sizeof (msg),
2508             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2509
2510         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2511         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2512             NULL)) == NULL)
2513                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2514
2515         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2516
2517         if (avail_spare)
2518                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2519
2520         zc.zc_cookie = VDEV_STATE_OFFLINE;
2521         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2522
2523         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2524                 return (0);
2525
2526         switch (errno) {
2527         case EBUSY:
2528
2529                 /*
2530                  * There are no other replicas of this device.
2531                  */
2532                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2533
2534         case EEXIST:
2535                 /*
2536                  * The log device has unplayed logs
2537                  */
2538                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2539
2540         default:
2541                 return (zpool_standard_error(hdl, errno, msg));
2542         }
2543 }
2544
2545 /*
2546  * Mark the given vdev faulted.
2547  */
2548 int
2549 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2550 {
2551         zfs_cmd_t zc = { 0 };
2552         char msg[1024];
2553         libzfs_handle_t *hdl = zhp->zpool_hdl;
2554
2555         (void) snprintf(msg, sizeof (msg),
2556             dgettext(TEXT_DOMAIN, "cannot fault %llu"), guid);
2557
2558         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2559         zc.zc_guid = guid;
2560         zc.zc_cookie = VDEV_STATE_FAULTED;
2561         zc.zc_obj = aux;
2562
2563         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2564                 return (0);
2565
2566         switch (errno) {
2567         case EBUSY:
2568
2569                 /*
2570                  * There are no other replicas of this device.
2571                  */
2572                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2573
2574         default:
2575                 return (zpool_standard_error(hdl, errno, msg));
2576         }
2577
2578 }
2579
2580 /*
2581  * Mark the given vdev degraded.
2582  */
2583 int
2584 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2585 {
2586         zfs_cmd_t zc = { 0 };
2587         char msg[1024];
2588         libzfs_handle_t *hdl = zhp->zpool_hdl;
2589
2590         (void) snprintf(msg, sizeof (msg),
2591             dgettext(TEXT_DOMAIN, "cannot degrade %llu"), guid);
2592
2593         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2594         zc.zc_guid = guid;
2595         zc.zc_cookie = VDEV_STATE_DEGRADED;
2596         zc.zc_obj = aux;
2597
2598         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2599                 return (0);
2600
2601         return (zpool_standard_error(hdl, errno, msg));
2602 }
2603
2604 /*
2605  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2606  * a hot spare.
2607  */
2608 static boolean_t
2609 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2610 {
2611         nvlist_t **child;
2612         uint_t c, children;
2613         char *type;
2614
2615         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2616             &children) == 0) {
2617                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2618                     &type) == 0);
2619
2620                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2621                     children == 2 && child[which] == tgt)
2622                         return (B_TRUE);
2623
2624                 for (c = 0; c < children; c++)
2625                         if (is_replacing_spare(child[c], tgt, which))
2626                                 return (B_TRUE);
2627         }
2628
2629         return (B_FALSE);
2630 }
2631
2632 /*
2633  * Attach new_disk (fully described by nvroot) to old_disk.
2634  * If 'replacing' is specified, the new disk will replace the old one.
2635  */
2636 int
2637 zpool_vdev_attach(zpool_handle_t *zhp,
2638     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2639 {
2640         zfs_cmd_t zc = { 0 };
2641         char msg[1024];
2642         int ret;
2643         nvlist_t *tgt;
2644         boolean_t avail_spare, l2cache, islog;
2645         uint64_t val;
2646         char *newname;
2647         nvlist_t **child;
2648         uint_t children;
2649         nvlist_t *config_root;
2650         libzfs_handle_t *hdl = zhp->zpool_hdl;
2651         boolean_t rootpool = zpool_is_bootable(zhp);
2652
2653         if (replacing)
2654                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2655                     "cannot replace %s with %s"), old_disk, new_disk);
2656         else
2657                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2658                     "cannot attach %s to %s"), new_disk, old_disk);
2659
2660         /*
2661          * If this is a root pool, make sure that we're not attaching an
2662          * EFI labeled device.
2663          */
2664         if (rootpool && pool_uses_efi(nvroot)) {
2665                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2666                     "EFI labeled devices are not supported on root pools."));
2667                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
2668         }
2669
2670         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2671         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2672             &islog)) == 0)
2673                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2674
2675         if (avail_spare)
2676                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2677
2678         if (l2cache)
2679                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2680
2681         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2682         zc.zc_cookie = replacing;
2683
2684         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2685             &child, &children) != 0 || children != 1) {
2686                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2687                     "new device must be a single disk"));
2688                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2689         }
2690
2691         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2692             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2693
2694         if ((newname = zpool_vdev_name(NULL, NULL, child[0], B_FALSE)) == NULL)
2695                 return (-1);
2696
2697         /*
2698          * If the target is a hot spare that has been swapped in, we can only
2699          * replace it with another hot spare.
2700          */
2701         if (replacing &&
2702             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2703             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2704             NULL) == NULL || !avail_spare) &&
2705             is_replacing_spare(config_root, tgt, 1)) {
2706                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2707                     "can only be replaced by another hot spare"));
2708                 free(newname);
2709                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2710         }
2711
2712         free(newname);
2713
2714         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2715                 return (-1);
2716
2717         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2718
2719         zcmd_free_nvlists(&zc);
2720
2721         if (ret == 0) {
2722                 if (rootpool) {
2723                         /*
2724                          * XXX need a better way to prevent user from
2725                          * booting up a half-baked vdev.
2726                          */
2727                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2728                             "sure to wait until resilver is done "
2729                             "before rebooting.\n"));
2730                         (void) fprintf(stderr, "\n");
2731                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "If "
2732                             "you boot from pool '%s', you may need to update\n"
2733                             "boot code on newly attached disk '%s'.\n\n"
2734                             "Assuming you use GPT partitioning and 'da0' is "
2735                             "your new boot disk\n"
2736                             "you may use the following command:\n\n"
2737                             "\tgpart bootcode -b /boot/pmbr -p "
2738                             "/boot/gptzfsboot -i 1 da0\n\n"),
2739                             zhp->zpool_name, new_disk);
2740                 }
2741                 return (0);
2742         }
2743
2744         switch (errno) {
2745         case ENOTSUP:
2746                 /*
2747                  * Can't attach to or replace this type of vdev.
2748                  */
2749                 if (replacing) {
2750                         uint64_t version = zpool_get_prop_int(zhp,
2751                             ZPOOL_PROP_VERSION, NULL);
2752
2753                         if (islog)
2754                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2755                                     "cannot replace a log with a spare"));
2756                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2757                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2758                                     "already in replacing/spare config; wait "
2759                                     "for completion or use 'zpool detach'"));
2760                         else
2761                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2762                                     "cannot replace a replacing device"));
2763                 } else {
2764                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2765                             "can only attach to mirrors and top-level "
2766                             "disks"));
2767                 }
2768                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2769                 break;
2770
2771         case EINVAL:
2772                 /*
2773                  * The new device must be a single disk.
2774                  */
2775                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2776                     "new device must be a single disk"));
2777                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2778                 break;
2779
2780         case EBUSY:
2781                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2782                     new_disk);
2783                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2784                 break;
2785
2786         case EOVERFLOW:
2787                 /*
2788                  * The new device is too small.
2789                  */
2790                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2791                     "device is too small"));
2792                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2793                 break;
2794
2795         case EDOM:
2796                 /*
2797                  * The new device has a different alignment requirement.
2798                  */
2799                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2800                     "devices have different sector alignment"));
2801                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2802                 break;
2803
2804         case ENAMETOOLONG:
2805                 /*
2806                  * The resulting top-level vdev spec won't fit in the label.
2807                  */
2808                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2809                 break;
2810
2811         default:
2812                 (void) zpool_standard_error(hdl, errno, msg);
2813         }
2814
2815         return (-1);
2816 }
2817
2818 /*
2819  * Detach the specified device.
2820  */
2821 int
2822 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2823 {
2824         zfs_cmd_t zc = { 0 };
2825         char msg[1024];
2826         nvlist_t *tgt;
2827         boolean_t avail_spare, l2cache;
2828         libzfs_handle_t *hdl = zhp->zpool_hdl;
2829
2830         (void) snprintf(msg, sizeof (msg),
2831             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2832
2833         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2834         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2835             NULL)) == 0)
2836                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2837
2838         if (avail_spare)
2839                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2840
2841         if (l2cache)
2842                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2843
2844         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2845
2846         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2847                 return (0);
2848
2849         switch (errno) {
2850
2851         case ENOTSUP:
2852                 /*
2853                  * Can't detach from this type of vdev.
2854                  */
2855                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2856                     "applicable to mirror and replacing vdevs"));
2857                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2858                 break;
2859
2860         case EBUSY:
2861                 /*
2862                  * There are no other replicas of this device.
2863                  */
2864                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2865                 break;
2866
2867         default:
2868                 (void) zpool_standard_error(hdl, errno, msg);
2869         }
2870
2871         return (-1);
2872 }
2873
2874 /*
2875  * Find a mirror vdev in the source nvlist.
2876  *
2877  * The mchild array contains a list of disks in one of the top-level mirrors
2878  * of the source pool.  The schild array contains a list of disks that the
2879  * user specified on the command line.  We loop over the mchild array to
2880  * see if any entry in the schild array matches.
2881  *
2882  * If a disk in the mchild array is found in the schild array, we return
2883  * the index of that entry.  Otherwise we return -1.
2884  */
2885 static int
2886 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2887     nvlist_t **schild, uint_t schildren)
2888 {
2889         uint_t mc;
2890
2891         for (mc = 0; mc < mchildren; mc++) {
2892                 uint_t sc;
2893                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2894                     mchild[mc], B_FALSE);
2895
2896                 for (sc = 0; sc < schildren; sc++) {
2897                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2898                             schild[sc], B_FALSE);
2899                         boolean_t result = (strcmp(mpath, spath) == 0);
2900
2901                         free(spath);
2902                         if (result) {
2903                                 free(mpath);
2904                                 return (mc);
2905                         }
2906                 }
2907
2908                 free(mpath);
2909         }
2910
2911         return (-1);
2912 }
2913
2914 /*
2915  * Split a mirror pool.  If newroot points to null, then a new nvlist
2916  * is generated and it is the responsibility of the caller to free it.
2917  */
2918 int
2919 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2920     nvlist_t *props, splitflags_t flags)
2921 {
2922         zfs_cmd_t zc = { 0 };
2923         char msg[1024];
2924         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2925         nvlist_t **varray = NULL, *zc_props = NULL;
2926         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2927         libzfs_handle_t *hdl = zhp->zpool_hdl;
2928         uint64_t vers;
2929         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2930         int retval = 0;
2931
2932         (void) snprintf(msg, sizeof (msg),
2933             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2934
2935         if (!zpool_name_valid(hdl, B_FALSE, newname))
2936                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2937
2938         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2939                 (void) fprintf(stderr, gettext("Internal error: unable to "
2940                     "retrieve pool configuration\n"));
2941                 return (-1);
2942         }
2943
2944         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2945             == 0);
2946         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2947
2948         if (props) {
2949                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2950                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2951                     props, vers, flags, msg)) == NULL)
2952                         return (-1);
2953         }
2954
2955         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2956             &children) != 0) {
2957                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2958                     "Source pool is missing vdev tree"));
2959                 if (zc_props)
2960                         nvlist_free(zc_props);
2961                 return (-1);
2962         }
2963
2964         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2965         vcount = 0;
2966
2967         if (*newroot == NULL ||
2968             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2969             &newchild, &newchildren) != 0)
2970                 newchildren = 0;
2971
2972         for (c = 0; c < children; c++) {
2973                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2974                 char *type;
2975                 nvlist_t **mchild, *vdev;
2976                 uint_t mchildren;
2977                 int entry;
2978
2979                 /*
2980                  * Unlike cache & spares, slogs are stored in the
2981                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2982                  */
2983                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2984                     &is_log);
2985                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2986                     &is_hole);
2987                 if (is_log || is_hole) {
2988                         /*
2989                          * Create a hole vdev and put it in the config.
2990                          */
2991                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2992                                 goto out;
2993                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2994                             VDEV_TYPE_HOLE) != 0)
2995                                 goto out;
2996                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2997                             1) != 0)
2998                                 goto out;
2999                         if (lastlog == 0)
3000                                 lastlog = vcount;
3001                         varray[vcount++] = vdev;
3002                         continue;
3003                 }
3004                 lastlog = 0;
3005                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
3006                     == 0);
3007                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
3008                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3009                             "Source pool must be composed only of mirrors\n"));
3010                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3011                         goto out;
3012                 }
3013
3014                 verify(nvlist_lookup_nvlist_array(child[c],
3015                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
3016
3017                 /* find or add an entry for this top-level vdev */
3018                 if (newchildren > 0 &&
3019                     (entry = find_vdev_entry(zhp, mchild, mchildren,
3020                     newchild, newchildren)) >= 0) {
3021                         /* We found a disk that the user specified. */
3022                         vdev = mchild[entry];
3023                         ++found;
3024                 } else {
3025                         /* User didn't specify a disk for this vdev. */
3026                         vdev = mchild[mchildren - 1];
3027                 }
3028
3029                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
3030                         goto out;
3031         }
3032
3033         /* did we find every disk the user specified? */
3034         if (found != newchildren) {
3035                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
3036                     "include at most one disk from each mirror"));
3037                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3038                 goto out;
3039         }
3040
3041         /* Prepare the nvlist for populating. */
3042         if (*newroot == NULL) {
3043                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
3044                         goto out;
3045                 freelist = B_TRUE;
3046                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
3047                     VDEV_TYPE_ROOT) != 0)
3048                         goto out;
3049         } else {
3050                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
3051         }
3052
3053         /* Add all the children we found */
3054         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
3055             lastlog == 0 ? vcount : lastlog) != 0)
3056                 goto out;
3057
3058         /*
3059          * If we're just doing a dry run, exit now with success.
3060          */
3061         if (flags.dryrun) {
3062                 memory_err = B_FALSE;
3063                 freelist = B_FALSE;
3064                 goto out;
3065         }
3066
3067         /* now build up the config list & call the ioctl */
3068         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3069                 goto out;
3070
3071         if (nvlist_add_nvlist(newconfig,
3072             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3073             nvlist_add_string(newconfig,
3074             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3075             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3076                 goto out;
3077
3078         /*
3079          * The new pool is automatically part of the namespace unless we
3080          * explicitly export it.
3081          */
3082         if (!flags.import)
3083                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3084         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3085         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3086         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
3087                 goto out;
3088         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
3089                 goto out;
3090
3091         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3092                 retval = zpool_standard_error(hdl, errno, msg);
3093                 goto out;
3094         }
3095
3096         freelist = B_FALSE;
3097         memory_err = B_FALSE;
3098
3099 out:
3100         if (varray != NULL) {
3101                 int v;
3102
3103                 for (v = 0; v < vcount; v++)
3104                         nvlist_free(varray[v]);
3105                 free(varray);
3106         }
3107         zcmd_free_nvlists(&zc);
3108         if (zc_props)
3109                 nvlist_free(zc_props);
3110         if (newconfig)
3111                 nvlist_free(newconfig);
3112         if (freelist) {
3113                 nvlist_free(*newroot);
3114                 *newroot = NULL;
3115         }
3116
3117         if (retval != 0)
3118                 return (retval);
3119
3120         if (memory_err)
3121                 return (no_memory(hdl));
3122
3123         return (0);
3124 }
3125
3126 /*
3127  * Remove the given device.  Currently, this is supported only for hot spares
3128  * and level 2 cache devices.
3129  */
3130 int
3131 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3132 {
3133         zfs_cmd_t zc = { 0 };
3134         char msg[1024];
3135         nvlist_t *tgt;
3136         boolean_t avail_spare, l2cache, islog;
3137         libzfs_handle_t *hdl = zhp->zpool_hdl;
3138         uint64_t version;
3139
3140         (void) snprintf(msg, sizeof (msg),
3141             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3142
3143         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3144         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3145             &islog)) == 0)
3146                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3147         /*
3148          * XXX - this should just go away.
3149          */
3150         if (!avail_spare && !l2cache && !islog) {
3151                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3152                     "only inactive hot spares, cache, top-level, "
3153                     "or log devices can be removed"));
3154                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3155         }
3156
3157         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3158         if (islog && version < SPA_VERSION_HOLES) {
3159                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3160                     "pool must be upgrade to support log removal"));
3161                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
3162         }
3163
3164         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
3165
3166         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3167                 return (0);
3168
3169         return (zpool_standard_error(hdl, errno, msg));
3170 }
3171
3172 /*
3173  * Clear the errors for the pool, or the particular device if specified.
3174  */
3175 int
3176 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3177 {
3178         zfs_cmd_t zc = { 0 };
3179         char msg[1024];
3180         nvlist_t *tgt;
3181         zpool_rewind_policy_t policy;
3182         boolean_t avail_spare, l2cache;
3183         libzfs_handle_t *hdl = zhp->zpool_hdl;
3184         nvlist_t *nvi = NULL;
3185         int error;
3186
3187         if (path)
3188                 (void) snprintf(msg, sizeof (msg),
3189                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3190                     path);
3191         else
3192                 (void) snprintf(msg, sizeof (msg),
3193                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3194                     zhp->zpool_name);
3195
3196         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3197         if (path) {
3198                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3199                     &l2cache, NULL)) == 0)
3200                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
3201
3202                 /*
3203                  * Don't allow error clearing for hot spares.  Do allow
3204                  * error clearing for l2cache devices.
3205                  */
3206                 if (avail_spare)
3207                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
3208
3209                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
3210                     &zc.zc_guid) == 0);
3211         }
3212
3213         zpool_get_rewind_policy(rewindnvl, &policy);
3214         zc.zc_cookie = policy.zrp_request;
3215
3216         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
3217                 return (-1);
3218
3219         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
3220                 return (-1);
3221
3222         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3223             errno == ENOMEM) {
3224                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3225                         zcmd_free_nvlists(&zc);
3226                         return (-1);
3227                 }
3228         }
3229
3230         if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
3231             errno != EPERM && errno != EACCES)) {
3232                 if (policy.zrp_request &
3233                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
3234                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3235                         zpool_rewind_exclaim(hdl, zc.zc_name,
3236                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
3237                             nvi);
3238                         nvlist_free(nvi);
3239                 }
3240                 zcmd_free_nvlists(&zc);
3241                 return (0);
3242         }
3243
3244         zcmd_free_nvlists(&zc);
3245         return (zpool_standard_error(hdl, errno, msg));
3246 }
3247
3248 /*
3249  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3250  */
3251 int
3252 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3253 {
3254         zfs_cmd_t zc = { 0 };
3255         char msg[1024];
3256         libzfs_handle_t *hdl = zhp->zpool_hdl;
3257
3258         (void) snprintf(msg, sizeof (msg),
3259             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3260             guid);
3261
3262         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3263         zc.zc_guid = guid;
3264         zc.zc_cookie = ZPOOL_NO_REWIND;
3265
3266         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
3267                 return (0);
3268
3269         return (zpool_standard_error(hdl, errno, msg));
3270 }
3271
3272 /*
3273  * Change the GUID for a pool.
3274  */
3275 int
3276 zpool_reguid(zpool_handle_t *zhp)
3277 {
3278         char msg[1024];
3279         libzfs_handle_t *hdl = zhp->zpool_hdl;
3280         zfs_cmd_t zc = { 0 };
3281
3282         (void) snprintf(msg, sizeof (msg),
3283             dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3284
3285         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3286         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3287                 return (0);
3288
3289         return (zpool_standard_error(hdl, errno, msg));
3290 }
3291
3292 /*
3293  * Reopen the pool.
3294  */
3295 int
3296 zpool_reopen(zpool_handle_t *zhp)
3297 {
3298         zfs_cmd_t zc = { 0 };
3299         char msg[1024];
3300         libzfs_handle_t *hdl = zhp->zpool_hdl;
3301
3302         (void) snprintf(msg, sizeof (msg),
3303             dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3304             zhp->zpool_name);
3305
3306         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3307         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3308                 return (0);
3309         return (zpool_standard_error(hdl, errno, msg));
3310 }
3311
3312 /*
3313  * Convert from a devid string to a path.
3314  */
3315 static char *
3316 devid_to_path(char *devid_str)
3317 {
3318         ddi_devid_t devid;
3319         char *minor;
3320         char *path;
3321         devid_nmlist_t *list = NULL;
3322         int ret;
3323
3324         if (devid_str_decode(devid_str, &devid, &minor) != 0)
3325                 return (NULL);
3326
3327         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3328
3329         devid_str_free(minor);
3330         devid_free(devid);
3331
3332         if (ret != 0)
3333                 return (NULL);
3334
3335         if ((path = strdup(list[0].devname)) == NULL)
3336                 return (NULL);
3337
3338         devid_free_nmlist(list);
3339
3340         return (path);
3341 }
3342
3343 /*
3344  * Convert from a path to a devid string.
3345  */
3346 static char *
3347 path_to_devid(const char *path)
3348 {
3349 #ifdef have_devid
3350         int fd;
3351         ddi_devid_t devid;
3352         char *minor, *ret;
3353
3354         if ((fd = open(path, O_RDONLY)) < 0)
3355                 return (NULL);
3356
3357         minor = NULL;
3358         ret = NULL;
3359         if (devid_get(fd, &devid) == 0) {
3360                 if (devid_get_minor_name(fd, &minor) == 0)
3361                         ret = devid_str_encode(devid, minor);
3362                 if (minor != NULL)
3363                         devid_str_free(minor);
3364                 devid_free(devid);
3365         }
3366         (void) close(fd);
3367
3368         return (ret);
3369 #else
3370         return (NULL);
3371 #endif
3372 }
3373
3374 /*
3375  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3376  * ignore any failure here, since a common case is for an unprivileged user to
3377  * type 'zpool status', and we'll display the correct information anyway.
3378  */
3379 static void
3380 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3381 {
3382         zfs_cmd_t zc = { 0 };
3383
3384         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3385         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3386         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3387             &zc.zc_guid) == 0);
3388
3389         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3390 }
3391
3392 /*
3393  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3394  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3395  * We also check if this is a whole disk, in which case we strip off the
3396  * trailing 's0' slice name.
3397  *
3398  * This routine is also responsible for identifying when disks have been
3399  * reconfigured in a new location.  The kernel will have opened the device by
3400  * devid, but the path will still refer to the old location.  To catch this, we
3401  * first do a path -> devid translation (which is fast for the common case).  If
3402  * the devid matches, we're done.  If not, we do a reverse devid -> path
3403  * translation and issue the appropriate ioctl() to update the path of the vdev.
3404  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3405  * of these checks.
3406  */
3407 char *
3408 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3409     boolean_t verbose)
3410 {
3411         char *path, *devid;
3412         uint64_t value;
3413         char buf[64];
3414         vdev_stat_t *vs;
3415         uint_t vsc;
3416         int have_stats;
3417         int have_path;
3418
3419         have_stats = nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3420             (uint64_t **)&vs, &vsc) == 0;
3421         have_path = nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0;
3422
3423         /*
3424          * If the device is not currently present, assume it will not
3425          * come back at the same device path.  Display the device by GUID.
3426          */
3427         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &value) == 0 ||
3428             have_path && have_stats && vs->vs_state <= VDEV_STATE_CANT_OPEN) {
3429                 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3430                     &value) == 0);
3431                 (void) snprintf(buf, sizeof (buf), "%llu",
3432                     (u_longlong_t)value);
3433                 path = buf;
3434         } else if (have_path) {
3435
3436                 /*
3437                  * If the device is dead (faulted, offline, etc) then don't
3438                  * bother opening it.  Otherwise we may be forcing the user to
3439                  * open a misbehaving device, which can have undesirable
3440                  * effects.
3441                  */
3442                 if ((have_stats == 0 ||
3443                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3444                     zhp != NULL &&
3445                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3446                         /*
3447                          * Determine if the current path is correct.
3448                          */
3449                         char *newdevid = path_to_devid(path);
3450
3451                         if (newdevid == NULL ||
3452                             strcmp(devid, newdevid) != 0) {
3453                                 char *newpath;
3454
3455                                 if ((newpath = devid_to_path(devid)) != NULL) {
3456                                         /*
3457                                          * Update the path appropriately.
3458                                          */
3459                                         set_path(zhp, nv, newpath);
3460                                         if (nvlist_add_string(nv,
3461                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3462                                                 verify(nvlist_lookup_string(nv,
3463                                                     ZPOOL_CONFIG_PATH,
3464                                                     &path) == 0);
3465                                         free(newpath);
3466                                 }
3467                         }
3468
3469                         if (newdevid)
3470                                 devid_str_free(newdevid);
3471                 }
3472
3473 #ifdef sun
3474                 if (strncmp(path, "/dev/dsk/", 9) == 0)
3475                         path += 9;
3476
3477                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
3478                     &value) == 0 && value) {
3479                         int pathlen = strlen(path);
3480                         char *tmp = zfs_strdup(hdl, path);
3481
3482                         /*
3483                          * If it starts with c#, and ends with "s0", chop
3484                          * the "s0" off, or if it ends with "s0/old", remove
3485                          * the "s0" from the middle.
3486                          */
3487                         if (CTD_CHECK(tmp)) {
3488                                 if (strcmp(&tmp[pathlen - 2], "s0") == 0) {
3489                                         tmp[pathlen - 2] = '\0';
3490                                 } else if (pathlen > 6 &&
3491                                     strcmp(&tmp[pathlen - 6], "s0/old") == 0) {
3492                                         (void) strcpy(&tmp[pathlen - 6],
3493                                             "/old");
3494                                 }
3495                         }
3496                         return (tmp);
3497                 }
3498 #else   /* !sun */
3499                 if (strncmp(path, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
3500                         path += sizeof(_PATH_DEV) - 1;
3501 #endif  /* !sun */
3502         } else {
3503                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3504
3505                 /*
3506                  * If it's a raidz device, we need to stick in the parity level.
3507                  */
3508                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3509                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3510                             &value) == 0);
3511                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3512                             (u_longlong_t)value);
3513                         path = buf;
3514                 }
3515
3516                 /*
3517                  * We identify each top-level vdev by using a <type-id>
3518                  * naming convention.
3519                  */
3520                 if (verbose) {
3521                         uint64_t id;
3522
3523                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3524                             &id) == 0);
3525                         (void) snprintf(buf, sizeof (buf), "%s-%llu", path,
3526                             (u_longlong_t)id);
3527                         path = buf;
3528                 }
3529         }
3530
3531         return (zfs_strdup(hdl, path));
3532 }
3533
3534 static int
3535 zbookmark_compare(const void *a, const void *b)
3536 {
3537         return (memcmp(a, b, sizeof (zbookmark_phys_t)));
3538 }
3539
3540 /*
3541  * Retrieve the persistent error log, uniquify the members, and return to the
3542  * caller.
3543  */
3544 int
3545 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3546 {
3547         zfs_cmd_t zc = { 0 };
3548         uint64_t count;
3549         zbookmark_phys_t *zb = NULL;
3550         int i;
3551
3552         /*
3553          * Retrieve the raw error list from the kernel.  If the number of errors
3554          * has increased, allocate more space and continue until we get the
3555          * entire list.
3556          */
3557         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3558             &count) == 0);
3559         if (count == 0)
3560                 return (0);
3561         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3562             count * sizeof (zbookmark_phys_t))) == (uintptr_t)NULL)
3563                 return (-1);
3564         zc.zc_nvlist_dst_size = count;
3565         (void) strcpy(zc.zc_name, zhp->zpool_name);
3566         for (;;) {
3567                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3568                     &zc) != 0) {
3569                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3570                         if (errno == ENOMEM) {
3571                                 void *dst;
3572
3573                                 count = zc.zc_nvlist_dst_size;
3574                                 dst = zfs_alloc(zhp->zpool_hdl, count *
3575                                     sizeof (zbookmark_phys_t));
3576                                 if (dst == NULL)
3577                                         return (-1);
3578                                 zc.zc_nvlist_dst = (uintptr_t)dst;
3579                         } else {
3580                                 return (-1);
3581                         }
3582                 } else {
3583                         break;
3584                 }
3585         }
3586
3587         /*
3588          * Sort the resulting bookmarks.  This is a little confusing due to the
3589          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3590          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3591          * _not_ copied as part of the process.  So we point the start of our
3592          * array appropriate and decrement the total number of elements.
3593          */
3594         zb = ((zbookmark_phys_t *)(uintptr_t)zc.zc_nvlist_dst) +
3595             zc.zc_nvlist_dst_size;
3596         count -= zc.zc_nvlist_dst_size;
3597
3598         qsort(zb, count, sizeof (zbookmark_phys_t), zbookmark_compare);
3599
3600         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3601
3602         /*
3603          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3604          */
3605         for (i = 0; i < count; i++) {
3606                 nvlist_t *nv;
3607
3608                 /* ignoring zb_blkid and zb_level for now */
3609                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3610                     zb[i-1].zb_object == zb[i].zb_object)
3611                         continue;
3612
3613                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3614                         goto nomem;
3615                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3616                     zb[i].zb_objset) != 0) {
3617                         nvlist_free(nv);
3618                         goto nomem;
3619                 }
3620                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3621                     zb[i].zb_object) != 0) {
3622                         nvlist_free(nv);
3623                         goto nomem;
3624                 }
3625                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3626                         nvlist_free(nv);
3627                         goto nomem;
3628                 }
3629                 nvlist_free(nv);
3630         }
3631
3632         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3633         return (0);
3634
3635 nomem:
3636         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3637         return (no_memory(zhp->zpool_hdl));
3638 }
3639
3640 /*
3641  * Upgrade a ZFS pool to the latest on-disk version.
3642  */
3643 int
3644 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3645 {
3646         zfs_cmd_t zc = { 0 };
3647         libzfs_handle_t *hdl = zhp->zpool_hdl;
3648
3649         (void) strcpy(zc.zc_name, zhp->zpool_name);
3650         zc.zc_cookie = new_version;
3651
3652         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3653                 return (zpool_standard_error_fmt(hdl, errno,
3654                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3655                     zhp->zpool_name));
3656         return (0);
3657 }
3658
3659 void
3660 zfs_save_arguments(int argc, char **argv, char *string, int len)
3661 {
3662         (void) strlcpy(string, basename(argv[0]), len);
3663         for (int i = 1; i < argc; i++) {
3664                 (void) strlcat(string, " ", len);
3665                 (void) strlcat(string, argv[i], len);
3666         }
3667 }
3668
3669 int
3670 zpool_log_history(libzfs_handle_t *hdl, const char *message)
3671 {
3672         zfs_cmd_t zc = { 0 };
3673         nvlist_t *args;
3674         int err;
3675
3676         args = fnvlist_alloc();
3677         fnvlist_add_string(args, "message", message);
3678         err = zcmd_write_src_nvlist(hdl, &zc, args);
3679         if (err == 0)
3680                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_LOG_HISTORY, &zc);
3681         nvlist_free(args);
3682         zcmd_free_nvlists(&zc);
3683         return (err);
3684 }
3685
3686 /*
3687  * Perform ioctl to get some command history of a pool.
3688  *
3689  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3690  * logical offset of the history buffer to start reading from.
3691  *
3692  * Upon return, 'off' is the next logical offset to read from and
3693  * 'len' is the actual amount of bytes read into 'buf'.
3694  */
3695 static int
3696 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3697 {
3698         zfs_cmd_t zc = { 0 };
3699         libzfs_handle_t *hdl = zhp->zpool_hdl;
3700
3701         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3702
3703         zc.zc_history = (uint64_t)(uintptr_t)buf;
3704         zc.zc_history_len = *len;
3705         zc.zc_history_offset = *off;
3706
3707         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3708                 switch (errno) {
3709                 case EPERM:
3710                         return (zfs_error_fmt(hdl, EZFS_PERM,
3711                             dgettext(TEXT_DOMAIN,
3712                             "cannot show history for pool '%s'"),
3713                             zhp->zpool_name));
3714                 case ENOENT:
3715                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3716                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3717                             "'%s'"), zhp->zpool_name));
3718                 case ENOTSUP:
3719                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3720                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3721                             "'%s', pool must be upgraded"), zhp->zpool_name));
3722                 default:
3723                         return (zpool_standard_error_fmt(hdl, errno,
3724                             dgettext(TEXT_DOMAIN,
3725                             "cannot get history for '%s'"), zhp->zpool_name));
3726                 }
3727         }
3728
3729         *len = zc.zc_history_len;
3730         *off = zc.zc_history_offset;
3731
3732         return (0);
3733 }
3734
3735 /*
3736  * Process the buffer of nvlists, unpacking and storing each nvlist record
3737  * into 'records'.  'leftover' is set to the number of bytes that weren't
3738  * processed as there wasn't a complete record.
3739  */
3740 int
3741 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3742     nvlist_t ***records, uint_t *numrecords)
3743 {
3744         uint64_t reclen;
3745         nvlist_t *nv;
3746         int i;
3747
3748         while (bytes_read > sizeof (reclen)) {
3749
3750                 /* get length of packed record (stored as little endian) */
3751                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3752                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3753
3754                 if (bytes_read < sizeof (reclen) + reclen)
3755                         break;
3756
3757                 /* unpack record */
3758                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3759                         return (ENOMEM);
3760                 bytes_read -= sizeof (reclen) + reclen;
3761                 buf += sizeof (reclen) + reclen;
3762
3763                 /* add record to nvlist array */
3764                 (*numrecords)++;
3765                 if (ISP2(*numrecords + 1)) {
3766                         *records = realloc(*records,
3767                             *numrecords * 2 * sizeof (nvlist_t *));
3768                 }
3769                 (*records)[*numrecords - 1] = nv;
3770         }
3771
3772         *leftover = bytes_read;
3773         return (0);
3774 }
3775
3776 /* from spa_history.c: spa_history_create_obj() */
3777 #define HIS_BUF_LEN_DEF (128 << 10)
3778 #define HIS_BUF_LEN_MAX (1 << 30)
3779
3780 /*
3781  * Retrieve the command history of a pool.
3782  */
3783 int
3784 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3785 {
3786         char *buf = NULL;
3787         uint64_t bufsize = HIS_BUF_LEN_DEF;
3788         uint64_t off = 0;
3789         nvlist_t **records = NULL;
3790         uint_t numrecords = 0;
3791         int err, i;
3792
3793         if ((buf = malloc(bufsize)) == NULL)
3794                 return (ENOMEM);
3795         do {
3796                 uint64_t bytes_read = bufsize;
3797                 uint64_t leftover;
3798
3799                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3800                         break;
3801
3802                 /* if nothing else was read in, we're at EOF, just return */
3803                 if (bytes_read == 0)
3804                         break;
3805
3806                 if ((err = zpool_history_unpack(buf, bytes_read,
3807                     &leftover, &records, &numrecords)) != 0)
3808                         break;
3809                 off -= leftover;
3810
3811                 /*
3812                  * If the history block is too big, double the buffer
3813                  * size and try again.
3814                  */
3815                 if (leftover == bytes_read) {
3816                         free(buf);
3817                         buf = NULL;
3818
3819                         bufsize <<= 1;
3820                         if ((bufsize >= HIS_BUF_LEN_MAX) ||
3821                             ((buf = malloc(bufsize)) == NULL)) {
3822                                 err = ENOMEM;
3823                                 break;
3824                         }
3825                 }
3826
3827                 /* CONSTCOND */
3828         } while (1);
3829         free(buf);
3830
3831         if (!err) {
3832                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3833                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3834                     records, numrecords) == 0);
3835         }
3836         for (i = 0; i < numrecords; i++)
3837                 nvlist_free(records[i]);
3838         free(records);
3839
3840         return (err);
3841 }
3842
3843 void
3844 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
3845     char *pathname, size_t len)
3846 {
3847         zfs_cmd_t zc = { 0 };
3848         boolean_t mounted = B_FALSE;
3849         char *mntpnt = NULL;
3850         char dsname[MAXNAMELEN];
3851
3852         if (dsobj == 0) {
3853                 /* special case for the MOS */
3854                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>", obj);
3855                 return;
3856         }
3857
3858         /* get the dataset's name */
3859         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3860         zc.zc_obj = dsobj;
3861         if (ioctl(zhp->zpool_hdl->libzfs_fd,
3862             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
3863                 /* just write out a path of two object numbers */
3864                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
3865                     dsobj, obj);
3866                 return;
3867         }
3868         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
3869
3870         /* find out if the dataset is mounted */
3871         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
3872
3873         /* get the corrupted object's path */
3874         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
3875         zc.zc_obj = obj;
3876         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
3877             &zc) == 0) {
3878                 if (mounted) {
3879                         (void) snprintf(pathname, len, "%s%s", mntpnt,
3880                             zc.zc_value);
3881                 } else {
3882                         (void) snprintf(pathname, len, "%s:%s",
3883                             dsname, zc.zc_value);
3884                 }
3885         } else {
3886                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname, obj);
3887         }
3888         free(mntpnt);
3889 }
3890
3891 #ifdef sun
3892 /*
3893  * Read the EFI label from the config, if a label does not exist then
3894  * pass back the error to the caller. If the caller has passed a non-NULL
3895  * diskaddr argument then we set it to the starting address of the EFI
3896  * partition.
3897  */
3898 static int
3899 read_efi_label(nvlist_t *config, diskaddr_t *sb)
3900 {
3901         char *path;
3902         int fd;
3903         char diskname[MAXPATHLEN];
3904         int err = -1;
3905
3906         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
3907                 return (err);
3908
3909         (void) snprintf(diskname, sizeof (diskname), "%s%s", RDISK_ROOT,
3910             strrchr(path, '/'));
3911         if ((fd = open(diskname, O_RDONLY|O_NDELAY)) >= 0) {
3912                 struct dk_gpt *vtoc;
3913
3914                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
3915                         if (sb != NULL)
3916                                 *sb = vtoc->efi_parts[0].p_start;
3917                         efi_free(vtoc);
3918                 }
3919                 (void) close(fd);
3920         }
3921         return (err);
3922 }
3923
3924 /*
3925  * determine where a partition starts on a disk in the current
3926  * configuration
3927  */
3928 static diskaddr_t
3929 find_start_block(nvlist_t *config)
3930 {
3931         nvlist_t **child;
3932         uint_t c, children;
3933         diskaddr_t sb = MAXOFFSET_T;
3934         uint64_t wholedisk;
3935
3936         if (nvlist_lookup_nvlist_array(config,
3937             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
3938                 if (nvlist_lookup_uint64(config,
3939                     ZPOOL_CONFIG_WHOLE_DISK,
3940                     &wholedisk) != 0 || !wholedisk) {
3941                         return (MAXOFFSET_T);
3942                 }
3943                 if (read_efi_label(config, &sb) < 0)
3944                         sb = MAXOFFSET_T;
3945                 return (sb);
3946         }
3947
3948         for (c = 0; c < children; c++) {
3949                 sb = find_start_block(child[c]);
3950                 if (sb != MAXOFFSET_T) {
3951                         return (sb);
3952                 }
3953         }
3954         return (MAXOFFSET_T);
3955 }
3956 #endif /* sun */
3957
3958 /*
3959  * Label an individual disk.  The name provided is the short name,
3960  * stripped of any leading /dev path.
3961  */
3962 int
3963 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, const char *name)
3964 {
3965 #ifdef sun
3966         char path[MAXPATHLEN];
3967         struct dk_gpt *vtoc;
3968         int fd;
3969         size_t resv = EFI_MIN_RESV_SIZE;
3970         uint64_t slice_size;
3971         diskaddr_t start_block;
3972         char errbuf[1024];
3973
3974         /* prepare an error message just in case */
3975         (void) snprintf(errbuf, sizeof (errbuf),
3976             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
3977
3978         if (zhp) {
3979                 nvlist_t *nvroot;
3980
3981                 if (zpool_is_bootable(zhp)) {
3982                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3983                             "EFI labeled devices are not supported on root "
3984                             "pools."));
3985                         return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
3986                 }
3987
3988                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
3989                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3990
3991                 if (zhp->zpool_start_block == 0)
3992                         start_block = find_start_block(nvroot);
3993                 else
3994                         start_block = zhp->zpool_start_block;
3995                 zhp->zpool_start_block = start_block;
3996         } else {
3997                 /* new pool */
3998                 start_block = NEW_START_BLOCK;
3999         }
4000
4001         (void) snprintf(path, sizeof (path), "%s/%s%s", RDISK_ROOT, name,
4002             BACKUP_SLICE);
4003
4004         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
4005                 /*
4006                  * This shouldn't happen.  We've long since verified that this
4007                  * is a valid device.
4008                  */
4009                 zfs_error_aux(hdl,
4010                     dgettext(TEXT_DOMAIN, "unable to open device"));
4011                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
4012         }
4013
4014         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
4015                 /*
4016                  * The only way this can fail is if we run out of memory, or we
4017                  * were unable to read the disk's capacity
4018                  */
4019                 if (errno == ENOMEM)
4020                         (void) no_memory(hdl);
4021
4022                 (void) close(fd);
4023                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4024                     "unable to read disk capacity"), name);
4025
4026                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
4027         }
4028
4029         slice_size = vtoc->efi_last_u_lba + 1;
4030         slice_size -= EFI_MIN_RESV_SIZE;
4031         if (start_block == MAXOFFSET_T)
4032                 start_block = NEW_START_BLOCK;
4033         slice_size -= start_block;
4034
4035         vtoc->efi_parts[0].p_start = start_block;
4036         vtoc->efi_parts[0].p_size = slice_size;
4037
4038         /*
4039          * Why we use V_USR: V_BACKUP confuses users, and is considered
4040          * disposable by some EFI utilities (since EFI doesn't have a backup
4041          * slice).  V_UNASSIGNED is supposed to be used only for zero size
4042          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
4043          * etc. were all pretty specific.  V_USR is as close to reality as we
4044          * can get, in the absence of V_OTHER.
4045          */
4046         vtoc->efi_parts[0].p_tag = V_USR;
4047         (void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
4048
4049         vtoc->efi_parts[8].p_start = slice_size + start_block;
4050         vtoc->efi_parts[8].p_size = resv;
4051         vtoc->efi_parts[8].p_tag = V_RESERVED;
4052
4053         if (efi_write(fd, vtoc) != 0) {
4054                 /*
4055                  * Some block drivers (like pcata) may not support EFI
4056                  * GPT labels.  Print out a helpful error message dir-
4057                  * ecting the user to manually label the disk and give
4058                  * a specific slice.
4059                  */
4060                 (void) close(fd);
4061                 efi_free(vtoc);
4062
4063                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4064                     "try using fdisk(1M) and then provide a specific slice"));
4065                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4066         }
4067
4068         (void) close(fd);
4069         efi_free(vtoc);
4070 #endif /* sun */
4071         return (0);
4072 }
4073
4074 static boolean_t
4075 supported_dump_vdev_type(libzfs_handle_t *hdl, nvlist_t *config, char *errbuf)
4076 {
4077         char *type;
4078         nvlist_t **child;
4079         uint_t children, c;
4080
4081         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_TYPE, &type) == 0);
4082         if (strcmp(type, VDEV_TYPE_FILE) == 0 ||
4083             strcmp(type, VDEV_TYPE_HOLE) == 0 ||
4084             strcmp(type, VDEV_TYPE_MISSING) == 0) {
4085                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4086                     "vdev type '%s' is not supported"), type);
4087                 (void) zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf);
4088                 return (B_FALSE);
4089         }
4090         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
4091             &child, &children) == 0) {
4092                 for (c = 0; c < children; c++) {
4093                         if (!supported_dump_vdev_type(hdl, child[c], errbuf))
4094                                 return (B_FALSE);
4095                 }
4096         }
4097         return (B_TRUE);
4098 }
4099
4100 /*
4101  * Check if this zvol is allowable for use as a dump device; zero if
4102  * it is, > 0 if it isn't, < 0 if it isn't a zvol.
4103  *
4104  * Allowable storage configurations include mirrors, all raidz variants, and
4105  * pools with log, cache, and spare devices.  Pools which are backed by files or
4106  * have missing/hole vdevs are not suitable.
4107  */
4108 int
4109 zvol_check_dump_config(char *arg)
4110 {
4111         zpool_handle_t *zhp = NULL;
4112         nvlist_t *config, *nvroot;
4113         char *p, *volname;
4114         nvlist_t **top;
4115         uint_t toplevels;
4116         libzfs_handle_t *hdl;
4117         char errbuf[1024];
4118         char poolname[ZPOOL_MAXNAMELEN];
4119         int pathlen = strlen(ZVOL_FULL_DEV_DIR);
4120         int ret = 1;
4121
4122         if (strncmp(arg, ZVOL_FULL_DEV_DIR, pathlen)) {
4123                 return (-1);
4124         }
4125
4126         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
4127             "dump is not supported on device '%s'"), arg);
4128
4129         if ((hdl = libzfs_init()) == NULL)
4130                 return (1);
4131         libzfs_print_on_error(hdl, B_TRUE);
4132
4133         volname = arg + pathlen;
4134
4135         /* check the configuration of the pool */
4136         if ((p = strchr(volname, '/')) == NULL) {
4137                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4138                     "malformed dataset name"));
4139                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4140                 return (1);
4141         } else if (p - volname >= ZFS_MAXNAMELEN) {
4142                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4143                     "dataset name is too long"));
4144                 (void) zfs_error(hdl, EZFS_NAMETOOLONG, errbuf);
4145                 return (1);
4146         } else {
4147                 (void) strncpy(poolname, volname, p - volname);
4148                 poolname[p - volname] = '\0';
4149         }
4150
4151         if ((zhp = zpool_open(hdl, poolname)) == NULL) {
4152                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4153                     "could not open pool '%s'"), poolname);
4154                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
4155                 goto out;
4156         }
4157         config = zpool_get_config(zhp, NULL);
4158         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4159             &nvroot) != 0) {
4160                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4161                     "could not obtain vdev configuration for  '%s'"), poolname);
4162                 (void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
4163                 goto out;
4164         }
4165
4166         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4167             &top, &toplevels) == 0);
4168
4169         if (!supported_dump_vdev_type(hdl, top[0], errbuf)) {
4170                 goto out;
4171         }
4172         ret = 0;
4173
4174 out:
4175         if (zhp)
4176                 zpool_close(zhp);
4177         libzfs_fini(hdl);
4178         return (ret);
4179 }