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