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

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  25  * Copyright (c) 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) {
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                 } else if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1713                         nvlist_free(props);
1714                         return (-1);
1715                 }
1716         }
1717
1718         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1719
1720         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1721             &zc.zc_guid) == 0);
1722
1723         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1724                 nvlist_free(props);
1725                 return (-1);
1726         }
1727         if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1728                 nvlist_free(props);
1729                 return (-1);
1730         }
1731
1732         zc.zc_cookie = flags;
1733         while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1734             errno == ENOMEM) {
1735                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1736                         zcmd_free_nvlists(&zc);
1737                         return (-1);
1738                 }
1739         }
1740         if (ret != 0)
1741                 error = errno;
1742
1743         (void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1744         zpool_get_rewind_policy(config, &policy);
1745
1746         if (error) {
1747                 char desc[1024];
1748
1749                 /*
1750                  * Dry-run failed, but we print out what success
1751                  * looks like if we found a best txg
1752                  */
1753                 if (policy.zrp_request & ZPOOL_TRY_REWIND) {
1754                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1755                             B_TRUE, nv);
1756                         nvlist_free(nv);
1757                         return (-1);
1758                 }
1759
1760                 if (newname == NULL)
1761                         (void) snprintf(desc, sizeof (desc),
1762                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1763                             thename);
1764                 else
1765                         (void) snprintf(desc, sizeof (desc),
1766                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1767                             origname, thename);
1768
1769                 switch (error) {
1770                 case ENOTSUP:
1771                         if (nv != NULL && nvlist_lookup_nvlist(nv,
1772                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1773                             nvlist_exists(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT)) {
1774                                 (void) printf(dgettext(TEXT_DOMAIN, "This "
1775                                     "pool uses the following feature(s) not "
1776                                     "supported by this system:\n"));
1777                                 zpool_print_unsup_feat(nv);
1778                                 if (nvlist_exists(nvinfo,
1779                                     ZPOOL_CONFIG_CAN_RDONLY)) {
1780                                         (void) printf(dgettext(TEXT_DOMAIN,
1781                                             "All unsupported features are only "
1782                                             "required for writing to the pool."
1783                                             "\nThe pool can be imported using "
1784                                             "'-o readonly=on'.\n"));
1785                                 }
1786                         }
1787                         /*
1788                          * Unsupported version.
1789                          */
1790                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1791                         break;
1792
1793                 case EINVAL:
1794                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1795                         break;
1796
1797                 case EROFS:
1798                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1799                             "one or more devices is read only"));
1800                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1801                         break;
1802
1803                 case ENXIO:
1804                         if (nv && nvlist_lookup_nvlist(nv,
1805                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1806                             nvlist_lookup_nvlist(nvinfo,
1807                             ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1808                                 (void) printf(dgettext(TEXT_DOMAIN,
1809                                     "The devices below are missing, use "
1810                                     "'-m' to import the pool anyway:\n"));
1811                                 print_vdev_tree(hdl, NULL, missing, 2);
1812                                 (void) printf("\n");
1813                         }
1814                         (void) zpool_standard_error(hdl, error, desc);
1815                         break;
1816
1817                 case EEXIST:
1818                         (void) zpool_standard_error(hdl, error, desc);
1819                         break;
1820
1821                 default:
1822                         (void) zpool_standard_error(hdl, error, desc);
1823                         zpool_explain_recover(hdl,
1824                             newname ? origname : thename, -error, nv);
1825                         break;
1826                 }
1827
1828                 nvlist_free(nv);
1829                 ret = -1;
1830         } else {
1831                 zpool_handle_t *zhp;
1832
1833                 /*
1834                  * This should never fail, but play it safe anyway.
1835                  */
1836                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1837                         ret = -1;
1838                 else if (zhp != NULL)
1839                         zpool_close(zhp);
1840                 if (policy.zrp_request &
1841                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1842                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1843                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1844                 }
1845                 nvlist_free(nv);
1846                 return (0);
1847         }
1848
1849         zcmd_free_nvlists(&zc);
1850         nvlist_free(props);
1851
1852         return (ret);
1853 }
1854
1855 /*
1856  * Scan the pool.
1857  */
1858 int
1859 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1860 {
1861         zfs_cmd_t zc = { 0 };
1862         char msg[1024];
1863         libzfs_handle_t *hdl = zhp->zpool_hdl;
1864
1865         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1866         zc.zc_cookie = func;
1867
1868         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1869             (errno == ENOENT && func != POOL_SCAN_NONE))
1870                 return (0);
1871
1872         if (func == POOL_SCAN_SCRUB) {
1873                 (void) snprintf(msg, sizeof (msg),
1874                     dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1875         } else if (func == POOL_SCAN_NONE) {
1876                 (void) snprintf(msg, sizeof (msg),
1877                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1878                     zc.zc_name);
1879         } else {
1880                 assert(!"unexpected result");
1881         }
1882
1883         if (errno == EBUSY) {
1884                 nvlist_t *nvroot;
1885                 pool_scan_stat_t *ps = NULL;
1886                 uint_t psc;
1887
1888                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
1889                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1890                 (void) nvlist_lookup_uint64_array(nvroot,
1891                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1892                 if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1893                         return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1894                 else
1895                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
1896         } else if (errno == ENOENT) {
1897                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1898         } else {
1899                 return (zpool_standard_error(hdl, errno, msg));
1900         }
1901 }
1902
1903 /*
1904  * This provides a very minimal check whether a given string is likely a
1905  * c#t#d# style string.  Users of this are expected to do their own
1906  * verification of the s# part.
1907  */
1908 #define CTD_CHECK(str)  (str && str[0] == 'c' && isdigit(str[1]))
1909
1910 /*
1911  * More elaborate version for ones which may start with "/dev/dsk/"
1912  * and the like.
1913  */
1914 static int
1915 ctd_check_path(char *str) {
1916         /*
1917          * If it starts with a slash, check the last component.
1918          */
1919         if (str && str[0] == '/') {
1920                 char *tmp = strrchr(str, '/');
1921
1922                 /*
1923                  * If it ends in "/old", check the second-to-last
1924                  * component of the string instead.
1925                  */
1926                 if (tmp != str && strcmp(tmp, "/old") == 0) {
1927                         for (tmp--; *tmp != '/'; tmp--)
1928                                 ;
1929                 }
1930                 str = tmp + 1;
1931         }
1932         return (CTD_CHECK(str));
1933 }
1934
1935 /*
1936  * Find a vdev that matches the search criteria specified. We use the
1937  * the nvpair name to determine how we should look for the device.
1938  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1939  * spare; but FALSE if its an INUSE spare.
1940  */
1941 static nvlist_t *
1942 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1943     boolean_t *l2cache, boolean_t *log)
1944 {
1945         uint_t c, children;
1946         nvlist_t **child;
1947         nvlist_t *ret;
1948         uint64_t is_log;
1949         char *srchkey;
1950         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
1951
1952         /* Nothing to look for */
1953         if (search == NULL || pair == NULL)
1954                 return (NULL);
1955
1956         /* Obtain the key we will use to search */
1957         srchkey = nvpair_name(pair);
1958
1959         switch (nvpair_type(pair)) {
1960         case DATA_TYPE_UINT64:
1961                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
1962                         uint64_t srchval, theguid;
1963
1964                         verify(nvpair_value_uint64(pair, &srchval) == 0);
1965                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
1966                             &theguid) == 0);
1967                         if (theguid == srchval)
1968                                 return (nv);
1969                 }
1970                 break;
1971
1972         case DATA_TYPE_STRING: {
1973                 char *srchval, *val;
1974
1975                 verify(nvpair_value_string(pair, &srchval) == 0);
1976                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
1977                         break;
1978
1979                 /*
1980                  * Search for the requested value. Special cases:
1981                  *
1982                  * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
1983                  *   "s0" or "s0/old".  The "s0" part is hidden from the user,
1984                  *   but included in the string, so this matches around it.
1985                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
1986                  *
1987                  * Otherwise, all other searches are simple string compares.
1988                  */
1989                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0 &&
1990                     ctd_check_path(val)) {
1991                         uint64_t wholedisk = 0;
1992
1993                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1994                             &wholedisk);
1995                         if (wholedisk) {
1996                                 int slen = strlen(srchval);
1997                                 int vlen = strlen(val);
1998
1999                                 if (slen != vlen - 2)
2000                                         break;
2001
2002                                 /*
2003                                  * make_leaf_vdev() should only set
2004                                  * wholedisk for ZPOOL_CONFIG_PATHs which
2005                                  * will include "/dev/dsk/", giving plenty of
2006                                  * room for the indices used next.
2007                                  */
2008                                 ASSERT(vlen >= 6);
2009
2010                                 /*
2011                                  * strings identical except trailing "s0"
2012                                  */
2013                                 if (strcmp(&val[vlen - 2], "s0") == 0 &&
2014                                     strncmp(srchval, val, slen) == 0)
2015                                         return (nv);
2016
2017                                 /*
2018                                  * strings identical except trailing "s0/old"
2019                                  */
2020                                 if (strcmp(&val[vlen - 6], "s0/old") == 0 &&
2021                                     strcmp(&srchval[slen - 4], "/old") == 0 &&
2022                                     strncmp(srchval, val, slen - 4) == 0)
2023                                         return (nv);
2024
2025                                 break;
2026                         }
2027                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2028                         char *type, *idx, *end, *p;
2029                         uint64_t id, vdev_id;
2030
2031                         /*
2032                          * Determine our vdev type, keeping in mind
2033                          * that the srchval is composed of a type and
2034                          * vdev id pair (i.e. mirror-4).
2035                          */
2036                         if ((type = strdup(srchval)) == NULL)
2037                                 return (NULL);
2038
2039                         if ((p = strrchr(type, '-')) == NULL) {
2040                                 free(type);
2041                                 break;
2042                         }
2043                         idx = p + 1;
2044                         *p = '\0';
2045
2046                         /*
2047                          * If the types don't match then keep looking.
2048                          */
2049                         if (strncmp(val, type, strlen(val)) != 0) {
2050                                 free(type);
2051                                 break;
2052                         }
2053
2054                         verify(strncmp(type, VDEV_TYPE_RAIDZ,
2055                             strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2056                             strncmp(type, VDEV_TYPE_MIRROR,
2057                             strlen(VDEV_TYPE_MIRROR)) == 0);
2058                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
2059                             &id) == 0);
2060
2061                         errno = 0;
2062                         vdev_id = strtoull(idx, &end, 10);
2063
2064                         free(type);
2065                         if (errno != 0)
2066                                 return (NULL);
2067
2068                         /*
2069                          * Now verify that we have the correct vdev id.
2070                          */
2071                         if (vdev_id == id)
2072                                 return (nv);
2073                 }
2074
2075                 /*
2076                  * Common case
2077                  */
2078                 if (strcmp(srchval, val) == 0)
2079                         return (nv);
2080                 break;
2081         }
2082
2083         default:
2084                 break;
2085         }
2086
2087         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2088             &child, &children) != 0)
2089                 return (NULL);
2090
2091         for (c = 0; c < children; c++) {
2092                 if ((ret = vdev_to_nvlist_iter(child[c], search,
2093                     avail_spare, l2cache, NULL)) != NULL) {
2094                         /*
2095                          * The 'is_log' value is only set for the toplevel
2096                          * vdev, not the leaf vdevs.  So we always lookup the
2097                          * log device from the root of the vdev tree (where
2098                          * 'log' is non-NULL).
2099                          */
2100                         if (log != NULL &&
2101                             nvlist_lookup_uint64(child[c],
2102                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2103                             is_log) {
2104                                 *log = B_TRUE;
2105                         }
2106                         return (ret);
2107                 }
2108         }
2109
2110         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2111             &child, &children) == 0) {
2112                 for (c = 0; c < children; c++) {
2113                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2114                             avail_spare, l2cache, NULL)) != NULL) {
2115                                 *avail_spare = B_TRUE;
2116                                 return (ret);
2117                         }
2118                 }
2119         }
2120
2121         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2122             &child, &children) == 0) {
2123                 for (c = 0; c < children; c++) {
2124                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2125                             avail_spare, l2cache, NULL)) != NULL) {
2126                                 *l2cache = B_TRUE;
2127                                 return (ret);
2128                         }
2129                 }
2130         }
2131
2132         return (NULL);
2133 }
2134
2135 /*
2136  * Given a physical path (minus the "/devices" prefix), find the
2137  * associated vdev.
2138  */
2139 nvlist_t *
2140 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2141     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2142 {
2143         nvlist_t *search, *nvroot, *ret;
2144
2145         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2146         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
2147
2148         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2149             &nvroot) == 0);
2150
2151         *avail_spare = B_FALSE;
2152         *l2cache = B_FALSE;
2153         if (log != NULL)
2154                 *log = B_FALSE;
2155         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2156         nvlist_free(search);
2157
2158         return (ret);
2159 }
2160
2161 /*
2162  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2163  */
2164 boolean_t
2165 zpool_vdev_is_interior(const char *name)
2166 {
2167         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2168             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2169                 return (B_TRUE);
2170         return (B_FALSE);
2171 }
2172
2173 nvlist_t *
2174 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2175     boolean_t *l2cache, boolean_t *log)
2176 {
2177         char buf[MAXPATHLEN];
2178         char *end;
2179         nvlist_t *nvroot, *search, *ret;
2180         uint64_t guid;
2181
2182         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2183
2184         guid = strtoull(path, &end, 10);
2185         if (guid != 0 && *end == '\0') {
2186                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
2187         } else if (zpool_vdev_is_interior(path)) {
2188                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
2189         } else if (path[0] != '/') {
2190                 (void) snprintf(buf, sizeof (buf), "%s%s", _PATH_DEV, path);
2191                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, buf) == 0);
2192         } else {
2193                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
2194         }
2195
2196         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2197             &nvroot) == 0);
2198
2199         *avail_spare = B_FALSE;
2200         *l2cache = B_FALSE;
2201         if (log != NULL)
2202                 *log = B_FALSE;
2203         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2204         nvlist_free(search);
2205
2206         return (ret);
2207 }
2208
2209 static int
2210 vdev_online(nvlist_t *nv)
2211 {
2212         uint64_t ival;
2213
2214         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2215             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2216             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2217                 return (0);
2218
2219         return (1);
2220 }
2221
2222 /*
2223  * Helper function for zpool_get_physpaths().
2224  */
2225 static int
2226 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2227     size_t *bytes_written)
2228 {
2229         size_t bytes_left, pos, rsz;
2230         char *tmppath;
2231         const char *format;
2232
2233         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2234             &tmppath) != 0)
2235                 return (EZFS_NODEVICE);
2236
2237         pos = *bytes_written;
2238         bytes_left = physpath_size - pos;
2239         format = (pos == 0) ? "%s" : " %s";
2240
2241         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2242         *bytes_written += rsz;
2243
2244         if (rsz >= bytes_left) {
2245                 /* if physpath was not copied properly, clear it */
2246                 if (bytes_left != 0) {
2247                         physpath[pos] = 0;
2248                 }
2249                 return (EZFS_NOSPC);
2250         }
2251         return (0);
2252 }
2253
2254 static int
2255 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2256     size_t *rsz, boolean_t is_spare)
2257 {
2258         char *type;
2259         int ret;
2260
2261         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2262                 return (EZFS_INVALCONFIG);
2263
2264         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2265                 /*
2266                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2267                  * For a spare vdev, we only want to boot from the active
2268                  * spare device.
2269                  */
2270                 if (is_spare) {
2271                         uint64_t spare = 0;
2272                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2273                             &spare);
2274                         if (!spare)
2275                                 return (EZFS_INVALCONFIG);
2276                 }
2277
2278                 if (vdev_online(nv)) {
2279                         if ((ret = vdev_get_one_physpath(nv, physpath,
2280                             phypath_size, rsz)) != 0)
2281                                 return (ret);
2282                 }
2283         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2284             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2285             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2286                 nvlist_t **child;
2287                 uint_t count;
2288                 int i, ret;
2289
2290                 if (nvlist_lookup_nvlist_array(nv,
2291                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2292                         return (EZFS_INVALCONFIG);
2293
2294                 for (i = 0; i < count; i++) {
2295                         ret = vdev_get_physpaths(child[i], physpath,
2296                             phypath_size, rsz, is_spare);
2297                         if (ret == EZFS_NOSPC)
2298                                 return (ret);
2299                 }
2300         }
2301
2302         return (EZFS_POOL_INVALARG);
2303 }
2304
2305 /*
2306  * Get phys_path for a root pool config.
2307  * Return 0 on success; non-zero on failure.
2308  */
2309 static int
2310 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2311 {
2312         size_t rsz;
2313         nvlist_t *vdev_root;
2314         nvlist_t **child;
2315         uint_t count;
2316         char *type;
2317
2318         rsz = 0;
2319
2320         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2321             &vdev_root) != 0)
2322                 return (EZFS_INVALCONFIG);
2323
2324         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2325             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2326             &child, &count) != 0)
2327                 return (EZFS_INVALCONFIG);
2328
2329         /*
2330          * root pool can not have EFI labeled disks and can only have
2331          * a single top-level vdev.
2332          */
2333         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1 ||
2334             pool_uses_efi(vdev_root))
2335                 return (EZFS_POOL_INVALARG);
2336
2337         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2338             B_FALSE);
2339
2340         /* No online devices */
2341         if (rsz == 0)
2342                 return (EZFS_NODEVICE);
2343
2344         return (0);
2345 }
2346
2347 /*
2348  * Get phys_path for a root pool
2349  * Return 0 on success; non-zero on failure.
2350  */
2351 int
2352 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2353 {
2354         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2355             phypath_size));
2356 }
2357
2358 /*
2359  * If the device has being dynamically expanded then we need to relabel
2360  * the disk to use the new unallocated space.
2361  */
2362 static int
2363 zpool_relabel_disk(libzfs_handle_t *hdl, const char *name)
2364 {
2365 #ifdef sun
2366         char path[MAXPATHLEN];
2367         char errbuf[1024];
2368         int fd, error;
2369         int (*_efi_use_whole_disk)(int);
2370
2371         if ((_efi_use_whole_disk = (int (*)(int))dlsym(RTLD_DEFAULT,
2372             "efi_use_whole_disk")) == NULL)
2373                 return (-1);
2374
2375         (void) snprintf(path, sizeof (path), "%s/%s", RDISK_ROOT, name);
2376
2377         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
2378                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2379                     "relabel '%s': unable to open device"), name);
2380                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
2381         }
2382
2383         /*
2384          * It's possible that we might encounter an error if the device
2385          * does not have any unallocated space left. If so, we simply
2386          * ignore that error and continue on.
2387          */
2388         error = _efi_use_whole_disk(fd);
2389         (void) close(fd);
2390         if (error && error != VT_ENOSPC) {
2391                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2392                     "relabel '%s': unable to read disk capacity"), name);
2393                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
2394         }
2395 #endif  /* sun */
2396         return (0);
2397 }
2398
2399 /*
2400  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2401  * ZFS_ONLINE_* flags.
2402  */
2403 int
2404 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2405     vdev_state_t *newstate)
2406 {
2407         zfs_cmd_t zc = { 0 };
2408         char msg[1024];
2409         nvlist_t *tgt;
2410         boolean_t avail_spare, l2cache, islog;
2411         libzfs_handle_t *hdl = zhp->zpool_hdl;
2412
2413         if (flags & ZFS_ONLINE_EXPAND) {
2414                 (void) snprintf(msg, sizeof (msg),
2415                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2416         } else {
2417                 (void) snprintf(msg, sizeof (msg),
2418                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2419         }
2420
2421         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2422         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2423             &islog)) == NULL)
2424                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2425
2426         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2427
2428         if (avail_spare)
2429                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2430
2431         if (flags & ZFS_ONLINE_EXPAND ||
2432             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2433                 char *pathname = NULL;
2434                 uint64_t wholedisk = 0;
2435
2436                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2437                     &wholedisk);
2438                 verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
2439                     &pathname) == 0);
2440
2441                 /*
2442                  * XXX - L2ARC 1.0 devices can't support expansion.
2443                  */
2444                 if (l2cache) {
2445                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2446                             "cannot expand cache devices"));
2447                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2448                 }
2449
2450                 if (wholedisk) {
2451                         pathname += strlen(DISK_ROOT) + 1;
2452                         (void) zpool_relabel_disk(hdl, pathname);
2453                 }
2454         }
2455
2456         zc.zc_cookie = VDEV_STATE_ONLINE;
2457         zc.zc_obj = flags;
2458
2459         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2460                 if (errno == EINVAL) {
2461                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2462                             "from this pool into a new one.  Use '%s' "
2463                             "instead"), "zpool detach");
2464                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2465                 }
2466                 return (zpool_standard_error(hdl, errno, msg));
2467         }
2468
2469         *newstate = zc.zc_cookie;
2470         return (0);
2471 }
2472
2473 /*
2474  * Take the specified vdev offline
2475  */
2476 int
2477 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2478 {
2479         zfs_cmd_t zc = { 0 };
2480         char msg[1024];
2481         nvlist_t *tgt;
2482         boolean_t avail_spare, l2cache;
2483         libzfs_handle_t *hdl = zhp->zpool_hdl;
2484
2485         (void) snprintf(msg, sizeof (msg),
2486             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2487
2488         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2489         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2490             NULL)) == NULL)
2491                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2492
2493         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2494
2495         if (avail_spare)
2496                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2497
2498         zc.zc_cookie = VDEV_STATE_OFFLINE;
2499         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2500
2501         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2502                 return (0);
2503
2504         switch (errno) {
2505         case EBUSY:
2506
2507                 /*
2508                  * There are no other replicas of this device.
2509                  */
2510                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2511
2512         case EEXIST:
2513                 /*
2514                  * The log device has unplayed logs
2515                  */
2516                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2517
2518         default:
2519                 return (zpool_standard_error(hdl, errno, msg));
2520         }
2521 }
2522
2523 /*
2524  * Mark the given vdev faulted.
2525  */
2526 int
2527 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2528 {
2529         zfs_cmd_t zc = { 0 };
2530         char msg[1024];
2531         libzfs_handle_t *hdl = zhp->zpool_hdl;
2532
2533         (void) snprintf(msg, sizeof (msg),
2534             dgettext(TEXT_DOMAIN, "cannot fault %llu"), guid);
2535
2536         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2537         zc.zc_guid = guid;
2538         zc.zc_cookie = VDEV_STATE_FAULTED;
2539         zc.zc_obj = aux;
2540
2541         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2542                 return (0);
2543
2544         switch (errno) {
2545         case EBUSY:
2546
2547                 /*
2548                  * There are no other replicas of this device.
2549                  */
2550                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2551
2552         default:
2553                 return (zpool_standard_error(hdl, errno, msg));
2554         }
2555
2556 }
2557
2558 /*
2559  * Mark the given vdev degraded.
2560  */
2561 int
2562 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2563 {
2564         zfs_cmd_t zc = { 0 };
2565         char msg[1024];
2566         libzfs_handle_t *hdl = zhp->zpool_hdl;
2567
2568         (void) snprintf(msg, sizeof (msg),
2569             dgettext(TEXT_DOMAIN, "cannot degrade %llu"), guid);
2570
2571         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2572         zc.zc_guid = guid;
2573         zc.zc_cookie = VDEV_STATE_DEGRADED;
2574         zc.zc_obj = aux;
2575
2576         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2577                 return (0);
2578
2579         return (zpool_standard_error(hdl, errno, msg));
2580 }
2581
2582 /*
2583  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2584  * a hot spare.
2585  */
2586 static boolean_t
2587 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2588 {
2589         nvlist_t **child;
2590         uint_t c, children;
2591         char *type;
2592
2593         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2594             &children) == 0) {
2595                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2596                     &type) == 0);
2597
2598                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2599                     children == 2 && child[which] == tgt)
2600                         return (B_TRUE);
2601
2602                 for (c = 0; c < children; c++)
2603                         if (is_replacing_spare(child[c], tgt, which))
2604                                 return (B_TRUE);
2605         }
2606
2607         return (B_FALSE);
2608 }
2609
2610 /*
2611  * Attach new_disk (fully described by nvroot) to old_disk.
2612  * If 'replacing' is specified, the new disk will replace the old one.
2613  */
2614 int
2615 zpool_vdev_attach(zpool_handle_t *zhp,
2616     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2617 {
2618         zfs_cmd_t zc = { 0 };
2619         char msg[1024];
2620         int ret;
2621         nvlist_t *tgt;
2622         boolean_t avail_spare, l2cache, islog;
2623         uint64_t val;
2624         char *newname;
2625         nvlist_t **child;
2626         uint_t children;
2627         nvlist_t *config_root;
2628         libzfs_handle_t *hdl = zhp->zpool_hdl;
2629         boolean_t rootpool = zpool_is_bootable(zhp);
2630
2631         if (replacing)
2632                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2633                     "cannot replace %s with %s"), old_disk, new_disk);
2634         else
2635                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2636                     "cannot attach %s to %s"), new_disk, old_disk);
2637
2638         /*
2639          * If this is a root pool, make sure that we're not attaching an
2640          * EFI labeled device.
2641          */
2642         if (rootpool && pool_uses_efi(nvroot)) {
2643                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2644                     "EFI labeled devices are not supported on root pools."));
2645                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
2646         }
2647
2648         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2649         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2650             &islog)) == 0)
2651                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2652
2653         if (avail_spare)
2654                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2655
2656         if (l2cache)
2657                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2658
2659         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2660         zc.zc_cookie = replacing;
2661
2662         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2663             &child, &children) != 0 || children != 1) {
2664                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2665                     "new device must be a single disk"));
2666                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2667         }
2668
2669         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2670             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2671
2672         if ((newname = zpool_vdev_name(NULL, NULL, child[0], B_FALSE)) == NULL)
2673                 return (-1);
2674
2675         /*
2676          * If the target is a hot spare that has been swapped in, we can only
2677          * replace it with another hot spare.
2678          */
2679         if (replacing &&
2680             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2681             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2682             NULL) == NULL || !avail_spare) &&
2683             is_replacing_spare(config_root, tgt, 1)) {
2684                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2685                     "can only be replaced by another hot spare"));
2686                 free(newname);
2687                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2688         }
2689
2690         free(newname);
2691
2692         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2693                 return (-1);
2694
2695         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2696
2697         zcmd_free_nvlists(&zc);
2698
2699         if (ret == 0) {
2700                 if (rootpool) {
2701                         /*
2702                          * XXX need a better way to prevent user from
2703                          * booting up a half-baked vdev.
2704                          */
2705                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2706                             "sure to wait until resilver is done "
2707                             "before rebooting.\n"));
2708                         (void) fprintf(stderr, "\n");
2709                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "If "
2710                             "you boot from pool '%s', you may need to update\n"
2711                             "boot code on newly attached disk '%s'.\n\n"
2712                             "Assuming you use GPT partitioning and 'da0' is "
2713                             "your new boot disk\n"
2714                             "you may use the following command:\n\n"
2715                             "\tgpart bootcode -b /boot/pmbr -p "
2716                             "/boot/gptzfsboot -i 1 da0\n\n"),
2717                             zhp->zpool_name, new_disk);
2718                 }
2719                 return (0);
2720         }
2721
2722         switch (errno) {
2723         case ENOTSUP:
2724                 /*
2725                  * Can't attach to or replace this type of vdev.
2726                  */
2727                 if (replacing) {
2728                         uint64_t version = zpool_get_prop_int(zhp,
2729                             ZPOOL_PROP_VERSION, NULL);
2730
2731                         if (islog)
2732                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2733                                     "cannot replace a log with a spare"));
2734                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2735                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2736                                     "already in replacing/spare config; wait "
2737                                     "for completion or use 'zpool detach'"));
2738                         else
2739                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2740                                     "cannot replace a replacing device"));
2741                 } else {
2742                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2743                             "can only attach to mirrors and top-level "
2744                             "disks"));
2745                 }
2746                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2747                 break;
2748
2749         case EINVAL:
2750                 /*
2751                  * The new device must be a single disk.
2752                  */
2753                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2754                     "new device must be a single disk"));
2755                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2756                 break;
2757
2758         case EBUSY:
2759                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2760                     new_disk);
2761                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2762                 break;
2763
2764         case EOVERFLOW:
2765                 /*
2766                  * The new device is too small.
2767                  */
2768                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2769                     "device is too small"));
2770                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2771                 break;
2772
2773         case EDOM:
2774                 /*
2775                  * The new device has a different alignment requirement.
2776                  */
2777                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2778                     "devices have different sector alignment"));
2779                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2780                 break;
2781
2782         case ENAMETOOLONG:
2783                 /*
2784                  * The resulting top-level vdev spec won't fit in the label.
2785                  */
2786                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2787                 break;
2788
2789         default:
2790                 (void) zpool_standard_error(hdl, errno, msg);
2791         }
2792
2793         return (-1);
2794 }
2795
2796 /*
2797  * Detach the specified device.
2798  */
2799 int
2800 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2801 {
2802         zfs_cmd_t zc = { 0 };
2803         char msg[1024];
2804         nvlist_t *tgt;
2805         boolean_t avail_spare, l2cache;
2806         libzfs_handle_t *hdl = zhp->zpool_hdl;
2807
2808         (void) snprintf(msg, sizeof (msg),
2809             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2810
2811         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2812         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2813             NULL)) == 0)
2814                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2815
2816         if (avail_spare)
2817                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2818
2819         if (l2cache)
2820                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2821
2822         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2823
2824         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2825                 return (0);
2826
2827         switch (errno) {
2828
2829         case ENOTSUP:
2830                 /*
2831                  * Can't detach from this type of vdev.
2832                  */
2833                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2834                     "applicable to mirror and replacing vdevs"));
2835                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2836                 break;
2837
2838         case EBUSY:
2839                 /*
2840                  * There are no other replicas of this device.
2841                  */
2842                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2843                 break;
2844
2845         default:
2846                 (void) zpool_standard_error(hdl, errno, msg);
2847         }
2848
2849         return (-1);
2850 }
2851
2852 /*
2853  * Find a mirror vdev in the source nvlist.
2854  *
2855  * The mchild array contains a list of disks in one of the top-level mirrors
2856  * of the source pool.  The schild array contains a list of disks that the
2857  * user specified on the command line.  We loop over the mchild array to
2858  * see if any entry in the schild array matches.
2859  *
2860  * If a disk in the mchild array is found in the schild array, we return
2861  * the index of that entry.  Otherwise we return -1.
2862  */
2863 static int
2864 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2865     nvlist_t **schild, uint_t schildren)
2866 {
2867         uint_t mc;
2868
2869         for (mc = 0; mc < mchildren; mc++) {
2870                 uint_t sc;
2871                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2872                     mchild[mc], B_FALSE);
2873
2874                 for (sc = 0; sc < schildren; sc++) {
2875                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2876                             schild[sc], B_FALSE);
2877                         boolean_t result = (strcmp(mpath, spath) == 0);
2878
2879                         free(spath);
2880                         if (result) {
2881                                 free(mpath);
2882                                 return (mc);
2883                         }
2884                 }
2885
2886                 free(mpath);
2887         }
2888
2889         return (-1);
2890 }
2891
2892 /*
2893  * Split a mirror pool.  If newroot points to null, then a new nvlist
2894  * is generated and it is the responsibility of the caller to free it.
2895  */
2896 int
2897 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2898     nvlist_t *props, splitflags_t flags)
2899 {
2900         zfs_cmd_t zc = { 0 };
2901         char msg[1024];
2902         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2903         nvlist_t **varray = NULL, *zc_props = NULL;
2904         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2905         libzfs_handle_t *hdl = zhp->zpool_hdl;
2906         uint64_t vers;
2907         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2908         int retval = 0;
2909
2910         (void) snprintf(msg, sizeof (msg),
2911             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2912
2913         if (!zpool_name_valid(hdl, B_FALSE, newname))
2914                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2915
2916         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2917                 (void) fprintf(stderr, gettext("Internal error: unable to "
2918                     "retrieve pool configuration\n"));
2919                 return (-1);
2920         }
2921
2922         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2923             == 0);
2924         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2925
2926         if (props) {
2927                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2928                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2929                     props, vers, flags, msg)) == NULL)
2930                         return (-1);
2931         }
2932
2933         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2934             &children) != 0) {
2935                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2936                     "Source pool is missing vdev tree"));
2937                 if (zc_props)
2938                         nvlist_free(zc_props);
2939                 return (-1);
2940         }
2941
2942         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2943         vcount = 0;
2944
2945         if (*newroot == NULL ||
2946             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2947             &newchild, &newchildren) != 0)
2948                 newchildren = 0;
2949
2950         for (c = 0; c < children; c++) {
2951                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2952                 char *type;
2953                 nvlist_t **mchild, *vdev;
2954                 uint_t mchildren;
2955                 int entry;
2956
2957                 /*
2958                  * Unlike cache & spares, slogs are stored in the
2959                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2960                  */
2961                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2962                     &is_log);
2963                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2964                     &is_hole);
2965                 if (is_log || is_hole) {
2966                         /*
2967                          * Create a hole vdev and put it in the config.
2968                          */
2969                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2970                                 goto out;
2971                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2972                             VDEV_TYPE_HOLE) != 0)
2973                                 goto out;
2974                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2975                             1) != 0)
2976                                 goto out;
2977                         if (lastlog == 0)
2978                                 lastlog = vcount;
2979                         varray[vcount++] = vdev;
2980                         continue;
2981                 }
2982                 lastlog = 0;
2983                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
2984                     == 0);
2985                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
2986                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2987                             "Source pool must be composed only of mirrors\n"));
2988                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2989                         goto out;
2990                 }
2991
2992                 verify(nvlist_lookup_nvlist_array(child[c],
2993                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2994
2995                 /* find or add an entry for this top-level vdev */
2996                 if (newchildren > 0 &&
2997                     (entry = find_vdev_entry(zhp, mchild, mchildren,
2998                     newchild, newchildren)) >= 0) {
2999                         /* We found a disk that the user specified. */
3000                         vdev = mchild[entry];
3001                         ++found;
3002                 } else {
3003                         /* User didn't specify a disk for this vdev. */
3004                         vdev = mchild[mchildren - 1];
3005                 }
3006
3007                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
3008                         goto out;
3009         }
3010
3011         /* did we find every disk the user specified? */
3012         if (found != newchildren) {
3013                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
3014                     "include at most one disk from each mirror"));
3015                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3016                 goto out;
3017         }
3018
3019         /* Prepare the nvlist for populating. */
3020         if (*newroot == NULL) {
3021                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
3022                         goto out;
3023                 freelist = B_TRUE;
3024                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
3025                     VDEV_TYPE_ROOT) != 0)
3026                         goto out;
3027         } else {
3028                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
3029         }
3030
3031         /* Add all the children we found */
3032         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
3033             lastlog == 0 ? vcount : lastlog) != 0)
3034                 goto out;
3035
3036         /*
3037          * If we're just doing a dry run, exit now with success.
3038          */
3039         if (flags.dryrun) {
3040                 memory_err = B_FALSE;
3041                 freelist = B_FALSE;
3042                 goto out;
3043         }
3044
3045         /* now build up the config list & call the ioctl */
3046         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3047                 goto out;
3048
3049         if (nvlist_add_nvlist(newconfig,
3050             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3051             nvlist_add_string(newconfig,
3052             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3053             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3054                 goto out;
3055
3056         /*
3057          * The new pool is automatically part of the namespace unless we
3058          * explicitly export it.
3059          */
3060         if (!flags.import)
3061                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3062         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3063         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3064         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
3065                 goto out;
3066         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
3067                 goto out;
3068
3069         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3070                 retval = zpool_standard_error(hdl, errno, msg);
3071                 goto out;
3072         }
3073
3074         freelist = B_FALSE;
3075         memory_err = B_FALSE;
3076
3077 out:
3078         if (varray != NULL) {
3079                 int v;
3080
3081                 for (v = 0; v < vcount; v++)
3082                         nvlist_free(varray[v]);
3083                 free(varray);
3084         }
3085         zcmd_free_nvlists(&zc);
3086         if (zc_props)
3087                 nvlist_free(zc_props);
3088         if (newconfig)
3089                 nvlist_free(newconfig);
3090         if (freelist) {
3091                 nvlist_free(*newroot);
3092                 *newroot = NULL;
3093         }
3094
3095         if (retval != 0)
3096                 return (retval);
3097
3098         if (memory_err)
3099                 return (no_memory(hdl));
3100
3101         return (0);
3102 }
3103
3104 /*
3105  * Remove the given device.  Currently, this is supported only for hot spares
3106  * and level 2 cache devices.
3107  */
3108 int
3109 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3110 {
3111         zfs_cmd_t zc = { 0 };
3112         char msg[1024];
3113         nvlist_t *tgt;
3114         boolean_t avail_spare, l2cache, islog;
3115         libzfs_handle_t *hdl = zhp->zpool_hdl;
3116         uint64_t version;
3117
3118         (void) snprintf(msg, sizeof (msg),
3119             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3120
3121         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3122         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3123             &islog)) == 0)
3124                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3125         /*
3126          * XXX - this should just go away.
3127          */
3128         if (!avail_spare && !l2cache && !islog) {
3129                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3130                     "only inactive hot spares, cache, top-level, "
3131                     "or log devices can be removed"));
3132                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3133         }
3134
3135         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3136         if (islog && version < SPA_VERSION_HOLES) {
3137                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3138                     "pool must be upgrade to support log removal"));
3139                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
3140         }
3141
3142         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
3143
3144         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3145                 return (0);
3146
3147         return (zpool_standard_error(hdl, errno, msg));
3148 }
3149
3150 /*
3151  * Clear the errors for the pool, or the particular device if specified.
3152  */
3153 int
3154 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3155 {
3156         zfs_cmd_t zc = { 0 };
3157         char msg[1024];
3158         nvlist_t *tgt;
3159         zpool_rewind_policy_t policy;
3160         boolean_t avail_spare, l2cache;
3161         libzfs_handle_t *hdl = zhp->zpool_hdl;
3162         nvlist_t *nvi = NULL;
3163         int error;
3164
3165         if (path)
3166                 (void) snprintf(msg, sizeof (msg),
3167                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3168                     path);
3169         else
3170                 (void) snprintf(msg, sizeof (msg),
3171                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3172                     zhp->zpool_name);
3173
3174         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3175         if (path) {
3176                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3177                     &l2cache, NULL)) == 0)
3178                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
3179
3180                 /*
3181                  * Don't allow error clearing for hot spares.  Do allow
3182                  * error clearing for l2cache devices.
3183                  */
3184                 if (avail_spare)
3185                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
3186
3187                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
3188                     &zc.zc_guid) == 0);
3189         }
3190
3191         zpool_get_rewind_policy(rewindnvl, &policy);
3192         zc.zc_cookie = policy.zrp_request;
3193
3194         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
3195                 return (-1);
3196
3197         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
3198                 return (-1);
3199
3200         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3201             errno == ENOMEM) {
3202                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3203                         zcmd_free_nvlists(&zc);
3204                         return (-1);
3205                 }
3206         }
3207
3208         if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
3209             errno != EPERM && errno != EACCES)) {
3210                 if (policy.zrp_request &
3211                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
3212                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3213                         zpool_rewind_exclaim(hdl, zc.zc_name,
3214                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
3215                             nvi);
3216                         nvlist_free(nvi);
3217                 }
3218                 zcmd_free_nvlists(&zc);
3219                 return (0);
3220         }
3221
3222         zcmd_free_nvlists(&zc);
3223         return (zpool_standard_error(hdl, errno, msg));
3224 }
3225
3226 /*
3227  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3228  */
3229 int
3230 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3231 {
3232         zfs_cmd_t zc = { 0 };
3233         char msg[1024];
3234         libzfs_handle_t *hdl = zhp->zpool_hdl;
3235
3236         (void) snprintf(msg, sizeof (msg),
3237             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3238             guid);
3239
3240         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3241         zc.zc_guid = guid;
3242         zc.zc_cookie = ZPOOL_NO_REWIND;
3243
3244         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
3245                 return (0);
3246
3247         return (zpool_standard_error(hdl, errno, msg));
3248 }
3249
3250 /*
3251  * Change the GUID for a pool.
3252  */
3253 int
3254 zpool_reguid(zpool_handle_t *zhp)
3255 {
3256         char msg[1024];
3257         libzfs_handle_t *hdl = zhp->zpool_hdl;
3258         zfs_cmd_t zc = { 0 };
3259
3260         (void) snprintf(msg, sizeof (msg),
3261             dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3262
3263         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3264         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3265                 return (0);
3266
3267         return (zpool_standard_error(hdl, errno, msg));
3268 }
3269
3270 /*
3271  * Reopen the pool.
3272  */
3273 int
3274 zpool_reopen(zpool_handle_t *zhp)
3275 {
3276         zfs_cmd_t zc = { 0 };
3277         char msg[1024];
3278         libzfs_handle_t *hdl = zhp->zpool_hdl;
3279
3280         (void) snprintf(msg, sizeof (msg),
3281             dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3282             zhp->zpool_name);
3283
3284         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3285         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3286                 return (0);
3287         return (zpool_standard_error(hdl, errno, msg));
3288 }
3289
3290 /*
3291  * Convert from a devid string to a path.
3292  */
3293 static char *
3294 devid_to_path(char *devid_str)
3295 {
3296         ddi_devid_t devid;
3297         char *minor;
3298         char *path;
3299         devid_nmlist_t *list = NULL;
3300         int ret;
3301
3302         if (devid_str_decode(devid_str, &devid, &minor) != 0)
3303                 return (NULL);
3304
3305         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3306
3307         devid_str_free(minor);
3308         devid_free(devid);
3309
3310         if (ret != 0)
3311                 return (NULL);
3312
3313         if ((path = strdup(list[0].devname)) == NULL)
3314                 return (NULL);
3315
3316         devid_free_nmlist(list);
3317
3318         return (path);
3319 }
3320
3321 /*
3322  * Convert from a path to a devid string.
3323  */
3324 static char *
3325 path_to_devid(const char *path)
3326 {
3327         int fd;
3328         ddi_devid_t devid;
3329         char *minor, *ret;
3330
3331         if ((fd = open(path, O_RDONLY)) < 0)
3332                 return (NULL);
3333
3334         minor = NULL;
3335         ret = NULL;
3336         if (devid_get(fd, &devid) == 0) {
3337                 if (devid_get_minor_name(fd, &minor) == 0)
3338                         ret = devid_str_encode(devid, minor);
3339                 if (minor != NULL)
3340                         devid_str_free(minor);
3341                 devid_free(devid);
3342         }
3343         (void) close(fd);
3344
3345         return (ret);
3346 }
3347
3348 /*
3349  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3350  * ignore any failure here, since a common case is for an unprivileged user to
3351  * type 'zpool status', and we'll display the correct information anyway.
3352  */
3353 static void
3354 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3355 {
3356         zfs_cmd_t zc = { 0 };
3357
3358         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3359         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3360         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3361             &zc.zc_guid) == 0);
3362
3363         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3364 }
3365
3366 /*
3367  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3368  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3369  * We also check if this is a whole disk, in which case we strip off the
3370  * trailing 's0' slice name.
3371  *
3372  * This routine is also responsible for identifying when disks have been
3373  * reconfigured in a new location.  The kernel will have opened the device by
3374  * devid, but the path will still refer to the old location.  To catch this, we
3375  * first do a path -> devid translation (which is fast for the common case).  If
3376  * the devid matches, we're done.  If not, we do a reverse devid -> path
3377  * translation and issue the appropriate ioctl() to update the path of the vdev.
3378  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3379  * of these checks.
3380  */
3381 char *
3382 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3383     boolean_t verbose)
3384 {
3385         char *path, *devid;
3386         uint64_t value;
3387         char buf[64];
3388         vdev_stat_t *vs;
3389         uint_t vsc;
3390         int have_stats;
3391         int have_path;
3392
3393         have_stats = nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3394             (uint64_t **)&vs, &vsc) == 0;
3395         have_path = nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0;
3396
3397         /*
3398          * If the device is not currently present, assume it will not
3399          * come back at the same device path.  Display the device by GUID.
3400          */
3401         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &value) == 0 ||
3402             have_path && have_stats && vs->vs_state <= VDEV_STATE_CANT_OPEN) {
3403                 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3404                     &value) == 0);
3405                 (void) snprintf(buf, sizeof (buf), "%llu",
3406                     (u_longlong_t)value);
3407                 path = buf;
3408         } else if (have_path) {
3409
3410                 /*
3411                  * If the device is dead (faulted, offline, etc) then don't
3412                  * bother opening it.  Otherwise we may be forcing the user to
3413                  * open a misbehaving device, which can have undesirable
3414                  * effects.
3415                  */
3416                 if ((have_stats == 0 ||
3417                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3418                     zhp != NULL &&
3419                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3420                         /*
3421                          * Determine if the current path is correct.
3422                          */
3423                         char *newdevid = path_to_devid(path);
3424
3425                         if (newdevid == NULL ||
3426                             strcmp(devid, newdevid) != 0) {
3427                                 char *newpath;
3428
3429                                 if ((newpath = devid_to_path(devid)) != NULL) {
3430                                         /*
3431                                          * Update the path appropriately.
3432                                          */
3433                                         set_path(zhp, nv, newpath);
3434                                         if (nvlist_add_string(nv,
3435                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3436                                                 verify(nvlist_lookup_string(nv,
3437                                                     ZPOOL_CONFIG_PATH,
3438                                                     &path) == 0);
3439                                         free(newpath);
3440                                 }
3441                         }
3442
3443                         if (newdevid)
3444                                 devid_str_free(newdevid);
3445                 }
3446
3447 #ifdef sun
3448                 if (strncmp(path, "/dev/dsk/", 9) == 0)
3449                         path += 9;
3450
3451                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
3452                     &value) == 0 && value) {
3453                         int pathlen = strlen(path);
3454                         char *tmp = zfs_strdup(hdl, path);
3455
3456                         /*
3457                          * If it starts with c#, and ends with "s0", chop
3458                          * the "s0" off, or if it ends with "s0/old", remove
3459                          * the "s0" from the middle.
3460                          */
3461                         if (CTD_CHECK(tmp)) {
3462                                 if (strcmp(&tmp[pathlen - 2], "s0") == 0) {
3463                                         tmp[pathlen - 2] = '\0';
3464                                 } else if (pathlen > 6 &&
3465                                     strcmp(&tmp[pathlen - 6], "s0/old") == 0) {
3466                                         (void) strcpy(&tmp[pathlen - 6],
3467                                             "/old");
3468                                 }
3469                         }
3470                         return (tmp);
3471                 }
3472 #else   /* !sun */
3473                 if (strncmp(path, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
3474                         path += sizeof(_PATH_DEV) - 1;
3475 #endif  /* !sun */
3476         } else {
3477                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3478
3479                 /*
3480                  * If it's a raidz device, we need to stick in the parity level.
3481                  */
3482                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3483                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3484                             &value) == 0);
3485                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3486                             (u_longlong_t)value);
3487                         path = buf;
3488                 }
3489
3490                 /*
3491                  * We identify each top-level vdev by using a <type-id>
3492                  * naming convention.
3493                  */
3494                 if (verbose) {
3495                         uint64_t id;
3496
3497                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3498                             &id) == 0);
3499                         (void) snprintf(buf, sizeof (buf), "%s-%llu", path,
3500                             (u_longlong_t)id);
3501                         path = buf;
3502                 }
3503         }
3504
3505         return (zfs_strdup(hdl, path));
3506 }
3507
3508 static int
3509 zbookmark_compare(const void *a, const void *b)
3510 {
3511         return (memcmp(a, b, sizeof (zbookmark_t)));
3512 }
3513
3514 /*
3515  * Retrieve the persistent error log, uniquify the members, and return to the
3516  * caller.
3517  */
3518 int
3519 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3520 {
3521         zfs_cmd_t zc = { 0 };
3522         uint64_t count;
3523         zbookmark_t *zb = NULL;
3524         int i;
3525
3526         /*
3527          * Retrieve the raw error list from the kernel.  If the number of errors
3528          * has increased, allocate more space and continue until we get the
3529          * entire list.
3530          */
3531         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3532             &count) == 0);
3533         if (count == 0)
3534                 return (0);
3535         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3536             count * sizeof (zbookmark_t))) == (uintptr_t)NULL)
3537                 return (-1);
3538         zc.zc_nvlist_dst_size = count;
3539         (void) strcpy(zc.zc_name, zhp->zpool_name);
3540         for (;;) {
3541                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3542                     &zc) != 0) {
3543                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3544                         if (errno == ENOMEM) {
3545                                 count = zc.zc_nvlist_dst_size;
3546                                 if ((zc.zc_nvlist_dst = (uintptr_t)
3547                                     zfs_alloc(zhp->zpool_hdl, count *
3548                                     sizeof (zbookmark_t))) == (uintptr_t)NULL)
3549                                         return (-1);
3550                         } else {
3551                                 return (-1);
3552                         }
3553                 } else {
3554                         break;
3555                 }
3556         }
3557
3558         /*
3559          * Sort the resulting bookmarks.  This is a little confusing due to the
3560          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3561          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3562          * _not_ copied as part of the process.  So we point the start of our
3563          * array appropriate and decrement the total number of elements.
3564          */
3565         zb = ((zbookmark_t *)(uintptr_t)zc.zc_nvlist_dst) +
3566             zc.zc_nvlist_dst_size;
3567         count -= zc.zc_nvlist_dst_size;
3568
3569         qsort(zb, count, sizeof (zbookmark_t), zbookmark_compare);
3570
3571         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3572
3573         /*
3574          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3575          */
3576         for (i = 0; i < count; i++) {
3577                 nvlist_t *nv;
3578
3579                 /* ignoring zb_blkid and zb_level for now */
3580                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3581                     zb[i-1].zb_object == zb[i].zb_object)
3582                         continue;
3583
3584                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3585                         goto nomem;
3586                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3587                     zb[i].zb_objset) != 0) {
3588                         nvlist_free(nv);
3589                         goto nomem;
3590                 }
3591                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3592                     zb[i].zb_object) != 0) {
3593                         nvlist_free(nv);
3594                         goto nomem;
3595                 }
3596                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3597                         nvlist_free(nv);
3598                         goto nomem;
3599                 }
3600                 nvlist_free(nv);
3601         }
3602
3603         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3604         return (0);
3605
3606 nomem:
3607         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3608         return (no_memory(zhp->zpool_hdl));
3609 }
3610
3611 /*
3612  * Upgrade a ZFS pool to the latest on-disk version.
3613  */
3614 int
3615 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3616 {
3617         zfs_cmd_t zc = { 0 };
3618         libzfs_handle_t *hdl = zhp->zpool_hdl;
3619
3620         (void) strcpy(zc.zc_name, zhp->zpool_name);
3621         zc.zc_cookie = new_version;
3622
3623         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3624                 return (zpool_standard_error_fmt(hdl, errno,
3625                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3626                     zhp->zpool_name));
3627         return (0);
3628 }
3629
3630 void
3631 zfs_save_arguments(int argc, char **argv, char *string, int len)
3632 {
3633         (void) strlcpy(string, basename(argv[0]), len);
3634         for (int i = 1; i < argc; i++) {
3635                 (void) strlcat(string, " ", len);
3636                 (void) strlcat(string, argv[i], len);
3637         }
3638 }
3639
3640 int
3641 zpool_log_history(libzfs_handle_t *hdl, const char *message)
3642 {
3643         zfs_cmd_t zc = { 0 };
3644         nvlist_t *args;
3645         int err;
3646
3647         args = fnvlist_alloc();
3648         fnvlist_add_string(args, "message", message);
3649         err = zcmd_write_src_nvlist(hdl, &zc, args);
3650         if (err == 0)
3651                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_LOG_HISTORY, &zc);
3652         nvlist_free(args);
3653         zcmd_free_nvlists(&zc);
3654         return (err);
3655 }
3656
3657 /*
3658  * Perform ioctl to get some command history of a pool.
3659  *
3660  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3661  * logical offset of the history buffer to start reading from.
3662  *
3663  * Upon return, 'off' is the next logical offset to read from and
3664  * 'len' is the actual amount of bytes read into 'buf'.
3665  */
3666 static int
3667 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3668 {
3669         zfs_cmd_t zc = { 0 };
3670         libzfs_handle_t *hdl = zhp->zpool_hdl;
3671
3672         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3673
3674         zc.zc_history = (uint64_t)(uintptr_t)buf;
3675         zc.zc_history_len = *len;
3676         zc.zc_history_offset = *off;
3677
3678         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3679                 switch (errno) {
3680                 case EPERM:
3681                         return (zfs_error_fmt(hdl, EZFS_PERM,
3682                             dgettext(TEXT_DOMAIN,
3683                             "cannot show history for pool '%s'"),
3684                             zhp->zpool_name));
3685                 case ENOENT:
3686                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3687                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3688                             "'%s'"), zhp->zpool_name));
3689                 case ENOTSUP:
3690                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3691                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3692                             "'%s', pool must be upgraded"), zhp->zpool_name));
3693                 default:
3694                         return (zpool_standard_error_fmt(hdl, errno,
3695                             dgettext(TEXT_DOMAIN,
3696                             "cannot get history for '%s'"), zhp->zpool_name));
3697                 }
3698         }
3699
3700         *len = zc.zc_history_len;
3701         *off = zc.zc_history_offset;
3702
3703         return (0);
3704 }
3705
3706 /*
3707  * Process the buffer of nvlists, unpacking and storing each nvlist record
3708  * into 'records'.  'leftover' is set to the number of bytes that weren't
3709  * processed as there wasn't a complete record.
3710  */
3711 int
3712 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3713     nvlist_t ***records, uint_t *numrecords)
3714 {
3715         uint64_t reclen;
3716         nvlist_t *nv;
3717         int i;
3718
3719         while (bytes_read > sizeof (reclen)) {
3720
3721                 /* get length of packed record (stored as little endian) */
3722                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3723                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3724
3725                 if (bytes_read < sizeof (reclen) + reclen)
3726                         break;
3727
3728                 /* unpack record */
3729                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3730                         return (ENOMEM);
3731                 bytes_read -= sizeof (reclen) + reclen;
3732                 buf += sizeof (reclen) + reclen;
3733
3734                 /* add record to nvlist array */
3735                 (*numrecords)++;
3736                 if (ISP2(*numrecords + 1)) {
3737                         *records = realloc(*records,
3738                             *numrecords * 2 * sizeof (nvlist_t *));
3739                 }
3740                 (*records)[*numrecords - 1] = nv;
3741         }
3742
3743         *leftover = bytes_read;
3744         return (0);
3745 }
3746
3747 /* from spa_history.c: spa_history_create_obj() */
3748 #define HIS_BUF_LEN_DEF (128 << 10)
3749 #define HIS_BUF_LEN_MAX (1 << 30)
3750
3751 /*
3752  * Retrieve the command history of a pool.
3753  */
3754 int
3755 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3756 {
3757         char *buf = NULL;
3758         uint64_t bufsize = HIS_BUF_LEN_DEF;
3759         uint64_t off = 0;
3760         nvlist_t **records = NULL;
3761         uint_t numrecords = 0;
3762         int err, i;
3763
3764         if ((buf = malloc(bufsize)) == NULL)
3765                 return (ENOMEM);
3766         do {
3767                 uint64_t bytes_read = bufsize;
3768                 uint64_t leftover;
3769
3770                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3771                         break;
3772
3773                 /* if nothing else was read in, we're at EOF, just return */
3774                 if (bytes_read == 0)
3775                         break;
3776
3777                 if ((err = zpool_history_unpack(buf, bytes_read,
3778                     &leftover, &records, &numrecords)) != 0)
3779                         break;
3780                 off -= leftover;
3781
3782                 /*
3783                  * If the history block is too big, double the buffer
3784                  * size and try again.
3785                  */
3786                 if (leftover == bytes_read) {
3787                         free(buf);
3788                         buf = NULL;
3789
3790                         bufsize <<= 1;
3791                         if ((bufsize >= HIS_BUF_LEN_MAX) ||
3792                             ((buf = malloc(bufsize)) == NULL)) {
3793                                 err = ENOMEM;
3794                                 break;
3795                         }
3796                 }
3797
3798                 /* CONSTCOND */
3799         } while (1);
3800         free(buf);
3801
3802         if (!err) {
3803                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3804                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3805                     records, numrecords) == 0);
3806         }
3807         for (i = 0; i < numrecords; i++)
3808                 nvlist_free(records[i]);
3809         free(records);
3810
3811         return (err);
3812 }
3813
3814 void
3815 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
3816     char *pathname, size_t len)
3817 {
3818         zfs_cmd_t zc = { 0 };
3819         boolean_t mounted = B_FALSE;
3820         char *mntpnt = NULL;
3821         char dsname[MAXNAMELEN];
3822
3823         if (dsobj == 0) {
3824                 /* special case for the MOS */
3825                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>", obj);
3826                 return;
3827         }
3828
3829         /* get the dataset's name */
3830         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3831         zc.zc_obj = dsobj;
3832         if (ioctl(zhp->zpool_hdl->libzfs_fd,
3833             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
3834                 /* just write out a path of two object numbers */
3835                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
3836                     dsobj, obj);
3837                 return;
3838         }
3839         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
3840
3841         /* find out if the dataset is mounted */
3842         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
3843
3844         /* get the corrupted object's path */
3845         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
3846         zc.zc_obj = obj;
3847         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
3848             &zc) == 0) {
3849                 if (mounted) {
3850                         (void) snprintf(pathname, len, "%s%s", mntpnt,
3851                             zc.zc_value);
3852                 } else {
3853                         (void) snprintf(pathname, len, "%s:%s",
3854                             dsname, zc.zc_value);
3855                 }
3856         } else {
3857                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname, obj);
3858         }
3859         free(mntpnt);
3860 }
3861
3862 #ifdef sun
3863 /*
3864  * Read the EFI label from the config, if a label does not exist then
3865  * pass back the error to the caller. If the caller has passed a non-NULL
3866  * diskaddr argument then we set it to the starting address of the EFI
3867  * partition.
3868  */
3869 static int
3870 read_efi_label(nvlist_t *config, diskaddr_t *sb)
3871 {
3872         char *path;
3873         int fd;
3874         char diskname[MAXPATHLEN];
3875         int err = -1;
3876
3877         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
3878                 return (err);
3879
3880         (void) snprintf(diskname, sizeof (diskname), "%s%s", RDISK_ROOT,
3881             strrchr(path, '/'));
3882         if ((fd = open(diskname, O_RDONLY|O_NDELAY)) >= 0) {
3883                 struct dk_gpt *vtoc;
3884
3885                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
3886                         if (sb != NULL)
3887                                 *sb = vtoc->efi_parts[0].p_start;
3888                         efi_free(vtoc);
3889                 }
3890                 (void) close(fd);
3891         }
3892         return (err);
3893 }
3894
3895 /*
3896  * determine where a partition starts on a disk in the current
3897  * configuration
3898  */
3899 static diskaddr_t
3900 find_start_block(nvlist_t *config)
3901 {
3902         nvlist_t **child;
3903         uint_t c, children;
3904         diskaddr_t sb = MAXOFFSET_T;
3905         uint64_t wholedisk;
3906
3907         if (nvlist_lookup_nvlist_array(config,
3908             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
3909                 if (nvlist_lookup_uint64(config,
3910                     ZPOOL_CONFIG_WHOLE_DISK,
3911                     &wholedisk) != 0 || !wholedisk) {
3912                         return (MAXOFFSET_T);
3913                 }
3914                 if (read_efi_label(config, &sb) < 0)
3915                         sb = MAXOFFSET_T;
3916                 return (sb);
3917         }
3918
3919         for (c = 0; c < children; c++) {
3920                 sb = find_start_block(child[c]);
3921                 if (sb != MAXOFFSET_T) {
3922                         return (sb);
3923                 }
3924         }
3925         return (MAXOFFSET_T);
3926 }
3927 #endif /* sun */
3928
3929 /*
3930  * Label an individual disk.  The name provided is the short name,
3931  * stripped of any leading /dev path.
3932  */
3933 int
3934 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, const char *name)
3935 {
3936 #ifdef sun
3937         char path[MAXPATHLEN];
3938         struct dk_gpt *vtoc;
3939         int fd;
3940         size_t resv = EFI_MIN_RESV_SIZE;
3941         uint64_t slice_size;
3942         diskaddr_t start_block;
3943         char errbuf[1024];
3944
3945         /* prepare an error message just in case */
3946         (void) snprintf(errbuf, sizeof (errbuf),
3947             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
3948
3949         if (zhp) {
3950                 nvlist_t *nvroot;
3951
3952                 if (zpool_is_bootable(zhp)) {
3953                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3954                             "EFI labeled devices are not supported on root "
3955                             "pools."));
3956                         return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
3957                 }
3958
3959                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
3960                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3961
3962                 if (zhp->zpool_start_block == 0)
3963                         start_block = find_start_block(nvroot);
3964                 else
3965                         start_block = zhp->zpool_start_block;
3966                 zhp->zpool_start_block = start_block;
3967         } else {
3968                 /* new pool */
3969                 start_block = NEW_START_BLOCK;
3970         }
3971
3972         (void) snprintf(path, sizeof (path), "%s/%s%s", RDISK_ROOT, name,
3973             BACKUP_SLICE);
3974
3975         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
3976                 /*
3977                  * This shouldn't happen.  We've long since verified that this
3978                  * is a valid device.
3979                  */
3980                 zfs_error_aux(hdl,
3981                     dgettext(TEXT_DOMAIN, "unable to open device"));
3982                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
3983         }
3984
3985         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
3986                 /*
3987                  * The only way this can fail is if we run out of memory, or we
3988                  * were unable to read the disk's capacity
3989                  */
3990                 if (errno == ENOMEM)
3991                         (void) no_memory(hdl);
3992
3993                 (void) close(fd);
3994                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3995                     "unable to read disk capacity"), name);
3996
3997                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
3998         }
3999
4000         slice_size = vtoc->efi_last_u_lba + 1;
4001         slice_size -= EFI_MIN_RESV_SIZE;
4002         if (start_block == MAXOFFSET_T)
4003                 start_block = NEW_START_BLOCK;
4004         slice_size -= start_block;
4005
4006         vtoc->efi_parts[0].p_start = start_block;
4007         vtoc->efi_parts[0].p_size = slice_size;
4008
4009         /*
4010          * Why we use V_USR: V_BACKUP confuses users, and is considered
4011          * disposable by some EFI utilities (since EFI doesn't have a backup
4012          * slice).  V_UNASSIGNED is supposed to be used only for zero size
4013          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
4014          * etc. were all pretty specific.  V_USR is as close to reality as we
4015          * can get, in the absence of V_OTHER.
4016          */
4017         vtoc->efi_parts[0].p_tag = V_USR;
4018         (void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
4019
4020         vtoc->efi_parts[8].p_start = slice_size + start_block;
4021         vtoc->efi_parts[8].p_size = resv;
4022         vtoc->efi_parts[8].p_tag = V_RESERVED;
4023
4024         if (efi_write(fd, vtoc) != 0) {
4025                 /*
4026                  * Some block drivers (like pcata) may not support EFI
4027                  * GPT labels.  Print out a helpful error message dir-
4028                  * ecting the user to manually label the disk and give
4029                  * a specific slice.
4030                  */
4031                 (void) close(fd);
4032                 efi_free(vtoc);
4033
4034                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4035                     "try using fdisk(1M) and then provide a specific slice"));
4036                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4037         }
4038
4039         (void) close(fd);
4040         efi_free(vtoc);
4041 #endif /* sun */
4042         return (0);
4043 }
4044
4045 static boolean_t
4046 supported_dump_vdev_type(libzfs_handle_t *hdl, nvlist_t *config, char *errbuf)
4047 {
4048         char *type;
4049         nvlist_t **child;
4050         uint_t children, c;
4051
4052         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_TYPE, &type) == 0);
4053         if (strcmp(type, VDEV_TYPE_FILE) == 0 ||
4054             strcmp(type, VDEV_TYPE_HOLE) == 0 ||
4055             strcmp(type, VDEV_TYPE_MISSING) == 0) {
4056                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4057                     "vdev type '%s' is not supported"), type);
4058                 (void) zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf);
4059                 return (B_FALSE);
4060         }
4061         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
4062             &child, &children) == 0) {
4063                 for (c = 0; c < children; c++) {
4064                         if (!supported_dump_vdev_type(hdl, child[c], errbuf))
4065                                 return (B_FALSE);
4066                 }
4067         }
4068         return (B_TRUE);
4069 }
4070
4071 /*
4072  * Check if this zvol is allowable for use as a dump device; zero if
4073  * it is, > 0 if it isn't, < 0 if it isn't a zvol.
4074  *
4075  * Allowable storage configurations include mirrors, all raidz variants, and
4076  * pools with log, cache, and spare devices.  Pools which are backed by files or
4077  * have missing/hole vdevs are not suitable.
4078  */
4079 int
4080 zvol_check_dump_config(char *arg)
4081 {
4082         zpool_handle_t *zhp = NULL;
4083         nvlist_t *config, *nvroot;
4084         char *p, *volname;
4085         nvlist_t **top;
4086         uint_t toplevels;
4087         libzfs_handle_t *hdl;
4088         char errbuf[1024];
4089         char poolname[ZPOOL_MAXNAMELEN];
4090         int pathlen = strlen(ZVOL_FULL_DEV_DIR);
4091         int ret = 1;
4092
4093         if (strncmp(arg, ZVOL_FULL_DEV_DIR, pathlen)) {
4094                 return (-1);
4095         }
4096
4097         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
4098             "dump is not supported on device '%s'"), arg);
4099
4100         if ((hdl = libzfs_init()) == NULL)
4101                 return (1);
4102         libzfs_print_on_error(hdl, B_TRUE);
4103
4104         volname = arg + pathlen;
4105
4106         /* check the configuration of the pool */
4107         if ((p = strchr(volname, '/')) == NULL) {
4108                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4109                     "malformed dataset name"));
4110                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4111                 return (1);
4112         } else if (p - volname >= ZFS_MAXNAMELEN) {
4113                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4114                     "dataset name is too long"));
4115                 (void) zfs_error(hdl, EZFS_NAMETOOLONG, errbuf);
4116                 return (1);
4117         } else {
4118                 (void) strncpy(poolname, volname, p - volname);
4119                 poolname[p - volname] = '\0';
4120         }
4121
4122         if ((zhp = zpool_open(hdl, poolname)) == NULL) {
4123                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4124                     "could not open pool '%s'"), poolname);
4125                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
4126                 goto out;
4127         }
4128         config = zpool_get_config(zhp, NULL);
4129         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4130             &nvroot) != 0) {
4131                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4132                     "could not obtain vdev configuration for  '%s'"), poolname);
4133                 (void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
4134                 goto out;
4135         }
4136
4137         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4138             &top, &toplevels) == 0);
4139
4140         if (!supported_dump_vdev_type(hdl, top[0], errbuf)) {
4141                 goto out;
4142         }
4143         ret = 0;
4144
4145 out:
4146         if (zhp)
4147                 zpool_close(zhp);
4148         libzfs_fini(hdl);
4149         return (ret);
4150 }