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

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  25  * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
  26  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  27  */
  28
  29 #include <sys/types.h>
  30 #include <sys/stat.h>
  31 #include <ctype.h>
  32 #include <errno.h>
  33 #include <devid.h>
  34 #include <fcntl.h>
  35 #include <libintl.h>
  36 #include <stdio.h>
  37 #include <stdlib.h>
  38 #include <strings.h>
  39 #include <unistd.h>
  40 #include <libgen.h>
  41 #include <sys/zfs_ioctl.h>
  42 #include <dlfcn.h>
  43
  44 #include "zfs_namecheck.h"
  45 #include "zfs_prop.h"
  46 #include "libzfs_impl.h"
  47 #include "zfs_comutil.h"
  48 #include "zfeature_common.h"
  49
  50 static int read_efi_label(nvlist_t *config, diskaddr_t *sb);
  51
  52 #define DISK_ROOT       "/dev/dsk"
  53 #define RDISK_ROOT      "/dev/rdsk"
  54 #define BACKUP_SLICE    "s2"
  55
  56 typedef struct prop_flags {
  57         int create:1;   /* Validate property on creation */
  58         int import:1;   /* Validate property on import */
  59 } prop_flags_t;
  60
  61 /*
  62  * ====================================================================
  63  *   zpool property functions
  64  * ====================================================================
  65  */
  66
  67 static int
  68 zpool_get_all_props(zpool_handle_t *zhp)
  69 {
  70         zfs_cmd_t zc = { 0 };
  71         libzfs_handle_t *hdl = zhp->zpool_hdl;
  72
  73         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
  74
  75         if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
  76                 return (-1);
  77
  78         while (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
  79                 if (errno == ENOMEM) {
  80                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
  81                                 zcmd_free_nvlists(&zc);
  82                                 return (-1);
  83                         }
  84                 } else {
  85                         zcmd_free_nvlists(&zc);
  86                         return (-1);
  87                 }
  88         }
  89
  90         if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
  91                 zcmd_free_nvlists(&zc);
  92                 return (-1);
  93         }
  94
  95         zcmd_free_nvlists(&zc);
  96
  97         return (0);
  98 }
  99
 100 static int
 101 zpool_props_refresh(zpool_handle_t *zhp)
 102 {
 103         nvlist_t *old_props;
 104
 105         old_props = zhp->zpool_props;
 106
 107         if (zpool_get_all_props(zhp) != 0)
 108                 return (-1);
 109
 110         nvlist_free(old_props);
 111         return (0);
 112 }
 113
 114 static char *
 115 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
 116     zprop_source_t *src)
 117 {
 118         nvlist_t *nv, *nvl;
 119         uint64_t ival;
 120         char *value;
 121         zprop_source_t source;
 122
 123         nvl = zhp->zpool_props;
 124         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 125                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &ival) == 0);
 126                 source = ival;
 127                 verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
 128         } else {
 129                 source = ZPROP_SRC_DEFAULT;
 130                 if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
 131                         value = "-";
 132         }
 133
 134         if (src)
 135                 *src = source;
 136
 137         return (value);
 138 }
 139
 140 uint64_t
 141 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
 142 {
 143         nvlist_t *nv, *nvl;
 144         uint64_t value;
 145         zprop_source_t source;
 146
 147         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
 148                 /*
 149                  * zpool_get_all_props() has most likely failed because
 150                  * the pool is faulted, but if all we need is the top level
 151                  * vdev's guid then get it from the zhp config nvlist.
 152                  */
 153                 if ((prop == ZPOOL_PROP_GUID) &&
 154                     (nvlist_lookup_nvlist(zhp->zpool_config,
 155                     ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
 156                     (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
 157                     == 0)) {
 158                         return (value);
 159                 }
 160                 return (zpool_prop_default_numeric(prop));
 161         }
 162
 163         nvl = zhp->zpool_props;
 164         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 165                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &value) == 0);
 166                 source = value;
 167                 verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
 168         } else {
 169                 source = ZPROP_SRC_DEFAULT;
 170                 value = zpool_prop_default_numeric(prop);
 171         }
 172
 173         if (src)
 174                 *src = source;
 175
 176         return (value);
 177 }
 178
 179 /*
 180  * Map VDEV STATE to printed strings.
 181  */
 182 const char *
 183 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
 184 {
 185         switch (state) {
 186         case VDEV_STATE_CLOSED:
 187         case VDEV_STATE_OFFLINE:
 188                 return (gettext("OFFLINE"));
 189         case VDEV_STATE_REMOVED:
 190                 return (gettext("REMOVED"));
 191         case VDEV_STATE_CANT_OPEN:
 192                 if (aux == VDEV_AUX_CORRUPT_DATA || aux == VDEV_AUX_BAD_LOG)
 193                         return (gettext("FAULTED"));
 194                 else if (aux == VDEV_AUX_SPLIT_POOL)
 195                         return (gettext("SPLIT"));
 196                 else
 197                         return (gettext("UNAVAIL"));
 198         case VDEV_STATE_FAULTED:
 199                 return (gettext("FAULTED"));
 200         case VDEV_STATE_DEGRADED:
 201                 return (gettext("DEGRADED"));
 202         case VDEV_STATE_HEALTHY:
 203                 return (gettext("ONLINE"));
 204         }
 205
 206         return (gettext("UNKNOWN"));
 207 }
 208
 209 /*
 210  * Map POOL STATE to printed strings.
 211  */
 212 const char *
 213 zpool_pool_state_to_name(pool_state_t state)
 214 {
 215         switch (state) {
 216         case POOL_STATE_ACTIVE:
 217                 return (gettext("ACTIVE"));
 218         case POOL_STATE_EXPORTED:
 219                 return (gettext("EXPORTED"));
 220         case POOL_STATE_DESTROYED:
 221                 return (gettext("DESTROYED"));
 222         case POOL_STATE_SPARE:
 223                 return (gettext("SPARE"));
 224         case POOL_STATE_L2CACHE:
 225                 return (gettext("L2CACHE"));
 226         case POOL_STATE_UNINITIALIZED:
 227                 return (gettext("UNINITIALIZED"));
 228         case POOL_STATE_UNAVAIL:
 229                 return (gettext("UNAVAIL"));
 230         case POOL_STATE_POTENTIALLY_ACTIVE:
 231                 return (gettext("POTENTIALLY_ACTIVE"));
 232         }
 233
 234         return (gettext("UNKNOWN"));
 235 }
 236
 237 /*
 238  * Get a zpool property value for 'prop' and return the value in
 239  * a pre-allocated buffer.
 240  */
 241 int
 242 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf, size_t len,
 243     zprop_source_t *srctype, boolean_t literal)
 244 {
 245         uint64_t intval;
 246         const char *strval;
 247         zprop_source_t src = ZPROP_SRC_NONE;
 248         nvlist_t *nvroot;
 249         vdev_stat_t *vs;
 250         uint_t vsc;
 251
 252         if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
 253                 switch (prop) {
 254                 case ZPOOL_PROP_NAME:
 255                         (void) strlcpy(buf, zpool_get_name(zhp), len);
 256                         break;
 257
 258                 case ZPOOL_PROP_HEALTH:
 259                         (void) strlcpy(buf, "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_LEAKED:
 307                 case ZPOOL_PROP_EXPANDSZ:
 308                         if (literal) {
 309                                 (void) snprintf(buf, len, "%llu",
 310                                     (u_longlong_t)intval);
 311                         } else {
 312                                 (void) zfs_nicenum(intval, buf, len);
 313                         }
 314                         break;
 315
 316                 case ZPOOL_PROP_CAPACITY:
 317                         if (literal) {
 318                                 (void) snprintf(buf, len, "%llu",
 319                                     (u_longlong_t)intval);
 320                         } else {
 321                                 (void) snprintf(buf, len, "%llu%%",
 322                                     (u_longlong_t)intval);
 323                         }
 324                         break;
 325                 case ZPOOL_PROP_FRAGMENTATION:
 326                         if (intval == UINT64_MAX) {
 327                                 (void) strlcpy(buf, "-", len);
 328                         } else {
 329                                 (void) snprintf(buf, len, "%llu%%",
 330                                     (u_longlong_t)intval);
 331                         }
 332                         break;
 333
 334                 case ZPOOL_PROP_DEDUPRATIO:
 335                         (void) snprintf(buf, len, "%llu.%02llux",
 336                             (u_longlong_t)(intval / 100),
 337                             (u_longlong_t)(intval % 100));
 338                         break;
 339
 340                 case ZPOOL_PROP_HEALTH:
 341                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 342                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 343                         verify(nvlist_lookup_uint64_array(nvroot,
 344                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
 345                             == 0);
 346
 347                         (void) strlcpy(buf, zpool_state_to_name(intval,
 348                             vs->vs_aux), len);
 349                         break;
 350                 case ZPOOL_PROP_VERSION:
 351                         if (intval >= SPA_VERSION_FEATURES) {
 352                                 (void) snprintf(buf, len, "-");
 353                                 break;
 354                         }
 355                         /* FALLTHROUGH */
 356                 default:
 357                         (void) snprintf(buf, len, "%llu", intval);
 358                 }
 359                 break;
 360
 361         case PROP_TYPE_INDEX:
 362                 intval = zpool_get_prop_int(zhp, prop, &src);
 363                 if (zpool_prop_index_to_string(prop, intval, &strval)
 364                     != 0)
 365                         return (-1);
 366                 (void) strlcpy(buf, strval, len);
 367                 break;
 368
 369         default:
 370                 abort();
 371         }
 372
 373         if (srctype)
 374                 *srctype = src;
 375
 376         return (0);
 377 }
 378
 379 /*
 380  * Check if the bootfs name has the same pool name as it is set to.
 381  * Assuming bootfs is a valid dataset name.
 382  */
 383 static boolean_t
 384 bootfs_name_valid(const char *pool, char *bootfs)
 385 {
 386         int len = strlen(pool);
 387
 388         if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
 389                 return (B_FALSE);
 390
 391         if (strncmp(pool, bootfs, len) == 0 &&
 392             (bootfs[len] == '/' || bootfs[len] == '\0'))
 393                 return (B_TRUE);
 394
 395         return (B_FALSE);
 396 }
 397
 398 /*
 399  * Inspect the configuration to determine if any of the devices contain
 400  * an EFI label.
 401  */
 402 static boolean_t
 403 pool_uses_efi(nvlist_t *config)
 404 {
 405 #ifdef sun
 406         nvlist_t **child;
 407         uint_t c, children;
 408
 409         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
 410             &child, &children) != 0)
 411                 return (read_efi_label(config, NULL) >= 0);
 412
 413         for (c = 0; c < children; c++) {
 414                 if (pool_uses_efi(child[c]))
 415                         return (B_TRUE);
 416         }
 417 #endif  /* sun */
 418         return (B_FALSE);
 419 }
 420
 421 boolean_t
 422 zpool_is_bootable(zpool_handle_t *zhp)
 423 {
 424         char bootfs[ZPOOL_MAXNAMELEN];
 425
 426         return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
 427             sizeof (bootfs), NULL, B_FALSE) == 0 && strncmp(bootfs, "-",
 428             sizeof (bootfs)) != 0);
 429 }
 430
 431
 432 /*
 433  * Given an nvlist of zpool properties to be set, validate that they are
 434  * correct, and parse any numeric properties (index, boolean, etc) if they are
 435  * specified as strings.
 436  */
 437 static nvlist_t *
 438 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 439     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
 440 {
 441         nvpair_t *elem;
 442         nvlist_t *retprops;
 443         zpool_prop_t prop;
 444         char *strval;
 445         uint64_t intval;
 446         char *slash, *check;
 447         struct stat64 statbuf;
 448         zpool_handle_t *zhp;
 449         nvlist_t *nvroot;
 450
 451         if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
 452                 (void) no_memory(hdl);
 453                 return (NULL);
 454         }
 455
 456         elem = NULL;
 457         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 458                 const char *propname = nvpair_name(elem);
 459
 460                 prop = zpool_name_to_prop(propname);
 461                 if (prop == ZPROP_INVAL && zpool_prop_feature(propname)) {
 462                         int err;
 463                         char *fname = strchr(propname, '@') + 1;
 464
 465                         err = zfeature_lookup_name(fname, NULL);
 466                         if (err != 0) {
 467                                 ASSERT3U(err, ==, ENOENT);
 468                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 469                                     "invalid feature '%s'"), fname);
 470                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 471                                 goto error;
 472                         }
 473
 474                         if (nvpair_type(elem) != DATA_TYPE_STRING) {
 475                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 476                                     "'%s' must be a string"), propname);
 477                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 478                                 goto error;
 479                         }
 480
 481                         (void) nvpair_value_string(elem, &strval);
 482                         if (strcmp(strval, ZFS_FEATURE_ENABLED) != 0) {
 483                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 484                                     "property '%s' can only be set to "
 485                                     "'enabled'"), propname);
 486                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 487                                 goto error;
 488                         }
 489
 490                         if (nvlist_add_uint64(retprops, propname, 0) != 0) {
 491                                 (void) no_memory(hdl);
 492                                 goto error;
 493                         }
 494                         continue;
 495                 }
 496
 497                 /*
 498                  * Make sure this property is valid and applies to this type.
 499                  */
 500                 if (prop == ZPROP_INVAL) {
 501                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 502                             "invalid property '%s'"), propname);
 503                         (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 504                         goto error;
 505                 }
 506
 507                 if (zpool_prop_readonly(prop)) {
 508                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 509                             "is readonly"), propname);
 510                         (void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
 511                         goto error;
 512                 }
 513
 514                 if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
 515                     &strval, &intval, errbuf) != 0)
 516                         goto error;
 517
 518                 /*
 519                  * Perform additional checking for specific properties.
 520                  */
 521                 switch (prop) {
 522                 case ZPOOL_PROP_VERSION:
 523                         if (intval < version ||
 524                             !SPA_VERSION_IS_SUPPORTED(intval)) {
 525                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 526                                     "property '%s' number %d is invalid."),
 527                                     propname, intval);
 528                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 529                                 goto error;
 530                         }
 531                         break;
 532
 533                 case ZPOOL_PROP_BOOTFS:
 534                         if (flags.create || flags.import) {
 535                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 536                                     "property '%s' cannot be set at creation "
 537                                     "or import time"), propname);
 538                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 539                                 goto error;
 540                         }
 541
 542                         if (version < SPA_VERSION_BOOTFS) {
 543                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 544                                     "pool must be upgraded to support "
 545                                     "'%s' property"), propname);
 546                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 547                                 goto error;
 548                         }
 549
 550                         /*
 551                          * bootfs property value has to be a dataset name and
 552                          * the dataset has to be in the same pool as it sets to.
 553                          */
 554                         if (strval[0] != '\0' && !bootfs_name_valid(poolname,
 555                             strval)) {
 556                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 557                                     "is an invalid name"), strval);
 558                                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
 559                                 goto error;
 560                         }
 561
 562                         if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
 563                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 564                                     "could not open pool '%s'"), poolname);
 565                                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
 566                                 goto error;
 567                         }
 568                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 569                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 570
 571 #ifdef sun
 572                         /*
 573                          * bootfs property cannot be set on a disk which has
 574                          * been EFI labeled.
 575                          */
 576                         if (pool_uses_efi(nvroot)) {
 577                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 578                                     "property '%s' not supported on "
 579                                     "EFI labeled devices"), propname);
 580                                 (void) zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf);
 581                                 zpool_close(zhp);
 582                                 goto error;
 583                         }
 584 #endif  /* sun */
 585                         zpool_close(zhp);
 586                         break;
 587
 588                 case ZPOOL_PROP_ALTROOT:
 589                         if (!flags.create && !flags.import) {
 590                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 591                                     "property '%s' can only be set during pool "
 592                                     "creation or import"), propname);
 593                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 594                                 goto error;
 595                         }
 596
 597                         if (strval[0] != '/') {
 598                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 599                                     "bad alternate root '%s'"), strval);
 600                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 601                                 goto error;
 602                         }
 603                         break;
 604
 605                 case ZPOOL_PROP_CACHEFILE:
 606                         if (strval[0] == '\0')
 607                                 break;
 608
 609                         if (strcmp(strval, "none") == 0)
 610                                 break;
 611
 612                         if (strval[0] != '/') {
 613                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 614                                     "property '%s' must be empty, an "
 615                                     "absolute path, or 'none'"), propname);
 616                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 617                                 goto error;
 618                         }
 619
 620                         slash = strrchr(strval, '/');
 621
 622                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 623                             strcmp(slash, "/..") == 0) {
 624                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 625                                     "'%s' is not a valid file"), strval);
 626                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 627                                 goto error;
 628                         }
 629
 630                         *slash = '\0';
 631
 632                         if (strval[0] != '\0' &&
 633                             (stat64(strval, &statbuf) != 0 ||
 634                             !S_ISDIR(statbuf.st_mode))) {
 635                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 636                                     "'%s' is not a valid directory"),
 637                                     strval);
 638                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 639                                 goto error;
 640                         }
 641
 642                         *slash = '/';
 643                         break;
 644
 645                 case ZPOOL_PROP_COMMENT:
 646                         for (check = strval; *check != '\0'; check++) {
 647                                 if (!isprint(*check)) {
 648                                         zfs_error_aux(hdl,
 649                                             dgettext(TEXT_DOMAIN,
 650                                             "comment may only have printable "
 651                                             "characters"));
 652                                         (void) zfs_error(hdl, EZFS_BADPROP,
 653                                             errbuf);
 654                                         goto error;
 655                                 }
 656                         }
 657                         if (strlen(strval) > ZPROP_MAX_COMMENT) {
 658                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 659                                     "comment must not exceed %d characters"),
 660                                     ZPROP_MAX_COMMENT);
 661                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 662                                 goto error;
 663                         }
 664                         break;
 665                 case ZPOOL_PROP_READONLY:
 666                         if (!flags.import) {
 667                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 668                                     "property '%s' can only be set at "
 669                                     "import time"), propname);
 670                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 671                                 goto error;
 672                         }
 673                         break;
 674                 }
 675         }
 676
 677         return (retprops);
 678 error:
 679         nvlist_free(retprops);
 680         return (NULL);
 681 }
 682
 683 /*
 684  * Set zpool property : propname=propval.
 685  */
 686 int
 687 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
 688 {
 689         zfs_cmd_t zc = { 0 };
 690         int ret = -1;
 691         char errbuf[1024];
 692         nvlist_t *nvl = NULL;
 693         nvlist_t *realprops;
 694         uint64_t version;
 695         prop_flags_t flags = { 0 };
 696
 697         (void) snprintf(errbuf, sizeof (errbuf),
 698             dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
 699             zhp->zpool_name);
 700
 701         if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
 702                 return (no_memory(zhp->zpool_hdl));
 703
 704         if (nvlist_add_string(nvl, propname, propval) != 0) {
 705                 nvlist_free(nvl);
 706                 return (no_memory(zhp->zpool_hdl));
 707         }
 708
 709         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
 710         if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
 711             zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
 712                 nvlist_free(nvl);
 713                 return (-1);
 714         }
 715
 716         nvlist_free(nvl);
 717         nvl = realprops;
 718
 719         /*
 720          * Execute the corresponding ioctl() to set this property.
 721          */
 722         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 723
 724         if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
 725                 nvlist_free(nvl);
 726                 return (-1);
 727         }
 728
 729         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
 730
 731         zcmd_free_nvlists(&zc);
 732         nvlist_free(nvl);
 733
 734         if (ret)
 735                 (void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
 736         else
 737                 (void) zpool_props_refresh(zhp);
 738
 739         return (ret);
 740 }
 741
 742 int
 743 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
 744 {
 745         libzfs_handle_t *hdl = zhp->zpool_hdl;
 746         zprop_list_t *entry;
 747         char buf[ZFS_MAXPROPLEN];
 748         nvlist_t *features = NULL;
 749         zprop_list_t **last;
 750         boolean_t firstexpand = (NULL == *plp);
 751
 752         if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
 753                 return (-1);
 754
 755         last = plp;
 756         while (*last != NULL)
 757                 last = &(*last)->pl_next;
 758
 759         if ((*plp)->pl_all)
 760                 features = zpool_get_features(zhp);
 761
 762         if ((*plp)->pl_all && firstexpand) {
 763                 for (int i = 0; i < SPA_FEATURES; i++) {
 764                         zprop_list_t *entry = zfs_alloc(hdl,
 765                             sizeof (zprop_list_t));
 766                         entry->pl_prop = ZPROP_INVAL;
 767                         entry->pl_user_prop = zfs_asprintf(hdl, "feature@%s",
 768                             spa_feature_table[i].fi_uname);
 769                         entry->pl_width = strlen(entry->pl_user_prop);
 770                         entry->pl_all = B_TRUE;
 771
 772                         *last = entry;
 773                         last = &entry->pl_next;
 774                 }
 775         }
 776
 777         /* add any unsupported features */
 778         for (nvpair_t *nvp = nvlist_next_nvpair(features, NULL);
 779             nvp != NULL; nvp = nvlist_next_nvpair(features, nvp)) {
 780                 char *propname;
 781                 boolean_t found;
 782                 zprop_list_t *entry;
 783
 784                 if (zfeature_is_supported(nvpair_name(nvp)))
 785                         continue;
 786
 787                 propname = zfs_asprintf(hdl, "unsupported@%s",
 788                     nvpair_name(nvp));
 789
 790                 /*
 791                  * Before adding the property to the list make sure that no
 792                  * other pool already added the same property.
 793                  */
 794                 found = B_FALSE;
 795                 entry = *plp;
 796                 while (entry != NULL) {
 797                         if (entry->pl_user_prop != NULL &&
 798                             strcmp(propname, entry->pl_user_prop) == 0) {
 799                                 found = B_TRUE;
 800                                 break;
 801                         }
 802                         entry = entry->pl_next;
 803                 }
 804                 if (found) {
 805                         free(propname);
 806                         continue;
 807                 }
 808
 809                 entry = zfs_alloc(hdl, sizeof (zprop_list_t));
 810                 entry->pl_prop = ZPROP_INVAL;
 811                 entry->pl_user_prop = propname;
 812                 entry->pl_width = strlen(entry->pl_user_prop);
 813                 entry->pl_all = B_TRUE;
 814
 815                 *last = entry;
 816                 last = &entry->pl_next;
 817         }
 818
 819         for (entry = *plp; entry != NULL; entry = entry->pl_next) {
 820
 821                 if (entry->pl_fixed)
 822                         continue;
 823
 824                 if (entry->pl_prop != ZPROP_INVAL &&
 825                     zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
 826                     NULL, B_FALSE) == 0) {
 827                         if (strlen(buf) > entry->pl_width)
 828                                 entry->pl_width = strlen(buf);
 829                 }
 830         }
 831
 832         return (0);
 833 }
 834
 835 /*
 836  * Get the state for the given feature on the given ZFS pool.
 837  */
 838 int
 839 zpool_prop_get_feature(zpool_handle_t *zhp, const char *propname, char *buf,
 840     size_t len)
 841 {
 842         uint64_t refcount;
 843         boolean_t found = B_FALSE;
 844         nvlist_t *features = zpool_get_features(zhp);
 845         boolean_t supported;
 846         const char *feature = strchr(propname, '@') + 1;
 847
 848         supported = zpool_prop_feature(propname);
 849         ASSERT(supported || zpool_prop_unsupported(propname));
 850
 851         /*
 852          * Convert from feature name to feature guid. This conversion is
 853          * unecessary for unsupported@... properties because they already
 854          * use guids.
 855          */
 856         if (supported) {
 857                 int ret;
 858                 spa_feature_t fid;
 859
 860                 ret = zfeature_lookup_name(feature, &fid);
 861                 if (ret != 0) {
 862                         (void) strlcpy(buf, "-", len);
 863                         return (ENOTSUP);
 864                 }
 865                 feature = spa_feature_table[fid].fi_guid;
 866         }
 867
 868         if (nvlist_lookup_uint64(features, feature, &refcount) == 0)
 869                 found = B_TRUE;
 870
 871         if (supported) {
 872                 if (!found) {
 873                         (void) strlcpy(buf, ZFS_FEATURE_DISABLED, len);
 874                 } else  {
 875                         if (refcount == 0)
 876                                 (void) strlcpy(buf, ZFS_FEATURE_ENABLED, len);
 877                         else
 878                                 (void) strlcpy(buf, ZFS_FEATURE_ACTIVE, len);
 879                 }
 880         } else {
 881                 if (found) {
 882                         if (refcount == 0) {
 883                                 (void) strcpy(buf, ZFS_UNSUPPORTED_INACTIVE);
 884                         } else {
 885                                 (void) strcpy(buf, ZFS_UNSUPPORTED_READONLY);
 886                         }
 887                 } else {
 888                         (void) strlcpy(buf, "-", len);
 889                         return (ENOTSUP);
 890                 }
 891         }
 892
 893         return (0);
 894 }
 895
 896 /*
 897  * Don't start the slice at the default block of 34; many storage
 898  * devices will use a stripe width of 128k, so start there instead.
 899  */
 900 #define NEW_START_BLOCK 256
 901
 902 /*
 903  * Validate the given pool name, optionally putting an extended error message in
 904  * 'buf'.
 905  */
 906 boolean_t
 907 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
 908 {
 909         namecheck_err_t why;
 910         char what;
 911         int ret;
 912
 913         ret = pool_namecheck(pool, &why, &what);
 914
 915         /*
 916          * The rules for reserved pool names were extended at a later point.
 917          * But we need to support users with existing pools that may now be
 918          * invalid.  So we only check for this expanded set of names during a
 919          * create (or import), and only in userland.
 920          */
 921         if (ret == 0 && !isopen &&
 922             (strncmp(pool, "mirror", 6) == 0 ||
 923             strncmp(pool, "raidz", 5) == 0 ||
 924             strncmp(pool, "spare", 5) == 0 ||
 925             strcmp(pool, "log") == 0)) {
 926                 if (hdl != NULL)
 927                         zfs_error_aux(hdl,
 928                             dgettext(TEXT_DOMAIN, "name is reserved"));
 929                 return (B_FALSE);
 930         }
 931
 932
 933         if (ret != 0) {
 934                 if (hdl != NULL) {
 935                         switch (why) {
 936                         case NAME_ERR_TOOLONG:
 937                                 zfs_error_aux(hdl,
 938                                     dgettext(TEXT_DOMAIN, "name is too long"));
 939                                 break;
 940
 941                         case NAME_ERR_INVALCHAR:
 942                                 zfs_error_aux(hdl,
 943                                     dgettext(TEXT_DOMAIN, "invalid character "
 944                                     "'%c' in pool name"), what);
 945                                 break;
 946
 947                         case NAME_ERR_NOLETTER:
 948                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 949                                     "name must begin with a letter"));
 950                                 break;
 951
 952                         case NAME_ERR_RESERVED:
 953                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 954                                     "name is reserved"));
 955                                 break;
 956
 957                         case NAME_ERR_DISKLIKE:
 958                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 959                                     "pool name is reserved"));
 960                                 break;
 961
 962                         case NAME_ERR_LEADING_SLASH:
 963                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 964                                     "leading slash in name"));
 965                                 break;
 966
 967                         case NAME_ERR_EMPTY_COMPONENT:
 968                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 969                                     "empty component in name"));
 970                                 break;
 971
 972                         case NAME_ERR_TRAILING_SLASH:
 973                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 974                                     "trailing slash in name"));
 975                                 break;
 976
 977                         case NAME_ERR_MULTIPLE_AT:
 978                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 979                                     "multiple '@' delimiters in name"));
 980                                 break;
 981
 982                         }
 983                 }
 984                 return (B_FALSE);
 985         }
 986
 987         return (B_TRUE);
 988 }
 989
 990 /*
 991  * Open a handle to the given pool, even if the pool is currently in the FAULTED
 992  * state.
 993  */
 994 zpool_handle_t *
 995 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
 996 {
 997         zpool_handle_t *zhp;
 998         boolean_t missing;
 999
1000         /*
1001          * Make sure the pool name is valid.
1002          */
1003         if (!zpool_name_valid(hdl, B_TRUE, pool)) {
1004                 (void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1005                     dgettext(TEXT_DOMAIN, "cannot open '%s'"),
1006                     pool);
1007                 return (NULL);
1008         }
1009
1010         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1011                 return (NULL);
1012
1013         zhp->zpool_hdl = hdl;
1014         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1015
1016         if (zpool_refresh_stats(zhp, &missing) != 0) {
1017                 zpool_close(zhp);
1018                 return (NULL);
1019         }
1020
1021         if (missing) {
1022                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
1023                 (void) zfs_error_fmt(hdl, EZFS_NOENT,
1024                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
1025                 zpool_close(zhp);
1026                 return (NULL);
1027         }
1028
1029         return (zhp);
1030 }
1031
1032 /*
1033  * Like the above, but silent on error.  Used when iterating over pools (because
1034  * the configuration cache may be out of date).
1035  */
1036 int
1037 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
1038 {
1039         zpool_handle_t *zhp;
1040         boolean_t missing;
1041
1042         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1043                 return (-1);
1044
1045         zhp->zpool_hdl = hdl;
1046         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1047
1048         if (zpool_refresh_stats(zhp, &missing) != 0) {
1049                 zpool_close(zhp);
1050                 return (-1);
1051         }
1052
1053         if (missing) {
1054                 zpool_close(zhp);
1055                 *ret = NULL;
1056                 return (0);
1057         }
1058
1059         *ret = zhp;
1060         return (0);
1061 }
1062
1063 /*
1064  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
1065  * state.
1066  */
1067 zpool_handle_t *
1068 zpool_open(libzfs_handle_t *hdl, const char *pool)
1069 {
1070         zpool_handle_t *zhp;
1071
1072         if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
1073                 return (NULL);
1074
1075         if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
1076                 (void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
1077                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
1078                 zpool_close(zhp);
1079                 return (NULL);
1080         }
1081
1082         return (zhp);
1083 }
1084
1085 /*
1086  * Close the handle.  Simply frees the memory associated with the handle.
1087  */
1088 void
1089 zpool_close(zpool_handle_t *zhp)
1090 {
1091         if (zhp->zpool_config)
1092                 nvlist_free(zhp->zpool_config);
1093         if (zhp->zpool_old_config)
1094                 nvlist_free(zhp->zpool_old_config);
1095         if (zhp->zpool_props)
1096                 nvlist_free(zhp->zpool_props);
1097         free(zhp);
1098 }
1099
1100 /*
1101  * Return the name of the pool.
1102  */
1103 const char *
1104 zpool_get_name(zpool_handle_t *zhp)
1105 {
1106         return (zhp->zpool_name);
1107 }
1108
1109
1110 /*
1111  * Return the state of the pool (ACTIVE or UNAVAILABLE)
1112  */
1113 int
1114 zpool_get_state(zpool_handle_t *zhp)
1115 {
1116         return (zhp->zpool_state);
1117 }
1118
1119 /*
1120  * Create the named pool, using the provided vdev list.  It is assumed
1121  * that the consumer has already validated the contents of the nvlist, so we
1122  * don't have to worry about error semantics.
1123  */
1124 int
1125 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
1126     nvlist_t *props, nvlist_t *fsprops)
1127 {
1128         zfs_cmd_t zc = { 0 };
1129         nvlist_t *zc_fsprops = NULL;
1130         nvlist_t *zc_props = NULL;
1131         char msg[1024];
1132         int ret = -1;
1133
1134         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1135             "cannot create '%s'"), pool);
1136
1137         if (!zpool_name_valid(hdl, B_FALSE, pool))
1138                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
1139
1140         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1141                 return (-1);
1142
1143         if (props) {
1144                 prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
1145
1146                 if ((zc_props = zpool_valid_proplist(hdl, pool, props,
1147                     SPA_VERSION_1, flags, msg)) == NULL) {
1148                         goto create_failed;
1149                 }
1150         }
1151
1152         if (fsprops) {
1153                 uint64_t zoned;
1154                 char *zonestr;
1155
1156                 zoned = ((nvlist_lookup_string(fsprops,
1157                     zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
1158                     strcmp(zonestr, "on") == 0);
1159
1160                 if ((zc_fsprops = zfs_valid_proplist(hdl,
1161                     ZFS_TYPE_FILESYSTEM, fsprops, zoned, NULL, msg)) == NULL) {
1162                         goto create_failed;
1163                 }
1164                 if (!zc_props &&
1165                     (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
1166                         goto create_failed;
1167                 }
1168                 if (nvlist_add_nvlist(zc_props,
1169                     ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
1170                         goto create_failed;
1171                 }
1172         }
1173
1174         if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
1175                 goto create_failed;
1176
1177         (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
1178
1179         if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
1180
1181                 zcmd_free_nvlists(&zc);
1182                 nvlist_free(zc_props);
1183                 nvlist_free(zc_fsprops);
1184
1185                 switch (errno) {
1186                 case EBUSY:
1187                         /*
1188                          * This can happen if the user has specified the same
1189                          * device multiple times.  We can't reliably detect this
1190                          * until we try to add it and see we already have a
1191                          * label.
1192                          */
1193                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1194                             "one or more vdevs refer to the same device"));
1195                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1196
1197                 case EOVERFLOW:
1198                         /*
1199                          * This occurs when one of the devices is below
1200                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1201                          * device was the problem device since there's no
1202                          * reliable way to determine device size from userland.
1203                          */
1204                         {
1205                                 char buf[64];
1206
1207                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1208
1209                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1210                                     "one or more devices is less than the "
1211                                     "minimum size (%s)"), buf);
1212                         }
1213                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1214
1215                 case ENOSPC:
1216                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1217                             "one or more devices is out of space"));
1218                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1219
1220                 case ENOTBLK:
1221                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1222                             "cache device must be a disk or disk slice"));
1223                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1224
1225                 default:
1226                         return (zpool_standard_error(hdl, errno, msg));
1227                 }
1228         }
1229
1230 create_failed:
1231         zcmd_free_nvlists(&zc);
1232         nvlist_free(zc_props);
1233         nvlist_free(zc_fsprops);
1234         return (ret);
1235 }
1236
1237 /*
1238  * Destroy the given pool.  It is up to the caller to ensure that there are no
1239  * datasets left in the pool.
1240  */
1241 int
1242 zpool_destroy(zpool_handle_t *zhp, const char *log_str)
1243 {
1244         zfs_cmd_t zc = { 0 };
1245         zfs_handle_t *zfp = NULL;
1246         libzfs_handle_t *hdl = zhp->zpool_hdl;
1247         char msg[1024];
1248
1249         if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1250             (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1251                 return (-1);
1252
1253         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1254         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1255
1256         if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1257                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1258                     "cannot destroy '%s'"), zhp->zpool_name);
1259
1260                 if (errno == EROFS) {
1261                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1262                             "one or more devices is read only"));
1263                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1264                 } else {
1265                         (void) zpool_standard_error(hdl, errno, msg);
1266                 }
1267
1268                 if (zfp)
1269                         zfs_close(zfp);
1270                 return (-1);
1271         }
1272
1273         if (zfp) {
1274                 remove_mountpoint(zfp);
1275                 zfs_close(zfp);
1276         }
1277
1278         return (0);
1279 }
1280
1281 /*
1282  * Add the given vdevs to the pool.  The caller must have already performed the
1283  * necessary verification to ensure that the vdev specification is well-formed.
1284  */
1285 int
1286 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1287 {
1288         zfs_cmd_t zc = { 0 };
1289         int ret;
1290         libzfs_handle_t *hdl = zhp->zpool_hdl;
1291         char msg[1024];
1292         nvlist_t **spares, **l2cache;
1293         uint_t nspares, nl2cache;
1294
1295         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1296             "cannot add to '%s'"), zhp->zpool_name);
1297
1298         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1299             SPA_VERSION_SPARES &&
1300             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1301             &spares, &nspares) == 0) {
1302                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1303                     "upgraded to add hot spares"));
1304                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1305         }
1306
1307         if (zpool_is_bootable(zhp) && nvlist_lookup_nvlist_array(nvroot,
1308             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) {
1309                 uint64_t s;
1310
1311                 for (s = 0; s < nspares; s++) {
1312                         char *path;
1313
1314                         if (nvlist_lookup_string(spares[s], ZPOOL_CONFIG_PATH,
1315                             &path) == 0 && pool_uses_efi(spares[s])) {
1316                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1317                                     "device '%s' contains an EFI label and "
1318                                     "cannot be used on root pools."),
1319                                     zpool_vdev_name(hdl, NULL, spares[s],
1320                                     B_FALSE));
1321                                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
1322                         }
1323                 }
1324         }
1325
1326         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1327             SPA_VERSION_L2CACHE &&
1328             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1329             &l2cache, &nl2cache) == 0) {
1330                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1331                     "upgraded to add cache devices"));
1332                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1333         }
1334
1335         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1336                 return (-1);
1337         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1338
1339         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1340                 switch (errno) {
1341                 case EBUSY:
1342                         /*
1343                          * This can happen if the user has specified the same
1344                          * device multiple times.  We can't reliably detect this
1345                          * until we try to add it and see we already have a
1346                          * label.
1347                          */
1348                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1349                             "one or more vdevs refer to the same device"));
1350                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1351                         break;
1352
1353                 case EOVERFLOW:
1354                         /*
1355                          * This occurrs when one of the devices is below
1356                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1357                          * device was the problem device since there's no
1358                          * reliable way to determine device size from userland.
1359                          */
1360                         {
1361                                 char buf[64];
1362
1363                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1364
1365                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1366                                     "device is less than the minimum "
1367                                     "size (%s)"), buf);
1368                         }
1369                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1370                         break;
1371
1372                 case ENOTSUP:
1373                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1374                             "pool must be upgraded to add these vdevs"));
1375                         (void) zfs_error(hdl, EZFS_BADVERSION, msg);
1376                         break;
1377
1378                 case EDOM:
1379                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1380                             "root pool can not have multiple vdevs"
1381                             " or separate logs"));
1382                         (void) zfs_error(hdl, EZFS_POOL_NOTSUP, msg);
1383                         break;
1384
1385                 case ENOTBLK:
1386                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1387                             "cache device must be a disk or disk slice"));
1388                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1389                         break;
1390
1391                 default:
1392                         (void) zpool_standard_error(hdl, errno, msg);
1393                 }
1394
1395                 ret = -1;
1396         } else {
1397                 ret = 0;
1398         }
1399
1400         zcmd_free_nvlists(&zc);
1401
1402         return (ret);
1403 }
1404
1405 /*
1406  * Exports the pool from the system.  The caller must ensure that there are no
1407  * mounted datasets in the pool.
1408  */
1409 static int
1410 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce,
1411     const char *log_str)
1412 {
1413         zfs_cmd_t zc = { 0 };
1414         char msg[1024];
1415
1416         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1417             "cannot export '%s'"), zhp->zpool_name);
1418
1419         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1420         zc.zc_cookie = force;
1421         zc.zc_guid = hardforce;
1422         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1423
1424         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1425                 switch (errno) {
1426                 case EXDEV:
1427                         zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1428                             "use '-f' to override the following errors:\n"
1429                             "'%s' has an active shared spare which could be"
1430                             " used by other pools once '%s' is exported."),
1431                             zhp->zpool_name, zhp->zpool_name);
1432                         return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1433                             msg));
1434                 default:
1435                         return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1436                             msg));
1437                 }
1438         }
1439
1440         return (0);
1441 }
1442
1443 int
1444 zpool_export(zpool_handle_t *zhp, boolean_t force, const char *log_str)
1445 {
1446         return (zpool_export_common(zhp, force, B_FALSE, log_str));
1447 }
1448
1449 int
1450 zpool_export_force(zpool_handle_t *zhp, const char *log_str)
1451 {
1452         return (zpool_export_common(zhp, B_TRUE, B_TRUE, log_str));
1453 }
1454
1455 static void
1456 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1457     nvlist_t *config)
1458 {
1459         nvlist_t *nv = NULL;
1460         uint64_t rewindto;
1461         int64_t loss = -1;
1462         struct tm t;
1463         char timestr[128];
1464
1465         if (!hdl->libzfs_printerr || config == NULL)
1466                 return;
1467
1468         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1469             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0) {
1470                 return;
1471         }
1472
1473         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1474                 return;
1475         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1476
1477         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1478             strftime(timestr, 128, 0, &t) != 0) {
1479                 if (dryrun) {
1480                         (void) printf(dgettext(TEXT_DOMAIN,
1481                             "Would be able to return %s "
1482                             "to its state as of %s.\n"),
1483                             name, timestr);
1484                 } else {
1485                         (void) printf(dgettext(TEXT_DOMAIN,
1486                             "Pool %s returned to its state as of %s.\n"),
1487                             name, timestr);
1488                 }
1489                 if (loss > 120) {
1490                         (void) printf(dgettext(TEXT_DOMAIN,
1491                             "%s approximately %lld "),
1492                             dryrun ? "Would discard" : "Discarded",
1493                             (loss + 30) / 60);
1494                         (void) printf(dgettext(TEXT_DOMAIN,
1495                             "minutes of transactions.\n"));
1496                 } else if (loss > 0) {
1497                         (void) printf(dgettext(TEXT_DOMAIN,
1498                             "%s approximately %lld "),
1499                             dryrun ? "Would discard" : "Discarded", loss);
1500                         (void) printf(dgettext(TEXT_DOMAIN,
1501                             "seconds of transactions.\n"));
1502                 }
1503         }
1504 }
1505
1506 void
1507 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1508     nvlist_t *config)
1509 {
1510         nvlist_t *nv = NULL;
1511         int64_t loss = -1;
1512         uint64_t edata = UINT64_MAX;
1513         uint64_t rewindto;
1514         struct tm t;
1515         char timestr[128];
1516
1517         if (!hdl->libzfs_printerr)
1518                 return;
1519
1520         if (reason >= 0)
1521                 (void) printf(dgettext(TEXT_DOMAIN, "action: "));
1522         else
1523                 (void) printf(dgettext(TEXT_DOMAIN, "\t"));
1524
1525         /* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1526         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1527             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0 ||
1528             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1529                 goto no_info;
1530
1531         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1532         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1533             &edata);
1534
1535         (void) printf(dgettext(TEXT_DOMAIN,
1536             "Recovery is possible, but will result in some data loss.\n"));
1537
1538         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1539             strftime(timestr, 128, 0, &t) != 0) {
1540                 (void) printf(dgettext(TEXT_DOMAIN,
1541                     "\tReturning the pool to its state as of %s\n"
1542                     "\tshould correct the problem.  "),
1543                     timestr);
1544         } else {
1545                 (void) printf(dgettext(TEXT_DOMAIN,
1546                     "\tReverting the pool to an earlier state "
1547                     "should correct the problem.\n\t"));
1548         }
1549
1550         if (loss > 120) {
1551                 (void) printf(dgettext(TEXT_DOMAIN,
1552                     "Approximately %lld minutes of data\n"
1553                     "\tmust be discarded, irreversibly.  "), (loss + 30) / 60);
1554         } else if (loss > 0) {
1555                 (void) printf(dgettext(TEXT_DOMAIN,
1556                     "Approximately %lld seconds of data\n"
1557                     "\tmust be discarded, irreversibly.  "), loss);
1558         }
1559         if (edata != 0 && edata != UINT64_MAX) {
1560                 if (edata == 1) {
1561                         (void) printf(dgettext(TEXT_DOMAIN,
1562                             "After rewind, at least\n"
1563                             "\tone persistent user-data error will remain.  "));
1564                 } else {
1565                         (void) printf(dgettext(TEXT_DOMAIN,
1566                             "After rewind, several\n"
1567                             "\tpersistent user-data errors will remain.  "));
1568                 }
1569         }
1570         (void) printf(dgettext(TEXT_DOMAIN,
1571             "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1572             reason >= 0 ? "clear" : "import", name);
1573
1574         (void) printf(dgettext(TEXT_DOMAIN,
1575             "A scrub of the pool\n"
1576             "\tis strongly recommended after recovery.\n"));
1577         return;
1578
1579 no_info:
1580         (void) printf(dgettext(TEXT_DOMAIN,
1581             "Destroy and re-create the pool from\n\ta backup source.\n"));
1582 }
1583
1584 /*
1585  * zpool_import() is a contracted interface. Should be kept the same
1586  * if possible.
1587  *
1588  * Applications should use zpool_import_props() to import a pool with
1589  * new properties value to be set.
1590  */
1591 int
1592 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1593     char *altroot)
1594 {
1595         nvlist_t *props = NULL;
1596         int ret;
1597
1598         if (altroot != NULL) {
1599                 if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1600                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1601                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1602                             newname));
1603                 }
1604
1605                 if (nvlist_add_string(props,
1606                     zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1607                     nvlist_add_string(props,
1608                     zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1609                         nvlist_free(props);
1610                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1611                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1612                             newname));
1613                 }
1614         }
1615
1616         ret = zpool_import_props(hdl, config, newname, props,
1617             ZFS_IMPORT_NORMAL);
1618         if (props)
1619                 nvlist_free(props);
1620         return (ret);
1621 }
1622
1623 static void
1624 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1625     int indent)
1626 {
1627         nvlist_t **child;
1628         uint_t c, children;
1629         char *vname;
1630         uint64_t is_log = 0;
1631
1632         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1633             &is_log);
1634
1635         if (name != NULL)
1636                 (void) printf("\t%*s%s%s\n", indent, "", name,
1637                     is_log ? " [log]" : "");
1638
1639         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1640             &child, &children) != 0)
1641                 return;
1642
1643         for (c = 0; c < children; c++) {
1644                 vname = zpool_vdev_name(hdl, NULL, child[c], B_TRUE);
1645                 print_vdev_tree(hdl, vname, child[c], indent + 2);
1646                 free(vname);
1647         }
1648 }
1649
1650 void
1651 zpool_print_unsup_feat(nvlist_t *config)
1652 {
1653         nvlist_t *nvinfo, *unsup_feat;
1654
1655         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nvinfo) ==
1656             0);
1657         verify(nvlist_lookup_nvlist(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT,
1658             &unsup_feat) == 0);
1659
1660         for (nvpair_t *nvp = nvlist_next_nvpair(unsup_feat, NULL); nvp != NULL;
1661             nvp = nvlist_next_nvpair(unsup_feat, nvp)) {
1662                 char *desc;
1663
1664                 verify(nvpair_type(nvp) == DATA_TYPE_STRING);
1665                 verify(nvpair_value_string(nvp, &desc) == 0);
1666
1667                 if (strlen(desc) > 0)
1668                         (void) printf("\t%s (%s)\n", nvpair_name(nvp), desc);
1669                 else
1670                         (void) printf("\t%s\n", nvpair_name(nvp));
1671         }
1672 }
1673
1674 /*
1675  * Import the given pool using the known configuration and a list of
1676  * properties to be set. The configuration should have come from
1677  * zpool_find_import(). The 'newname' parameters control whether the pool
1678  * is imported with a different name.
1679  */
1680 int
1681 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1682     nvlist_t *props, int flags)
1683 {
1684         zfs_cmd_t zc = { 0 };
1685         zpool_rewind_policy_t policy;
1686         nvlist_t *nv = NULL;
1687         nvlist_t *nvinfo = NULL;
1688         nvlist_t *missing = NULL;
1689         char *thename;
1690         char *origname;
1691         int ret;
1692         int error = 0;
1693         char errbuf[1024];
1694
1695         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1696             &origname) == 0);
1697
1698         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1699             "cannot import pool '%s'"), origname);
1700
1701         if (newname != NULL) {
1702                 if (!zpool_name_valid(hdl, B_FALSE, newname))
1703                         return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1704                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1705                             newname));
1706                 thename = (char *)newname;
1707         } else {
1708                 thename = origname;
1709         }
1710
1711         if (props) {
1712                 uint64_t version;
1713                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1714
1715                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1716                     &version) == 0);
1717
1718                 if ((props = zpool_valid_proplist(hdl, origname,
1719                     props, version, flags, errbuf)) == NULL) {
1720                         return (-1);
1721                 } else if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1722                         nvlist_free(props);
1723                         return (-1);
1724                 }
1725         }
1726
1727         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1728
1729         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1730             &zc.zc_guid) == 0);
1731
1732         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1733                 nvlist_free(props);
1734                 return (-1);
1735         }
1736         if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1737                 nvlist_free(props);
1738                 return (-1);
1739         }
1740
1741         zc.zc_cookie = flags;
1742         while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1743             errno == ENOMEM) {
1744                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1745                         zcmd_free_nvlists(&zc);
1746                         return (-1);
1747                 }
1748         }
1749         if (ret != 0)
1750                 error = errno;
1751
1752         (void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1753         zpool_get_rewind_policy(config, &policy);
1754
1755         if (error) {
1756                 char desc[1024];
1757
1758                 /*
1759                  * Dry-run failed, but we print out what success
1760                  * looks like if we found a best txg
1761                  */
1762                 if (policy.zrp_request & ZPOOL_TRY_REWIND) {
1763                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1764                             B_TRUE, nv);
1765                         nvlist_free(nv);
1766                         return (-1);
1767                 }
1768
1769                 if (newname == NULL)
1770                         (void) snprintf(desc, sizeof (desc),
1771                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1772                             thename);
1773                 else
1774                         (void) snprintf(desc, sizeof (desc),
1775                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1776                             origname, thename);
1777
1778                 switch (error) {
1779                 case ENOTSUP:
1780                         if (nv != NULL && nvlist_lookup_nvlist(nv,
1781                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1782                             nvlist_exists(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT)) {
1783                                 (void) printf(dgettext(TEXT_DOMAIN, "This "
1784                                     "pool uses the following feature(s) not "
1785                                     "supported by this system:\n"));
1786                                 zpool_print_unsup_feat(nv);
1787                                 if (nvlist_exists(nvinfo,
1788                                     ZPOOL_CONFIG_CAN_RDONLY)) {
1789                                         (void) printf(dgettext(TEXT_DOMAIN,
1790                                             "All unsupported features are only "
1791                                             "required for writing to the pool."
1792                                             "\nThe pool can be imported using "
1793                                             "'-o readonly=on'.\n"));
1794                                 }
1795                         }
1796                         /*
1797                          * Unsupported version.
1798                          */
1799                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1800                         break;
1801
1802                 case EINVAL:
1803                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1804                         break;
1805
1806                 case EROFS:
1807                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1808                             "one or more devices is read only"));
1809                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1810                         break;
1811
1812                 case ENXIO:
1813                         if (nv && nvlist_lookup_nvlist(nv,
1814                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1815                             nvlist_lookup_nvlist(nvinfo,
1816                             ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1817                                 (void) printf(dgettext(TEXT_DOMAIN,
1818                                     "The devices below are missing, use "
1819                                     "'-m' to import the pool anyway:\n"));
1820                                 print_vdev_tree(hdl, NULL, missing, 2);
1821                                 (void) printf("\n");
1822                         }
1823                         (void) zpool_standard_error(hdl, error, desc);
1824                         break;
1825
1826                 case EEXIST:
1827                         (void) zpool_standard_error(hdl, error, desc);
1828                         break;
1829
1830                 default:
1831                         (void) zpool_standard_error(hdl, error, desc);
1832                         zpool_explain_recover(hdl,
1833                             newname ? origname : thename, -error, nv);
1834                         break;
1835                 }
1836
1837                 nvlist_free(nv);
1838                 ret = -1;
1839         } else {
1840                 zpool_handle_t *zhp;
1841
1842                 /*
1843                  * This should never fail, but play it safe anyway.
1844                  */
1845                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1846                         ret = -1;
1847                 else if (zhp != NULL)
1848                         zpool_close(zhp);
1849                 if (policy.zrp_request &
1850                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1851                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1852                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1853                 }
1854                 nvlist_free(nv);
1855                 return (0);
1856         }
1857
1858         zcmd_free_nvlists(&zc);
1859         nvlist_free(props);
1860
1861         return (ret);
1862 }
1863
1864 /*
1865  * Scan the pool.
1866  */
1867 int
1868 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1869 {
1870         zfs_cmd_t zc = { 0 };
1871         char msg[1024];
1872         libzfs_handle_t *hdl = zhp->zpool_hdl;
1873
1874         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1875         zc.zc_cookie = func;
1876
1877         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1878             (errno == ENOENT && func != POOL_SCAN_NONE))
1879                 return (0);
1880
1881         if (func == POOL_SCAN_SCRUB) {
1882                 (void) snprintf(msg, sizeof (msg),
1883                     dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1884         } else if (func == POOL_SCAN_NONE) {
1885                 (void) snprintf(msg, sizeof (msg),
1886                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1887                     zc.zc_name);
1888         } else {
1889                 assert(!"unexpected result");
1890         }
1891
1892         if (errno == EBUSY) {
1893                 nvlist_t *nvroot;
1894                 pool_scan_stat_t *ps = NULL;
1895                 uint_t psc;
1896
1897                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
1898                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1899                 (void) nvlist_lookup_uint64_array(nvroot,
1900                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1901                 if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1902                         return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1903                 else
1904                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
1905         } else if (errno == ENOENT) {
1906                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1907         } else {
1908                 return (zpool_standard_error(hdl, errno, msg));
1909         }
1910 }
1911
1912 /*
1913  * This provides a very minimal check whether a given string is likely a
1914  * c#t#d# style string.  Users of this are expected to do their own
1915  * verification of the s# part.
1916  */
1917 #define CTD_CHECK(str)  (str && str[0] == 'c' && isdigit(str[1]))
1918
1919 /*
1920  * More elaborate version for ones which may start with "/dev/dsk/"
1921  * and the like.
1922  */
1923 static int
1924 ctd_check_path(char *str) {
1925         /*
1926          * If it starts with a slash, check the last component.
1927          */
1928         if (str && str[0] == '/') {
1929                 char *tmp = strrchr(str, '/');
1930
1931                 /*
1932                  * If it ends in "/old", check the second-to-last
1933                  * component of the string instead.
1934                  */
1935                 if (tmp != str && strcmp(tmp, "/old") == 0) {
1936                         for (tmp--; *tmp != '/'; tmp--)
1937                                 ;
1938                 }
1939                 str = tmp + 1;
1940         }
1941         return (CTD_CHECK(str));
1942 }
1943
1944 /*
1945  * Find a vdev that matches the search criteria specified. We use the
1946  * the nvpair name to determine how we should look for the device.
1947  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1948  * spare; but FALSE if its an INUSE spare.
1949  */
1950 static nvlist_t *
1951 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1952     boolean_t *l2cache, boolean_t *log)
1953 {
1954         uint_t c, children;
1955         nvlist_t **child;
1956         nvlist_t *ret;
1957         uint64_t is_log;
1958         char *srchkey;
1959         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
1960
1961         /* Nothing to look for */
1962         if (search == NULL || pair == NULL)
1963                 return (NULL);
1964
1965         /* Obtain the key we will use to search */
1966         srchkey = nvpair_name(pair);
1967
1968         switch (nvpair_type(pair)) {
1969         case DATA_TYPE_UINT64:
1970                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
1971                         uint64_t srchval, theguid;
1972
1973                         verify(nvpair_value_uint64(pair, &srchval) == 0);
1974                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
1975                             &theguid) == 0);
1976                         if (theguid == srchval)
1977                                 return (nv);
1978                 }
1979                 break;
1980
1981         case DATA_TYPE_STRING: {
1982                 char *srchval, *val;
1983
1984                 verify(nvpair_value_string(pair, &srchval) == 0);
1985                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
1986                         break;
1987
1988                 /*
1989                  * Search for the requested value. Special cases:
1990                  *
1991                  * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
1992                  *   "s0" or "s0/old".  The "s0" part is hidden from the user,
1993                  *   but included in the string, so this matches around it.
1994                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
1995                  *
1996                  * Otherwise, all other searches are simple string compares.
1997                  */
1998                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0 &&
1999                     ctd_check_path(val)) {
2000                         uint64_t wholedisk = 0;
2001
2002                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
2003                             &wholedisk);
2004                         if (wholedisk) {
2005                                 int slen = strlen(srchval);
2006                                 int vlen = strlen(val);
2007
2008                                 if (slen != vlen - 2)
2009                                         break;
2010
2011                                 /*
2012                                  * make_leaf_vdev() should only set
2013                                  * wholedisk for ZPOOL_CONFIG_PATHs which
2014                                  * will include "/dev/dsk/", giving plenty of
2015                                  * room for the indices used next.
2016                                  */
2017                                 ASSERT(vlen >= 6);
2018
2019                                 /*
2020                                  * strings identical except trailing "s0"
2021                                  */
2022                                 if (strcmp(&val[vlen - 2], "s0") == 0 &&
2023                                     strncmp(srchval, val, slen) == 0)
2024                                         return (nv);
2025
2026                                 /*
2027                                  * strings identical except trailing "s0/old"
2028                                  */
2029                                 if (strcmp(&val[vlen - 6], "s0/old") == 0 &&
2030                                     strcmp(&srchval[slen - 4], "/old") == 0 &&
2031                                     strncmp(srchval, val, slen - 4) == 0)
2032                                         return (nv);
2033
2034                                 break;
2035                         }
2036                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2037                         char *type, *idx, *end, *p;
2038                         uint64_t id, vdev_id;
2039
2040                         /*
2041                          * Determine our vdev type, keeping in mind
2042                          * that the srchval is composed of a type and
2043                          * vdev id pair (i.e. mirror-4).
2044                          */
2045                         if ((type = strdup(srchval)) == NULL)
2046                                 return (NULL);
2047
2048                         if ((p = strrchr(type, '-')) == NULL) {
2049                                 free(type);
2050                                 break;
2051                         }
2052                         idx = p + 1;
2053                         *p = '\0';
2054
2055                         /*
2056                          * If the types don't match then keep looking.
2057                          */
2058                         if (strncmp(val, type, strlen(val)) != 0) {
2059                                 free(type);
2060                                 break;
2061                         }
2062
2063                         verify(strncmp(type, VDEV_TYPE_RAIDZ,
2064                             strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2065                             strncmp(type, VDEV_TYPE_MIRROR,
2066                             strlen(VDEV_TYPE_MIRROR)) == 0);
2067                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
2068                             &id) == 0);
2069
2070                         errno = 0;
2071                         vdev_id = strtoull(idx, &end, 10);
2072
2073                         free(type);
2074                         if (errno != 0)
2075                                 return (NULL);
2076
2077                         /*
2078                          * Now verify that we have the correct vdev id.
2079                          */
2080                         if (vdev_id == id)
2081                                 return (nv);
2082                 }
2083
2084                 /*
2085                  * Common case
2086                  */
2087                 if (strcmp(srchval, val) == 0)
2088                         return (nv);
2089                 break;
2090         }
2091
2092         default:
2093                 break;
2094         }
2095
2096         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2097             &child, &children) != 0)
2098                 return (NULL);
2099
2100         for (c = 0; c < children; c++) {
2101                 if ((ret = vdev_to_nvlist_iter(child[c], search,
2102                     avail_spare, l2cache, NULL)) != NULL) {
2103                         /*
2104                          * The 'is_log' value is only set for the toplevel
2105                          * vdev, not the leaf vdevs.  So we always lookup the
2106                          * log device from the root of the vdev tree (where
2107                          * 'log' is non-NULL).
2108                          */
2109                         if (log != NULL &&
2110                             nvlist_lookup_uint64(child[c],
2111                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2112                             is_log) {
2113                                 *log = B_TRUE;
2114                         }
2115                         return (ret);
2116                 }
2117         }
2118
2119         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2120             &child, &children) == 0) {
2121                 for (c = 0; c < children; c++) {
2122                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2123                             avail_spare, l2cache, NULL)) != NULL) {
2124                                 *avail_spare = B_TRUE;
2125                                 return (ret);
2126                         }
2127                 }
2128         }
2129
2130         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2131             &child, &children) == 0) {
2132                 for (c = 0; c < children; c++) {
2133                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2134                             avail_spare, l2cache, NULL)) != NULL) {
2135                                 *l2cache = B_TRUE;
2136                                 return (ret);
2137                         }
2138                 }
2139         }
2140
2141         return (NULL);
2142 }
2143
2144 /*
2145  * Given a physical path (minus the "/devices" prefix), find the
2146  * associated vdev.
2147  */
2148 nvlist_t *
2149 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2150     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2151 {
2152         nvlist_t *search, *nvroot, *ret;
2153
2154         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2155         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
2156
2157         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2158             &nvroot) == 0);
2159
2160         *avail_spare = B_FALSE;
2161         *l2cache = B_FALSE;
2162         if (log != NULL)
2163                 *log = B_FALSE;
2164         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2165         nvlist_free(search);
2166
2167         return (ret);
2168 }
2169
2170 /*
2171  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2172  */
2173 boolean_t
2174 zpool_vdev_is_interior(const char *name)
2175 {
2176         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2177             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2178                 return (B_TRUE);
2179         return (B_FALSE);
2180 }
2181
2182 nvlist_t *
2183 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2184     boolean_t *l2cache, boolean_t *log)
2185 {
2186         char buf[MAXPATHLEN];
2187         char *end;
2188         nvlist_t *nvroot, *search, *ret;
2189         uint64_t guid;
2190
2191         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2192
2193         guid = strtoull(path, &end, 10);
2194         if (guid != 0 && *end == '\0') {
2195                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
2196         } else if (zpool_vdev_is_interior(path)) {
2197                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
2198         } else if (path[0] != '/') {
2199                 (void) snprintf(buf, sizeof (buf), "%s%s", _PATH_DEV, path);
2200                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, buf) == 0);
2201         } else {
2202                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
2203         }
2204
2205         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2206             &nvroot) == 0);
2207
2208         *avail_spare = B_FALSE;
2209         *l2cache = B_FALSE;
2210         if (log != NULL)
2211                 *log = B_FALSE;
2212         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2213         nvlist_free(search);
2214
2215         return (ret);
2216 }
2217
2218 static int
2219 vdev_online(nvlist_t *nv)
2220 {
2221         uint64_t ival;
2222
2223         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2224             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2225             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2226                 return (0);
2227
2228         return (1);
2229 }
2230
2231 /*
2232  * Helper function for zpool_get_physpaths().
2233  */
2234 static int
2235 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2236     size_t *bytes_written)
2237 {
2238         size_t bytes_left, pos, rsz;
2239         char *tmppath;
2240         const char *format;
2241
2242         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2243             &tmppath) != 0)
2244                 return (EZFS_NODEVICE);
2245
2246         pos = *bytes_written;
2247         bytes_left = physpath_size - pos;
2248         format = (pos == 0) ? "%s" : " %s";
2249
2250         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2251         *bytes_written += rsz;
2252
2253         if (rsz >= bytes_left) {
2254                 /* if physpath was not copied properly, clear it */
2255                 if (bytes_left != 0) {
2256                         physpath[pos] = 0;
2257                 }
2258                 return (EZFS_NOSPC);
2259         }
2260         return (0);
2261 }
2262
2263 static int
2264 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2265     size_t *rsz, boolean_t is_spare)
2266 {
2267         char *type;
2268         int ret;
2269
2270         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2271                 return (EZFS_INVALCONFIG);
2272
2273         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2274                 /*
2275                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2276                  * For a spare vdev, we only want to boot from the active
2277                  * spare device.
2278                  */
2279                 if (is_spare) {
2280                         uint64_t spare = 0;
2281                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2282                             &spare);
2283                         if (!spare)
2284                                 return (EZFS_INVALCONFIG);
2285                 }
2286
2287                 if (vdev_online(nv)) {
2288                         if ((ret = vdev_get_one_physpath(nv, physpath,
2289                             phypath_size, rsz)) != 0)
2290                                 return (ret);
2291                 }
2292         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2293             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2294             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2295                 nvlist_t **child;
2296                 uint_t count;
2297                 int i, ret;
2298
2299                 if (nvlist_lookup_nvlist_array(nv,
2300                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2301                         return (EZFS_INVALCONFIG);
2302
2303                 for (i = 0; i < count; i++) {
2304                         ret = vdev_get_physpaths(child[i], physpath,
2305                             phypath_size, rsz, is_spare);
2306                         if (ret == EZFS_NOSPC)
2307                                 return (ret);
2308                 }
2309         }
2310
2311         return (EZFS_POOL_INVALARG);
2312 }
2313
2314 /*
2315  * Get phys_path for a root pool config.
2316  * Return 0 on success; non-zero on failure.
2317  */
2318 static int
2319 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2320 {
2321         size_t rsz;
2322         nvlist_t *vdev_root;
2323         nvlist_t **child;
2324         uint_t count;
2325         char *type;
2326
2327         rsz = 0;
2328
2329         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2330             &vdev_root) != 0)
2331                 return (EZFS_INVALCONFIG);
2332
2333         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2334             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2335             &child, &count) != 0)
2336                 return (EZFS_INVALCONFIG);
2337
2338         /*
2339          * root pool can not have EFI labeled disks and can only have
2340          * a single top-level vdev.
2341          */
2342         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1 ||
2343             pool_uses_efi(vdev_root))
2344                 return (EZFS_POOL_INVALARG);
2345
2346         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2347             B_FALSE);
2348
2349         /* No online devices */
2350         if (rsz == 0)
2351                 return (EZFS_NODEVICE);
2352
2353         return (0);
2354 }
2355
2356 /*
2357  * Get phys_path for a root pool
2358  * Return 0 on success; non-zero on failure.
2359  */
2360 int
2361 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2362 {
2363         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2364             phypath_size));
2365 }
2366
2367 /*
2368  * If the device has being dynamically expanded then we need to relabel
2369  * the disk to use the new unallocated space.
2370  */
2371 static int
2372 zpool_relabel_disk(libzfs_handle_t *hdl, const char *name)
2373 {
2374 #ifdef sun
2375         char path[MAXPATHLEN];
2376         char errbuf[1024];
2377         int fd, error;
2378         int (*_efi_use_whole_disk)(int);
2379
2380         if ((_efi_use_whole_disk = (int (*)(int))dlsym(RTLD_DEFAULT,
2381             "efi_use_whole_disk")) == NULL)
2382                 return (-1);
2383
2384         (void) snprintf(path, sizeof (path), "%s/%s", RDISK_ROOT, name);
2385
2386         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
2387                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2388                     "relabel '%s': unable to open device"), name);
2389                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
2390         }
2391
2392         /*
2393          * It's possible that we might encounter an error if the device
2394          * does not have any unallocated space left. If so, we simply
2395          * ignore that error and continue on.
2396          */
2397         error = _efi_use_whole_disk(fd);
2398         (void) close(fd);
2399         if (error && error != VT_ENOSPC) {
2400                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2401                     "relabel '%s': unable to read disk capacity"), name);
2402                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
2403         }
2404 #endif  /* sun */
2405         return (0);
2406 }
2407
2408 /*
2409  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2410  * ZFS_ONLINE_* flags.
2411  */
2412 int
2413 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2414     vdev_state_t *newstate)
2415 {
2416         zfs_cmd_t zc = { 0 };
2417         char msg[1024];
2418         nvlist_t *tgt;
2419         boolean_t avail_spare, l2cache, islog;
2420         libzfs_handle_t *hdl = zhp->zpool_hdl;
2421
2422         if (flags & ZFS_ONLINE_EXPAND) {
2423                 (void) snprintf(msg, sizeof (msg),
2424                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2425         } else {
2426                 (void) snprintf(msg, sizeof (msg),
2427                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2428         }
2429
2430         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2431         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2432             &islog)) == NULL)
2433                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2434
2435         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2436
2437         if (avail_spare)
2438                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2439
2440         if (flags & ZFS_ONLINE_EXPAND ||
2441             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2442                 char *pathname = NULL;
2443                 uint64_t wholedisk = 0;
2444
2445                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2446                     &wholedisk);
2447                 verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
2448                     &pathname) == 0);
2449
2450                 /*
2451                  * XXX - L2ARC 1.0 devices can't support expansion.
2452                  */
2453                 if (l2cache) {
2454                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2455                             "cannot expand cache devices"));
2456                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2457                 }
2458
2459                 if (wholedisk) {
2460                         pathname += strlen(DISK_ROOT) + 1;
2461                         (void) zpool_relabel_disk(hdl, pathname);
2462                 }
2463         }
2464
2465         zc.zc_cookie = VDEV_STATE_ONLINE;
2466         zc.zc_obj = flags;
2467
2468         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2469                 if (errno == EINVAL) {
2470                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2471                             "from this pool into a new one.  Use '%s' "
2472                             "instead"), "zpool detach");
2473                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2474                 }
2475                 return (zpool_standard_error(hdl, errno, msg));
2476         }
2477
2478         *newstate = zc.zc_cookie;
2479         return (0);
2480 }
2481
2482 /*
2483  * Take the specified vdev offline
2484  */
2485 int
2486 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2487 {
2488         zfs_cmd_t zc = { 0 };
2489         char msg[1024];
2490         nvlist_t *tgt;
2491         boolean_t avail_spare, l2cache;
2492         libzfs_handle_t *hdl = zhp->zpool_hdl;
2493
2494         (void) snprintf(msg, sizeof (msg),
2495             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2496
2497         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2498         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2499             NULL)) == NULL)
2500                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2501
2502         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2503
2504         if (avail_spare)
2505                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2506
2507         zc.zc_cookie = VDEV_STATE_OFFLINE;
2508         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2509
2510         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2511                 return (0);
2512
2513         switch (errno) {
2514         case EBUSY:
2515
2516                 /*
2517                  * There are no other replicas of this device.
2518                  */
2519                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2520
2521         case EEXIST:
2522                 /*
2523                  * The log device has unplayed logs
2524                  */
2525                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2526
2527         default:
2528                 return (zpool_standard_error(hdl, errno, msg));
2529         }
2530 }
2531
2532 /*
2533  * Mark the given vdev faulted.
2534  */
2535 int
2536 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2537 {
2538         zfs_cmd_t zc = { 0 };
2539         char msg[1024];
2540         libzfs_handle_t *hdl = zhp->zpool_hdl;
2541
2542         (void) snprintf(msg, sizeof (msg),
2543             dgettext(TEXT_DOMAIN, "cannot fault %llu"), guid);
2544
2545         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2546         zc.zc_guid = guid;
2547         zc.zc_cookie = VDEV_STATE_FAULTED;
2548         zc.zc_obj = aux;
2549
2550         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2551                 return (0);
2552
2553         switch (errno) {
2554         case EBUSY:
2555
2556                 /*
2557                  * There are no other replicas of this device.
2558                  */
2559                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2560
2561         default:
2562                 return (zpool_standard_error(hdl, errno, msg));
2563         }
2564
2565 }
2566
2567 /*
2568  * Mark the given vdev degraded.
2569  */
2570 int
2571 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2572 {
2573         zfs_cmd_t zc = { 0 };
2574         char msg[1024];
2575         libzfs_handle_t *hdl = zhp->zpool_hdl;
2576
2577         (void) snprintf(msg, sizeof (msg),
2578             dgettext(TEXT_DOMAIN, "cannot degrade %llu"), guid);
2579
2580         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2581         zc.zc_guid = guid;
2582         zc.zc_cookie = VDEV_STATE_DEGRADED;
2583         zc.zc_obj = aux;
2584
2585         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2586                 return (0);
2587
2588         return (zpool_standard_error(hdl, errno, msg));
2589 }
2590
2591 /*
2592  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2593  * a hot spare.
2594  */
2595 static boolean_t
2596 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2597 {
2598         nvlist_t **child;
2599         uint_t c, children;
2600         char *type;
2601
2602         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2603             &children) == 0) {
2604                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2605                     &type) == 0);
2606
2607                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2608                     children == 2 && child[which] == tgt)
2609                         return (B_TRUE);
2610
2611                 for (c = 0; c < children; c++)
2612                         if (is_replacing_spare(child[c], tgt, which))
2613                                 return (B_TRUE);
2614         }
2615
2616         return (B_FALSE);
2617 }
2618
2619 /*
2620  * Attach new_disk (fully described by nvroot) to old_disk.
2621  * If 'replacing' is specified, the new disk will replace the old one.
2622  */
2623 int
2624 zpool_vdev_attach(zpool_handle_t *zhp,
2625     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2626 {
2627         zfs_cmd_t zc = { 0 };
2628         char msg[1024];
2629         int ret;
2630         nvlist_t *tgt;
2631         boolean_t avail_spare, l2cache, islog;
2632         uint64_t val;
2633         char *newname;
2634         nvlist_t **child;
2635         uint_t children;
2636         nvlist_t *config_root;
2637         libzfs_handle_t *hdl = zhp->zpool_hdl;
2638         boolean_t rootpool = zpool_is_bootable(zhp);
2639
2640         if (replacing)
2641                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2642                     "cannot replace %s with %s"), old_disk, new_disk);
2643         else
2644                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2645                     "cannot attach %s to %s"), new_disk, old_disk);
2646
2647         /*
2648          * If this is a root pool, make sure that we're not attaching an
2649          * EFI labeled device.
2650          */
2651         if (rootpool && pool_uses_efi(nvroot)) {
2652                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2653                     "EFI labeled devices are not supported on root pools."));
2654                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
2655         }
2656
2657         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2658         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2659             &islog)) == 0)
2660                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2661
2662         if (avail_spare)
2663                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2664
2665         if (l2cache)
2666                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2667
2668         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2669         zc.zc_cookie = replacing;
2670
2671         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2672             &child, &children) != 0 || children != 1) {
2673                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2674                     "new device must be a single disk"));
2675                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2676         }
2677
2678         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2679             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2680
2681         if ((newname = zpool_vdev_name(NULL, NULL, child[0], B_FALSE)) == NULL)
2682                 return (-1);
2683
2684         /*
2685          * If the target is a hot spare that has been swapped in, we can only
2686          * replace it with another hot spare.
2687          */
2688         if (replacing &&
2689             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2690             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2691             NULL) == NULL || !avail_spare) &&
2692             is_replacing_spare(config_root, tgt, 1)) {
2693                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2694                     "can only be replaced by another hot spare"));
2695                 free(newname);
2696                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2697         }
2698
2699         free(newname);
2700
2701         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2702                 return (-1);
2703
2704         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2705
2706         zcmd_free_nvlists(&zc);
2707
2708         if (ret == 0) {
2709                 if (rootpool) {
2710                         /*
2711                          * XXX need a better way to prevent user from
2712                          * booting up a half-baked vdev.
2713                          */
2714                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2715                             "sure to wait until resilver is done "
2716                             "before rebooting.\n"));
2717                         (void) fprintf(stderr, "\n");
2718                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "If "
2719                             "you boot from pool '%s', you may need to update\n"
2720                             "boot code on newly attached disk '%s'.\n\n"
2721                             "Assuming you use GPT partitioning and 'da0' is "
2722                             "your new boot disk\n"
2723                             "you may use the following command:\n\n"
2724                             "\tgpart bootcode -b /boot/pmbr -p "
2725                             "/boot/gptzfsboot -i 1 da0\n\n"),
2726                             zhp->zpool_name, new_disk);
2727                 }
2728                 return (0);
2729         }
2730
2731         switch (errno) {
2732         case ENOTSUP:
2733                 /*
2734                  * Can't attach to or replace this type of vdev.
2735                  */
2736                 if (replacing) {
2737                         uint64_t version = zpool_get_prop_int(zhp,
2738                             ZPOOL_PROP_VERSION, NULL);
2739
2740                         if (islog)
2741                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2742                                     "cannot replace a log with a spare"));
2743                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2744                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2745                                     "already in replacing/spare config; wait "
2746                                     "for completion or use 'zpool detach'"));
2747                         else
2748                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2749                                     "cannot replace a replacing device"));
2750                 } else {
2751                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2752                             "can only attach to mirrors and top-level "
2753                             "disks"));
2754                 }
2755                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2756                 break;
2757
2758         case EINVAL:
2759                 /*
2760                  * The new device must be a single disk.
2761                  */
2762                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2763                     "new device must be a single disk"));
2764                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2765                 break;
2766
2767         case EBUSY:
2768                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2769                     new_disk);
2770                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2771                 break;
2772
2773         case EOVERFLOW:
2774                 /*
2775                  * The new device is too small.
2776                  */
2777                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2778                     "device is too small"));
2779                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2780                 break;
2781
2782         case EDOM:
2783                 /*
2784                  * The new device has a different alignment requirement.
2785                  */
2786                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2787                     "devices have different sector alignment"));
2788                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2789                 break;
2790
2791         case ENAMETOOLONG:
2792                 /*
2793                  * The resulting top-level vdev spec won't fit in the label.
2794                  */
2795                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2796                 break;
2797
2798         default:
2799                 (void) zpool_standard_error(hdl, errno, msg);
2800         }
2801
2802         return (-1);
2803 }
2804
2805 /*
2806  * Detach the specified device.
2807  */
2808 int
2809 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2810 {
2811         zfs_cmd_t zc = { 0 };
2812         char msg[1024];
2813         nvlist_t *tgt;
2814         boolean_t avail_spare, l2cache;
2815         libzfs_handle_t *hdl = zhp->zpool_hdl;
2816
2817         (void) snprintf(msg, sizeof (msg),
2818             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2819
2820         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2821         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2822             NULL)) == 0)
2823                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2824
2825         if (avail_spare)
2826                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2827
2828         if (l2cache)
2829                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2830
2831         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2832
2833         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2834                 return (0);
2835
2836         switch (errno) {
2837
2838         case ENOTSUP:
2839                 /*
2840                  * Can't detach from this type of vdev.
2841                  */
2842                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2843                     "applicable to mirror and replacing vdevs"));
2844                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2845                 break;
2846
2847         case EBUSY:
2848                 /*
2849                  * There are no other replicas of this device.
2850                  */
2851                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2852                 break;
2853
2854         default:
2855                 (void) zpool_standard_error(hdl, errno, msg);
2856         }
2857
2858         return (-1);
2859 }
2860
2861 /*
2862  * Find a mirror vdev in the source nvlist.
2863  *
2864  * The mchild array contains a list of disks in one of the top-level mirrors
2865  * of the source pool.  The schild array contains a list of disks that the
2866  * user specified on the command line.  We loop over the mchild array to
2867  * see if any entry in the schild array matches.
2868  *
2869  * If a disk in the mchild array is found in the schild array, we return
2870  * the index of that entry.  Otherwise we return -1.
2871  */
2872 static int
2873 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2874     nvlist_t **schild, uint_t schildren)
2875 {
2876         uint_t mc;
2877
2878         for (mc = 0; mc < mchildren; mc++) {
2879                 uint_t sc;
2880                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2881                     mchild[mc], B_FALSE);
2882
2883                 for (sc = 0; sc < schildren; sc++) {
2884                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2885                             schild[sc], B_FALSE);
2886                         boolean_t result = (strcmp(mpath, spath) == 0);
2887
2888                         free(spath);
2889                         if (result) {
2890                                 free(mpath);
2891                                 return (mc);
2892                         }
2893                 }
2894
2895                 free(mpath);
2896         }
2897
2898         return (-1);
2899 }
2900
2901 /*
2902  * Split a mirror pool.  If newroot points to null, then a new nvlist
2903  * is generated and it is the responsibility of the caller to free it.
2904  */
2905 int
2906 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2907     nvlist_t *props, splitflags_t flags)
2908 {
2909         zfs_cmd_t zc = { 0 };
2910         char msg[1024];
2911         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2912         nvlist_t **varray = NULL, *zc_props = NULL;
2913         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2914         libzfs_handle_t *hdl = zhp->zpool_hdl;
2915         uint64_t vers;
2916         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2917         int retval = 0;
2918
2919         (void) snprintf(msg, sizeof (msg),
2920             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2921
2922         if (!zpool_name_valid(hdl, B_FALSE, newname))
2923                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2924
2925         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2926                 (void) fprintf(stderr, gettext("Internal error: unable to "
2927                     "retrieve pool configuration\n"));
2928                 return (-1);
2929         }
2930
2931         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2932             == 0);
2933         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2934
2935         if (props) {
2936                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2937                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2938                     props, vers, flags, msg)) == NULL)
2939                         return (-1);
2940         }
2941
2942         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2943             &children) != 0) {
2944                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2945                     "Source pool is missing vdev tree"));
2946                 if (zc_props)
2947                         nvlist_free(zc_props);
2948                 return (-1);
2949         }
2950
2951         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2952         vcount = 0;
2953
2954         if (*newroot == NULL ||
2955             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2956             &newchild, &newchildren) != 0)
2957                 newchildren = 0;
2958
2959         for (c = 0; c < children; c++) {
2960                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2961                 char *type;
2962                 nvlist_t **mchild, *vdev;
2963                 uint_t mchildren;
2964                 int entry;
2965
2966                 /*
2967                  * Unlike cache & spares, slogs are stored in the
2968                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2969                  */
2970                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2971                     &is_log);
2972                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2973                     &is_hole);
2974                 if (is_log || is_hole) {
2975                         /*
2976                          * Create a hole vdev and put it in the config.
2977                          */
2978                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2979                                 goto out;
2980                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2981                             VDEV_TYPE_HOLE) != 0)
2982                                 goto out;
2983                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2984                             1) != 0)
2985                                 goto out;
2986                         if (lastlog == 0)
2987                                 lastlog = vcount;
2988                         varray[vcount++] = vdev;
2989                         continue;
2990                 }
2991                 lastlog = 0;
2992                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
2993                     == 0);
2994                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
2995                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2996                             "Source pool must be composed only of mirrors\n"));
2997                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2998                         goto out;
2999                 }
3000
3001                 verify(nvlist_lookup_nvlist_array(child[c],
3002                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
3003
3004                 /* find or add an entry for this top-level vdev */
3005                 if (newchildren > 0 &&
3006                     (entry = find_vdev_entry(zhp, mchild, mchildren,
3007                     newchild, newchildren)) >= 0) {
3008                         /* We found a disk that the user specified. */
3009                         vdev = mchild[entry];
3010                         ++found;
3011                 } else {
3012                         /* User didn't specify a disk for this vdev. */
3013                         vdev = mchild[mchildren - 1];
3014                 }
3015
3016                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
3017                         goto out;
3018         }
3019
3020         /* did we find every disk the user specified? */
3021         if (found != newchildren) {
3022                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
3023                     "include at most one disk from each mirror"));
3024                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
3025                 goto out;
3026         }
3027
3028         /* Prepare the nvlist for populating. */
3029         if (*newroot == NULL) {
3030                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
3031                         goto out;
3032                 freelist = B_TRUE;
3033                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
3034                     VDEV_TYPE_ROOT) != 0)
3035                         goto out;
3036         } else {
3037                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
3038         }
3039
3040         /* Add all the children we found */
3041         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
3042             lastlog == 0 ? vcount : lastlog) != 0)
3043                 goto out;
3044
3045         /*
3046          * If we're just doing a dry run, exit now with success.
3047          */
3048         if (flags.dryrun) {
3049                 memory_err = B_FALSE;
3050                 freelist = B_FALSE;
3051                 goto out;
3052         }
3053
3054         /* now build up the config list & call the ioctl */
3055         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3056                 goto out;
3057
3058         if (nvlist_add_nvlist(newconfig,
3059             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3060             nvlist_add_string(newconfig,
3061             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3062             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3063                 goto out;
3064
3065         /*
3066          * The new pool is automatically part of the namespace unless we
3067          * explicitly export it.
3068          */
3069         if (!flags.import)
3070                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3071         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3072         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3073         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
3074                 goto out;
3075         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
3076                 goto out;
3077
3078         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3079                 retval = zpool_standard_error(hdl, errno, msg);
3080                 goto out;
3081         }
3082
3083         freelist = B_FALSE;
3084         memory_err = B_FALSE;
3085
3086 out:
3087         if (varray != NULL) {
3088                 int v;
3089
3090                 for (v = 0; v < vcount; v++)
3091                         nvlist_free(varray[v]);
3092                 free(varray);
3093         }
3094         zcmd_free_nvlists(&zc);
3095         if (zc_props)
3096                 nvlist_free(zc_props);
3097         if (newconfig)
3098                 nvlist_free(newconfig);
3099         if (freelist) {
3100                 nvlist_free(*newroot);
3101                 *newroot = NULL;
3102         }
3103
3104         if (retval != 0)
3105                 return (retval);
3106
3107         if (memory_err)
3108                 return (no_memory(hdl));
3109
3110         return (0);
3111 }
3112
3113 /*
3114  * Remove the given device.  Currently, this is supported only for hot spares
3115  * and level 2 cache devices.
3116  */
3117 int
3118 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3119 {
3120         zfs_cmd_t zc = { 0 };
3121         char msg[1024];
3122         nvlist_t *tgt;
3123         boolean_t avail_spare, l2cache, islog;
3124         libzfs_handle_t *hdl = zhp->zpool_hdl;
3125         uint64_t version;
3126
3127         (void) snprintf(msg, sizeof (msg),
3128             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3129
3130         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3131         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3132             &islog)) == 0)
3133                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3134         /*
3135          * XXX - this should just go away.
3136          */
3137         if (!avail_spare && !l2cache && !islog) {
3138                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3139                     "only inactive hot spares, cache, top-level, "
3140                     "or log devices can be removed"));
3141                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3142         }
3143
3144         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3145         if (islog && version < SPA_VERSION_HOLES) {
3146                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3147                     "pool must be upgrade to support log removal"));
3148                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
3149         }
3150
3151         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
3152
3153         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3154                 return (0);
3155
3156         return (zpool_standard_error(hdl, errno, msg));
3157 }
3158
3159 /*
3160  * Clear the errors for the pool, or the particular device if specified.
3161  */
3162 int
3163 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3164 {
3165         zfs_cmd_t zc = { 0 };
3166         char msg[1024];
3167         nvlist_t *tgt;
3168         zpool_rewind_policy_t policy;
3169         boolean_t avail_spare, l2cache;
3170         libzfs_handle_t *hdl = zhp->zpool_hdl;
3171         nvlist_t *nvi = NULL;
3172         int error;
3173
3174         if (path)
3175                 (void) snprintf(msg, sizeof (msg),
3176                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3177                     path);
3178         else
3179                 (void) snprintf(msg, sizeof (msg),
3180                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3181                     zhp->zpool_name);
3182
3183         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3184         if (path) {
3185                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3186                     &l2cache, NULL)) == 0)
3187                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
3188
3189                 /*
3190                  * Don't allow error clearing for hot spares.  Do allow
3191                  * error clearing for l2cache devices.
3192                  */
3193                 if (avail_spare)
3194                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
3195
3196                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
3197                     &zc.zc_guid) == 0);
3198         }
3199
3200         zpool_get_rewind_policy(rewindnvl, &policy);
3201         zc.zc_cookie = policy.zrp_request;
3202
3203         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
3204                 return (-1);
3205
3206         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
3207                 return (-1);
3208
3209         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3210             errno == ENOMEM) {
3211                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3212                         zcmd_free_nvlists(&zc);
3213                         return (-1);
3214                 }
3215         }
3216
3217         if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
3218             errno != EPERM && errno != EACCES)) {
3219                 if (policy.zrp_request &
3220                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
3221                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3222                         zpool_rewind_exclaim(hdl, zc.zc_name,
3223                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
3224                             nvi);
3225                         nvlist_free(nvi);
3226                 }
3227                 zcmd_free_nvlists(&zc);
3228                 return (0);
3229         }
3230
3231         zcmd_free_nvlists(&zc);
3232         return (zpool_standard_error(hdl, errno, msg));
3233 }
3234
3235 /*
3236  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3237  */
3238 int
3239 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3240 {
3241         zfs_cmd_t zc = { 0 };
3242         char msg[1024];
3243         libzfs_handle_t *hdl = zhp->zpool_hdl;
3244
3245         (void) snprintf(msg, sizeof (msg),
3246             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3247             guid);
3248
3249         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3250         zc.zc_guid = guid;
3251         zc.zc_cookie = ZPOOL_NO_REWIND;
3252
3253         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
3254                 return (0);
3255
3256         return (zpool_standard_error(hdl, errno, msg));
3257 }
3258
3259 /*
3260  * Change the GUID for a pool.
3261  */
3262 int
3263 zpool_reguid(zpool_handle_t *zhp)
3264 {
3265         char msg[1024];
3266         libzfs_handle_t *hdl = zhp->zpool_hdl;
3267         zfs_cmd_t zc = { 0 };
3268
3269         (void) snprintf(msg, sizeof (msg),
3270             dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3271
3272         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3273         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3274                 return (0);
3275
3276         return (zpool_standard_error(hdl, errno, msg));
3277 }
3278
3279 /*
3280  * Reopen the pool.
3281  */
3282 int
3283 zpool_reopen(zpool_handle_t *zhp)
3284 {
3285         zfs_cmd_t zc = { 0 };
3286         char msg[1024];
3287         libzfs_handle_t *hdl = zhp->zpool_hdl;
3288
3289         (void) snprintf(msg, sizeof (msg),
3290             dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3291             zhp->zpool_name);
3292
3293         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3294         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3295                 return (0);
3296         return (zpool_standard_error(hdl, errno, msg));
3297 }
3298
3299 /*
3300  * Convert from a devid string to a path.
3301  */
3302 static char *
3303 devid_to_path(char *devid_str)
3304 {
3305         ddi_devid_t devid;
3306         char *minor;
3307         char *path;
3308         devid_nmlist_t *list = NULL;
3309         int ret;
3310
3311         if (devid_str_decode(devid_str, &devid, &minor) != 0)
3312                 return (NULL);
3313
3314         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3315
3316         devid_str_free(minor);
3317         devid_free(devid);
3318
3319         if (ret != 0)
3320                 return (NULL);
3321
3322         if ((path = strdup(list[0].devname)) == NULL)
3323                 return (NULL);
3324
3325         devid_free_nmlist(list);
3326
3327         return (path);
3328 }
3329
3330 /*
3331  * Convert from a path to a devid string.
3332  */
3333 static char *
3334 path_to_devid(const char *path)
3335 {
3336 #ifdef have_devid
3337         int fd;
3338         ddi_devid_t devid;
3339         char *minor, *ret;
3340
3341         if ((fd = open(path, O_RDONLY)) < 0)
3342                 return (NULL);
3343
3344         minor = NULL;
3345         ret = NULL;
3346         if (devid_get(fd, &devid) == 0) {
3347                 if (devid_get_minor_name(fd, &minor) == 0)
3348                         ret = devid_str_encode(devid, minor);
3349                 if (minor != NULL)
3350                         devid_str_free(minor);
3351                 devid_free(devid);
3352         }
3353         (void) close(fd);
3354
3355         return (ret);
3356 #else
3357         return (NULL);
3358 #endif
3359 }
3360
3361 /*
3362  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3363  * ignore any failure here, since a common case is for an unprivileged user to
3364  * type 'zpool status', and we'll display the correct information anyway.
3365  */
3366 static void
3367 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3368 {
3369         zfs_cmd_t zc = { 0 };
3370
3371         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3372         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3373         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3374             &zc.zc_guid) == 0);
3375
3376         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3377 }
3378
3379 /*
3380  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3381  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3382  * We also check if this is a whole disk, in which case we strip off the
3383  * trailing 's0' slice name.
3384  *
3385  * This routine is also responsible for identifying when disks have been
3386  * reconfigured in a new location.  The kernel will have opened the device by
3387  * devid, but the path will still refer to the old location.  To catch this, we
3388  * first do a path -> devid translation (which is fast for the common case).  If
3389  * the devid matches, we're done.  If not, we do a reverse devid -> path
3390  * translation and issue the appropriate ioctl() to update the path of the vdev.
3391  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3392  * of these checks.
3393  */
3394 char *
3395 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3396     boolean_t verbose)
3397 {
3398         char *path, *devid;
3399         uint64_t value;
3400         char buf[64];
3401         vdev_stat_t *vs;
3402         uint_t vsc;
3403         int have_stats;
3404         int have_path;
3405
3406         have_stats = nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3407             (uint64_t **)&vs, &vsc) == 0;
3408         have_path = nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0;
3409
3410         /*
3411          * If the device is not currently present, assume it will not
3412          * come back at the same device path.  Display the device by GUID.
3413          */
3414         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &value) == 0 ||
3415             have_path && have_stats && vs->vs_state <= VDEV_STATE_CANT_OPEN) {
3416                 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3417                     &value) == 0);
3418                 (void) snprintf(buf, sizeof (buf), "%llu",
3419                     (u_longlong_t)value);
3420                 path = buf;
3421         } else if (have_path) {
3422
3423                 /*
3424                  * If the device is dead (faulted, offline, etc) then don't
3425                  * bother opening it.  Otherwise we may be forcing the user to
3426                  * open a misbehaving device, which can have undesirable
3427                  * effects.
3428                  */
3429                 if ((have_stats == 0 ||
3430                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3431                     zhp != NULL &&
3432                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3433                         /*
3434                          * Determine if the current path is correct.
3435                          */
3436                         char *newdevid = path_to_devid(path);
3437
3438                         if (newdevid == NULL ||
3439                             strcmp(devid, newdevid) != 0) {
3440                                 char *newpath;
3441
3442                                 if ((newpath = devid_to_path(devid)) != NULL) {
3443                                         /*
3444                                          * Update the path appropriately.
3445                                          */
3446                                         set_path(zhp, nv, newpath);
3447                                         if (nvlist_add_string(nv,
3448                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3449                                                 verify(nvlist_lookup_string(nv,
3450                                                     ZPOOL_CONFIG_PATH,
3451                                                     &path) == 0);
3452                                         free(newpath);
3453                                 }
3454                         }
3455
3456                         if (newdevid)
3457                                 devid_str_free(newdevid);
3458                 }
3459
3460 #ifdef sun
3461                 if (strncmp(path, "/dev/dsk/", 9) == 0)
3462                         path += 9;
3463
3464                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
3465                     &value) == 0 && value) {
3466                         int pathlen = strlen(path);
3467                         char *tmp = zfs_strdup(hdl, path);
3468
3469                         /*
3470                          * If it starts with c#, and ends with "s0", chop
3471                          * the "s0" off, or if it ends with "s0/old", remove
3472                          * the "s0" from the middle.
3473                          */
3474                         if (CTD_CHECK(tmp)) {
3475                                 if (strcmp(&tmp[pathlen - 2], "s0") == 0) {
3476                                         tmp[pathlen - 2] = '\0';
3477                                 } else if (pathlen > 6 &&
3478                                     strcmp(&tmp[pathlen - 6], "s0/old") == 0) {
3479                                         (void) strcpy(&tmp[pathlen - 6],
3480                                             "/old");
3481                                 }
3482                         }
3483                         return (tmp);
3484                 }
3485 #else   /* !sun */
3486                 if (strncmp(path, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
3487                         path += sizeof(_PATH_DEV) - 1;
3488 #endif  /* !sun */
3489         } else {
3490                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3491
3492                 /*
3493                  * If it's a raidz device, we need to stick in the parity level.
3494                  */
3495                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3496                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3497                             &value) == 0);
3498                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3499                             (u_longlong_t)value);
3500                         path = buf;
3501                 }
3502
3503                 /*
3504                  * We identify each top-level vdev by using a <type-id>
3505                  * naming convention.
3506                  */
3507                 if (verbose) {
3508                         uint64_t id;
3509
3510                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3511                             &id) == 0);
3512                         (void) snprintf(buf, sizeof (buf), "%s-%llu", path,
3513                             (u_longlong_t)id);
3514                         path = buf;
3515                 }
3516         }
3517
3518         return (zfs_strdup(hdl, path));
3519 }
3520
3521 static int
3522 zbookmark_compare(const void *a, const void *b)
3523 {
3524         return (memcmp(a, b, sizeof (zbookmark_phys_t)));
3525 }
3526
3527 /*
3528  * Retrieve the persistent error log, uniquify the members, and return to the
3529  * caller.
3530  */
3531 int
3532 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3533 {
3534         zfs_cmd_t zc = { 0 };
3535         uint64_t count;
3536         zbookmark_phys_t *zb = NULL;
3537         int i;
3538
3539         /*
3540          * Retrieve the raw error list from the kernel.  If the number of errors
3541          * has increased, allocate more space and continue until we get the
3542          * entire list.
3543          */
3544         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3545             &count) == 0);
3546         if (count == 0)
3547                 return (0);
3548         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3549             count * sizeof (zbookmark_phys_t))) == (uintptr_t)NULL)
3550                 return (-1);
3551         zc.zc_nvlist_dst_size = count;
3552         (void) strcpy(zc.zc_name, zhp->zpool_name);
3553         for (;;) {
3554                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3555                     &zc) != 0) {
3556                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3557                         if (errno == ENOMEM) {
3558                                 void *dst;
3559
3560                                 count = zc.zc_nvlist_dst_size;
3561                                 dst = zfs_alloc(zhp->zpool_hdl, count *
3562                                     sizeof (zbookmark_phys_t));
3563                                 if (dst == NULL)
3564                                         return (-1);
3565                                 zc.zc_nvlist_dst = (uintptr_t)dst;
3566                         } else {
3567                                 return (-1);
3568                         }
3569                 } else {
3570                         break;
3571                 }
3572         }
3573
3574         /*
3575          * Sort the resulting bookmarks.  This is a little confusing due to the
3576          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3577          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3578          * _not_ copied as part of the process.  So we point the start of our
3579          * array appropriate and decrement the total number of elements.
3580          */
3581         zb = ((zbookmark_phys_t *)(uintptr_t)zc.zc_nvlist_dst) +
3582             zc.zc_nvlist_dst_size;
3583         count -= zc.zc_nvlist_dst_size;
3584
3585         qsort(zb, count, sizeof (zbookmark_phys_t), zbookmark_compare);
3586
3587         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3588
3589         /*
3590          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3591          */
3592         for (i = 0; i < count; i++) {
3593                 nvlist_t *nv;
3594
3595                 /* ignoring zb_blkid and zb_level for now */
3596                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3597                     zb[i-1].zb_object == zb[i].zb_object)
3598                         continue;
3599
3600                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3601                         goto nomem;
3602                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3603                     zb[i].zb_objset) != 0) {
3604                         nvlist_free(nv);
3605                         goto nomem;
3606                 }
3607                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3608                     zb[i].zb_object) != 0) {
3609                         nvlist_free(nv);
3610                         goto nomem;
3611                 }
3612                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3613                         nvlist_free(nv);
3614                         goto nomem;
3615                 }
3616                 nvlist_free(nv);
3617         }
3618
3619         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3620         return (0);
3621
3622 nomem:
3623         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3624         return (no_memory(zhp->zpool_hdl));
3625 }
3626
3627 /*
3628  * Upgrade a ZFS pool to the latest on-disk version.
3629  */
3630 int
3631 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3632 {
3633         zfs_cmd_t zc = { 0 };
3634         libzfs_handle_t *hdl = zhp->zpool_hdl;
3635
3636         (void) strcpy(zc.zc_name, zhp->zpool_name);
3637         zc.zc_cookie = new_version;
3638
3639         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3640                 return (zpool_standard_error_fmt(hdl, errno,
3641                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3642                     zhp->zpool_name));
3643         return (0);
3644 }
3645
3646 void
3647 zfs_save_arguments(int argc, char **argv, char *string, int len)
3648 {
3649         (void) strlcpy(string, basename(argv[0]), len);
3650         for (int i = 1; i < argc; i++) {
3651                 (void) strlcat(string, " ", len);
3652                 (void) strlcat(string, argv[i], len);
3653         }
3654 }
3655
3656 int
3657 zpool_log_history(libzfs_handle_t *hdl, const char *message)
3658 {
3659         zfs_cmd_t zc = { 0 };
3660         nvlist_t *args;
3661         int err;
3662
3663         args = fnvlist_alloc();
3664         fnvlist_add_string(args, "message", message);
3665         err = zcmd_write_src_nvlist(hdl, &zc, args);
3666         if (err == 0)
3667                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_LOG_HISTORY, &zc);
3668         nvlist_free(args);
3669         zcmd_free_nvlists(&zc);
3670         return (err);
3671 }
3672
3673 /*
3674  * Perform ioctl to get some command history of a pool.
3675  *
3676  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3677  * logical offset of the history buffer to start reading from.
3678  *
3679  * Upon return, 'off' is the next logical offset to read from and
3680  * 'len' is the actual amount of bytes read into 'buf'.
3681  */
3682 static int
3683 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3684 {
3685         zfs_cmd_t zc = { 0 };
3686         libzfs_handle_t *hdl = zhp->zpool_hdl;
3687
3688         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3689
3690         zc.zc_history = (uint64_t)(uintptr_t)buf;
3691         zc.zc_history_len = *len;
3692         zc.zc_history_offset = *off;
3693
3694         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3695                 switch (errno) {
3696                 case EPERM:
3697                         return (zfs_error_fmt(hdl, EZFS_PERM,
3698                             dgettext(TEXT_DOMAIN,
3699                             "cannot show history for pool '%s'"),
3700                             zhp->zpool_name));
3701                 case ENOENT:
3702                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3703                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3704                             "'%s'"), zhp->zpool_name));
3705                 case ENOTSUP:
3706                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3707                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3708                             "'%s', pool must be upgraded"), zhp->zpool_name));
3709                 default:
3710                         return (zpool_standard_error_fmt(hdl, errno,
3711                             dgettext(TEXT_DOMAIN,
3712                             "cannot get history for '%s'"), zhp->zpool_name));
3713                 }
3714         }
3715
3716         *len = zc.zc_history_len;
3717         *off = zc.zc_history_offset;
3718
3719         return (0);
3720 }
3721
3722 /*
3723  * Process the buffer of nvlists, unpacking and storing each nvlist record
3724  * into 'records'.  'leftover' is set to the number of bytes that weren't
3725  * processed as there wasn't a complete record.
3726  */
3727 int
3728 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3729     nvlist_t ***records, uint_t *numrecords)
3730 {
3731         uint64_t reclen;
3732         nvlist_t *nv;
3733         int i;
3734
3735         while (bytes_read > sizeof (reclen)) {
3736
3737                 /* get length of packed record (stored as little endian) */
3738                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3739                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3740
3741                 if (bytes_read < sizeof (reclen) + reclen)
3742                         break;
3743
3744                 /* unpack record */
3745                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3746                         return (ENOMEM);
3747                 bytes_read -= sizeof (reclen) + reclen;
3748                 buf += sizeof (reclen) + reclen;
3749
3750                 /* add record to nvlist array */
3751                 (*numrecords)++;
3752                 if (ISP2(*numrecords + 1)) {
3753                         *records = realloc(*records,
3754                             *numrecords * 2 * sizeof (nvlist_t *));
3755                 }
3756                 (*records)[*numrecords - 1] = nv;
3757         }
3758
3759         *leftover = bytes_read;
3760         return (0);
3761 }
3762
3763 /* from spa_history.c: spa_history_create_obj() */
3764 #define HIS_BUF_LEN_DEF (128 << 10)
3765 #define HIS_BUF_LEN_MAX (1 << 30)
3766
3767 /*
3768  * Retrieve the command history of a pool.
3769  */
3770 int
3771 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3772 {
3773         char *buf = NULL;
3774         uint64_t bufsize = HIS_BUF_LEN_DEF;
3775         uint64_t off = 0;
3776         nvlist_t **records = NULL;
3777         uint_t numrecords = 0;
3778         int err, i;
3779
3780         if ((buf = malloc(bufsize)) == NULL)
3781                 return (ENOMEM);
3782         do {
3783                 uint64_t bytes_read = bufsize;
3784                 uint64_t leftover;
3785
3786                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3787                         break;
3788
3789                 /* if nothing else was read in, we're at EOF, just return */
3790                 if (bytes_read == 0)
3791                         break;
3792
3793                 if ((err = zpool_history_unpack(buf, bytes_read,
3794                     &leftover, &records, &numrecords)) != 0)
3795                         break;
3796                 off -= leftover;
3797
3798                 /*
3799                  * If the history block is too big, double the buffer
3800                  * size and try again.
3801                  */
3802                 if (leftover == bytes_read) {
3803                         free(buf);
3804                         buf = NULL;
3805
3806                         bufsize <<= 1;
3807                         if ((bufsize >= HIS_BUF_LEN_MAX) ||
3808                             ((buf = malloc(bufsize)) == NULL)) {
3809                                 err = ENOMEM;
3810                                 break;
3811                         }
3812                 }
3813
3814                 /* CONSTCOND */
3815         } while (1);
3816         free(buf);
3817
3818         if (!err) {
3819                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3820                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3821                     records, numrecords) == 0);
3822         }
3823         for (i = 0; i < numrecords; i++)
3824                 nvlist_free(records[i]);
3825         free(records);
3826
3827         return (err);
3828 }
3829
3830 void
3831 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
3832     char *pathname, size_t len)
3833 {
3834         zfs_cmd_t zc = { 0 };
3835         boolean_t mounted = B_FALSE;
3836         char *mntpnt = NULL;
3837         char dsname[MAXNAMELEN];
3838
3839         if (dsobj == 0) {
3840                 /* special case for the MOS */
3841                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>", obj);
3842                 return;
3843         }
3844
3845         /* get the dataset's name */
3846         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3847         zc.zc_obj = dsobj;
3848         if (ioctl(zhp->zpool_hdl->libzfs_fd,
3849             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
3850                 /* just write out a path of two object numbers */
3851                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
3852                     dsobj, obj);
3853                 return;
3854         }
3855         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
3856
3857         /* find out if the dataset is mounted */
3858         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
3859
3860         /* get the corrupted object's path */
3861         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
3862         zc.zc_obj = obj;
3863         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
3864             &zc) == 0) {
3865                 if (mounted) {
3866                         (void) snprintf(pathname, len, "%s%s", mntpnt,
3867                             zc.zc_value);
3868                 } else {
3869                         (void) snprintf(pathname, len, "%s:%s",
3870                             dsname, zc.zc_value);
3871                 }
3872         } else {
3873                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname, obj);
3874         }
3875         free(mntpnt);
3876 }
3877
3878 #ifdef sun
3879 /*
3880  * Read the EFI label from the config, if a label does not exist then
3881  * pass back the error to the caller. If the caller has passed a non-NULL
3882  * diskaddr argument then we set it to the starting address of the EFI
3883  * partition.
3884  */
3885 static int
3886 read_efi_label(nvlist_t *config, diskaddr_t *sb)
3887 {
3888         char *path;
3889         int fd;
3890         char diskname[MAXPATHLEN];
3891         int err = -1;
3892
3893         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
3894                 return (err);
3895
3896         (void) snprintf(diskname, sizeof (diskname), "%s%s", RDISK_ROOT,
3897             strrchr(path, '/'));
3898         if ((fd = open(diskname, O_RDONLY|O_NDELAY)) >= 0) {
3899                 struct dk_gpt *vtoc;
3900
3901                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
3902                         if (sb != NULL)
3903                                 *sb = vtoc->efi_parts[0].p_start;
3904                         efi_free(vtoc);
3905                 }
3906                 (void) close(fd);
3907         }
3908         return (err);
3909 }
3910
3911 /*
3912  * determine where a partition starts on a disk in the current
3913  * configuration
3914  */
3915 static diskaddr_t
3916 find_start_block(nvlist_t *config)
3917 {
3918         nvlist_t **child;
3919         uint_t c, children;
3920         diskaddr_t sb = MAXOFFSET_T;
3921         uint64_t wholedisk;
3922
3923         if (nvlist_lookup_nvlist_array(config,
3924             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
3925                 if (nvlist_lookup_uint64(config,
3926                     ZPOOL_CONFIG_WHOLE_DISK,
3927                     &wholedisk) != 0 || !wholedisk) {
3928                         return (MAXOFFSET_T);
3929                 }
3930                 if (read_efi_label(config, &sb) < 0)
3931                         sb = MAXOFFSET_T;
3932                 return (sb);
3933         }
3934
3935         for (c = 0; c < children; c++) {
3936                 sb = find_start_block(child[c]);
3937                 if (sb != MAXOFFSET_T) {
3938                         return (sb);
3939                 }
3940         }
3941         return (MAXOFFSET_T);
3942 }
3943 #endif /* sun */
3944
3945 /*
3946  * Label an individual disk.  The name provided is the short name,
3947  * stripped of any leading /dev path.
3948  */
3949 int
3950 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, const char *name)
3951 {
3952 #ifdef sun
3953         char path[MAXPATHLEN];
3954         struct dk_gpt *vtoc;
3955         int fd;
3956         size_t resv = EFI_MIN_RESV_SIZE;
3957         uint64_t slice_size;
3958         diskaddr_t start_block;
3959         char errbuf[1024];
3960
3961         /* prepare an error message just in case */
3962         (void) snprintf(errbuf, sizeof (errbuf),
3963             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
3964
3965         if (zhp) {
3966                 nvlist_t *nvroot;
3967
3968                 if (zpool_is_bootable(zhp)) {
3969                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3970                             "EFI labeled devices are not supported on root "
3971                             "pools."));
3972                         return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
3973                 }
3974
3975                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
3976                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3977
3978                 if (zhp->zpool_start_block == 0)
3979                         start_block = find_start_block(nvroot);
3980                 else
3981                         start_block = zhp->zpool_start_block;
3982                 zhp->zpool_start_block = start_block;
3983         } else {
3984                 /* new pool */
3985                 start_block = NEW_START_BLOCK;
3986         }
3987
3988         (void) snprintf(path, sizeof (path), "%s/%s%s", RDISK_ROOT, name,
3989             BACKUP_SLICE);
3990
3991         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
3992                 /*
3993                  * This shouldn't happen.  We've long since verified that this
3994                  * is a valid device.
3995                  */
3996                 zfs_error_aux(hdl,
3997                     dgettext(TEXT_DOMAIN, "unable to open device"));
3998                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
3999         }
4000
4001         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
4002                 /*
4003                  * The only way this can fail is if we run out of memory, or we
4004                  * were unable to read the disk's capacity
4005                  */
4006                 if (errno == ENOMEM)
4007                         (void) no_memory(hdl);
4008
4009                 (void) close(fd);
4010                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4011                     "unable to read disk capacity"), name);
4012
4013                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
4014         }
4015
4016         slice_size = vtoc->efi_last_u_lba + 1;
4017         slice_size -= EFI_MIN_RESV_SIZE;
4018         if (start_block == MAXOFFSET_T)
4019                 start_block = NEW_START_BLOCK;
4020         slice_size -= start_block;
4021
4022         vtoc->efi_parts[0].p_start = start_block;
4023         vtoc->efi_parts[0].p_size = slice_size;
4024
4025         /*
4026          * Why we use V_USR: V_BACKUP confuses users, and is considered
4027          * disposable by some EFI utilities (since EFI doesn't have a backup
4028          * slice).  V_UNASSIGNED is supposed to be used only for zero size
4029          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
4030          * etc. were all pretty specific.  V_USR is as close to reality as we
4031          * can get, in the absence of V_OTHER.
4032          */
4033         vtoc->efi_parts[0].p_tag = V_USR;
4034         (void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
4035
4036         vtoc->efi_parts[8].p_start = slice_size + start_block;
4037         vtoc->efi_parts[8].p_size = resv;
4038         vtoc->efi_parts[8].p_tag = V_RESERVED;
4039
4040         if (efi_write(fd, vtoc) != 0) {
4041                 /*
4042                  * Some block drivers (like pcata) may not support EFI
4043                  * GPT labels.  Print out a helpful error message dir-
4044                  * ecting the user to manually label the disk and give
4045                  * a specific slice.
4046                  */
4047                 (void) close(fd);
4048                 efi_free(vtoc);
4049
4050                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4051                     "try using fdisk(1M) and then provide a specific slice"));
4052                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4053         }
4054
4055         (void) close(fd);
4056         efi_free(vtoc);
4057 #endif /* sun */
4058         return (0);
4059 }
4060
4061 static boolean_t
4062 supported_dump_vdev_type(libzfs_handle_t *hdl, nvlist_t *config, char *errbuf)
4063 {
4064         char *type;
4065         nvlist_t **child;
4066         uint_t children, c;
4067
4068         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_TYPE, &type) == 0);
4069         if (strcmp(type, VDEV_TYPE_FILE) == 0 ||
4070             strcmp(type, VDEV_TYPE_HOLE) == 0 ||
4071             strcmp(type, VDEV_TYPE_MISSING) == 0) {
4072                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4073                     "vdev type '%s' is not supported"), type);
4074                 (void) zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf);
4075                 return (B_FALSE);
4076         }
4077         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
4078             &child, &children) == 0) {
4079                 for (c = 0; c < children; c++) {
4080                         if (!supported_dump_vdev_type(hdl, child[c], errbuf))
4081                                 return (B_FALSE);
4082                 }
4083         }
4084         return (B_TRUE);
4085 }
4086
4087 /*
4088  * Check if this zvol is allowable for use as a dump device; zero if
4089  * it is, > 0 if it isn't, < 0 if it isn't a zvol.
4090  *
4091  * Allowable storage configurations include mirrors, all raidz variants, and
4092  * pools with log, cache, and spare devices.  Pools which are backed by files or
4093  * have missing/hole vdevs are not suitable.
4094  */
4095 int
4096 zvol_check_dump_config(char *arg)
4097 {
4098         zpool_handle_t *zhp = NULL;
4099         nvlist_t *config, *nvroot;
4100         char *p, *volname;
4101         nvlist_t **top;
4102         uint_t toplevels;
4103         libzfs_handle_t *hdl;
4104         char errbuf[1024];
4105         char poolname[ZPOOL_MAXNAMELEN];
4106         int pathlen = strlen(ZVOL_FULL_DEV_DIR);
4107         int ret = 1;
4108
4109         if (strncmp(arg, ZVOL_FULL_DEV_DIR, pathlen)) {
4110                 return (-1);
4111         }
4112
4113         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
4114             "dump is not supported on device '%s'"), arg);
4115
4116         if ((hdl = libzfs_init()) == NULL)
4117                 return (1);
4118         libzfs_print_on_error(hdl, B_TRUE);
4119
4120         volname = arg + pathlen;
4121
4122         /* check the configuration of the pool */
4123         if ((p = strchr(volname, '/')) == NULL) {
4124                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4125                     "malformed dataset name"));
4126                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4127                 return (1);
4128         } else if (p - volname >= ZFS_MAXNAMELEN) {
4129                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4130                     "dataset name is too long"));
4131                 (void) zfs_error(hdl, EZFS_NAMETOOLONG, errbuf);
4132                 return (1);
4133         } else {
4134                 (void) strncpy(poolname, volname, p - volname);
4135                 poolname[p - volname] = '\0';
4136         }
4137
4138         if ((zhp = zpool_open(hdl, poolname)) == NULL) {
4139                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4140                     "could not open pool '%s'"), poolname);
4141                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
4142                 goto out;
4143         }
4144         config = zpool_get_config(zhp, NULL);
4145         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4146             &nvroot) != 0) {
4147                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4148                     "could not obtain vdev configuration for  '%s'"), poolname);
4149                 (void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
4150                 goto out;
4151         }
4152
4153         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4154             &top, &toplevels) == 0);
4155
4156         if (!supported_dump_vdev_type(hdl, top[0], errbuf)) {
4157                 goto out;
4158         }
4159         ret = 0;
4160
4161 out:
4162         if (zhp)
4163                 zpool_close(zhp);
4164         libzfs_fini(hdl);
4165         return (ret);
4166 }