lib/libzfs/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  */
  25
  26 #include <ctype.h>
  27 #include <errno.h>
  28 #include <devid.h>
  29 #include <fcntl.h>
  30 #include <libintl.h>
  31 #include <stdio.h>
  32 #include <stdlib.h>
  33 #include <strings.h>
  34 #include <unistd.h>
  35 #include <sys/efi_partition.h>
  36 #include <sys/vtoc.h>
  37 #include <sys/zfs_ioctl.h>
  38 #include <dlfcn.h>
  39
  40 #include "zfs_namecheck.h"
  41 #include "zfs_prop.h"
  42 #include "libzfs_impl.h"
  43 #include "zfs_comutil.h"
  44
  45 static int read_efi_label(nvlist_t *config, diskaddr_t *sb);
  46
  47 #define DISK_ROOT       "/dev/dsk"
  48 #define RDISK_ROOT      "/dev/rdsk"
  49 #define BACKUP_SLICE    "s2"
  50
  51 typedef struct prop_flags {
  52         int create:1;   /* Validate property on creation */
  53         int import:1;   /* Validate property on import */
  54 } prop_flags_t;
  55
  56 /*
  57  * ====================================================================
  58  *   zpool property functions
  59  * ====================================================================
  60  */
  61
  62 static int
  63 zpool_get_all_props(zpool_handle_t *zhp)
  64 {
  65         zfs_cmd_t zc = { 0 };
  66         libzfs_handle_t *hdl = zhp->zpool_hdl;
  67
  68         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
  69
  70         if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
  71                 return (-1);
  72
  73         while (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
  74                 if (errno == ENOMEM) {
  75                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
  76                                 zcmd_free_nvlists(&zc);
  77                                 return (-1);
  78                         }
  79                 } else {
  80                         zcmd_free_nvlists(&zc);
  81                         return (-1);
  82                 }
  83         }
  84
  85         if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
  86                 zcmd_free_nvlists(&zc);
  87                 return (-1);
  88         }
  89
  90         zcmd_free_nvlists(&zc);
  91
  92         return (0);
  93 }
  94
  95 static int
  96 zpool_props_refresh(zpool_handle_t *zhp)
  97 {
  98         nvlist_t *old_props;
  99
 100         old_props = zhp->zpool_props;
 101
 102         if (zpool_get_all_props(zhp) != 0)
 103                 return (-1);
 104
 105         nvlist_free(old_props);
 106         return (0);
 107 }
 108
 109 static char *
 110 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
 111     zprop_source_t *src)
 112 {
 113         nvlist_t *nv, *nvl;
 114         uint64_t ival;
 115         char *value;
 116         zprop_source_t source;
 117
 118         nvl = zhp->zpool_props;
 119         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 120                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &ival) == 0);
 121                 source = ival;
 122                 verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
 123         } else {
 124                 source = ZPROP_SRC_DEFAULT;
 125                 if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
 126                         value = "-";
 127         }
 128
 129         if (src)
 130                 *src = source;
 131
 132         return (value);
 133 }
 134
 135 uint64_t
 136 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
 137 {
 138         nvlist_t *nv, *nvl;
 139         uint64_t value;
 140         zprop_source_t source;
 141
 142         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
 143                 /*
 144                  * zpool_get_all_props() has most likely failed because
 145                  * the pool is faulted, but if all we need is the top level
 146                  * vdev's guid then get it from the zhp config nvlist.
 147                  */
 148                 if ((prop == ZPOOL_PROP_GUID) &&
 149                     (nvlist_lookup_nvlist(zhp->zpool_config,
 150                     ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
 151                     (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
 152                     == 0)) {
 153                         return (value);
 154                 }
 155                 return (zpool_prop_default_numeric(prop));
 156         }
 157
 158         nvl = zhp->zpool_props;
 159         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 160                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &value) == 0);
 161                 source = value;
 162                 verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
 163         } else {
 164                 source = ZPROP_SRC_DEFAULT;
 165                 value = zpool_prop_default_numeric(prop);
 166         }
 167
 168         if (src)
 169                 *src = source;
 170
 171         return (value);
 172 }
 173
 174 /*
 175  * Map VDEV STATE to printed strings.
 176  */
 177 char *
 178 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
 179 {
 180         switch (state) {
 181         case VDEV_STATE_CLOSED:
 182         case VDEV_STATE_OFFLINE:
 183                 return (gettext("OFFLINE"));
 184         case VDEV_STATE_REMOVED:
 185                 return (gettext("REMOVED"));
 186         case VDEV_STATE_CANT_OPEN:
 187                 if (aux == VDEV_AUX_CORRUPT_DATA || aux == VDEV_AUX_BAD_LOG)
 188                         return (gettext("FAULTED"));
 189                 else if (aux == VDEV_AUX_SPLIT_POOL)
 190                         return (gettext("SPLIT"));
 191                 else
 192                         return (gettext("UNAVAIL"));
 193         case VDEV_STATE_FAULTED:
 194                 return (gettext("FAULTED"));
 195         case VDEV_STATE_DEGRADED:
 196                 return (gettext("DEGRADED"));
 197         case VDEV_STATE_HEALTHY:
 198                 return (gettext("ONLINE"));
 199         }
 200
 201         return (gettext("UNKNOWN"));
 202 }
 203
 204 /*
 205  * Get a zpool property value for 'prop' and return the value in
 206  * a pre-allocated buffer.
 207  */
 208 int
 209 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf, size_t len,
 210     zprop_source_t *srctype)
 211 {
 212         uint64_t intval;
 213         const char *strval;
 214         zprop_source_t src = ZPROP_SRC_NONE;
 215         nvlist_t *nvroot;
 216         vdev_stat_t *vs;
 217         uint_t vsc;
 218
 219         if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
 220                 switch (prop) {
 221                 case ZPOOL_PROP_NAME:
 222                         (void) strlcpy(buf, zpool_get_name(zhp), len);
 223                         break;
 224
 225                 case ZPOOL_PROP_HEALTH:
 226                         (void) strlcpy(buf, "FAULTED", len);
 227                         break;
 228
 229                 case ZPOOL_PROP_GUID:
 230                         intval = zpool_get_prop_int(zhp, prop, &src);
 231                         (void) snprintf(buf, len, "%llu", intval);
 232                         break;
 233
 234                 case ZPOOL_PROP_ALTROOT:
 235                 case ZPOOL_PROP_CACHEFILE:
 236                         if (zhp->zpool_props != NULL ||
 237                             zpool_get_all_props(zhp) == 0) {
 238                                 (void) strlcpy(buf,
 239                                     zpool_get_prop_string(zhp, prop, &src),
 240                                     len);
 241                                 if (srctype != NULL)
 242                                         *srctype = src;
 243                                 return (0);
 244                         }
 245                         /* FALLTHROUGH */
 246                 default:
 247                         (void) strlcpy(buf, "-", len);
 248                         break;
 249                 }
 250
 251                 if (srctype != NULL)
 252                         *srctype = src;
 253                 return (0);
 254         }
 255
 256         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
 257             prop != ZPOOL_PROP_NAME)
 258                 return (-1);
 259
 260         switch (zpool_prop_get_type(prop)) {
 261         case PROP_TYPE_STRING:
 262                 (void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
 263                     len);
 264                 break;
 265
 266         case PROP_TYPE_NUMBER:
 267                 intval = zpool_get_prop_int(zhp, prop, &src);
 268
 269                 switch (prop) {
 270                 case ZPOOL_PROP_SIZE:
 271                 case ZPOOL_PROP_ALLOCATED:
 272                 case ZPOOL_PROP_FREE:
 273                         (void) zfs_nicenum(intval, buf, len);
 274                         break;
 275
 276                 case ZPOOL_PROP_CAPACITY:
 277                         (void) snprintf(buf, len, "%llu%%",
 278                             (u_longlong_t)intval);
 279                         break;
 280
 281                 case ZPOOL_PROP_DEDUPRATIO:
 282                         (void) snprintf(buf, len, "%llu.%02llux",
 283                             (u_longlong_t)(intval / 100),
 284                             (u_longlong_t)(intval % 100));
 285                         break;
 286
 287                 case ZPOOL_PROP_HEALTH:
 288                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 289                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 290                         verify(nvlist_lookup_uint64_array(nvroot,
 291                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
 292                             == 0);
 293
 294                         (void) strlcpy(buf, zpool_state_to_name(intval,
 295                             vs->vs_aux), len);
 296                         break;
 297                 default:
 298                         (void) snprintf(buf, len, "%llu", intval);
 299                 }
 300                 break;
 301
 302         case PROP_TYPE_INDEX:
 303                 intval = zpool_get_prop_int(zhp, prop, &src);
 304                 if (zpool_prop_index_to_string(prop, intval, &strval)
 305                     != 0)
 306                         return (-1);
 307                 (void) strlcpy(buf, strval, len);
 308                 break;
 309
 310         default:
 311                 abort();
 312         }
 313
 314         if (srctype)
 315                 *srctype = src;
 316
 317         return (0);
 318 }
 319
 320 /*
 321  * Check if the bootfs name has the same pool name as it is set to.
 322  * Assuming bootfs is a valid dataset name.
 323  */
 324 static boolean_t
 325 bootfs_name_valid(const char *pool, char *bootfs)
 326 {
 327         int len = strlen(pool);
 328
 329         if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
 330                 return (B_FALSE);
 331
 332         if (strncmp(pool, bootfs, len) == 0 &&
 333             (bootfs[len] == '/' || bootfs[len] == '\0'))
 334                 return (B_TRUE);
 335
 336         return (B_FALSE);
 337 }
 338
 339 /*
 340  * Inspect the configuration to determine if any of the devices contain
 341  * an EFI label.
 342  */
 343 static boolean_t
 344 pool_uses_efi(nvlist_t *config)
 345 {
 346         nvlist_t **child;
 347         uint_t c, children;
 348
 349         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
 350             &child, &children) != 0)
 351                 return (read_efi_label(config, NULL) >= 0);
 352
 353         for (c = 0; c < children; c++) {
 354                 if (pool_uses_efi(child[c]))
 355                         return (B_TRUE);
 356         }
 357         return (B_FALSE);
 358 }
 359
 360 static boolean_t
 361 pool_is_bootable(zpool_handle_t *zhp)
 362 {
 363         char bootfs[ZPOOL_MAXNAMELEN];
 364
 365         return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
 366             sizeof (bootfs), NULL) == 0 && strncmp(bootfs, "-",
 367             sizeof (bootfs)) != 0);
 368 }
 369
 370
 371 /*
 372  * Given an nvlist of zpool properties to be set, validate that they are
 373  * correct, and parse any numeric properties (index, boolean, etc) if they are
 374  * specified as strings.
 375  */
 376 static nvlist_t *
 377 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 378     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
 379 {
 380         nvpair_t *elem;
 381         nvlist_t *retprops;
 382         zpool_prop_t prop;
 383         char *strval;
 384         uint64_t intval;
 385         char *slash;
 386         struct stat64 statbuf;
 387         zpool_handle_t *zhp;
 388         nvlist_t *nvroot;
 389
 390         if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
 391                 (void) no_memory(hdl);
 392                 return (NULL);
 393         }
 394
 395         elem = NULL;
 396         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 397                 const char *propname = nvpair_name(elem);
 398
 399                 /*
 400                  * Make sure this property is valid and applies to this type.
 401                  */
 402                 if ((prop = zpool_name_to_prop(propname)) == ZPROP_INVAL) {
 403                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 404                             "invalid property '%s'"), propname);
 405                         (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 406                         goto error;
 407                 }
 408
 409                 if (zpool_prop_readonly(prop)) {
 410                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 411                             "is readonly"), propname);
 412                         (void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
 413                         goto error;
 414                 }
 415
 416                 if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
 417                     &strval, &intval, errbuf) != 0)
 418                         goto error;
 419
 420                 /*
 421                  * Perform additional checking for specific properties.
 422                  */
 423                 switch (prop) {
 424                 case ZPOOL_PROP_VERSION:
 425                         if (intval < version || intval > SPA_VERSION) {
 426                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 427                                     "property '%s' number %d is invalid."),
 428                                     propname, intval);
 429                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 430                                 goto error;
 431                         }
 432                         break;
 433
 434                 case ZPOOL_PROP_BOOTFS:
 435                         if (flags.create || flags.import) {
 436                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 437                                     "property '%s' cannot be set at creation "
 438                                     "or import time"), propname);
 439                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 440                                 goto error;
 441                         }
 442
 443                         if (version < SPA_VERSION_BOOTFS) {
 444                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 445                                     "pool must be upgraded to support "
 446                                     "'%s' property"), propname);
 447                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 448                                 goto error;
 449                         }
 450
 451                         /*
 452                          * bootfs property value has to be a dataset name and
 453                          * the dataset has to be in the same pool as it sets to.
 454                          */
 455                         if (strval[0] != '\0' && !bootfs_name_valid(poolname,
 456                             strval)) {
 457                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 458                                     "is an invalid name"), strval);
 459                                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
 460                                 goto error;
 461                         }
 462
 463                         if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
 464                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 465                                     "could not open pool '%s'"), poolname);
 466                                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
 467                                 goto error;
 468                         }
 469                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 470                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 471
 472                         /*
 473                          * bootfs property cannot be set on a disk which has
 474                          * been EFI labeled.
 475                          */
 476                         if (pool_uses_efi(nvroot)) {
 477                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 478                                     "property '%s' not supported on "
 479                                     "EFI labeled devices"), propname);
 480                                 (void) zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf);
 481                                 zpool_close(zhp);
 482                                 goto error;
 483                         }
 484                         zpool_close(zhp);
 485                         break;
 486
 487                 case ZPOOL_PROP_ALTROOT:
 488                         if (!flags.create && !flags.import) {
 489                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 490                                     "property '%s' can only be set during pool "
 491                                     "creation or import"), propname);
 492                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 493                                 goto error;
 494                         }
 495
 496                         if (strval[0] != '/') {
 497                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 498                                     "bad alternate root '%s'"), strval);
 499                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 500                                 goto error;
 501                         }
 502                         break;
 503
 504                 case ZPOOL_PROP_CACHEFILE:
 505                         if (strval[0] == '\0')
 506                                 break;
 507
 508                         if (strcmp(strval, "none") == 0)
 509                                 break;
 510
 511                         if (strval[0] != '/') {
 512                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 513                                     "property '%s' must be empty, an "
 514                                     "absolute path, or 'none'"), propname);
 515                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 516                                 goto error;
 517                         }
 518
 519                         slash = strrchr(strval, '/');
 520
 521                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 522                             strcmp(slash, "/..") == 0) {
 523                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 524                                     "'%s' is not a valid file"), strval);
 525                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 526                                 goto error;
 527                         }
 528
 529                         *slash = '\0';
 530
 531                         if (strval[0] != '\0' &&
 532                             (stat64(strval, &statbuf) != 0 ||
 533                             !S_ISDIR(statbuf.st_mode))) {
 534                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 535                                     "'%s' is not a valid directory"),
 536                                     strval);
 537                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 538                                 goto error;
 539                         }
 540
 541                         *slash = '/';
 542                         break;
 543
 544                 case ZPOOL_PROP_READONLY:
 545                         if (!flags.import) {
 546                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 547                                     "property '%s' can only be set at "
 548                                     "import time"), propname);
 549                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 550                                 goto error;
 551                         }
 552                         break;
 553                 }
 554         }
 555
 556         return (retprops);
 557 error:
 558         nvlist_free(retprops);
 559         return (NULL);
 560 }
 561
 562 /*
 563  * Set zpool property : propname=propval.
 564  */
 565 int
 566 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
 567 {
 568         zfs_cmd_t zc = { 0 };
 569         int ret = -1;
 570         char errbuf[1024];
 571         nvlist_t *nvl = NULL;
 572         nvlist_t *realprops;
 573         uint64_t version;
 574         prop_flags_t flags = { 0 };
 575
 576         (void) snprintf(errbuf, sizeof (errbuf),
 577             dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
 578             zhp->zpool_name);
 579
 580         if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
 581                 return (no_memory(zhp->zpool_hdl));
 582
 583         if (nvlist_add_string(nvl, propname, propval) != 0) {
 584                 nvlist_free(nvl);
 585                 return (no_memory(zhp->zpool_hdl));
 586         }
 587
 588         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
 589         if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
 590             zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
 591                 nvlist_free(nvl);
 592                 return (-1);
 593         }
 594
 595         nvlist_free(nvl);
 596         nvl = realprops;
 597
 598         /*
 599          * Execute the corresponding ioctl() to set this property.
 600          */
 601         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 602
 603         if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
 604                 nvlist_free(nvl);
 605                 return (-1);
 606         }
 607
 608         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
 609
 610         zcmd_free_nvlists(&zc);
 611         nvlist_free(nvl);
 612
 613         if (ret)
 614                 (void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
 615         else
 616                 (void) zpool_props_refresh(zhp);
 617
 618         return (ret);
 619 }
 620
 621 int
 622 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
 623 {
 624         libzfs_handle_t *hdl = zhp->zpool_hdl;
 625         zprop_list_t *entry;
 626         char buf[ZFS_MAXPROPLEN];
 627
 628         if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
 629                 return (-1);
 630
 631         for (entry = *plp; entry != NULL; entry = entry->pl_next) {
 632
 633                 if (entry->pl_fixed)
 634                         continue;
 635
 636                 if (entry->pl_prop != ZPROP_INVAL &&
 637                     zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
 638                     NULL) == 0) {
 639                         if (strlen(buf) > entry->pl_width)
 640                                 entry->pl_width = strlen(buf);
 641                 }
 642         }
 643
 644         return (0);
 645 }
 646
 647
 648 /*
 649  * Don't start the slice at the default block of 34; many storage
 650  * devices will use a stripe width of 128k, so start there instead.
 651  */
 652 #define NEW_START_BLOCK 256
 653
 654 /*
 655  * Validate the given pool name, optionally putting an extended error message in
 656  * 'buf'.
 657  */
 658 boolean_t
 659 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
 660 {
 661         namecheck_err_t why;
 662         char what;
 663         int ret;
 664
 665         ret = pool_namecheck(pool, &why, &what);
 666
 667         /*
 668          * The rules for reserved pool names were extended at a later point.
 669          * But we need to support users with existing pools that may now be
 670          * invalid.  So we only check for this expanded set of names during a
 671          * create (or import), and only in userland.
 672          */
 673         if (ret == 0 && !isopen &&
 674             (strncmp(pool, "mirror", 6) == 0 ||
 675             strncmp(pool, "raidz", 5) == 0 ||
 676             strncmp(pool, "spare", 5) == 0 ||
 677             strcmp(pool, "log") == 0)) {
 678                 if (hdl != NULL)
 679                         zfs_error_aux(hdl,
 680                             dgettext(TEXT_DOMAIN, "name is reserved"));
 681                 return (B_FALSE);
 682         }
 683
 684
 685         if (ret != 0) {
 686                 if (hdl != NULL) {
 687                         switch (why) {
 688                         case NAME_ERR_TOOLONG:
 689                                 zfs_error_aux(hdl,
 690                                     dgettext(TEXT_DOMAIN, "name is too long"));
 691                                 break;
 692
 693                         case NAME_ERR_INVALCHAR:
 694                                 zfs_error_aux(hdl,
 695                                     dgettext(TEXT_DOMAIN, "invalid character "
 696                                     "'%c' in pool name"), what);
 697                                 break;
 698
 699                         case NAME_ERR_NOLETTER:
 700                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 701                                     "name must begin with a letter"));
 702                                 break;
 703
 704                         case NAME_ERR_RESERVED:
 705                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 706                                     "name is reserved"));
 707                                 break;
 708
 709                         case NAME_ERR_DISKLIKE:
 710                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 711                                     "pool name is reserved"));
 712                                 break;
 713
 714                         case NAME_ERR_LEADING_SLASH:
 715                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 716                                     "leading slash in name"));
 717                                 break;
 718
 719                         case NAME_ERR_EMPTY_COMPONENT:
 720                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 721                                     "empty component in name"));
 722                                 break;
 723
 724                         case NAME_ERR_TRAILING_SLASH:
 725                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 726                                     "trailing slash in name"));
 727                                 break;
 728
 729                         case NAME_ERR_MULTIPLE_AT:
 730                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 731                                     "multiple '@' delimiters in name"));
 732                                 break;
 733
 734                         }
 735                 }
 736                 return (B_FALSE);
 737         }
 738
 739         return (B_TRUE);
 740 }
 741
 742 /*
 743  * Open a handle to the given pool, even if the pool is currently in the FAULTED
 744  * state.
 745  */
 746 zpool_handle_t *
 747 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
 748 {
 749         zpool_handle_t *zhp;
 750         boolean_t missing;
 751
 752         /*
 753          * Make sure the pool name is valid.
 754          */
 755         if (!zpool_name_valid(hdl, B_TRUE, pool)) {
 756                 (void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
 757                     dgettext(TEXT_DOMAIN, "cannot open '%s'"),
 758                     pool);
 759                 return (NULL);
 760         }
 761
 762         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
 763                 return (NULL);
 764
 765         zhp->zpool_hdl = hdl;
 766         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
 767
 768         if (zpool_refresh_stats(zhp, &missing) != 0) {
 769                 zpool_close(zhp);
 770                 return (NULL);
 771         }
 772
 773         if (missing) {
 774                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
 775                 (void) zfs_error_fmt(hdl, EZFS_NOENT,
 776                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
 777                 zpool_close(zhp);
 778                 return (NULL);
 779         }
 780
 781         return (zhp);
 782 }
 783
 784 /*
 785  * Like the above, but silent on error.  Used when iterating over pools (because
 786  * the configuration cache may be out of date).
 787  */
 788 int
 789 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
 790 {
 791         zpool_handle_t *zhp;
 792         boolean_t missing;
 793
 794         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
 795                 return (-1);
 796
 797         zhp->zpool_hdl = hdl;
 798         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
 799
 800         if (zpool_refresh_stats(zhp, &missing) != 0) {
 801                 zpool_close(zhp);
 802                 return (-1);
 803         }
 804
 805         if (missing) {
 806                 zpool_close(zhp);
 807                 *ret = NULL;
 808                 return (0);
 809         }
 810
 811         *ret = zhp;
 812         return (0);
 813 }
 814
 815 /*
 816  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
 817  * state.
 818  */
 819 zpool_handle_t *
 820 zpool_open(libzfs_handle_t *hdl, const char *pool)
 821 {
 822         zpool_handle_t *zhp;
 823
 824         if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
 825                 return (NULL);
 826
 827         if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
 828                 (void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
 829                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
 830                 zpool_close(zhp);
 831                 return (NULL);
 832         }
 833
 834         return (zhp);
 835 }
 836
 837 /*
 838  * Close the handle.  Simply frees the memory associated with the handle.
 839  */
 840 void
 841 zpool_close(zpool_handle_t *zhp)
 842 {
 843         if (zhp->zpool_config)
 844                 nvlist_free(zhp->zpool_config);
 845         if (zhp->zpool_old_config)
 846                 nvlist_free(zhp->zpool_old_config);
 847         if (zhp->zpool_props)
 848                 nvlist_free(zhp->zpool_props);
 849         free(zhp);
 850 }
 851
 852 /*
 853  * Return the name of the pool.
 854  */
 855 const char *
 856 zpool_get_name(zpool_handle_t *zhp)
 857 {
 858         return (zhp->zpool_name);
 859 }
 860
 861
 862 /*
 863  * Return the state of the pool (ACTIVE or UNAVAILABLE)
 864  */
 865 int
 866 zpool_get_state(zpool_handle_t *zhp)
 867 {
 868         return (zhp->zpool_state);
 869 }
 870
 871 /*
 872  * Create the named pool, using the provided vdev list.  It is assumed
 873  * that the consumer has already validated the contents of the nvlist, so we
 874  * don't have to worry about error semantics.
 875  */
 876 int
 877 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
 878     nvlist_t *props, nvlist_t *fsprops)
 879 {
 880         zfs_cmd_t zc = { 0 };
 881         nvlist_t *zc_fsprops = NULL;
 882         nvlist_t *zc_props = NULL;
 883         char msg[1024];
 884         char *altroot;
 885         int ret = -1;
 886
 887         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
 888             "cannot create '%s'"), pool);
 889
 890         if (!zpool_name_valid(hdl, B_FALSE, pool))
 891                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
 892
 893         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
 894                 return (-1);
 895
 896         if (props) {
 897                 prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
 898
 899                 if ((zc_props = zpool_valid_proplist(hdl, pool, props,
 900                     SPA_VERSION_1, flags, msg)) == NULL) {
 901                         goto create_failed;
 902                 }
 903         }
 904
 905         if (fsprops) {
 906                 uint64_t zoned;
 907                 char *zonestr;
 908
 909                 zoned = ((nvlist_lookup_string(fsprops,
 910                     zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
 911                     strcmp(zonestr, "on") == 0);
 912
 913                 if ((zc_fsprops = zfs_valid_proplist(hdl,
 914                     ZFS_TYPE_FILESYSTEM, fsprops, zoned, NULL, msg)) == NULL) {
 915                         goto create_failed;
 916                 }
 917                 if (!zc_props &&
 918                     (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
 919                         goto create_failed;
 920                 }
 921                 if (nvlist_add_nvlist(zc_props,
 922                     ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
 923                         goto create_failed;
 924                 }
 925         }
 926
 927         if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
 928                 goto create_failed;
 929
 930         (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
 931
 932         if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
 933
 934                 zcmd_free_nvlists(&zc);
 935                 nvlist_free(zc_props);
 936                 nvlist_free(zc_fsprops);
 937
 938                 switch (errno) {
 939                 case EBUSY:
 940                         /*
 941                          * This can happen if the user has specified the same
 942                          * device multiple times.  We can't reliably detect this
 943                          * until we try to add it and see we already have a
 944                          * label.
 945                          */
 946                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 947                             "one or more vdevs refer to the same device"));
 948                         return (zfs_error(hdl, EZFS_BADDEV, msg));
 949
 950                 case EOVERFLOW:
 951                         /*
 952                          * This occurs when one of the devices is below
 953                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
 954                          * device was the problem device since there's no
 955                          * reliable way to determine device size from userland.
 956                          */
 957                         {
 958                                 char buf[64];
 959
 960                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
 961
 962                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 963                                     "one or more devices is less than the "
 964                                     "minimum size (%s)"), buf);
 965                         }
 966                         return (zfs_error(hdl, EZFS_BADDEV, msg));
 967
 968                 case ENOSPC:
 969                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 970                             "one or more devices is out of space"));
 971                         return (zfs_error(hdl, EZFS_BADDEV, msg));
 972
 973                 case ENOTBLK:
 974                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 975                             "cache device must be a disk or disk slice"));
 976                         return (zfs_error(hdl, EZFS_BADDEV, msg));
 977
 978                 default:
 979                         return (zpool_standard_error(hdl, errno, msg));
 980                 }
 981         }
 982
 983         /*
 984          * If this is an alternate root pool, then we automatically set the
 985          * mountpoint of the root dataset to be '/'.
 986          */
 987         if (nvlist_lookup_string(props, zpool_prop_to_name(ZPOOL_PROP_ALTROOT),
 988             &altroot) == 0) {
 989                 zfs_handle_t *zhp;
 990
 991                 verify((zhp = zfs_open(hdl, pool, ZFS_TYPE_DATASET)) != NULL);
 992                 verify(zfs_prop_set(zhp, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT),
 993                     "/") == 0);
 994
 995                 zfs_close(zhp);
 996         }
 997
 998 create_failed:
 999         zcmd_free_nvlists(&zc);
1000         nvlist_free(zc_props);
1001         nvlist_free(zc_fsprops);
1002         return (ret);
1003 }
1004
1005 /*
1006  * Destroy the given pool.  It is up to the caller to ensure that there are no
1007  * datasets left in the pool.
1008  */
1009 int
1010 zpool_destroy(zpool_handle_t *zhp)
1011 {
1012         zfs_cmd_t zc = { 0 };
1013         zfs_handle_t *zfp = NULL;
1014         libzfs_handle_t *hdl = zhp->zpool_hdl;
1015         char msg[1024];
1016
1017         if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1018             (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1019                 return (-1);
1020
1021         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1022
1023         if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1024                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1025                     "cannot destroy '%s'"), zhp->zpool_name);
1026
1027                 if (errno == EROFS) {
1028                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1029                             "one or more devices is read only"));
1030                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1031                 } else {
1032                         (void) zpool_standard_error(hdl, errno, msg);
1033                 }
1034
1035                 if (zfp)
1036                         zfs_close(zfp);
1037                 return (-1);
1038         }
1039
1040         if (zfp) {
1041                 remove_mountpoint(zfp);
1042                 zfs_close(zfp);
1043         }
1044
1045         return (0);
1046 }
1047
1048 /*
1049  * Add the given vdevs to the pool.  The caller must have already performed the
1050  * necessary verification to ensure that the vdev specification is well-formed.
1051  */
1052 int
1053 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1054 {
1055         zfs_cmd_t zc = { 0 };
1056         int ret;
1057         libzfs_handle_t *hdl = zhp->zpool_hdl;
1058         char msg[1024];
1059         nvlist_t **spares, **l2cache;
1060         uint_t nspares, nl2cache;
1061
1062         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1063             "cannot add to '%s'"), zhp->zpool_name);
1064
1065         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1066             SPA_VERSION_SPARES &&
1067             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1068             &spares, &nspares) == 0) {
1069                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1070                     "upgraded to add hot spares"));
1071                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1072         }
1073
1074         if (pool_is_bootable(zhp) && nvlist_lookup_nvlist_array(nvroot,
1075             ZPOOL_CONFIG_SPARES, &spares, &nspares) == 0) {
1076                 uint64_t s;
1077
1078                 for (s = 0; s < nspares; s++) {
1079                         char *path;
1080
1081                         if (nvlist_lookup_string(spares[s], ZPOOL_CONFIG_PATH,
1082                             &path) == 0 && pool_uses_efi(spares[s])) {
1083                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1084                                     "device '%s' contains an EFI label and "
1085                                     "cannot be used on root pools."),
1086                                     zpool_vdev_name(hdl, NULL, spares[s],
1087                                     B_FALSE));
1088                                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
1089                         }
1090                 }
1091         }
1092
1093         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1094             SPA_VERSION_L2CACHE &&
1095             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1096             &l2cache, &nl2cache) == 0) {
1097                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1098                     "upgraded to add cache devices"));
1099                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1100         }
1101
1102         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1103                 return (-1);
1104         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1105
1106         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1107                 switch (errno) {
1108                 case EBUSY:
1109                         /*
1110                          * This can happen if the user has specified the same
1111                          * device multiple times.  We can't reliably detect this
1112                          * until we try to add it and see we already have a
1113                          * label.
1114                          */
1115                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1116                             "one or more vdevs refer to the same device"));
1117                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1118                         break;
1119
1120                 case EOVERFLOW:
1121                         /*
1122                          * This occurrs when one of the devices is below
1123                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1124                          * device was the problem device since there's no
1125                          * reliable way to determine device size from userland.
1126                          */
1127                         {
1128                                 char buf[64];
1129
1130                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1131
1132                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1133                                     "device is less than the minimum "
1134                                     "size (%s)"), buf);
1135                         }
1136                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1137                         break;
1138
1139                 case ENOTSUP:
1140                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1141                             "pool must be upgraded to add these vdevs"));
1142                         (void) zfs_error(hdl, EZFS_BADVERSION, msg);
1143                         break;
1144
1145                 case EDOM:
1146                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1147                             "root pool can not have multiple vdevs"
1148                             " or separate logs"));
1149                         (void) zfs_error(hdl, EZFS_POOL_NOTSUP, msg);
1150                         break;
1151
1152                 case ENOTBLK:
1153                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1154                             "cache device must be a disk or disk slice"));
1155                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1156                         break;
1157
1158                 default:
1159                         (void) zpool_standard_error(hdl, errno, msg);
1160                 }
1161
1162                 ret = -1;
1163         } else {
1164                 ret = 0;
1165         }
1166
1167         zcmd_free_nvlists(&zc);
1168
1169         return (ret);
1170 }
1171
1172 /*
1173  * Exports the pool from the system.  The caller must ensure that there are no
1174  * mounted datasets in the pool.
1175  */
1176 int
1177 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce)
1178 {
1179         zfs_cmd_t zc = { 0 };
1180         char msg[1024];
1181
1182         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1183             "cannot export '%s'"), zhp->zpool_name);
1184
1185         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1186         zc.zc_cookie = force;
1187         zc.zc_guid = hardforce;
1188
1189         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1190                 switch (errno) {
1191                 case EXDEV:
1192                         zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1193                             "use '-f' to override the following errors:\n"
1194                             "'%s' has an active shared spare which could be"
1195                             " used by other pools once '%s' is exported."),
1196                             zhp->zpool_name, zhp->zpool_name);
1197                         return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1198                             msg));
1199                 default:
1200                         return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1201                             msg));
1202                 }
1203         }
1204
1205         return (0);
1206 }
1207
1208 int
1209 zpool_export(zpool_handle_t *zhp, boolean_t force)
1210 {
1211         return (zpool_export_common(zhp, force, B_FALSE));
1212 }
1213
1214 int
1215 zpool_export_force(zpool_handle_t *zhp)
1216 {
1217         return (zpool_export_common(zhp, B_TRUE, B_TRUE));
1218 }
1219
1220 static void
1221 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1222     nvlist_t *config)
1223 {
1224         nvlist_t *nv = NULL;
1225         uint64_t rewindto;
1226         int64_t loss = -1;
1227         struct tm t;
1228         char timestr[128];
1229
1230         if (!hdl->libzfs_printerr || config == NULL)
1231                 return;
1232
1233         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0)
1234                 return;
1235
1236         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1237                 return;
1238         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1239
1240         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1241             strftime(timestr, 128, 0, &t) != 0) {
1242                 if (dryrun) {
1243                         (void) printf(dgettext(TEXT_DOMAIN,
1244                             "Would be able to return %s "
1245                             "to its state as of %s.\n"),
1246                             name, timestr);
1247                 } else {
1248                         (void) printf(dgettext(TEXT_DOMAIN,
1249                             "Pool %s returned to its state as of %s.\n"),
1250                             name, timestr);
1251                 }
1252                 if (loss > 120) {
1253                         (void) printf(dgettext(TEXT_DOMAIN,
1254                             "%s approximately %lld "),
1255                             dryrun ? "Would discard" : "Discarded",
1256                             (loss + 30) / 60);
1257                         (void) printf(dgettext(TEXT_DOMAIN,
1258                             "minutes of transactions.\n"));
1259                 } else if (loss > 0) {
1260                         (void) printf(dgettext(TEXT_DOMAIN,
1261                             "%s approximately %lld "),
1262                             dryrun ? "Would discard" : "Discarded", loss);
1263                         (void) printf(dgettext(TEXT_DOMAIN,
1264                             "seconds of transactions.\n"));
1265                 }
1266         }
1267 }
1268
1269 void
1270 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1271     nvlist_t *config)
1272 {
1273         nvlist_t *nv = NULL;
1274         int64_t loss = -1;
1275         uint64_t edata = UINT64_MAX;
1276         uint64_t rewindto;
1277         struct tm t;
1278         char timestr[128];
1279
1280         if (!hdl->libzfs_printerr)
1281                 return;
1282
1283         if (reason >= 0)
1284                 (void) printf(dgettext(TEXT_DOMAIN, "action: "));
1285         else
1286                 (void) printf(dgettext(TEXT_DOMAIN, "\t"));
1287
1288         /* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1289         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1290             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1291                 goto no_info;
1292
1293         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1294         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1295             &edata);
1296
1297         (void) printf(dgettext(TEXT_DOMAIN,
1298             "Recovery is possible, but will result in some data loss.\n"));
1299
1300         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1301             strftime(timestr, 128, 0, &t) != 0) {
1302                 (void) printf(dgettext(TEXT_DOMAIN,
1303                     "\tReturning the pool to its state as of %s\n"
1304                     "\tshould correct the problem.  "),
1305                     timestr);
1306         } else {
1307                 (void) printf(dgettext(TEXT_DOMAIN,
1308                     "\tReverting the pool to an earlier state "
1309                     "should correct the problem.\n\t"));
1310         }
1311
1312         if (loss > 120) {
1313                 (void) printf(dgettext(TEXT_DOMAIN,
1314                     "Approximately %lld minutes of data\n"
1315                     "\tmust be discarded, irreversibly.  "), (loss + 30) / 60);
1316         } else if (loss > 0) {
1317                 (void) printf(dgettext(TEXT_DOMAIN,
1318                     "Approximately %lld seconds of data\n"
1319                     "\tmust be discarded, irreversibly.  "), loss);
1320         }
1321         if (edata != 0 && edata != UINT64_MAX) {
1322                 if (edata == 1) {
1323                         (void) printf(dgettext(TEXT_DOMAIN,
1324                             "After rewind, at least\n"
1325                             "\tone persistent user-data error will remain.  "));
1326                 } else {
1327                         (void) printf(dgettext(TEXT_DOMAIN,
1328                             "After rewind, several\n"
1329                             "\tpersistent user-data errors will remain.  "));
1330                 }
1331         }
1332         (void) printf(dgettext(TEXT_DOMAIN,
1333             "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1334             reason >= 0 ? "clear" : "import", name);
1335
1336         (void) printf(dgettext(TEXT_DOMAIN,
1337             "A scrub of the pool\n"
1338             "\tis strongly recommended after recovery.\n"));
1339         return;
1340
1341 no_info:
1342         (void) printf(dgettext(TEXT_DOMAIN,
1343             "Destroy and re-create the pool from\n\ta backup source.\n"));
1344 }
1345
1346 /*
1347  * zpool_import() is a contracted interface. Should be kept the same
1348  * if possible.
1349  *
1350  * Applications should use zpool_import_props() to import a pool with
1351  * new properties value to be set.
1352  */
1353 int
1354 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1355     char *altroot)
1356 {
1357         nvlist_t *props = NULL;
1358         int ret;
1359
1360         if (altroot != NULL) {
1361                 if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1362                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1363                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1364                             newname));
1365                 }
1366
1367                 if (nvlist_add_string(props,
1368                     zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1369                     nvlist_add_string(props,
1370                     zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1371                         nvlist_free(props);
1372                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1373                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1374                             newname));
1375                 }
1376         }
1377
1378         ret = zpool_import_props(hdl, config, newname, props,
1379             ZFS_IMPORT_NORMAL);
1380         if (props)
1381                 nvlist_free(props);
1382         return (ret);
1383 }
1384
1385 static void
1386 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1387     int indent)
1388 {
1389         nvlist_t **child;
1390         uint_t c, children;
1391         char *vname;
1392         uint64_t is_log = 0;
1393
1394         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1395             &is_log);
1396
1397         if (name != NULL)
1398                 (void) printf("\t%*s%s%s\n", indent, "", name,
1399                     is_log ? " [log]" : "");
1400
1401         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1402             &child, &children) != 0)
1403                 return;
1404
1405         for (c = 0; c < children; c++) {
1406                 vname = zpool_vdev_name(hdl, NULL, child[c], B_TRUE);
1407                 print_vdev_tree(hdl, vname, child[c], indent + 2);
1408                 free(vname);
1409         }
1410 }
1411
1412 /*
1413  * Import the given pool using the known configuration and a list of
1414  * properties to be set. The configuration should have come from
1415  * zpool_find_import(). The 'newname' parameters control whether the pool
1416  * is imported with a different name.
1417  */
1418 int
1419 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1420     nvlist_t *props, int flags)
1421 {
1422         zfs_cmd_t zc = { 0 };
1423         zpool_rewind_policy_t policy;
1424         nvlist_t *nv = NULL;
1425         nvlist_t *nvinfo = NULL;
1426         nvlist_t *missing = NULL;
1427         char *thename;
1428         char *origname;
1429         int ret;
1430         int error = 0;
1431         char errbuf[1024];
1432
1433         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1434             &origname) == 0);
1435
1436         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1437             "cannot import pool '%s'"), origname);
1438
1439         if (newname != NULL) {
1440                 if (!zpool_name_valid(hdl, B_FALSE, newname))
1441                         return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1442                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1443                             newname));
1444                 thename = (char *)newname;
1445         } else {
1446                 thename = origname;
1447         }
1448
1449         if (props) {
1450                 uint64_t version;
1451                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1452
1453                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1454                     &version) == 0);
1455
1456                 if ((props = zpool_valid_proplist(hdl, origname,
1457                     props, version, flags, errbuf)) == NULL) {
1458                         return (-1);
1459                 } else if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1460                         nvlist_free(props);
1461                         return (-1);
1462                 }
1463         }
1464
1465         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1466
1467         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1468             &zc.zc_guid) == 0);
1469
1470         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1471                 nvlist_free(props);
1472                 return (-1);
1473         }
1474         if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1475                 nvlist_free(props);
1476                 return (-1);
1477         }
1478
1479         zc.zc_cookie = flags;
1480         while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1481             errno == ENOMEM) {
1482                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1483                         zcmd_free_nvlists(&zc);
1484                         return (-1);
1485                 }
1486         }
1487         if (ret != 0)
1488                 error = errno;
1489
1490         (void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1491         zpool_get_rewind_policy(config, &policy);
1492
1493         if (error) {
1494                 char desc[1024];
1495
1496                 /*
1497                  * Dry-run failed, but we print out what success
1498                  * looks like if we found a best txg
1499                  */
1500                 if (policy.zrp_request & ZPOOL_TRY_REWIND) {
1501                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1502                             B_TRUE, nv);
1503                         nvlist_free(nv);
1504                         return (-1);
1505                 }
1506
1507                 if (newname == NULL)
1508                         (void) snprintf(desc, sizeof (desc),
1509                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1510                             thename);
1511                 else
1512                         (void) snprintf(desc, sizeof (desc),
1513                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1514                             origname, thename);
1515
1516                 switch (error) {
1517                 case ENOTSUP:
1518                         /*
1519                          * Unsupported version.
1520                          */
1521                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1522                         break;
1523
1524                 case EINVAL:
1525                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1526                         break;
1527
1528                 case EROFS:
1529                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1530                             "one or more devices is read only"));
1531                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1532                         break;
1533
1534                 case ENXIO:
1535                         if (nv && nvlist_lookup_nvlist(nv,
1536                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1537                             nvlist_lookup_nvlist(nvinfo,
1538                             ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1539                                 (void) printf(dgettext(TEXT_DOMAIN,
1540                                     "The devices below are missing, use "
1541                                     "'-m' to import the pool anyway:\n"));
1542                                 print_vdev_tree(hdl, NULL, missing, 2);
1543                                 (void) printf("\n");
1544                         }
1545                         (void) zpool_standard_error(hdl, error, desc);
1546                         break;
1547
1548                 case EEXIST:
1549                         (void) zpool_standard_error(hdl, error, desc);
1550                         break;
1551
1552                 default:
1553                         (void) zpool_standard_error(hdl, error, desc);
1554                         zpool_explain_recover(hdl,
1555                             newname ? origname : thename, -error, nv);
1556                         break;
1557                 }
1558
1559                 nvlist_free(nv);
1560                 ret = -1;
1561         } else {
1562                 zpool_handle_t *zhp;
1563
1564                 /*
1565                  * This should never fail, but play it safe anyway.
1566                  */
1567                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1568                         ret = -1;
1569                 else if (zhp != NULL)
1570                         zpool_close(zhp);
1571                 if (policy.zrp_request &
1572                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1573                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1574                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1575                 }
1576                 nvlist_free(nv);
1577                 return (0);
1578         }
1579
1580         zcmd_free_nvlists(&zc);
1581         nvlist_free(props);
1582
1583         return (ret);
1584 }
1585
1586 /*
1587  * Scan the pool.
1588  */
1589 int
1590 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1591 {
1592         zfs_cmd_t zc = { 0 };
1593         char msg[1024];
1594         libzfs_handle_t *hdl = zhp->zpool_hdl;
1595
1596         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1597         zc.zc_cookie = func;
1598
1599         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1600             (errno == ENOENT && func != POOL_SCAN_NONE))
1601                 return (0);
1602
1603         if (func == POOL_SCAN_SCRUB) {
1604                 (void) snprintf(msg, sizeof (msg),
1605                     dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1606         } else if (func == POOL_SCAN_NONE) {
1607                 (void) snprintf(msg, sizeof (msg),
1608                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1609                     zc.zc_name);
1610         } else {
1611                 assert(!"unexpected result");
1612         }
1613
1614         if (errno == EBUSY) {
1615                 nvlist_t *nvroot;
1616                 pool_scan_stat_t *ps = NULL;
1617                 uint_t psc;
1618
1619                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
1620                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1621                 (void) nvlist_lookup_uint64_array(nvroot,
1622                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1623                 if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1624                         return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1625                 else
1626                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
1627         } else if (errno == ENOENT) {
1628                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1629         } else {
1630                 return (zpool_standard_error(hdl, errno, msg));
1631         }
1632 }
1633
1634 /*
1635  * This provides a very minimal check whether a given string is likely a
1636  * c#t#d# style string.  Users of this are expected to do their own
1637  * verification of the s# part.
1638  */
1639 #define CTD_CHECK(str)  (str && str[0] == 'c' && isdigit(str[1]))
1640
1641 /*
1642  * More elaborate version for ones which may start with "/dev/dsk/"
1643  * and the like.
1644  */
1645 static int
1646 ctd_check_path(char *str) {
1647         /*
1648          * If it starts with a slash, check the last component.
1649          */
1650         if (str && str[0] == '/') {
1651                 char *tmp = strrchr(str, '/');
1652
1653                 /*
1654                  * If it ends in "/old", check the second-to-last
1655                  * component of the string instead.
1656                  */
1657                 if (tmp != str && strcmp(tmp, "/old") == 0) {
1658                         for (tmp--; *tmp != '/'; tmp--)
1659                                 ;
1660                 }
1661                 str = tmp + 1;
1662         }
1663         return (CTD_CHECK(str));
1664 }
1665
1666 /*
1667  * Find a vdev that matches the search criteria specified. We use the
1668  * the nvpair name to determine how we should look for the device.
1669  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1670  * spare; but FALSE if its an INUSE spare.
1671  */
1672 static nvlist_t *
1673 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1674     boolean_t *l2cache, boolean_t *log)
1675 {
1676         uint_t c, children;
1677         nvlist_t **child;
1678         nvlist_t *ret;
1679         uint64_t is_log;
1680         char *srchkey;
1681         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
1682
1683         /* Nothing to look for */
1684         if (search == NULL || pair == NULL)
1685                 return (NULL);
1686
1687         /* Obtain the key we will use to search */
1688         srchkey = nvpair_name(pair);
1689
1690         switch (nvpair_type(pair)) {
1691         case DATA_TYPE_UINT64:
1692                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
1693                         uint64_t srchval, theguid;
1694
1695                         verify(nvpair_value_uint64(pair, &srchval) == 0);
1696                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
1697                             &theguid) == 0);
1698                         if (theguid == srchval)
1699                                 return (nv);
1700                 }
1701                 break;
1702
1703         case DATA_TYPE_STRING: {
1704                 char *srchval, *val;
1705
1706                 verify(nvpair_value_string(pair, &srchval) == 0);
1707                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
1708                         break;
1709
1710                 /*
1711                  * Search for the requested value. Special cases:
1712                  *
1713                  * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
1714                  *   "s0" or "s0/old".  The "s0" part is hidden from the user,
1715                  *   but included in the string, so this matches around it.
1716                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
1717                  *
1718                  * Otherwise, all other searches are simple string compares.
1719                  */
1720                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0 &&
1721                     ctd_check_path(val)) {
1722                         uint64_t wholedisk = 0;
1723
1724                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1725                             &wholedisk);
1726                         if (wholedisk) {
1727                                 int slen = strlen(srchval);
1728                                 int vlen = strlen(val);
1729
1730                                 if (slen != vlen - 2)
1731                                         break;
1732
1733                                 /*
1734                                  * make_leaf_vdev() should only set
1735                                  * wholedisk for ZPOOL_CONFIG_PATHs which
1736                                  * will include "/dev/dsk/", giving plenty of
1737                                  * room for the indices used next.
1738                                  */
1739                                 ASSERT(vlen >= 6);
1740
1741                                 /*
1742                                  * strings identical except trailing "s0"
1743                                  */
1744                                 if (strcmp(&val[vlen - 2], "s0") == 0 &&
1745                                     strncmp(srchval, val, slen) == 0)
1746                                         return (nv);
1747
1748                                 /*
1749                                  * strings identical except trailing "s0/old"
1750                                  */
1751                                 if (strcmp(&val[vlen - 6], "s0/old") == 0 &&
1752                                     strcmp(&srchval[slen - 4], "/old") == 0 &&
1753                                     strncmp(srchval, val, slen - 4) == 0)
1754                                         return (nv);
1755
1756                                 break;
1757                         }
1758                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
1759                         char *type, *idx, *end, *p;
1760                         uint64_t id, vdev_id;
1761
1762                         /*
1763                          * Determine our vdev type, keeping in mind
1764                          * that the srchval is composed of a type and
1765                          * vdev id pair (i.e. mirror-4).
1766                          */
1767                         if ((type = strdup(srchval)) == NULL)
1768                                 return (NULL);
1769
1770                         if ((p = strrchr(type, '-')) == NULL) {
1771                                 free(type);
1772                                 break;
1773                         }
1774                         idx = p + 1;
1775                         *p = '\0';
1776
1777                         /*
1778                          * If the types don't match then keep looking.
1779                          */
1780                         if (strncmp(val, type, strlen(val)) != 0) {
1781                                 free(type);
1782                                 break;
1783                         }
1784
1785                         verify(strncmp(type, VDEV_TYPE_RAIDZ,
1786                             strlen(VDEV_TYPE_RAIDZ)) == 0 ||
1787                             strncmp(type, VDEV_TYPE_MIRROR,
1788                             strlen(VDEV_TYPE_MIRROR)) == 0);
1789                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
1790                             &id) == 0);
1791
1792                         errno = 0;
1793                         vdev_id = strtoull(idx, &end, 10);
1794
1795                         free(type);
1796                         if (errno != 0)
1797                                 return (NULL);
1798
1799                         /*
1800                          * Now verify that we have the correct vdev id.
1801                          */
1802                         if (vdev_id == id)
1803                                 return (nv);
1804                 }
1805
1806                 /*
1807                  * Common case
1808                  */
1809                 if (strcmp(srchval, val) == 0)
1810                         return (nv);
1811                 break;
1812         }
1813
1814         default:
1815                 break;
1816         }
1817
1818         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1819             &child, &children) != 0)
1820                 return (NULL);
1821
1822         for (c = 0; c < children; c++) {
1823                 if ((ret = vdev_to_nvlist_iter(child[c], search,
1824                     avail_spare, l2cache, NULL)) != NULL) {
1825                         /*
1826                          * The 'is_log' value is only set for the toplevel
1827                          * vdev, not the leaf vdevs.  So we always lookup the
1828                          * log device from the root of the vdev tree (where
1829                          * 'log' is non-NULL).
1830                          */
1831                         if (log != NULL &&
1832                             nvlist_lookup_uint64(child[c],
1833                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
1834                             is_log) {
1835                                 *log = B_TRUE;
1836                         }
1837                         return (ret);
1838                 }
1839         }
1840
1841         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
1842             &child, &children) == 0) {
1843                 for (c = 0; c < children; c++) {
1844                         if ((ret = vdev_to_nvlist_iter(child[c], search,
1845                             avail_spare, l2cache, NULL)) != NULL) {
1846                                 *avail_spare = B_TRUE;
1847                                 return (ret);
1848                         }
1849                 }
1850         }
1851
1852         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
1853             &child, &children) == 0) {
1854                 for (c = 0; c < children; c++) {
1855                         if ((ret = vdev_to_nvlist_iter(child[c], search,
1856                             avail_spare, l2cache, NULL)) != NULL) {
1857                                 *l2cache = B_TRUE;
1858                                 return (ret);
1859                         }
1860                 }
1861         }
1862
1863         return (NULL);
1864 }
1865
1866 /*
1867  * Given a physical path (minus the "/devices" prefix), find the
1868  * associated vdev.
1869  */
1870 nvlist_t *
1871 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
1872     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
1873 {
1874         nvlist_t *search, *nvroot, *ret;
1875
1876         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1877         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
1878
1879         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1880             &nvroot) == 0);
1881
1882         *avail_spare = B_FALSE;
1883         *l2cache = B_FALSE;
1884         if (log != NULL)
1885                 *log = B_FALSE;
1886         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
1887         nvlist_free(search);
1888
1889         return (ret);
1890 }
1891
1892 /*
1893  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
1894  */
1895 boolean_t
1896 zpool_vdev_is_interior(const char *name)
1897 {
1898         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
1899             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
1900                 return (B_TRUE);
1901         return (B_FALSE);
1902 }
1903
1904 nvlist_t *
1905 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
1906     boolean_t *l2cache, boolean_t *log)
1907 {
1908         char buf[MAXPATHLEN];
1909         char *end;
1910         nvlist_t *nvroot, *search, *ret;
1911         uint64_t guid;
1912
1913         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1914
1915         guid = strtoull(path, &end, 10);
1916         if (guid != 0 && *end == '\0') {
1917                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
1918         } else if (zpool_vdev_is_interior(path)) {
1919                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
1920         } else if (path[0] != '/') {
1921                 (void) snprintf(buf, sizeof (buf), "%s%s", "/dev/dsk/", path);
1922                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, buf) == 0);
1923         } else {
1924                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
1925         }
1926
1927         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
1928             &nvroot) == 0);
1929
1930         *avail_spare = B_FALSE;
1931         *l2cache = B_FALSE;
1932         if (log != NULL)
1933                 *log = B_FALSE;
1934         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
1935         nvlist_free(search);
1936
1937         return (ret);
1938 }
1939
1940 static int
1941 vdev_online(nvlist_t *nv)
1942 {
1943         uint64_t ival;
1944
1945         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
1946             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
1947             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
1948                 return (0);
1949
1950         return (1);
1951 }
1952
1953 /*
1954  * Helper function for zpool_get_physpaths().
1955  */
1956 static int
1957 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
1958     size_t *bytes_written)
1959 {
1960         size_t bytes_left, pos, rsz;
1961         char *tmppath;
1962         const char *format;
1963
1964         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
1965             &tmppath) != 0)
1966                 return (EZFS_NODEVICE);
1967
1968         pos = *bytes_written;
1969         bytes_left = physpath_size - pos;
1970         format = (pos == 0) ? "%s" : " %s";
1971
1972         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
1973         *bytes_written += rsz;
1974
1975         if (rsz >= bytes_left) {
1976                 /* if physpath was not copied properly, clear it */
1977                 if (bytes_left != 0) {
1978                         physpath[pos] = 0;
1979                 }
1980                 return (EZFS_NOSPC);
1981         }
1982         return (0);
1983 }
1984
1985 static int
1986 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
1987     size_t *rsz, boolean_t is_spare)
1988 {
1989         char *type;
1990         int ret;
1991
1992         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
1993                 return (EZFS_INVALCONFIG);
1994
1995         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
1996                 /*
1997                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
1998                  * For a spare vdev, we only want to boot from the active
1999                  * spare device.
2000                  */
2001                 if (is_spare) {
2002                         uint64_t spare = 0;
2003                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2004                             &spare);
2005                         if (!spare)
2006                                 return (EZFS_INVALCONFIG);
2007                 }
2008
2009                 if (vdev_online(nv)) {
2010                         if ((ret = vdev_get_one_physpath(nv, physpath,
2011                             phypath_size, rsz)) != 0)
2012                                 return (ret);
2013                 }
2014         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2015             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2016             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2017                 nvlist_t **child;
2018                 uint_t count;
2019                 int i, ret;
2020
2021                 if (nvlist_lookup_nvlist_array(nv,
2022                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2023                         return (EZFS_INVALCONFIG);
2024
2025                 for (i = 0; i < count; i++) {
2026                         ret = vdev_get_physpaths(child[i], physpath,
2027                             phypath_size, rsz, is_spare);
2028                         if (ret == EZFS_NOSPC)
2029                                 return (ret);
2030                 }
2031         }
2032
2033         return (EZFS_POOL_INVALARG);
2034 }
2035
2036 /*
2037  * Get phys_path for a root pool config.
2038  * Return 0 on success; non-zero on failure.
2039  */
2040 static int
2041 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2042 {
2043         size_t rsz;
2044         nvlist_t *vdev_root;
2045         nvlist_t **child;
2046         uint_t count;
2047         char *type;
2048
2049         rsz = 0;
2050
2051         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2052             &vdev_root) != 0)
2053                 return (EZFS_INVALCONFIG);
2054
2055         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2056             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2057             &child, &count) != 0)
2058                 return (EZFS_INVALCONFIG);
2059
2060         /*
2061          * root pool can not have EFI labeled disks and can only have
2062          * a single top-level vdev.
2063          */
2064         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1 ||
2065             pool_uses_efi(vdev_root))
2066                 return (EZFS_POOL_INVALARG);
2067
2068         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2069             B_FALSE);
2070
2071         /* No online devices */
2072         if (rsz == 0)
2073                 return (EZFS_NODEVICE);
2074
2075         return (0);
2076 }
2077
2078 /*
2079  * Get phys_path for a root pool
2080  * Return 0 on success; non-zero on failure.
2081  */
2082 int
2083 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2084 {
2085         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2086             phypath_size));
2087 }
2088
2089 /*
2090  * If the device has being dynamically expanded then we need to relabel
2091  * the disk to use the new unallocated space.
2092  */
2093 static int
2094 zpool_relabel_disk(libzfs_handle_t *hdl, const char *name)
2095 {
2096         char path[MAXPATHLEN];
2097         char errbuf[1024];
2098         int fd, error;
2099         int (*_efi_use_whole_disk)(int);
2100
2101         if ((_efi_use_whole_disk = (int (*)(int))dlsym(RTLD_DEFAULT,
2102             "efi_use_whole_disk")) == NULL)
2103                 return (-1);
2104
2105         (void) snprintf(path, sizeof (path), "%s/%s", RDISK_ROOT, name);
2106
2107         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
2108                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2109                     "relabel '%s': unable to open device"), name);
2110                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
2111         }
2112
2113         /*
2114          * It's possible that we might encounter an error if the device
2115          * does not have any unallocated space left. If so, we simply
2116          * ignore that error and continue on.
2117          */
2118         error = _efi_use_whole_disk(fd);
2119         (void) close(fd);
2120         if (error && error != VT_ENOSPC) {
2121                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2122                     "relabel '%s': unable to read disk capacity"), name);
2123                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
2124         }
2125         return (0);
2126 }
2127
2128 /*
2129  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2130  * ZFS_ONLINE_* flags.
2131  */
2132 int
2133 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2134     vdev_state_t *newstate)
2135 {
2136         zfs_cmd_t zc = { 0 };
2137         char msg[1024];
2138         nvlist_t *tgt;
2139         boolean_t avail_spare, l2cache, islog;
2140         libzfs_handle_t *hdl = zhp->zpool_hdl;
2141
2142         if (flags & ZFS_ONLINE_EXPAND) {
2143                 (void) snprintf(msg, sizeof (msg),
2144                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2145         } else {
2146                 (void) snprintf(msg, sizeof (msg),
2147                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2148         }
2149
2150         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2151         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2152             &islog)) == NULL)
2153                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2154
2155         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2156
2157         if (avail_spare)
2158                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2159
2160         if (flags & ZFS_ONLINE_EXPAND ||
2161             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2162                 char *pathname = NULL;
2163                 uint64_t wholedisk = 0;
2164
2165                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2166                     &wholedisk);
2167                 verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
2168                     &pathname) == 0);
2169
2170                 /*
2171                  * XXX - L2ARC 1.0 devices can't support expansion.
2172                  */
2173                 if (l2cache) {
2174                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2175                             "cannot expand cache devices"));
2176                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2177                 }
2178
2179                 if (wholedisk) {
2180                         pathname += strlen(DISK_ROOT) + 1;
2181                         (void) zpool_relabel_disk(hdl, pathname);
2182                 }
2183         }
2184
2185         zc.zc_cookie = VDEV_STATE_ONLINE;
2186         zc.zc_obj = flags;
2187
2188         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2189                 if (errno == EINVAL) {
2190                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2191                             "from this pool into a new one.  Use '%s' "
2192                             "instead"), "zpool detach");
2193                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2194                 }
2195                 return (zpool_standard_error(hdl, errno, msg));
2196         }
2197
2198         *newstate = zc.zc_cookie;
2199         return (0);
2200 }
2201
2202 /*
2203  * Take the specified vdev offline
2204  */
2205 int
2206 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2207 {
2208         zfs_cmd_t zc = { 0 };
2209         char msg[1024];
2210         nvlist_t *tgt;
2211         boolean_t avail_spare, l2cache;
2212         libzfs_handle_t *hdl = zhp->zpool_hdl;
2213
2214         (void) snprintf(msg, sizeof (msg),
2215             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2216
2217         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2218         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2219             NULL)) == NULL)
2220                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2221
2222         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2223
2224         if (avail_spare)
2225                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2226
2227         zc.zc_cookie = VDEV_STATE_OFFLINE;
2228         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2229
2230         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2231                 return (0);
2232
2233         switch (errno) {
2234         case EBUSY:
2235
2236                 /*
2237                  * There are no other replicas of this device.
2238                  */
2239                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2240
2241         case EEXIST:
2242                 /*
2243                  * The log device has unplayed logs
2244                  */
2245                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2246
2247         default:
2248                 return (zpool_standard_error(hdl, errno, msg));
2249         }
2250 }
2251
2252 /*
2253  * Mark the given vdev faulted.
2254  */
2255 int
2256 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2257 {
2258         zfs_cmd_t zc = { 0 };
2259         char msg[1024];
2260         libzfs_handle_t *hdl = zhp->zpool_hdl;
2261
2262         (void) snprintf(msg, sizeof (msg),
2263             dgettext(TEXT_DOMAIN, "cannot fault %llu"), guid);
2264
2265         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2266         zc.zc_guid = guid;
2267         zc.zc_cookie = VDEV_STATE_FAULTED;
2268         zc.zc_obj = aux;
2269
2270         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2271                 return (0);
2272
2273         switch (errno) {
2274         case EBUSY:
2275
2276                 /*
2277                  * There are no other replicas of this device.
2278                  */
2279                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2280
2281         default:
2282                 return (zpool_standard_error(hdl, errno, msg));
2283         }
2284
2285 }
2286
2287 /*
2288  * Mark the given vdev degraded.
2289  */
2290 int
2291 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2292 {
2293         zfs_cmd_t zc = { 0 };
2294         char msg[1024];
2295         libzfs_handle_t *hdl = zhp->zpool_hdl;
2296
2297         (void) snprintf(msg, sizeof (msg),
2298             dgettext(TEXT_DOMAIN, "cannot degrade %llu"), guid);
2299
2300         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2301         zc.zc_guid = guid;
2302         zc.zc_cookie = VDEV_STATE_DEGRADED;
2303         zc.zc_obj = aux;
2304
2305         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2306                 return (0);
2307
2308         return (zpool_standard_error(hdl, errno, msg));
2309 }
2310
2311 /*
2312  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2313  * a hot spare.
2314  */
2315 static boolean_t
2316 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2317 {
2318         nvlist_t **child;
2319         uint_t c, children;
2320         char *type;
2321
2322         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2323             &children) == 0) {
2324                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2325                     &type) == 0);
2326
2327                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2328                     children == 2 && child[which] == tgt)
2329                         return (B_TRUE);
2330
2331                 for (c = 0; c < children; c++)
2332                         if (is_replacing_spare(child[c], tgt, which))
2333                                 return (B_TRUE);
2334         }
2335
2336         return (B_FALSE);
2337 }
2338
2339 /*
2340  * Attach new_disk (fully described by nvroot) to old_disk.
2341  * If 'replacing' is specified, the new disk will replace the old one.
2342  */
2343 int
2344 zpool_vdev_attach(zpool_handle_t *zhp,
2345     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2346 {
2347         zfs_cmd_t zc = { 0 };
2348         char msg[1024];
2349         int ret;
2350         nvlist_t *tgt;
2351         boolean_t avail_spare, l2cache, islog;
2352         uint64_t val;
2353         char *newname;
2354         nvlist_t **child;
2355         uint_t children;
2356         nvlist_t *config_root;
2357         libzfs_handle_t *hdl = zhp->zpool_hdl;
2358         boolean_t rootpool = pool_is_bootable(zhp);
2359
2360         if (replacing)
2361                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2362                     "cannot replace %s with %s"), old_disk, new_disk);
2363         else
2364                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2365                     "cannot attach %s to %s"), new_disk, old_disk);
2366
2367         /*
2368          * If this is a root pool, make sure that we're not attaching an
2369          * EFI labeled device.
2370          */
2371         if (rootpool && pool_uses_efi(nvroot)) {
2372                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2373                     "EFI labeled devices are not supported on root pools."));
2374                 return (zfs_error(hdl, EZFS_POOL_NOTSUP, msg));
2375         }
2376
2377         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2378         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2379             &islog)) == 0)
2380                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2381
2382         if (avail_spare)
2383                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2384
2385         if (l2cache)
2386                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2387
2388         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2389         zc.zc_cookie = replacing;
2390
2391         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2392             &child, &children) != 0 || children != 1) {
2393                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2394                     "new device must be a single disk"));
2395                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2396         }
2397
2398         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2399             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2400
2401         if ((newname = zpool_vdev_name(NULL, NULL, child[0], B_FALSE)) == NULL)
2402                 return (-1);
2403
2404         /*
2405          * If the target is a hot spare that has been swapped in, we can only
2406          * replace it with another hot spare.
2407          */
2408         if (replacing &&
2409             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2410             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2411             NULL) == NULL || !avail_spare) &&
2412             is_replacing_spare(config_root, tgt, 1)) {
2413                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2414                     "can only be replaced by another hot spare"));
2415                 free(newname);
2416                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2417         }
2418
2419         free(newname);
2420
2421         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2422                 return (-1);
2423
2424         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2425
2426         zcmd_free_nvlists(&zc);
2427
2428         if (ret == 0) {
2429                 if (rootpool) {
2430                         /*
2431                          * XXX need a better way to prevent user from
2432                          * booting up a half-baked vdev.
2433                          */
2434                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2435                             "sure to wait until resilver is done "
2436                             "before rebooting.\n"));
2437                 }
2438                 return (0);
2439         }
2440
2441         switch (errno) {
2442         case ENOTSUP:
2443                 /*
2444                  * Can't attach to or replace this type of vdev.
2445                  */
2446                 if (replacing) {
2447                         uint64_t version = zpool_get_prop_int(zhp,
2448                             ZPOOL_PROP_VERSION, NULL);
2449
2450                         if (islog)
2451                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2452                                     "cannot replace a log with a spare"));
2453                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2454                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2455                                     "already in replacing/spare config; wait "
2456                                     "for completion or use 'zpool detach'"));
2457                         else
2458                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2459                                     "cannot replace a replacing device"));
2460                 } else {
2461                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2462                             "can only attach to mirrors and top-level "
2463                             "disks"));
2464                 }
2465                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2466                 break;
2467
2468         case EINVAL:
2469                 /*
2470                  * The new device must be a single disk.
2471                  */
2472                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2473                     "new device must be a single disk"));
2474                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2475                 break;
2476
2477         case EBUSY:
2478                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2479                     new_disk);
2480                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2481                 break;
2482
2483         case EOVERFLOW:
2484                 /*
2485                  * The new device is too small.
2486                  */
2487                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2488                     "device is too small"));
2489                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2490                 break;
2491
2492         case EDOM:
2493                 /*
2494                  * The new device has a different alignment requirement.
2495                  */
2496                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2497                     "devices have different sector alignment"));
2498                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2499                 break;
2500
2501         case ENAMETOOLONG:
2502                 /*
2503                  * The resulting top-level vdev spec won't fit in the label.
2504                  */
2505                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2506                 break;
2507
2508         default:
2509                 (void) zpool_standard_error(hdl, errno, msg);
2510         }
2511
2512         return (-1);
2513 }
2514
2515 /*
2516  * Detach the specified device.
2517  */
2518 int
2519 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2520 {
2521         zfs_cmd_t zc = { 0 };
2522         char msg[1024];
2523         nvlist_t *tgt;
2524         boolean_t avail_spare, l2cache;
2525         libzfs_handle_t *hdl = zhp->zpool_hdl;
2526
2527         (void) snprintf(msg, sizeof (msg),
2528             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2529
2530         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2531         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2532             NULL)) == 0)
2533                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2534
2535         if (avail_spare)
2536                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2537
2538         if (l2cache)
2539                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2540
2541         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2542
2543         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2544                 return (0);
2545
2546         switch (errno) {
2547
2548         case ENOTSUP:
2549                 /*
2550                  * Can't detach from this type of vdev.
2551                  */
2552                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2553                     "applicable to mirror and replacing vdevs"));
2554                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2555                 break;
2556
2557         case EBUSY:
2558                 /*
2559                  * There are no other replicas of this device.
2560                  */
2561                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2562                 break;
2563
2564         default:
2565                 (void) zpool_standard_error(hdl, errno, msg);
2566         }
2567
2568         return (-1);
2569 }
2570
2571 /*
2572  * Find a mirror vdev in the source nvlist.
2573  *
2574  * The mchild array contains a list of disks in one of the top-level mirrors
2575  * of the source pool.  The schild array contains a list of disks that the
2576  * user specified on the command line.  We loop over the mchild array to
2577  * see if any entry in the schild array matches.
2578  *
2579  * If a disk in the mchild array is found in the schild array, we return
2580  * the index of that entry.  Otherwise we return -1.
2581  */
2582 static int
2583 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2584     nvlist_t **schild, uint_t schildren)
2585 {
2586         uint_t mc;
2587
2588         for (mc = 0; mc < mchildren; mc++) {
2589                 uint_t sc;
2590                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2591                     mchild[mc], B_FALSE);
2592
2593                 for (sc = 0; sc < schildren; sc++) {
2594                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2595                             schild[sc], B_FALSE);
2596                         boolean_t result = (strcmp(mpath, spath) == 0);
2597
2598                         free(spath);
2599                         if (result) {
2600                                 free(mpath);
2601                                 return (mc);
2602                         }
2603                 }
2604
2605                 free(mpath);
2606         }
2607
2608         return (-1);
2609 }
2610
2611 /*
2612  * Split a mirror pool.  If newroot points to null, then a new nvlist
2613  * is generated and it is the responsibility of the caller to free it.
2614  */
2615 int
2616 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2617     nvlist_t *props, splitflags_t flags)
2618 {
2619         zfs_cmd_t zc = { 0 };
2620         char msg[1024];
2621         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2622         nvlist_t **varray = NULL, *zc_props = NULL;
2623         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2624         libzfs_handle_t *hdl = zhp->zpool_hdl;
2625         uint64_t vers;
2626         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2627         int retval = 0;
2628
2629         (void) snprintf(msg, sizeof (msg),
2630             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2631
2632         if (!zpool_name_valid(hdl, B_FALSE, newname))
2633                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2634
2635         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2636                 (void) fprintf(stderr, gettext("Internal error: unable to "
2637                     "retrieve pool configuration\n"));
2638                 return (-1);
2639         }
2640
2641         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2642             == 0);
2643         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2644
2645         if (props) {
2646                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2647                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2648                     props, vers, flags, msg)) == NULL)
2649                         return (-1);
2650         }
2651
2652         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2653             &children) != 0) {
2654                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2655                     "Source pool is missing vdev tree"));
2656                 if (zc_props)
2657                         nvlist_free(zc_props);
2658                 return (-1);
2659         }
2660
2661         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2662         vcount = 0;
2663
2664         if (*newroot == NULL ||
2665             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2666             &newchild, &newchildren) != 0)
2667                 newchildren = 0;
2668
2669         for (c = 0; c < children; c++) {
2670                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2671                 char *type;
2672                 nvlist_t **mchild, *vdev;
2673                 uint_t mchildren;
2674                 int entry;
2675
2676                 /*
2677                  * Unlike cache & spares, slogs are stored in the
2678                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2679                  */
2680                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2681                     &is_log);
2682                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2683                     &is_hole);
2684                 if (is_log || is_hole) {
2685                         /*
2686                          * Create a hole vdev and put it in the config.
2687                          */
2688                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2689                                 goto out;
2690                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2691                             VDEV_TYPE_HOLE) != 0)
2692                                 goto out;
2693                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2694                             1) != 0)
2695                                 goto out;
2696                         if (lastlog == 0)
2697                                 lastlog = vcount;
2698                         varray[vcount++] = vdev;
2699                         continue;
2700                 }
2701                 lastlog = 0;
2702                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
2703                     == 0);
2704                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
2705                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2706                             "Source pool must be composed only of mirrors\n"));
2707                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2708                         goto out;
2709                 }
2710
2711                 verify(nvlist_lookup_nvlist_array(child[c],
2712                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2713
2714                 /* find or add an entry for this top-level vdev */
2715                 if (newchildren > 0 &&
2716                     (entry = find_vdev_entry(zhp, mchild, mchildren,
2717                     newchild, newchildren)) >= 0) {
2718                         /* We found a disk that the user specified. */
2719                         vdev = mchild[entry];
2720                         ++found;
2721                 } else {
2722                         /* User didn't specify a disk for this vdev. */
2723                         vdev = mchild[mchildren - 1];
2724                 }
2725
2726                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
2727                         goto out;
2728         }
2729
2730         /* did we find every disk the user specified? */
2731         if (found != newchildren) {
2732                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
2733                     "include at most one disk from each mirror"));
2734                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2735                 goto out;
2736         }
2737
2738         /* Prepare the nvlist for populating. */
2739         if (*newroot == NULL) {
2740                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
2741                         goto out;
2742                 freelist = B_TRUE;
2743                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
2744                     VDEV_TYPE_ROOT) != 0)
2745                         goto out;
2746         } else {
2747                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
2748         }
2749
2750         /* Add all the children we found */
2751         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
2752             lastlog == 0 ? vcount : lastlog) != 0)
2753                 goto out;
2754
2755         /*
2756          * If we're just doing a dry run, exit now with success.
2757          */
2758         if (flags.dryrun) {
2759                 memory_err = B_FALSE;
2760                 freelist = B_FALSE;
2761                 goto out;
2762         }
2763
2764         /* now build up the config list & call the ioctl */
2765         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
2766                 goto out;
2767
2768         if (nvlist_add_nvlist(newconfig,
2769             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
2770             nvlist_add_string(newconfig,
2771             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
2772             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
2773                 goto out;
2774
2775         /*
2776          * The new pool is automatically part of the namespace unless we
2777          * explicitly export it.
2778          */
2779         if (!flags.import)
2780                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
2781         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2782         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
2783         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
2784                 goto out;
2785         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
2786                 goto out;
2787
2788         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
2789                 retval = zpool_standard_error(hdl, errno, msg);
2790                 goto out;
2791         }
2792
2793         freelist = B_FALSE;
2794         memory_err = B_FALSE;
2795
2796 out:
2797         if (varray != NULL) {
2798                 int v;
2799
2800                 for (v = 0; v < vcount; v++)
2801                         nvlist_free(varray[v]);
2802                 free(varray);
2803         }
2804         zcmd_free_nvlists(&zc);
2805         if (zc_props)
2806                 nvlist_free(zc_props);
2807         if (newconfig)
2808                 nvlist_free(newconfig);
2809         if (freelist) {
2810                 nvlist_free(*newroot);
2811                 *newroot = NULL;
2812         }
2813
2814         if (retval != 0)
2815                 return (retval);
2816
2817         if (memory_err)
2818                 return (no_memory(hdl));
2819
2820         return (0);
2821 }
2822
2823 /*
2824  * Remove the given device.  Currently, this is supported only for hot spares
2825  * and level 2 cache devices.
2826  */
2827 int
2828 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
2829 {
2830         zfs_cmd_t zc = { 0 };
2831         char msg[1024];
2832         nvlist_t *tgt;
2833         boolean_t avail_spare, l2cache, islog;
2834         libzfs_handle_t *hdl = zhp->zpool_hdl;
2835         uint64_t version;
2836
2837         (void) snprintf(msg, sizeof (msg),
2838             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
2839
2840         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2841         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2842             &islog)) == 0)
2843                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2844         /*
2845          * XXX - this should just go away.
2846          */
2847         if (!avail_spare && !l2cache && !islog) {
2848                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2849                     "only inactive hot spares, cache, top-level, "
2850                     "or log devices can be removed"));
2851                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2852         }
2853
2854         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
2855         if (islog && version < SPA_VERSION_HOLES) {
2856                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2857                     "pool must be upgrade to support log removal"));
2858                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
2859         }
2860
2861         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2862
2863         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
2864                 return (0);
2865
2866         return (zpool_standard_error(hdl, errno, msg));
2867 }
2868
2869 /*
2870  * Clear the errors for the pool, or the particular device if specified.
2871  */
2872 int
2873 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
2874 {
2875         zfs_cmd_t zc = { 0 };
2876         char msg[1024];
2877         nvlist_t *tgt;
2878         zpool_rewind_policy_t policy;
2879         boolean_t avail_spare, l2cache;
2880         libzfs_handle_t *hdl = zhp->zpool_hdl;
2881         nvlist_t *nvi = NULL;
2882         int error;
2883
2884         if (path)
2885                 (void) snprintf(msg, sizeof (msg),
2886                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
2887                     path);
2888         else
2889                 (void) snprintf(msg, sizeof (msg),
2890                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
2891                     zhp->zpool_name);
2892
2893         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2894         if (path) {
2895                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
2896                     &l2cache, NULL)) == 0)
2897                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
2898
2899                 /*
2900                  * Don't allow error clearing for hot spares.  Do allow
2901                  * error clearing for l2cache devices.
2902                  */
2903                 if (avail_spare)
2904                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
2905
2906                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
2907                     &zc.zc_guid) == 0);
2908         }
2909
2910         zpool_get_rewind_policy(rewindnvl, &policy);
2911         zc.zc_cookie = policy.zrp_request;
2912
2913         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
2914                 return (-1);
2915
2916         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
2917                 return (-1);
2918
2919         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
2920             errno == ENOMEM) {
2921                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
2922                         zcmd_free_nvlists(&zc);
2923                         return (-1);
2924                 }
2925         }
2926
2927         if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
2928             errno != EPERM && errno != EACCES)) {
2929                 if (policy.zrp_request &
2930                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
2931                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
2932                         zpool_rewind_exclaim(hdl, zc.zc_name,
2933                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
2934                             nvi);
2935                         nvlist_free(nvi);
2936                 }
2937                 zcmd_free_nvlists(&zc);
2938                 return (0);
2939         }
2940
2941         zcmd_free_nvlists(&zc);
2942         return (zpool_standard_error(hdl, errno, msg));
2943 }
2944
2945 /*
2946  * Similar to zpool_clear(), but takes a GUID (used by fmd).
2947  */
2948 int
2949 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
2950 {
2951         zfs_cmd_t zc = { 0 };
2952         char msg[1024];
2953         libzfs_handle_t *hdl = zhp->zpool_hdl;
2954
2955         (void) snprintf(msg, sizeof (msg),
2956             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
2957             guid);
2958
2959         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2960         zc.zc_guid = guid;
2961         zc.zc_cookie = ZPOOL_NO_REWIND;
2962
2963         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
2964                 return (0);
2965
2966         return (zpool_standard_error(hdl, errno, msg));
2967 }
2968
2969 /*
2970  * Convert from a devid string to a path.
2971  */
2972 static char *
2973 devid_to_path(char *devid_str)
2974 {
2975         ddi_devid_t devid;
2976         char *minor;
2977         char *path;
2978         devid_nmlist_t *list = NULL;
2979         int ret;
2980
2981         if (devid_str_decode(devid_str, &devid, &minor) != 0)
2982                 return (NULL);
2983
2984         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
2985
2986         devid_str_free(minor);
2987         devid_free(devid);
2988
2989         if (ret != 0)
2990                 return (NULL);
2991
2992         if ((path = strdup(list[0].devname)) == NULL)
2993                 return (NULL);
2994
2995         devid_free_nmlist(list);
2996
2997         return (path);
2998 }
2999
3000 /*
3001  * Convert from a path to a devid string.
3002  */
3003 static char *
3004 path_to_devid(const char *path)
3005 {
3006         int fd;
3007         ddi_devid_t devid;
3008         char *minor, *ret;
3009
3010         if ((fd = open(path, O_RDONLY)) < 0)
3011                 return (NULL);
3012
3013         minor = NULL;
3014         ret = NULL;
3015         if (devid_get(fd, &devid) == 0) {
3016                 if (devid_get_minor_name(fd, &minor) == 0)
3017                         ret = devid_str_encode(devid, minor);
3018                 if (minor != NULL)
3019                         devid_str_free(minor);
3020                 devid_free(devid);
3021         }
3022         (void) close(fd);
3023
3024         return (ret);
3025 }
3026
3027 /*
3028  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3029  * ignore any failure here, since a common case is for an unprivileged user to
3030  * type 'zpool status', and we'll display the correct information anyway.
3031  */
3032 static void
3033 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3034 {
3035         zfs_cmd_t zc = { 0 };
3036
3037         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3038         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3039         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3040             &zc.zc_guid) == 0);
3041
3042         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3043 }
3044
3045 /*
3046  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3047  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3048  * We also check if this is a whole disk, in which case we strip off the
3049  * trailing 's0' slice name.
3050  *
3051  * This routine is also responsible for identifying when disks have been
3052  * reconfigured in a new location.  The kernel will have opened the device by
3053  * devid, but the path will still refer to the old location.  To catch this, we
3054  * first do a path -> devid translation (which is fast for the common case).  If
3055  * the devid matches, we're done.  If not, we do a reverse devid -> path
3056  * translation and issue the appropriate ioctl() to update the path of the vdev.
3057  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3058  * of these checks.
3059  */
3060 char *
3061 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3062     boolean_t verbose)
3063 {
3064         char *path, *devid;
3065         uint64_t value;
3066         char buf[64];
3067         vdev_stat_t *vs;
3068         uint_t vsc;
3069
3070         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT,
3071             &value) == 0) {
3072                 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3073                     &value) == 0);
3074                 (void) snprintf(buf, sizeof (buf), "%llu",
3075                     (u_longlong_t)value);
3076                 path = buf;
3077         } else if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0) {
3078
3079                 /*
3080                  * If the device is dead (faulted, offline, etc) then don't
3081                  * bother opening it.  Otherwise we may be forcing the user to
3082                  * open a misbehaving device, which can have undesirable
3083                  * effects.
3084                  */
3085                 if ((nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3086                     (uint64_t **)&vs, &vsc) != 0 ||
3087                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3088                     zhp != NULL &&
3089                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3090                         /*
3091                          * Determine if the current path is correct.
3092                          */
3093                         char *newdevid = path_to_devid(path);
3094
3095                         if (newdevid == NULL ||
3096                             strcmp(devid, newdevid) != 0) {
3097                                 char *newpath;
3098
3099                                 if ((newpath = devid_to_path(devid)) != NULL) {
3100                                         /*
3101                                          * Update the path appropriately.
3102                                          */
3103                                         set_path(zhp, nv, newpath);
3104                                         if (nvlist_add_string(nv,
3105                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3106                                                 verify(nvlist_lookup_string(nv,
3107                                                     ZPOOL_CONFIG_PATH,
3108                                                     &path) == 0);
3109                                         free(newpath);
3110                                 }
3111                         }
3112
3113                         if (newdevid)
3114                                 devid_str_free(newdevid);
3115                 }
3116
3117                 if (strncmp(path, "/dev/dsk/", 9) == 0)
3118                         path += 9;
3119
3120                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
3121                     &value) == 0 && value) {
3122                         int pathlen = strlen(path);
3123                         char *tmp = zfs_strdup(hdl, path);
3124
3125                         /*
3126                          * If it starts with c#, and ends with "s0", chop
3127                          * the "s0" off, or if it ends with "s0/old", remove
3128                          * the "s0" from the middle.
3129                          */
3130                         if (CTD_CHECK(tmp)) {
3131                                 if (strcmp(&tmp[pathlen - 2], "s0") == 0) {
3132                                         tmp[pathlen - 2] = '\0';
3133                                 } else if (pathlen > 6 &&
3134                                     strcmp(&tmp[pathlen - 6], "s0/old") == 0) {
3135                                         (void) strcpy(&tmp[pathlen - 6],
3136                                             "/old");
3137                                 }
3138                         }
3139                         return (tmp);
3140                 }
3141         } else {
3142                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3143
3144                 /*
3145                  * If it's a raidz device, we need to stick in the parity level.
3146                  */
3147                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3148                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3149                             &value) == 0);
3150                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3151                             (u_longlong_t)value);
3152                         path = buf;
3153                 }
3154
3155                 /*
3156                  * We identify each top-level vdev by using a <type-id>
3157                  * naming convention.
3158                  */
3159                 if (verbose) {
3160                         uint64_t id;
3161
3162                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3163                             &id) == 0);
3164                         (void) snprintf(buf, sizeof (buf), "%s-%llu", path,
3165                             (u_longlong_t)id);
3166                         path = buf;
3167                 }
3168         }
3169
3170         return (zfs_strdup(hdl, path));
3171 }
3172
3173 static int
3174 zbookmark_compare(const void *a, const void *b)
3175 {
3176         return (memcmp(a, b, sizeof (zbookmark_t)));
3177 }
3178
3179 /*
3180  * Retrieve the persistent error log, uniquify the members, and return to the
3181  * caller.
3182  */
3183 int
3184 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3185 {
3186         zfs_cmd_t zc = { 0 };
3187         uint64_t count;
3188         zbookmark_t *zb = NULL;
3189         int i;
3190
3191         /*
3192          * Retrieve the raw error list from the kernel.  If the number of errors
3193          * has increased, allocate more space and continue until we get the
3194          * entire list.
3195          */
3196         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3197             &count) == 0);
3198         if (count == 0)
3199                 return (0);
3200         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3201             count * sizeof (zbookmark_t))) == (uintptr_t)NULL)
3202                 return (-1);
3203         zc.zc_nvlist_dst_size = count;
3204         (void) strcpy(zc.zc_name, zhp->zpool_name);
3205         for (;;) {
3206                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3207                     &zc) != 0) {
3208                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3209                         if (errno == ENOMEM) {
3210                                 count = zc.zc_nvlist_dst_size;
3211                                 if ((zc.zc_nvlist_dst = (uintptr_t)
3212                                     zfs_alloc(zhp->zpool_hdl, count *
3213                                     sizeof (zbookmark_t))) == (uintptr_t)NULL)
3214                                         return (-1);
3215                         } else {
3216                                 return (-1);
3217                         }
3218                 } else {
3219                         break;
3220                 }
3221         }
3222
3223         /*
3224          * Sort the resulting bookmarks.  This is a little confusing due to the
3225          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3226          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3227          * _not_ copied as part of the process.  So we point the start of our
3228          * array appropriate and decrement the total number of elements.
3229          */
3230         zb = ((zbookmark_t *)(uintptr_t)zc.zc_nvlist_dst) +
3231             zc.zc_nvlist_dst_size;
3232         count -= zc.zc_nvlist_dst_size;
3233
3234         qsort(zb, count, sizeof (zbookmark_t), zbookmark_compare);
3235
3236         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3237
3238         /*
3239          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3240          */
3241         for (i = 0; i < count; i++) {
3242                 nvlist_t *nv;
3243
3244                 /* ignoring zb_blkid and zb_level for now */
3245                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3246                     zb[i-1].zb_object == zb[i].zb_object)
3247                         continue;
3248
3249                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3250                         goto nomem;
3251                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3252                     zb[i].zb_objset) != 0) {
3253                         nvlist_free(nv);
3254                         goto nomem;
3255                 }
3256                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3257                     zb[i].zb_object) != 0) {
3258                         nvlist_free(nv);
3259                         goto nomem;
3260                 }
3261                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3262                         nvlist_free(nv);
3263                         goto nomem;
3264                 }
3265                 nvlist_free(nv);
3266         }
3267
3268         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3269         return (0);
3270
3271 nomem:
3272         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3273         return (no_memory(zhp->zpool_hdl));
3274 }
3275
3276 /*
3277  * Upgrade a ZFS pool to the latest on-disk version.
3278  */
3279 int
3280 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3281 {
3282         zfs_cmd_t zc = { 0 };
3283         libzfs_handle_t *hdl = zhp->zpool_hdl;
3284
3285         (void) strcpy(zc.zc_name, zhp->zpool_name);
3286         zc.zc_cookie = new_version;
3287
3288         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3289                 return (zpool_standard_error_fmt(hdl, errno,
3290                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3291                     zhp->zpool_name));
3292         return (0);
3293 }
3294
3295 void
3296 zpool_set_history_str(const char *subcommand, int argc, char **argv,
3297     char *history_str)
3298 {
3299         int i;
3300
3301         (void) strlcpy(history_str, subcommand, HIS_MAX_RECORD_LEN);
3302         for (i = 1; i < argc; i++) {
3303                 if (strlen(history_str) + 1 + strlen(argv[i]) >
3304                     HIS_MAX_RECORD_LEN)
3305                         break;
3306                 (void) strlcat(history_str, " ", HIS_MAX_RECORD_LEN);
3307                 (void) strlcat(history_str, argv[i], HIS_MAX_RECORD_LEN);
3308         }
3309 }
3310
3311 /*
3312  * Stage command history for logging.
3313  */
3314 int
3315 zpool_stage_history(libzfs_handle_t *hdl, const char *history_str)
3316 {
3317         if (history_str == NULL)
3318                 return (EINVAL);
3319
3320         if (strlen(history_str) > HIS_MAX_RECORD_LEN)
3321                 return (EINVAL);
3322
3323         if (hdl->libzfs_log_str != NULL)
3324                 free(hdl->libzfs_log_str);
3325
3326         if ((hdl->libzfs_log_str = strdup(history_str)) == NULL)
3327                 return (no_memory(hdl));
3328
3329         return (0);
3330 }
3331
3332 /*
3333  * Perform ioctl to get some command history of a pool.
3334  *
3335  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3336  * logical offset of the history buffer to start reading from.
3337  *
3338  * Upon return, 'off' is the next logical offset to read from and
3339  * 'len' is the actual amount of bytes read into 'buf'.
3340  */
3341 static int
3342 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3343 {
3344         zfs_cmd_t zc = { 0 };
3345         libzfs_handle_t *hdl = zhp->zpool_hdl;
3346
3347         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3348
3349         zc.zc_history = (uint64_t)(uintptr_t)buf;
3350         zc.zc_history_len = *len;
3351         zc.zc_history_offset = *off;
3352
3353         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3354                 switch (errno) {
3355                 case EPERM:
3356                         return (zfs_error_fmt(hdl, EZFS_PERM,
3357                             dgettext(TEXT_DOMAIN,
3358                             "cannot show history for pool '%s'"),
3359                             zhp->zpool_name));
3360                 case ENOENT:
3361                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3362                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3363                             "'%s'"), zhp->zpool_name));
3364                 case ENOTSUP:
3365                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3366                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3367                             "'%s', pool must be upgraded"), zhp->zpool_name));
3368                 default:
3369                         return (zpool_standard_error_fmt(hdl, errno,
3370                             dgettext(TEXT_DOMAIN,
3371                             "cannot get history for '%s'"), zhp->zpool_name));
3372                 }
3373         }
3374
3375         *len = zc.zc_history_len;
3376         *off = zc.zc_history_offset;
3377
3378         return (0);
3379 }
3380
3381 /*
3382  * Process the buffer of nvlists, unpacking and storing each nvlist record
3383  * into 'records'.  'leftover' is set to the number of bytes that weren't
3384  * processed as there wasn't a complete record.
3385  */
3386 int
3387 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3388     nvlist_t ***records, uint_t *numrecords)
3389 {
3390         uint64_t reclen;
3391         nvlist_t *nv;
3392         int i;
3393
3394         while (bytes_read > sizeof (reclen)) {
3395
3396                 /* get length of packed record (stored as little endian) */
3397                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3398                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3399
3400                 if (bytes_read < sizeof (reclen) + reclen)
3401                         break;
3402
3403                 /* unpack record */
3404                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3405                         return (ENOMEM);
3406                 bytes_read -= sizeof (reclen) + reclen;
3407                 buf += sizeof (reclen) + reclen;
3408
3409                 /* add record to nvlist array */
3410                 (*numrecords)++;
3411                 if (ISP2(*numrecords + 1)) {
3412                         *records = realloc(*records,
3413                             *numrecords * 2 * sizeof (nvlist_t *));
3414                 }
3415                 (*records)[*numrecords - 1] = nv;
3416         }
3417
3418         *leftover = bytes_read;
3419         return (0);
3420 }
3421
3422 #define HIS_BUF_LEN     (128*1024)
3423
3424 /*
3425  * Retrieve the command history of a pool.
3426  */
3427 int
3428 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3429 {
3430         char buf[HIS_BUF_LEN];
3431         uint64_t off = 0;
3432         nvlist_t **records = NULL;
3433         uint_t numrecords = 0;
3434         int err, i;
3435
3436         do {
3437                 uint64_t bytes_read = sizeof (buf);
3438                 uint64_t leftover;
3439
3440                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3441                         break;
3442
3443                 /* if nothing else was read in, we're at EOF, just return */
3444                 if (!bytes_read)
3445                         break;
3446
3447                 if ((err = zpool_history_unpack(buf, bytes_read,
3448                     &leftover, &records, &numrecords)) != 0)
3449                         break;
3450                 off -= leftover;
3451
3452                 /* CONSTCOND */
3453         } while (1);
3454
3455         if (!err) {
3456                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3457                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3458                     records, numrecords) == 0);
3459         }
3460         for (i = 0; i < numrecords; i++)
3461                 nvlist_free(records[i]);
3462         free(records);
3463
3464         return (err);
3465 }
3466
3467 void
3468 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
3469     char *pathname, size_t len)
3470 {
3471         zfs_cmd_t zc = { 0 };
3472         boolean_t mounted = B_FALSE;
3473         char *mntpnt = NULL;
3474         char dsname[MAXNAMELEN];
3475
3476         if (dsobj == 0) {
3477                 /* special case for the MOS */
3478                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>", obj);
3479                 return;
3480         }
3481
3482         /* get the dataset's name */
3483         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3484         zc.zc_obj = dsobj;
3485         if (ioctl(zhp->zpool_hdl->libzfs_fd,
3486             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
3487                 /* just write out a path of two object numbers */
3488                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
3489                     dsobj, obj);
3490                 return;
3491         }
3492         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
3493
3494         /* find out if the dataset is mounted */
3495         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
3496
3497         /* get the corrupted object's path */
3498         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
3499         zc.zc_obj = obj;
3500         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
3501             &zc) == 0) {
3502                 if (mounted) {
3503                         (void) snprintf(pathname, len, "%s%s", mntpnt,
3504                             zc.zc_value);
3505                 } else {
3506                         (void) snprintf(pathname, len, "%s:%s",
3507                             dsname, zc.zc_value);
3508                 }
3509         } else {
3510                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname, obj);
3511         }
3512         free(mntpnt);
3513 }
3514
3515 /*
3516  * Read the EFI label from the config, if a label does not exist then
3517  * pass back the error to the caller. If the caller has passed a non-NULL
3518  * diskaddr argument then we set it to the starting address of the EFI
3519  * partition.
3520  */
3521 static int
3522 read_efi_label(nvlist_t *config, diskaddr_t *sb)
3523 {
3524         char *path;
3525         int fd;
3526         char diskname[MAXPATHLEN];
3527         int err = -1;
3528
3529         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
3530                 return (err);
3531
3532         (void) snprintf(diskname, sizeof (diskname), "%s%s", RDISK_ROOT,
3533             strrchr(path, '/'));
3534         if ((fd = open(diskname, O_RDONLY|O_NDELAY)) >= 0) {
3535                 struct dk_gpt *vtoc;
3536
3537                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
3538                         if (sb != NULL)
3539                                 *sb = vtoc->efi_parts[0].p_start;
3540                         efi_free(vtoc);
3541                 }
3542                 (void) close(fd);
3543         }
3544         return (err);
3545 }
3546
3547 /*
3548  * determine where a partition starts on a disk in the current
3549  * configuration
3550  */
3551 static diskaddr_t
3552 find_start_block(nvlist_t *config)
3553 {
3554         nvlist_t **child;
3555         uint_t c, children;
3556         diskaddr_t sb = MAXOFFSET_T;
3557         uint64_t wholedisk;
3558
3559         if (nvlist_lookup_nvlist_array(config,
3560             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
3561                 if (nvlist_lookup_uint64(config,
3562                     ZPOOL_CONFIG_WHOLE_DISK,
3563                     &wholedisk) != 0 || !wholedisk) {
3564                         return (MAXOFFSET_T);
3565                 }
3566                 if (read_efi_label(config, &sb) < 0)
3567                         sb = MAXOFFSET_T;
3568                 return (sb);
3569         }
3570
3571         for (c = 0; c < children; c++) {
3572                 sb = find_start_block(child[c]);
3573                 if (sb != MAXOFFSET_T) {
3574                         return (sb);
3575                 }
3576         }
3577         return (MAXOFFSET_T);
3578 }
3579
3580 /*
3581  * Label an individual disk.  The name provided is the short name,
3582  * stripped of any leading /dev path.
3583  */
3584 int
3585 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, char *name)
3586 {
3587         char path[MAXPATHLEN];
3588         struct dk_gpt *vtoc;
3589         int fd;
3590         size_t resv = EFI_MIN_RESV_SIZE;
3591         uint64_t slice_size;
3592         diskaddr_t start_block;
3593         char errbuf[1024];
3594
3595         /* prepare an error message just in case */
3596         (void) snprintf(errbuf, sizeof (errbuf),
3597             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
3598
3599         if (zhp) {
3600                 nvlist_t *nvroot;
3601
3602                 if (pool_is_bootable(zhp)) {
3603                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3604                             "EFI labeled devices are not supported on root "
3605                             "pools."));
3606                         return (zfs_error(hdl, EZFS_POOL_NOTSUP, errbuf));
3607                 }
3608
3609                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
3610                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3611
3612                 if (zhp->zpool_start_block == 0)
3613                         start_block = find_start_block(nvroot);
3614                 else
3615                         start_block = zhp->zpool_start_block;
3616                 zhp->zpool_start_block = start_block;
3617         } else {
3618                 /* new pool */
3619                 start_block = NEW_START_BLOCK;
3620         }
3621
3622         (void) snprintf(path, sizeof (path), "%s/%s%s", RDISK_ROOT, name,
3623             BACKUP_SLICE);
3624
3625         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
3626                 /*
3627                  * This shouldn't happen.  We've long since verified that this
3628                  * is a valid device.
3629                  */
3630                 zfs_error_aux(hdl,
3631                     dgettext(TEXT_DOMAIN, "unable to open device"));
3632                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
3633         }
3634
3635         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
3636                 /*
3637                  * The only way this can fail is if we run out of memory, or we
3638                  * were unable to read the disk's capacity
3639                  */
3640                 if (errno == ENOMEM)
3641                         (void) no_memory(hdl);
3642
3643                 (void) close(fd);
3644                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3645                     "unable to read disk capacity"), name);
3646
3647                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
3648         }
3649
3650         slice_size = vtoc->efi_last_u_lba + 1;
3651         slice_size -= EFI_MIN_RESV_SIZE;
3652         if (start_block == MAXOFFSET_T)
3653                 start_block = NEW_START_BLOCK;
3654         slice_size -= start_block;
3655
3656         vtoc->efi_parts[0].p_start = start_block;
3657         vtoc->efi_parts[0].p_size = slice_size;
3658
3659         /*
3660          * Why we use V_USR: V_BACKUP confuses users, and is considered
3661          * disposable by some EFI utilities (since EFI doesn't have a backup
3662          * slice).  V_UNASSIGNED is supposed to be used only for zero size
3663          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
3664          * etc. were all pretty specific.  V_USR is as close to reality as we
3665          * can get, in the absence of V_OTHER.
3666          */
3667         vtoc->efi_parts[0].p_tag = V_USR;
3668         (void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
3669
3670         vtoc->efi_parts[8].p_start = slice_size + start_block;
3671         vtoc->efi_parts[8].p_size = resv;
3672         vtoc->efi_parts[8].p_tag = V_RESERVED;
3673
3674         if (efi_write(fd, vtoc) != 0) {
3675                 /*
3676                  * Some block drivers (like pcata) may not support EFI
3677                  * GPT labels.  Print out a helpful error message dir-
3678                  * ecting the user to manually label the disk and give
3679                  * a specific slice.
3680                  */
3681                 (void) close(fd);
3682                 efi_free(vtoc);
3683
3684                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3685                     "try using fdisk(1M) and then provide a specific slice"));
3686                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
3687         }
3688
3689         (void) close(fd);
3690         efi_free(vtoc);
3691         return (0);
3692 }
3693
3694 static boolean_t
3695 supported_dump_vdev_type(libzfs_handle_t *hdl, nvlist_t *config, char *errbuf)
3696 {
3697         char *type;
3698         nvlist_t **child;
3699         uint_t children, c;
3700
3701         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_TYPE, &type) == 0);
3702         if (strcmp(type, VDEV_TYPE_RAIDZ) == 0 ||
3703             strcmp(type, VDEV_TYPE_FILE) == 0 ||
3704             strcmp(type, VDEV_TYPE_LOG) == 0 ||
3705             strcmp(type, VDEV_TYPE_HOLE) == 0 ||
3706             strcmp(type, VDEV_TYPE_MISSING) == 0) {
3707                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3708                     "vdev type '%s' is not supported"), type);
3709                 (void) zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf);
3710                 return (B_FALSE);
3711         }
3712         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
3713             &child, &children) == 0) {
3714                 for (c = 0; c < children; c++) {
3715                         if (!supported_dump_vdev_type(hdl, child[c], errbuf))
3716                                 return (B_FALSE);
3717                 }
3718         }
3719         return (B_TRUE);
3720 }
3721
3722 /*
3723  * check if this zvol is allowable for use as a dump device; zero if
3724  * it is, > 0 if it isn't, < 0 if it isn't a zvol
3725  */
3726 int
3727 zvol_check_dump_config(char *arg)
3728 {
3729         zpool_handle_t *zhp = NULL;
3730         nvlist_t *config, *nvroot;
3731         char *p, *volname;
3732         nvlist_t **top;
3733         uint_t toplevels;
3734         libzfs_handle_t *hdl;
3735         char errbuf[1024];
3736         char poolname[ZPOOL_MAXNAMELEN];
3737         int pathlen = strlen(ZVOL_FULL_DEV_DIR);
3738         int ret = 1;
3739
3740         if (strncmp(arg, ZVOL_FULL_DEV_DIR, pathlen)) {
3741                 return (-1);
3742         }
3743
3744         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3745             "dump is not supported on device '%s'"), arg);
3746
3747         if ((hdl = libzfs_init()) == NULL)
3748                 return (1);
3749         libzfs_print_on_error(hdl, B_TRUE);
3750
3751         volname = arg + pathlen;
3752
3753         /* check the configuration of the pool */
3754         if ((p = strchr(volname, '/')) == NULL) {
3755                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3756                     "malformed dataset name"));
3757                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
3758                 return (1);
3759         } else if (p - volname >= ZFS_MAXNAMELEN) {
3760                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3761                     "dataset name is too long"));
3762                 (void) zfs_error(hdl, EZFS_NAMETOOLONG, errbuf);
3763                 return (1);
3764         } else {
3765                 (void) strncpy(poolname, volname, p - volname);
3766                 poolname[p - volname] = '\0';
3767         }
3768
3769         if ((zhp = zpool_open(hdl, poolname)) == NULL) {
3770                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3771                     "could not open pool '%s'"), poolname);
3772                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
3773                 goto out;
3774         }
3775         config = zpool_get_config(zhp, NULL);
3776         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
3777             &nvroot) != 0) {
3778                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3779                     "could not obtain vdev configuration for  '%s'"), poolname);
3780                 (void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
3781                 goto out;
3782         }
3783
3784         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
3785             &top, &toplevels) == 0);
3786         if (toplevels != 1) {
3787                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3788                     "'%s' has multiple top level vdevs"), poolname);
3789                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, errbuf);
3790                 goto out;
3791         }
3792
3793         if (!supported_dump_vdev_type(hdl, top[0], errbuf)) {
3794                 goto out;
3795         }
3796         ret = 0;
3797
3798 out:
3799         if (zhp)
3800                 zpool_close(zhp);
3801         libzfs_fini(hdl);
3802         return (ret);
3803 }