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

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21
  22 /*
  23  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
  24  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  25  * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
  26  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  27  */
  28
  29 #include <sys/types.h>
  30 #include <sys/stat.h>
  31 #include <ctype.h>
  32 #include <errno.h>
  33 #include <devid.h>
  34 #include <fcntl.h>
  35 #include <libintl.h>
  36 #include <stdio.h>
  37 #include <stdlib.h>
  38 #include <strings.h>
  39 #include <unistd.h>
  40 #include <libgen.h>
  41 #include <sys/zfs_ioctl.h>
  42 #include <dlfcn.h>
  43
  44 #include "zfs_namecheck.h"
  45 #include "zfs_prop.h"
  46 #include "libzfs_impl.h"
  47 #include "zfs_comutil.h"
  48 #include "zfeature_common.h"
  49
  50 static int read_efi_label(nvlist_t *config, diskaddr_t *sb);
  51
  52 #define DISK_ROOT       "/dev/dsk"
  53 #define RDISK_ROOT      "/dev/rdsk"
  54 #define BACKUP_SLICE    "s2"
  55
  56 typedef struct prop_flags {
  57         int create:1;   /* Validate property on creation */
  58         int import:1;   /* Validate property on import */
  59 } prop_flags_t;
  60
  61 /*
  62  * ====================================================================
  63  *   zpool property functions
  64  * ====================================================================
  65  */
  66
  67 static int
  68 zpool_get_all_props(zpool_handle_t *zhp)
  69 {
  70         zfs_cmd_t zc = { 0 };
  71         libzfs_handle_t *hdl = zhp->zpool_hdl;
  72
  73         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
  74
  75         if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
  76                 return (-1);
  77
  78         while (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_PROPS, &zc) != 0) {
  79                 if (errno == ENOMEM) {
  80                         if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
  81                                 zcmd_free_nvlists(&zc);
  82                                 return (-1);
  83                         }
  84                 } else {
  85                         zcmd_free_nvlists(&zc);
  86                         return (-1);
  87                 }
  88         }
  89
  90         if (zcmd_read_dst_nvlist(hdl, &zc, &zhp->zpool_props) != 0) {
  91                 zcmd_free_nvlists(&zc);
  92                 return (-1);
  93         }
  94
  95         zcmd_free_nvlists(&zc);
  96
  97         return (0);
  98 }
  99
 100 static int
 101 zpool_props_refresh(zpool_handle_t *zhp)
 102 {
 103         nvlist_t *old_props;
 104
 105         old_props = zhp->zpool_props;
 106
 107         if (zpool_get_all_props(zhp) != 0)
 108                 return (-1);
 109
 110         nvlist_free(old_props);
 111         return (0);
 112 }
 113
 114 static char *
 115 zpool_get_prop_string(zpool_handle_t *zhp, zpool_prop_t prop,
 116     zprop_source_t *src)
 117 {
 118         nvlist_t *nv, *nvl;
 119         uint64_t ival;
 120         char *value;
 121         zprop_source_t source;
 122
 123         nvl = zhp->zpool_props;
 124         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 125                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &ival) == 0);
 126                 source = ival;
 127                 verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
 128         } else {
 129                 source = ZPROP_SRC_DEFAULT;
 130                 if ((value = (char *)zpool_prop_default_string(prop)) == NULL)
 131                         value = "-";
 132         }
 133
 134         if (src)
 135                 *src = source;
 136
 137         return (value);
 138 }
 139
 140 uint64_t
 141 zpool_get_prop_int(zpool_handle_t *zhp, zpool_prop_t prop, zprop_source_t *src)
 142 {
 143         nvlist_t *nv, *nvl;
 144         uint64_t value;
 145         zprop_source_t source;
 146
 147         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp)) {
 148                 /*
 149                  * zpool_get_all_props() has most likely failed because
 150                  * the pool is faulted, but if all we need is the top level
 151                  * vdev's guid then get it from the zhp config nvlist.
 152                  */
 153                 if ((prop == ZPOOL_PROP_GUID) &&
 154                     (nvlist_lookup_nvlist(zhp->zpool_config,
 155                     ZPOOL_CONFIG_VDEV_TREE, &nv) == 0) &&
 156                     (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &value)
 157                     == 0)) {
 158                         return (value);
 159                 }
 160                 return (zpool_prop_default_numeric(prop));
 161         }
 162
 163         nvl = zhp->zpool_props;
 164         if (nvlist_lookup_nvlist(nvl, zpool_prop_to_name(prop), &nv) == 0) {
 165                 verify(nvlist_lookup_uint64(nv, ZPROP_SOURCE, &value) == 0);
 166                 source = value;
 167                 verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
 168         } else {
 169                 source = ZPROP_SRC_DEFAULT;
 170                 value = zpool_prop_default_numeric(prop);
 171         }
 172
 173         if (src)
 174                 *src = source;
 175
 176         return (value);
 177 }
 178
 179 /*
 180  * Map VDEV STATE to printed strings.
 181  */
 182 const char *
 183 zpool_state_to_name(vdev_state_t state, vdev_aux_t aux)
 184 {
 185         switch (state) {
 186         case VDEV_STATE_CLOSED:
 187         case VDEV_STATE_OFFLINE:
 188                 return (gettext("OFFLINE"));
 189         case VDEV_STATE_REMOVED:
 190                 return (gettext("REMOVED"));
 191         case VDEV_STATE_CANT_OPEN:
 192                 if (aux == VDEV_AUX_CORRUPT_DATA || aux == VDEV_AUX_BAD_LOG)
 193                         return (gettext("FAULTED"));
 194                 else if (aux == VDEV_AUX_SPLIT_POOL)
 195                         return (gettext("SPLIT"));
 196                 else
 197                         return (gettext("UNAVAIL"));
 198         case VDEV_STATE_FAULTED:
 199                 return (gettext("FAULTED"));
 200         case VDEV_STATE_DEGRADED:
 201                 return (gettext("DEGRADED"));
 202         case VDEV_STATE_HEALTHY:
 203                 return (gettext("ONLINE"));
 204         }
 205
 206         return (gettext("UNKNOWN"));
 207 }
 208
 209 /*
 210  * Map POOL STATE to printed strings.
 211  */
 212 const char *
 213 zpool_pool_state_to_name(pool_state_t state)
 214 {
 215         switch (state) {
 216         case POOL_STATE_ACTIVE:
 217                 return (gettext("ACTIVE"));
 218         case POOL_STATE_EXPORTED:
 219                 return (gettext("EXPORTED"));
 220         case POOL_STATE_DESTROYED:
 221                 return (gettext("DESTROYED"));
 222         case POOL_STATE_SPARE:
 223                 return (gettext("SPARE"));
 224         case POOL_STATE_L2CACHE:
 225                 return (gettext("L2CACHE"));
 226         case POOL_STATE_UNINITIALIZED:
 227                 return (gettext("UNINITIALIZED"));
 228         case POOL_STATE_UNAVAIL:
 229                 return (gettext("UNAVAIL"));
 230         case POOL_STATE_POTENTIALLY_ACTIVE:
 231                 return (gettext("POTENTIALLY_ACTIVE"));
 232         }
 233
 234         return (gettext("UNKNOWN"));
 235 }
 236
 237 /*
 238  * Get a zpool property value for 'prop' and return the value in
 239  * a pre-allocated buffer.
 240  */
 241 int
 242 zpool_get_prop(zpool_handle_t *zhp, zpool_prop_t prop, char *buf, size_t len,
 243     zprop_source_t *srctype, boolean_t literal)
 244 {
 245         uint64_t intval;
 246         const char *strval;
 247         zprop_source_t src = ZPROP_SRC_NONE;
 248         nvlist_t *nvroot;
 249         vdev_stat_t *vs;
 250         uint_t vsc;
 251
 252         if (zpool_get_state(zhp) == POOL_STATE_UNAVAIL) {
 253                 switch (prop) {
 254                 case ZPOOL_PROP_NAME:
 255                         (void) strlcpy(buf, zpool_get_name(zhp), len);
 256                         break;
 257
 258                 case ZPOOL_PROP_HEALTH:
 259                         (void) strlcpy(buf,
 260                             zpool_pool_state_to_name(POOL_STATE_UNAVAIL), len);
 261                         break;
 262
 263                 case ZPOOL_PROP_GUID:
 264                         intval = zpool_get_prop_int(zhp, prop, &src);
 265                         (void) snprintf(buf, len, "%llu", intval);
 266                         break;
 267
 268                 case ZPOOL_PROP_ALTROOT:
 269                 case ZPOOL_PROP_CACHEFILE:
 270                 case ZPOOL_PROP_COMMENT:
 271                         if (zhp->zpool_props != NULL ||
 272                             zpool_get_all_props(zhp) == 0) {
 273                                 (void) strlcpy(buf,
 274                                     zpool_get_prop_string(zhp, prop, &src),
 275                                     len);
 276                                 break;
 277                         }
 278                         /* FALLTHROUGH */
 279                 default:
 280                         (void) strlcpy(buf, "-", len);
 281                         break;
 282                 }
 283
 284                 if (srctype != NULL)
 285                         *srctype = src;
 286                 return (0);
 287         }
 288
 289         if (zhp->zpool_props == NULL && zpool_get_all_props(zhp) &&
 290             prop != ZPOOL_PROP_NAME)
 291                 return (-1);
 292
 293         switch (zpool_prop_get_type(prop)) {
 294         case PROP_TYPE_STRING:
 295                 (void) strlcpy(buf, zpool_get_prop_string(zhp, prop, &src),
 296                     len);
 297                 break;
 298
 299         case PROP_TYPE_NUMBER:
 300                 intval = zpool_get_prop_int(zhp, prop, &src);
 301
 302                 switch (prop) {
 303                 case ZPOOL_PROP_SIZE:
 304                 case ZPOOL_PROP_ALLOCATED:
 305                 case ZPOOL_PROP_FREE:
 306                 case ZPOOL_PROP_FREEING:
 307                 case ZPOOL_PROP_LEAKED:
 308                         if (literal) {
 309                                 (void) snprintf(buf, len, "%llu",
 310                                     (u_longlong_t)intval);
 311                         } else {
 312                                 (void) zfs_nicenum(intval, buf, len);
 313                         }
 314                         break;
 315                 case ZPOOL_PROP_EXPANDSZ:
 316                         if (intval == 0) {
 317                                 (void) strlcpy(buf, "-", len);
 318                         } else if (literal) {
 319                                 (void) snprintf(buf, len, "%llu",
 320                                     (u_longlong_t)intval);
 321                         } else {
 322                                 (void) zfs_nicenum(intval, buf, len);
 323                         }
 324                         break;
 325                 case ZPOOL_PROP_CAPACITY:
 326                         if (literal) {
 327                                 (void) snprintf(buf, len, "%llu",
 328                                     (u_longlong_t)intval);
 329                         } else {
 330                                 (void) snprintf(buf, len, "%llu%%",
 331                                     (u_longlong_t)intval);
 332                         }
 333                         break;
 334                 case ZPOOL_PROP_FRAGMENTATION:
 335                         if (intval == UINT64_MAX) {
 336                                 (void) strlcpy(buf, "-", len);
 337                         } else {
 338                                 (void) snprintf(buf, len, "%llu%%",
 339                                     (u_longlong_t)intval);
 340                         }
 341                         break;
 342                 case ZPOOL_PROP_DEDUPRATIO:
 343                         (void) snprintf(buf, len, "%llu.%02llux",
 344                             (u_longlong_t)(intval / 100),
 345                             (u_longlong_t)(intval % 100));
 346                         break;
 347                 case ZPOOL_PROP_HEALTH:
 348                         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
 349                             ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
 350                         verify(nvlist_lookup_uint64_array(nvroot,
 351                             ZPOOL_CONFIG_VDEV_STATS, (uint64_t **)&vs, &vsc)
 352                             == 0);
 353
 354                         (void) strlcpy(buf, zpool_state_to_name(intval,
 355                             vs->vs_aux), len);
 356                         break;
 357                 case ZPOOL_PROP_VERSION:
 358                         if (intval >= SPA_VERSION_FEATURES) {
 359                                 (void) snprintf(buf, len, "-");
 360                                 break;
 361                         }
 362                         /* FALLTHROUGH */
 363                 default:
 364                         (void) snprintf(buf, len, "%llu", intval);
 365                 }
 366                 break;
 367
 368         case PROP_TYPE_INDEX:
 369                 intval = zpool_get_prop_int(zhp, prop, &src);
 370                 if (zpool_prop_index_to_string(prop, intval, &strval)
 371                     != 0)
 372                         return (-1);
 373                 (void) strlcpy(buf, strval, len);
 374                 break;
 375
 376         default:
 377                 abort();
 378         }
 379
 380         if (srctype)
 381                 *srctype = src;
 382
 383         return (0);
 384 }
 385
 386 /*
 387  * Check if the bootfs name has the same pool name as it is set to.
 388  * Assuming bootfs is a valid dataset name.
 389  */
 390 static boolean_t
 391 bootfs_name_valid(const char *pool, char *bootfs)
 392 {
 393         int len = strlen(pool);
 394
 395         if (!zfs_name_valid(bootfs, ZFS_TYPE_FILESYSTEM|ZFS_TYPE_SNAPSHOT))
 396                 return (B_FALSE);
 397
 398         if (strncmp(pool, bootfs, len) == 0 &&
 399             (bootfs[len] == '/' || bootfs[len] == '\0'))
 400                 return (B_TRUE);
 401
 402         return (B_FALSE);
 403 }
 404
 405 boolean_t
 406 zpool_is_bootable(zpool_handle_t *zhp)
 407 {
 408         char bootfs[ZPOOL_MAXNAMELEN];
 409
 410         return (zpool_get_prop(zhp, ZPOOL_PROP_BOOTFS, bootfs,
 411             sizeof (bootfs), NULL, B_FALSE) == 0 && strncmp(bootfs, "-",
 412             sizeof (bootfs)) != 0);
 413 }
 414
 415
 416 /*
 417  * Given an nvlist of zpool properties to be set, validate that they are
 418  * correct, and parse any numeric properties (index, boolean, etc) if they are
 419  * specified as strings.
 420  */
 421 static nvlist_t *
 422 zpool_valid_proplist(libzfs_handle_t *hdl, const char *poolname,
 423     nvlist_t *props, uint64_t version, prop_flags_t flags, char *errbuf)
 424 {
 425         nvpair_t *elem;
 426         nvlist_t *retprops;
 427         zpool_prop_t prop;
 428         char *strval;
 429         uint64_t intval;
 430         char *slash, *check;
 431         struct stat64 statbuf;
 432         zpool_handle_t *zhp;
 433
 434         if (nvlist_alloc(&retprops, NV_UNIQUE_NAME, 0) != 0) {
 435                 (void) no_memory(hdl);
 436                 return (NULL);
 437         }
 438
 439         elem = NULL;
 440         while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
 441                 const char *propname = nvpair_name(elem);
 442
 443                 prop = zpool_name_to_prop(propname);
 444                 if (prop == ZPROP_INVAL && zpool_prop_feature(propname)) {
 445                         int err;
 446                         char *fname = strchr(propname, '@') + 1;
 447
 448                         err = zfeature_lookup_name(fname, NULL);
 449                         if (err != 0) {
 450                                 ASSERT3U(err, ==, ENOENT);
 451                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 452                                     "invalid feature '%s'"), fname);
 453                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 454                                 goto error;
 455                         }
 456
 457                         if (nvpair_type(elem) != DATA_TYPE_STRING) {
 458                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 459                                     "'%s' must be a string"), propname);
 460                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 461                                 goto error;
 462                         }
 463
 464                         (void) nvpair_value_string(elem, &strval);
 465                         if (strcmp(strval, ZFS_FEATURE_ENABLED) != 0) {
 466                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 467                                     "property '%s' can only be set to "
 468                                     "'enabled'"), propname);
 469                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 470                                 goto error;
 471                         }
 472
 473                         if (nvlist_add_uint64(retprops, propname, 0) != 0) {
 474                                 (void) no_memory(hdl);
 475                                 goto error;
 476                         }
 477                         continue;
 478                 }
 479
 480                 /*
 481                  * Make sure this property is valid and applies to this type.
 482                  */
 483                 if (prop == ZPROP_INVAL) {
 484                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 485                             "invalid property '%s'"), propname);
 486                         (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 487                         goto error;
 488                 }
 489
 490                 if (zpool_prop_readonly(prop)) {
 491                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 492                             "is readonly"), propname);
 493                         (void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
 494                         goto error;
 495                 }
 496
 497                 if (zprop_parse_value(hdl, elem, prop, ZFS_TYPE_POOL, retprops,
 498                     &strval, &intval, errbuf) != 0)
 499                         goto error;
 500
 501                 /*
 502                  * Perform additional checking for specific properties.
 503                  */
 504                 switch (prop) {
 505                 case ZPOOL_PROP_VERSION:
 506                         if (intval < version ||
 507                             !SPA_VERSION_IS_SUPPORTED(intval)) {
 508                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 509                                     "property '%s' number %d is invalid."),
 510                                     propname, intval);
 511                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 512                                 goto error;
 513                         }
 514                         break;
 515
 516                 case ZPOOL_PROP_BOOTFS:
 517                         if (flags.create || flags.import) {
 518                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 519                                     "property '%s' cannot be set at creation "
 520                                     "or import time"), propname);
 521                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 522                                 goto error;
 523                         }
 524
 525                         if (version < SPA_VERSION_BOOTFS) {
 526                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 527                                     "pool must be upgraded to support "
 528                                     "'%s' property"), propname);
 529                                 (void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
 530                                 goto error;
 531                         }
 532
 533                         /*
 534                          * bootfs property value has to be a dataset name and
 535                          * the dataset has to be in the same pool as it sets to.
 536                          */
 537                         if (strval[0] != '\0' && !bootfs_name_valid(poolname,
 538                             strval)) {
 539                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "'%s' "
 540                                     "is an invalid name"), strval);
 541                                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
 542                                 goto error;
 543                         }
 544
 545                         if ((zhp = zpool_open_canfail(hdl, poolname)) == NULL) {
 546                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 547                                     "could not open pool '%s'"), poolname);
 548                                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
 549                                 goto error;
 550                         }
 551                         zpool_close(zhp);
 552                         break;
 553
 554                 case ZPOOL_PROP_ALTROOT:
 555                         if (!flags.create && !flags.import) {
 556                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 557                                     "property '%s' can only be set during pool "
 558                                     "creation or import"), propname);
 559                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 560                                 goto error;
 561                         }
 562
 563                         if (strval[0] != '/') {
 564                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 565                                     "bad alternate root '%s'"), strval);
 566                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 567                                 goto error;
 568                         }
 569                         break;
 570
 571                 case ZPOOL_PROP_CACHEFILE:
 572                         if (strval[0] == '\0')
 573                                 break;
 574
 575                         if (strcmp(strval, "none") == 0)
 576                                 break;
 577
 578                         if (strval[0] != '/') {
 579                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 580                                     "property '%s' must be empty, an "
 581                                     "absolute path, or 'none'"), propname);
 582                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 583                                 goto error;
 584                         }
 585
 586                         slash = strrchr(strval, '/');
 587
 588                         if (slash[1] == '\0' || strcmp(slash, "/.") == 0 ||
 589                             strcmp(slash, "/..") == 0) {
 590                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 591                                     "'%s' is not a valid file"), strval);
 592                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 593                                 goto error;
 594                         }
 595
 596                         *slash = '\0';
 597
 598                         if (strval[0] != '\0' &&
 599                             (stat64(strval, &statbuf) != 0 ||
 600                             !S_ISDIR(statbuf.st_mode))) {
 601                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 602                                     "'%s' is not a valid directory"),
 603                                     strval);
 604                                 (void) zfs_error(hdl, EZFS_BADPATH, errbuf);
 605                                 goto error;
 606                         }
 607
 608                         *slash = '/';
 609                         break;
 610
 611                 case ZPOOL_PROP_COMMENT:
 612                         for (check = strval; *check != '\0'; check++) {
 613                                 if (!isprint(*check)) {
 614                                         zfs_error_aux(hdl,
 615                                             dgettext(TEXT_DOMAIN,
 616                                             "comment may only have printable "
 617                                             "characters"));
 618                                         (void) zfs_error(hdl, EZFS_BADPROP,
 619                                             errbuf);
 620                                         goto error;
 621                                 }
 622                         }
 623                         if (strlen(strval) > ZPROP_MAX_COMMENT) {
 624                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 625                                     "comment must not exceed %d characters"),
 626                                     ZPROP_MAX_COMMENT);
 627                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 628                                 goto error;
 629                         }
 630                         break;
 631                 case ZPOOL_PROP_READONLY:
 632                         if (!flags.import) {
 633                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 634                                     "property '%s' can only be set at "
 635                                     "import time"), propname);
 636                                 (void) zfs_error(hdl, EZFS_BADPROP, errbuf);
 637                                 goto error;
 638                         }
 639                         break;
 640                 }
 641         }
 642
 643         return (retprops);
 644 error:
 645         nvlist_free(retprops);
 646         return (NULL);
 647 }
 648
 649 /*
 650  * Set zpool property : propname=propval.
 651  */
 652 int
 653 zpool_set_prop(zpool_handle_t *zhp, const char *propname, const char *propval)
 654 {
 655         zfs_cmd_t zc = { 0 };
 656         int ret = -1;
 657         char errbuf[1024];
 658         nvlist_t *nvl = NULL;
 659         nvlist_t *realprops;
 660         uint64_t version;
 661         prop_flags_t flags = { 0 };
 662
 663         (void) snprintf(errbuf, sizeof (errbuf),
 664             dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
 665             zhp->zpool_name);
 666
 667         if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0)
 668                 return (no_memory(zhp->zpool_hdl));
 669
 670         if (nvlist_add_string(nvl, propname, propval) != 0) {
 671                 nvlist_free(nvl);
 672                 return (no_memory(zhp->zpool_hdl));
 673         }
 674
 675         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
 676         if ((realprops = zpool_valid_proplist(zhp->zpool_hdl,
 677             zhp->zpool_name, nvl, version, flags, errbuf)) == NULL) {
 678                 nvlist_free(nvl);
 679                 return (-1);
 680         }
 681
 682         nvlist_free(nvl);
 683         nvl = realprops;
 684
 685         /*
 686          * Execute the corresponding ioctl() to set this property.
 687          */
 688         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
 689
 690         if (zcmd_write_src_nvlist(zhp->zpool_hdl, &zc, nvl) != 0) {
 691                 nvlist_free(nvl);
 692                 return (-1);
 693         }
 694
 695         ret = zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_SET_PROPS, &zc);
 696
 697         zcmd_free_nvlists(&zc);
 698         nvlist_free(nvl);
 699
 700         if (ret)
 701                 (void) zpool_standard_error(zhp->zpool_hdl, errno, errbuf);
 702         else
 703                 (void) zpool_props_refresh(zhp);
 704
 705         return (ret);
 706 }
 707
 708 int
 709 zpool_expand_proplist(zpool_handle_t *zhp, zprop_list_t **plp)
 710 {
 711         libzfs_handle_t *hdl = zhp->zpool_hdl;
 712         zprop_list_t *entry;
 713         char buf[ZFS_MAXPROPLEN];
 714         nvlist_t *features = NULL;
 715         zprop_list_t **last;
 716         boolean_t firstexpand = (NULL == *plp);
 717
 718         if (zprop_expand_list(hdl, plp, ZFS_TYPE_POOL) != 0)
 719                 return (-1);
 720
 721         last = plp;
 722         while (*last != NULL)
 723                 last = &(*last)->pl_next;
 724
 725         if ((*plp)->pl_all)
 726                 features = zpool_get_features(zhp);
 727
 728         if ((*plp)->pl_all && firstexpand) {
 729                 for (int i = 0; i < SPA_FEATURES; i++) {
 730                         zprop_list_t *entry = zfs_alloc(hdl,
 731                             sizeof (zprop_list_t));
 732                         entry->pl_prop = ZPROP_INVAL;
 733                         entry->pl_user_prop = zfs_asprintf(hdl, "feature@%s",
 734                             spa_feature_table[i].fi_uname);
 735                         entry->pl_width = strlen(entry->pl_user_prop);
 736                         entry->pl_all = B_TRUE;
 737
 738                         *last = entry;
 739                         last = &entry->pl_next;
 740                 }
 741         }
 742
 743         /* add any unsupported features */
 744         for (nvpair_t *nvp = nvlist_next_nvpair(features, NULL);
 745             nvp != NULL; nvp = nvlist_next_nvpair(features, nvp)) {
 746                 char *propname;
 747                 boolean_t found;
 748                 zprop_list_t *entry;
 749
 750                 if (zfeature_is_supported(nvpair_name(nvp)))
 751                         continue;
 752
 753                 propname = zfs_asprintf(hdl, "unsupported@%s",
 754                     nvpair_name(nvp));
 755
 756                 /*
 757                  * Before adding the property to the list make sure that no
 758                  * other pool already added the same property.
 759                  */
 760                 found = B_FALSE;
 761                 entry = *plp;
 762                 while (entry != NULL) {
 763                         if (entry->pl_user_prop != NULL &&
 764                             strcmp(propname, entry->pl_user_prop) == 0) {
 765                                 found = B_TRUE;
 766                                 break;
 767                         }
 768                         entry = entry->pl_next;
 769                 }
 770                 if (found) {
 771                         free(propname);
 772                         continue;
 773                 }
 774
 775                 entry = zfs_alloc(hdl, sizeof (zprop_list_t));
 776                 entry->pl_prop = ZPROP_INVAL;
 777                 entry->pl_user_prop = propname;
 778                 entry->pl_width = strlen(entry->pl_user_prop);
 779                 entry->pl_all = B_TRUE;
 780
 781                 *last = entry;
 782                 last = &entry->pl_next;
 783         }
 784
 785         for (entry = *plp; entry != NULL; entry = entry->pl_next) {
 786
 787                 if (entry->pl_fixed)
 788                         continue;
 789
 790                 if (entry->pl_prop != ZPROP_INVAL &&
 791                     zpool_get_prop(zhp, entry->pl_prop, buf, sizeof (buf),
 792                     NULL, B_FALSE) == 0) {
 793                         if (strlen(buf) > entry->pl_width)
 794                                 entry->pl_width = strlen(buf);
 795                 }
 796         }
 797
 798         return (0);
 799 }
 800
 801 /*
 802  * Get the state for the given feature on the given ZFS pool.
 803  */
 804 int
 805 zpool_prop_get_feature(zpool_handle_t *zhp, const char *propname, char *buf,
 806     size_t len)
 807 {
 808         uint64_t refcount;
 809         boolean_t found = B_FALSE;
 810         nvlist_t *features = zpool_get_features(zhp);
 811         boolean_t supported;
 812         const char *feature = strchr(propname, '@') + 1;
 813
 814         supported = zpool_prop_feature(propname);
 815         ASSERT(supported || zpool_prop_unsupported(propname));
 816
 817         /*
 818          * Convert from feature name to feature guid. This conversion is
 819          * unecessary for unsupported@... properties because they already
 820          * use guids.
 821          */
 822         if (supported) {
 823                 int ret;
 824                 spa_feature_t fid;
 825
 826                 ret = zfeature_lookup_name(feature, &fid);
 827                 if (ret != 0) {
 828                         (void) strlcpy(buf, "-", len);
 829                         return (ENOTSUP);
 830                 }
 831                 feature = spa_feature_table[fid].fi_guid;
 832         }
 833
 834         if (nvlist_lookup_uint64(features, feature, &refcount) == 0)
 835                 found = B_TRUE;
 836
 837         if (supported) {
 838                 if (!found) {
 839                         (void) strlcpy(buf, ZFS_FEATURE_DISABLED, len);
 840                 } else  {
 841                         if (refcount == 0)
 842                                 (void) strlcpy(buf, ZFS_FEATURE_ENABLED, len);
 843                         else
 844                                 (void) strlcpy(buf, ZFS_FEATURE_ACTIVE, len);
 845                 }
 846         } else {
 847                 if (found) {
 848                         if (refcount == 0) {
 849                                 (void) strcpy(buf, ZFS_UNSUPPORTED_INACTIVE);
 850                         } else {
 851                                 (void) strcpy(buf, ZFS_UNSUPPORTED_READONLY);
 852                         }
 853                 } else {
 854                         (void) strlcpy(buf, "-", len);
 855                         return (ENOTSUP);
 856                 }
 857         }
 858
 859         return (0);
 860 }
 861
 862 /*
 863  * Don't start the slice at the default block of 34; many storage
 864  * devices will use a stripe width of 128k, so start there instead.
 865  */
 866 #define NEW_START_BLOCK 256
 867
 868 /*
 869  * Validate the given pool name, optionally putting an extended error message in
 870  * 'buf'.
 871  */
 872 boolean_t
 873 zpool_name_valid(libzfs_handle_t *hdl, boolean_t isopen, const char *pool)
 874 {
 875         namecheck_err_t why;
 876         char what;
 877         int ret;
 878
 879         ret = pool_namecheck(pool, &why, &what);
 880
 881         /*
 882          * The rules for reserved pool names were extended at a later point.
 883          * But we need to support users with existing pools that may now be
 884          * invalid.  So we only check for this expanded set of names during a
 885          * create (or import), and only in userland.
 886          */
 887         if (ret == 0 && !isopen &&
 888             (strncmp(pool, "mirror", 6) == 0 ||
 889             strncmp(pool, "raidz", 5) == 0 ||
 890             strncmp(pool, "spare", 5) == 0 ||
 891             strcmp(pool, "log") == 0)) {
 892                 if (hdl != NULL)
 893                         zfs_error_aux(hdl,
 894                             dgettext(TEXT_DOMAIN, "name is reserved"));
 895                 return (B_FALSE);
 896         }
 897
 898
 899         if (ret != 0) {
 900                 if (hdl != NULL) {
 901                         switch (why) {
 902                         case NAME_ERR_TOOLONG:
 903                                 zfs_error_aux(hdl,
 904                                     dgettext(TEXT_DOMAIN, "name is too long"));
 905                                 break;
 906
 907                         case NAME_ERR_INVALCHAR:
 908                                 zfs_error_aux(hdl,
 909                                     dgettext(TEXT_DOMAIN, "invalid character "
 910                                     "'%c' in pool name"), what);
 911                                 break;
 912
 913                         case NAME_ERR_NOLETTER:
 914                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 915                                     "name must begin with a letter"));
 916                                 break;
 917
 918                         case NAME_ERR_RESERVED:
 919                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 920                                     "name is reserved"));
 921                                 break;
 922
 923                         case NAME_ERR_DISKLIKE:
 924                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 925                                     "pool name is reserved"));
 926                                 break;
 927
 928                         case NAME_ERR_LEADING_SLASH:
 929                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 930                                     "leading slash in name"));
 931                                 break;
 932
 933                         case NAME_ERR_EMPTY_COMPONENT:
 934                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 935                                     "empty component in name"));
 936                                 break;
 937
 938                         case NAME_ERR_TRAILING_SLASH:
 939                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 940                                     "trailing slash in name"));
 941                                 break;
 942
 943                         case NAME_ERR_MULTIPLE_AT:
 944                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
 945                                     "multiple '@' delimiters in name"));
 946                                 break;
 947
 948                         }
 949                 }
 950                 return (B_FALSE);
 951         }
 952
 953         return (B_TRUE);
 954 }
 955
 956 /*
 957  * Open a handle to the given pool, even if the pool is currently in the FAULTED
 958  * state.
 959  */
 960 zpool_handle_t *
 961 zpool_open_canfail(libzfs_handle_t *hdl, const char *pool)
 962 {
 963         zpool_handle_t *zhp;
 964         boolean_t missing;
 965
 966         /*
 967          * Make sure the pool name is valid.
 968          */
 969         if (!zpool_name_valid(hdl, B_TRUE, pool)) {
 970                 (void) zfs_error_fmt(hdl, EZFS_INVALIDNAME,
 971                     dgettext(TEXT_DOMAIN, "cannot open '%s'"),
 972                     pool);
 973                 return (NULL);
 974         }
 975
 976         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
 977                 return (NULL);
 978
 979         zhp->zpool_hdl = hdl;
 980         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
 981
 982         if (zpool_refresh_stats(zhp, &missing) != 0) {
 983                 zpool_close(zhp);
 984                 return (NULL);
 985         }
 986
 987         if (missing) {
 988                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "no such pool"));
 989                 (void) zfs_error_fmt(hdl, EZFS_NOENT,
 990                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), pool);
 991                 zpool_close(zhp);
 992                 return (NULL);
 993         }
 994
 995         return (zhp);
 996 }
 997
 998 /*
 999  * Like the above, but silent on error.  Used when iterating over pools (because
1000  * the configuration cache may be out of date).
1001  */
1002 int
1003 zpool_open_silent(libzfs_handle_t *hdl, const char *pool, zpool_handle_t **ret)
1004 {
1005         zpool_handle_t *zhp;
1006         boolean_t missing;
1007
1008         if ((zhp = zfs_alloc(hdl, sizeof (zpool_handle_t))) == NULL)
1009                 return (-1);
1010
1011         zhp->zpool_hdl = hdl;
1012         (void) strlcpy(zhp->zpool_name, pool, sizeof (zhp->zpool_name));
1013
1014         if (zpool_refresh_stats(zhp, &missing) != 0) {
1015                 zpool_close(zhp);
1016                 return (-1);
1017         }
1018
1019         if (missing) {
1020                 zpool_close(zhp);
1021                 *ret = NULL;
1022                 return (0);
1023         }
1024
1025         *ret = zhp;
1026         return (0);
1027 }
1028
1029 /*
1030  * Similar to zpool_open_canfail(), but refuses to open pools in the faulted
1031  * state.
1032  */
1033 zpool_handle_t *
1034 zpool_open(libzfs_handle_t *hdl, const char *pool)
1035 {
1036         zpool_handle_t *zhp;
1037
1038         if ((zhp = zpool_open_canfail(hdl, pool)) == NULL)
1039                 return (NULL);
1040
1041         if (zhp->zpool_state == POOL_STATE_UNAVAIL) {
1042                 (void) zfs_error_fmt(hdl, EZFS_POOLUNAVAIL,
1043                     dgettext(TEXT_DOMAIN, "cannot open '%s'"), zhp->zpool_name);
1044                 zpool_close(zhp);
1045                 return (NULL);
1046         }
1047
1048         return (zhp);
1049 }
1050
1051 /*
1052  * Close the handle.  Simply frees the memory associated with the handle.
1053  */
1054 void
1055 zpool_close(zpool_handle_t *zhp)
1056 {
1057         if (zhp->zpool_config)
1058                 nvlist_free(zhp->zpool_config);
1059         if (zhp->zpool_old_config)
1060                 nvlist_free(zhp->zpool_old_config);
1061         if (zhp->zpool_props)
1062                 nvlist_free(zhp->zpool_props);
1063         free(zhp);
1064 }
1065
1066 /*
1067  * Return the name of the pool.
1068  */
1069 const char *
1070 zpool_get_name(zpool_handle_t *zhp)
1071 {
1072         return (zhp->zpool_name);
1073 }
1074
1075
1076 /*
1077  * Return the state of the pool (ACTIVE or UNAVAILABLE)
1078  */
1079 int
1080 zpool_get_state(zpool_handle_t *zhp)
1081 {
1082         return (zhp->zpool_state);
1083 }
1084
1085 /*
1086  * Create the named pool, using the provided vdev list.  It is assumed
1087  * that the consumer has already validated the contents of the nvlist, so we
1088  * don't have to worry about error semantics.
1089  */
1090 int
1091 zpool_create(libzfs_handle_t *hdl, const char *pool, nvlist_t *nvroot,
1092     nvlist_t *props, nvlist_t *fsprops)
1093 {
1094         zfs_cmd_t zc = { 0 };
1095         nvlist_t *zc_fsprops = NULL;
1096         nvlist_t *zc_props = NULL;
1097         char msg[1024];
1098         int ret = -1;
1099
1100         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1101             "cannot create '%s'"), pool);
1102
1103         if (!zpool_name_valid(hdl, B_FALSE, pool))
1104                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
1105
1106         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1107                 return (-1);
1108
1109         if (props) {
1110                 prop_flags_t flags = { .create = B_TRUE, .import = B_FALSE };
1111
1112                 if ((zc_props = zpool_valid_proplist(hdl, pool, props,
1113                     SPA_VERSION_1, flags, msg)) == NULL) {
1114                         goto create_failed;
1115                 }
1116         }
1117
1118         if (fsprops) {
1119                 uint64_t zoned;
1120                 char *zonestr;
1121
1122                 zoned = ((nvlist_lookup_string(fsprops,
1123                     zfs_prop_to_name(ZFS_PROP_ZONED), &zonestr) == 0) &&
1124                     strcmp(zonestr, "on") == 0);
1125
1126                 if ((zc_fsprops = zfs_valid_proplist(hdl, ZFS_TYPE_FILESYSTEM,
1127                     fsprops, zoned, NULL, NULL, msg)) == NULL) {
1128                         goto create_failed;
1129                 }
1130                 if (!zc_props &&
1131                     (nvlist_alloc(&zc_props, NV_UNIQUE_NAME, 0) != 0)) {
1132                         goto create_failed;
1133                 }
1134                 if (nvlist_add_nvlist(zc_props,
1135                     ZPOOL_ROOTFS_PROPS, zc_fsprops) != 0) {
1136                         goto create_failed;
1137                 }
1138         }
1139
1140         if (zc_props && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
1141                 goto create_failed;
1142
1143         (void) strlcpy(zc.zc_name, pool, sizeof (zc.zc_name));
1144
1145         if ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_CREATE, &zc)) != 0) {
1146
1147                 zcmd_free_nvlists(&zc);
1148                 nvlist_free(zc_props);
1149                 nvlist_free(zc_fsprops);
1150
1151                 switch (errno) {
1152                 case EBUSY:
1153                         /*
1154                          * This can happen if the user has specified the same
1155                          * device multiple times.  We can't reliably detect this
1156                          * until we try to add it and see we already have a
1157                          * label.
1158                          */
1159                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1160                             "one or more vdevs refer to the same device"));
1161                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1162
1163                 case ERANGE:
1164                         /*
1165                          * This happens if the record size is smaller or larger
1166                          * than the allowed size range, or not a power of 2.
1167                          *
1168                          * NOTE: although zfs_valid_proplist is called earlier,
1169                          * this case may have slipped through since the
1170                          * pool does not exist yet and it is therefore
1171                          * impossible to read properties e.g. max blocksize
1172                          * from the pool.
1173                          */
1174                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1175                             "record size invalid"));
1176                         return (zfs_error(hdl, EZFS_BADPROP, msg));
1177
1178                 case EOVERFLOW:
1179                         /*
1180                          * This occurs when one of the devices is below
1181                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1182                          * device was the problem device since there's no
1183                          * reliable way to determine device size from userland.
1184                          */
1185                         {
1186                                 char buf[64];
1187
1188                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1189
1190                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1191                                     "one or more devices is less than the "
1192                                     "minimum size (%s)"), buf);
1193                         }
1194                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1195
1196                 case ENOSPC:
1197                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1198                             "one or more devices is out of space"));
1199                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1200
1201                 case ENOTBLK:
1202                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1203                             "cache device must be a disk or disk slice"));
1204                         return (zfs_error(hdl, EZFS_BADDEV, msg));
1205
1206                 default:
1207                         return (zpool_standard_error(hdl, errno, msg));
1208                 }
1209         }
1210
1211 create_failed:
1212         zcmd_free_nvlists(&zc);
1213         nvlist_free(zc_props);
1214         nvlist_free(zc_fsprops);
1215         return (ret);
1216 }
1217
1218 /*
1219  * Destroy the given pool.  It is up to the caller to ensure that there are no
1220  * datasets left in the pool.
1221  */
1222 int
1223 zpool_destroy(zpool_handle_t *zhp, const char *log_str)
1224 {
1225         zfs_cmd_t zc = { 0 };
1226         zfs_handle_t *zfp = NULL;
1227         libzfs_handle_t *hdl = zhp->zpool_hdl;
1228         char msg[1024];
1229
1230         if (zhp->zpool_state == POOL_STATE_ACTIVE &&
1231             (zfp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_FILESYSTEM)) == NULL)
1232                 return (-1);
1233
1234         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1235         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1236
1237         if (zfs_ioctl(hdl, ZFS_IOC_POOL_DESTROY, &zc) != 0) {
1238                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1239                     "cannot destroy '%s'"), zhp->zpool_name);
1240
1241                 if (errno == EROFS) {
1242                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1243                             "one or more devices is read only"));
1244                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1245                 } else {
1246                         (void) zpool_standard_error(hdl, errno, msg);
1247                 }
1248
1249                 if (zfp)
1250                         zfs_close(zfp);
1251                 return (-1);
1252         }
1253
1254         if (zfp) {
1255                 remove_mountpoint(zfp);
1256                 zfs_close(zfp);
1257         }
1258
1259         return (0);
1260 }
1261
1262 /*
1263  * Add the given vdevs to the pool.  The caller must have already performed the
1264  * necessary verification to ensure that the vdev specification is well-formed.
1265  */
1266 int
1267 zpool_add(zpool_handle_t *zhp, nvlist_t *nvroot)
1268 {
1269         zfs_cmd_t zc = { 0 };
1270         int ret;
1271         libzfs_handle_t *hdl = zhp->zpool_hdl;
1272         char msg[1024];
1273         nvlist_t **spares, **l2cache;
1274         uint_t nspares, nl2cache;
1275
1276         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1277             "cannot add to '%s'"), zhp->zpool_name);
1278
1279         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1280             SPA_VERSION_SPARES &&
1281             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_SPARES,
1282             &spares, &nspares) == 0) {
1283                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1284                     "upgraded to add hot spares"));
1285                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1286         }
1287
1288         if (zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL) <
1289             SPA_VERSION_L2CACHE &&
1290             nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_L2CACHE,
1291             &l2cache, &nl2cache) == 0) {
1292                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "pool must be "
1293                     "upgraded to add cache devices"));
1294                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
1295         }
1296
1297         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
1298                 return (-1);
1299         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1300
1301         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_ADD, &zc) != 0) {
1302                 switch (errno) {
1303                 case EBUSY:
1304                         /*
1305                          * This can happen if the user has specified the same
1306                          * device multiple times.  We can't reliably detect this
1307                          * until we try to add it and see we already have a
1308                          * label.
1309                          */
1310                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1311                             "one or more vdevs refer to the same device"));
1312                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1313                         break;
1314
1315                 case EOVERFLOW:
1316                         /*
1317                          * This occurrs when one of the devices is below
1318                          * SPA_MINDEVSIZE.  Unfortunately, we can't detect which
1319                          * device was the problem device since there's no
1320                          * reliable way to determine device size from userland.
1321                          */
1322                         {
1323                                 char buf[64];
1324
1325                                 zfs_nicenum(SPA_MINDEVSIZE, buf, sizeof (buf));
1326
1327                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1328                                     "device is less than the minimum "
1329                                     "size (%s)"), buf);
1330                         }
1331                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1332                         break;
1333
1334                 case ENOTSUP:
1335                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1336                             "pool must be upgraded to add these vdevs"));
1337                         (void) zfs_error(hdl, EZFS_BADVERSION, msg);
1338                         break;
1339
1340                 case EDOM:
1341                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1342                             "root pool can not have multiple vdevs"
1343                             " or separate logs"));
1344                         (void) zfs_error(hdl, EZFS_POOL_NOTSUP, msg);
1345                         break;
1346
1347                 case ENOTBLK:
1348                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1349                             "cache device must be a disk or disk slice"));
1350                         (void) zfs_error(hdl, EZFS_BADDEV, msg);
1351                         break;
1352
1353                 default:
1354                         (void) zpool_standard_error(hdl, errno, msg);
1355                 }
1356
1357                 ret = -1;
1358         } else {
1359                 ret = 0;
1360         }
1361
1362         zcmd_free_nvlists(&zc);
1363
1364         return (ret);
1365 }
1366
1367 /*
1368  * Exports the pool from the system.  The caller must ensure that there are no
1369  * mounted datasets in the pool.
1370  */
1371 static int
1372 zpool_export_common(zpool_handle_t *zhp, boolean_t force, boolean_t hardforce,
1373     const char *log_str)
1374 {
1375         zfs_cmd_t zc = { 0 };
1376         char msg[1024];
1377
1378         (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
1379             "cannot export '%s'"), zhp->zpool_name);
1380
1381         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1382         zc.zc_cookie = force;
1383         zc.zc_guid = hardforce;
1384         zc.zc_history = (uint64_t)(uintptr_t)log_str;
1385
1386         if (zfs_ioctl(zhp->zpool_hdl, ZFS_IOC_POOL_EXPORT, &zc) != 0) {
1387                 switch (errno) {
1388                 case EXDEV:
1389                         zfs_error_aux(zhp->zpool_hdl, dgettext(TEXT_DOMAIN,
1390                             "use '-f' to override the following errors:\n"
1391                             "'%s' has an active shared spare which could be"
1392                             " used by other pools once '%s' is exported."),
1393                             zhp->zpool_name, zhp->zpool_name);
1394                         return (zfs_error(zhp->zpool_hdl, EZFS_ACTIVE_SPARE,
1395                             msg));
1396                 default:
1397                         return (zpool_standard_error_fmt(zhp->zpool_hdl, errno,
1398                             msg));
1399                 }
1400         }
1401
1402         return (0);
1403 }
1404
1405 int
1406 zpool_export(zpool_handle_t *zhp, boolean_t force, const char *log_str)
1407 {
1408         return (zpool_export_common(zhp, force, B_FALSE, log_str));
1409 }
1410
1411 int
1412 zpool_export_force(zpool_handle_t *zhp, const char *log_str)
1413 {
1414         return (zpool_export_common(zhp, B_TRUE, B_TRUE, log_str));
1415 }
1416
1417 static void
1418 zpool_rewind_exclaim(libzfs_handle_t *hdl, const char *name, boolean_t dryrun,
1419     nvlist_t *config)
1420 {
1421         nvlist_t *nv = NULL;
1422         uint64_t rewindto;
1423         int64_t loss = -1;
1424         struct tm t;
1425         char timestr[128];
1426
1427         if (!hdl->libzfs_printerr || config == NULL)
1428                 return;
1429
1430         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1431             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0) {
1432                 return;
1433         }
1434
1435         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1436                 return;
1437         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1438
1439         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1440             strftime(timestr, 128, 0, &t) != 0) {
1441                 if (dryrun) {
1442                         (void) printf(dgettext(TEXT_DOMAIN,
1443                             "Would be able to return %s "
1444                             "to its state as of %s.\n"),
1445                             name, timestr);
1446                 } else {
1447                         (void) printf(dgettext(TEXT_DOMAIN,
1448                             "Pool %s returned to its state as of %s.\n"),
1449                             name, timestr);
1450                 }
1451                 if (loss > 120) {
1452                         (void) printf(dgettext(TEXT_DOMAIN,
1453                             "%s approximately %lld "),
1454                             dryrun ? "Would discard" : "Discarded",
1455                             (loss + 30) / 60);
1456                         (void) printf(dgettext(TEXT_DOMAIN,
1457                             "minutes of transactions.\n"));
1458                 } else if (loss > 0) {
1459                         (void) printf(dgettext(TEXT_DOMAIN,
1460                             "%s approximately %lld "),
1461                             dryrun ? "Would discard" : "Discarded", loss);
1462                         (void) printf(dgettext(TEXT_DOMAIN,
1463                             "seconds of transactions.\n"));
1464                 }
1465         }
1466 }
1467
1468 void
1469 zpool_explain_recover(libzfs_handle_t *hdl, const char *name, int reason,
1470     nvlist_t *config)
1471 {
1472         nvlist_t *nv = NULL;
1473         int64_t loss = -1;
1474         uint64_t edata = UINT64_MAX;
1475         uint64_t rewindto;
1476         struct tm t;
1477         char timestr[128];
1478
1479         if (!hdl->libzfs_printerr)
1480                 return;
1481
1482         if (reason >= 0)
1483                 (void) printf(dgettext(TEXT_DOMAIN, "action: "));
1484         else
1485                 (void) printf(dgettext(TEXT_DOMAIN, "\t"));
1486
1487         /* All attempted rewinds failed if ZPOOL_CONFIG_LOAD_TIME missing */
1488         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nv) != 0 ||
1489             nvlist_lookup_nvlist(nv, ZPOOL_CONFIG_REWIND_INFO, &nv) != 0 ||
1490             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_TIME, &rewindto) != 0)
1491                 goto no_info;
1492
1493         (void) nvlist_lookup_int64(nv, ZPOOL_CONFIG_REWIND_TIME, &loss);
1494         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_LOAD_DATA_ERRORS,
1495             &edata);
1496
1497         (void) printf(dgettext(TEXT_DOMAIN,
1498             "Recovery is possible, but will result in some data loss.\n"));
1499
1500         if (localtime_r((time_t *)&rewindto, &t) != NULL &&
1501             strftime(timestr, 128, 0, &t) != 0) {
1502                 (void) printf(dgettext(TEXT_DOMAIN,
1503                     "\tReturning the pool to its state as of %s\n"
1504                     "\tshould correct the problem.  "),
1505                     timestr);
1506         } else {
1507                 (void) printf(dgettext(TEXT_DOMAIN,
1508                     "\tReverting the pool to an earlier state "
1509                     "should correct the problem.\n\t"));
1510         }
1511
1512         if (loss > 120) {
1513                 (void) printf(dgettext(TEXT_DOMAIN,
1514                     "Approximately %lld minutes of data\n"
1515                     "\tmust be discarded, irreversibly.  "), (loss + 30) / 60);
1516         } else if (loss > 0) {
1517                 (void) printf(dgettext(TEXT_DOMAIN,
1518                     "Approximately %lld seconds of data\n"
1519                     "\tmust be discarded, irreversibly.  "), loss);
1520         }
1521         if (edata != 0 && edata != UINT64_MAX) {
1522                 if (edata == 1) {
1523                         (void) printf(dgettext(TEXT_DOMAIN,
1524                             "After rewind, at least\n"
1525                             "\tone persistent user-data error will remain.  "));
1526                 } else {
1527                         (void) printf(dgettext(TEXT_DOMAIN,
1528                             "After rewind, several\n"
1529                             "\tpersistent user-data errors will remain.  "));
1530                 }
1531         }
1532         (void) printf(dgettext(TEXT_DOMAIN,
1533             "Recovery can be attempted\n\tby executing 'zpool %s -F %s'.  "),
1534             reason >= 0 ? "clear" : "import", name);
1535
1536         (void) printf(dgettext(TEXT_DOMAIN,
1537             "A scrub of the pool\n"
1538             "\tis strongly recommended after recovery.\n"));
1539         return;
1540
1541 no_info:
1542         (void) printf(dgettext(TEXT_DOMAIN,
1543             "Destroy and re-create the pool from\n\ta backup source.\n"));
1544 }
1545
1546 /*
1547  * zpool_import() is a contracted interface. Should be kept the same
1548  * if possible.
1549  *
1550  * Applications should use zpool_import_props() to import a pool with
1551  * new properties value to be set.
1552  */
1553 int
1554 zpool_import(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1555     char *altroot)
1556 {
1557         nvlist_t *props = NULL;
1558         int ret;
1559
1560         if (altroot != NULL) {
1561                 if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0) {
1562                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1563                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1564                             newname));
1565                 }
1566
1567                 if (nvlist_add_string(props,
1568                     zpool_prop_to_name(ZPOOL_PROP_ALTROOT), altroot) != 0 ||
1569                     nvlist_add_string(props,
1570                     zpool_prop_to_name(ZPOOL_PROP_CACHEFILE), "none") != 0) {
1571                         nvlist_free(props);
1572                         return (zfs_error_fmt(hdl, EZFS_NOMEM,
1573                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1574                             newname));
1575                 }
1576         }
1577
1578         ret = zpool_import_props(hdl, config, newname, props,
1579             ZFS_IMPORT_NORMAL);
1580         if (props)
1581                 nvlist_free(props);
1582         return (ret);
1583 }
1584
1585 static void
1586 print_vdev_tree(libzfs_handle_t *hdl, const char *name, nvlist_t *nv,
1587     int indent)
1588 {
1589         nvlist_t **child;
1590         uint_t c, children;
1591         char *vname;
1592         uint64_t is_log = 0;
1593
1594         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG,
1595             &is_log);
1596
1597         if (name != NULL)
1598                 (void) printf("\t%*s%s%s\n", indent, "", name,
1599                     is_log ? " [log]" : "");
1600
1601         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
1602             &child, &children) != 0)
1603                 return;
1604
1605         for (c = 0; c < children; c++) {
1606                 vname = zpool_vdev_name(hdl, NULL, child[c], B_TRUE);
1607                 print_vdev_tree(hdl, vname, child[c], indent + 2);
1608                 free(vname);
1609         }
1610 }
1611
1612 void
1613 zpool_print_unsup_feat(nvlist_t *config)
1614 {
1615         nvlist_t *nvinfo, *unsup_feat;
1616
1617         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_LOAD_INFO, &nvinfo) ==
1618             0);
1619         verify(nvlist_lookup_nvlist(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT,
1620             &unsup_feat) == 0);
1621
1622         for (nvpair_t *nvp = nvlist_next_nvpair(unsup_feat, NULL); nvp != NULL;
1623             nvp = nvlist_next_nvpair(unsup_feat, nvp)) {
1624                 char *desc;
1625
1626                 verify(nvpair_type(nvp) == DATA_TYPE_STRING);
1627                 verify(nvpair_value_string(nvp, &desc) == 0);
1628
1629                 if (strlen(desc) > 0)
1630                         (void) printf("\t%s (%s)\n", nvpair_name(nvp), desc);
1631                 else
1632                         (void) printf("\t%s\n", nvpair_name(nvp));
1633         }
1634 }
1635
1636 /*
1637  * Import the given pool using the known configuration and a list of
1638  * properties to be set. The configuration should have come from
1639  * zpool_find_import(). The 'newname' parameters control whether the pool
1640  * is imported with a different name.
1641  */
1642 int
1643 zpool_import_props(libzfs_handle_t *hdl, nvlist_t *config, const char *newname,
1644     nvlist_t *props, int flags)
1645 {
1646         zfs_cmd_t zc = { 0 };
1647         zpool_rewind_policy_t policy;
1648         nvlist_t *nv = NULL;
1649         nvlist_t *nvinfo = NULL;
1650         nvlist_t *missing = NULL;
1651         char *thename;
1652         char *origname;
1653         int ret;
1654         int error = 0;
1655         char errbuf[1024];
1656
1657         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME,
1658             &origname) == 0);
1659
1660         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1661             "cannot import pool '%s'"), origname);
1662
1663         if (newname != NULL) {
1664                 if (!zpool_name_valid(hdl, B_FALSE, newname))
1665                         return (zfs_error_fmt(hdl, EZFS_INVALIDNAME,
1666                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1667                             newname));
1668                 thename = (char *)newname;
1669         } else {
1670                 thename = origname;
1671         }
1672
1673         if (props != NULL) {
1674                 uint64_t version;
1675                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
1676
1677                 verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
1678                     &version) == 0);
1679
1680                 if ((props = zpool_valid_proplist(hdl, origname,
1681                     props, version, flags, errbuf)) == NULL)
1682                         return (-1);
1683                 if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
1684                         nvlist_free(props);
1685                         return (-1);
1686                 }
1687                 nvlist_free(props);
1688         }
1689
1690         (void) strlcpy(zc.zc_name, thename, sizeof (zc.zc_name));
1691
1692         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID,
1693             &zc.zc_guid) == 0);
1694
1695         if (zcmd_write_conf_nvlist(hdl, &zc, config) != 0) {
1696                 zcmd_free_nvlists(&zc);
1697                 return (-1);
1698         }
1699         if (zcmd_alloc_dst_nvlist(hdl, &zc, zc.zc_nvlist_conf_size * 2) != 0) {
1700                 zcmd_free_nvlists(&zc);
1701                 return (-1);
1702         }
1703
1704         zc.zc_cookie = flags;
1705         while ((ret = zfs_ioctl(hdl, ZFS_IOC_POOL_IMPORT, &zc)) != 0 &&
1706             errno == ENOMEM) {
1707                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
1708                         zcmd_free_nvlists(&zc);
1709                         return (-1);
1710                 }
1711         }
1712         if (ret != 0)
1713                 error = errno;
1714
1715         (void) zcmd_read_dst_nvlist(hdl, &zc, &nv);
1716
1717         zcmd_free_nvlists(&zc);
1718
1719         zpool_get_rewind_policy(config, &policy);
1720
1721         if (error) {
1722                 char desc[1024];
1723
1724                 /*
1725                  * Dry-run failed, but we print out what success
1726                  * looks like if we found a best txg
1727                  */
1728                 if (policy.zrp_request & ZPOOL_TRY_REWIND) {
1729                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1730                             B_TRUE, nv);
1731                         nvlist_free(nv);
1732                         return (-1);
1733                 }
1734
1735                 if (newname == NULL)
1736                         (void) snprintf(desc, sizeof (desc),
1737                             dgettext(TEXT_DOMAIN, "cannot import '%s'"),
1738                             thename);
1739                 else
1740                         (void) snprintf(desc, sizeof (desc),
1741                             dgettext(TEXT_DOMAIN, "cannot import '%s' as '%s'"),
1742                             origname, thename);
1743
1744                 switch (error) {
1745                 case ENOTSUP:
1746                         if (nv != NULL && nvlist_lookup_nvlist(nv,
1747                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1748                             nvlist_exists(nvinfo, ZPOOL_CONFIG_UNSUP_FEAT)) {
1749                                 (void) printf(dgettext(TEXT_DOMAIN, "This "
1750                                     "pool uses the following feature(s) not "
1751                                     "supported by this system:\n"));
1752                                 zpool_print_unsup_feat(nv);
1753                                 if (nvlist_exists(nvinfo,
1754                                     ZPOOL_CONFIG_CAN_RDONLY)) {
1755                                         (void) printf(dgettext(TEXT_DOMAIN,
1756                                             "All unsupported features are only "
1757                                             "required for writing to the pool."
1758                                             "\nThe pool can be imported using "
1759                                             "'-o readonly=on'.\n"));
1760                                 }
1761                         }
1762                         /*
1763                          * Unsupported version.
1764                          */
1765                         (void) zfs_error(hdl, EZFS_BADVERSION, desc);
1766                         break;
1767
1768                 case EINVAL:
1769                         (void) zfs_error(hdl, EZFS_INVALCONFIG, desc);
1770                         break;
1771
1772                 case EROFS:
1773                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1774                             "one or more devices is read only"));
1775                         (void) zfs_error(hdl, EZFS_BADDEV, desc);
1776                         break;
1777
1778                 case ENXIO:
1779                         if (nv && nvlist_lookup_nvlist(nv,
1780                             ZPOOL_CONFIG_LOAD_INFO, &nvinfo) == 0 &&
1781                             nvlist_lookup_nvlist(nvinfo,
1782                             ZPOOL_CONFIG_MISSING_DEVICES, &missing) == 0) {
1783                                 (void) printf(dgettext(TEXT_DOMAIN,
1784                                     "The devices below are missing, use "
1785                                     "'-m' to import the pool anyway:\n"));
1786                                 print_vdev_tree(hdl, NULL, missing, 2);
1787                                 (void) printf("\n");
1788                         }
1789                         (void) zpool_standard_error(hdl, error, desc);
1790                         break;
1791
1792                 case EEXIST:
1793                         (void) zpool_standard_error(hdl, error, desc);
1794                         break;
1795
1796                 default:
1797                         (void) zpool_standard_error(hdl, error, desc);
1798                         zpool_explain_recover(hdl,
1799                             newname ? origname : thename, -error, nv);
1800                         break;
1801                 }
1802
1803                 nvlist_free(nv);
1804                 ret = -1;
1805         } else {
1806                 zpool_handle_t *zhp;
1807
1808                 /*
1809                  * This should never fail, but play it safe anyway.
1810                  */
1811                 if (zpool_open_silent(hdl, thename, &zhp) != 0)
1812                         ret = -1;
1813                 else if (zhp != NULL)
1814                         zpool_close(zhp);
1815                 if (policy.zrp_request &
1816                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
1817                         zpool_rewind_exclaim(hdl, newname ? origname : thename,
1818                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0), nv);
1819                 }
1820                 nvlist_free(nv);
1821                 return (0);
1822         }
1823
1824         return (ret);
1825 }
1826
1827 /*
1828  * Scan the pool.
1829  */
1830 int
1831 zpool_scan(zpool_handle_t *zhp, pool_scan_func_t func)
1832 {
1833         zfs_cmd_t zc = { 0 };
1834         char msg[1024];
1835         libzfs_handle_t *hdl = zhp->zpool_hdl;
1836
1837         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
1838         zc.zc_cookie = func;
1839
1840         if (zfs_ioctl(hdl, ZFS_IOC_POOL_SCAN, &zc) == 0 ||
1841             (errno == ENOENT && func != POOL_SCAN_NONE))
1842                 return (0);
1843
1844         if (func == POOL_SCAN_SCRUB) {
1845                 (void) snprintf(msg, sizeof (msg),
1846                     dgettext(TEXT_DOMAIN, "cannot scrub %s"), zc.zc_name);
1847         } else if (func == POOL_SCAN_NONE) {
1848                 (void) snprintf(msg, sizeof (msg),
1849                     dgettext(TEXT_DOMAIN, "cannot cancel scrubbing %s"),
1850                     zc.zc_name);
1851         } else {
1852                 assert(!"unexpected result");
1853         }
1854
1855         if (errno == EBUSY) {
1856                 nvlist_t *nvroot;
1857                 pool_scan_stat_t *ps = NULL;
1858                 uint_t psc;
1859
1860                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
1861                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
1862                 (void) nvlist_lookup_uint64_array(nvroot,
1863                     ZPOOL_CONFIG_SCAN_STATS, (uint64_t **)&ps, &psc);
1864                 if (ps && ps->pss_func == POOL_SCAN_SCRUB)
1865                         return (zfs_error(hdl, EZFS_SCRUBBING, msg));
1866                 else
1867                         return (zfs_error(hdl, EZFS_RESILVERING, msg));
1868         } else if (errno == ENOENT) {
1869                 return (zfs_error(hdl, EZFS_NO_SCRUB, msg));
1870         } else {
1871                 return (zpool_standard_error(hdl, errno, msg));
1872         }
1873 }
1874
1875 #ifdef illumos
1876 /*
1877  * This provides a very minimal check whether a given string is likely a
1878  * c#t#d# style string.  Users of this are expected to do their own
1879  * verification of the s# part.
1880  */
1881 #define CTD_CHECK(str)  (str && str[0] == 'c' && isdigit(str[1]))
1882
1883 /*
1884  * More elaborate version for ones which may start with "/dev/dsk/"
1885  * and the like.
1886  */
1887 static int
1888 ctd_check_path(char *str)
1889 {
1890         /*
1891          * If it starts with a slash, check the last component.
1892          */
1893         if (str && str[0] == '/') {
1894                 char *tmp = strrchr(str, '/');
1895
1896                 /*
1897                  * If it ends in "/old", check the second-to-last
1898                  * component of the string instead.
1899                  */
1900                 if (tmp != str && strcmp(tmp, "/old") == 0) {
1901                         for (tmp--; *tmp != '/'; tmp--)
1902                                 ;
1903                 }
1904                 str = tmp + 1;
1905         }
1906         return (CTD_CHECK(str));
1907 }
1908 #endif
1909
1910 /*
1911  * Find a vdev that matches the search criteria specified. We use the
1912  * the nvpair name to determine how we should look for the device.
1913  * 'avail_spare' is set to TRUE if the provided guid refers to an AVAIL
1914  * spare; but FALSE if its an INUSE spare.
1915  */
1916 static nvlist_t *
1917 vdev_to_nvlist_iter(nvlist_t *nv, nvlist_t *search, boolean_t *avail_spare,
1918     boolean_t *l2cache, boolean_t *log)
1919 {
1920         uint_t c, children;
1921         nvlist_t **child;
1922         nvlist_t *ret;
1923         uint64_t is_log;
1924         char *srchkey;
1925         nvpair_t *pair = nvlist_next_nvpair(search, NULL);
1926
1927         /* Nothing to look for */
1928         if (search == NULL || pair == NULL)
1929                 return (NULL);
1930
1931         /* Obtain the key we will use to search */
1932         srchkey = nvpair_name(pair);
1933
1934         switch (nvpair_type(pair)) {
1935         case DATA_TYPE_UINT64:
1936                 if (strcmp(srchkey, ZPOOL_CONFIG_GUID) == 0) {
1937                         uint64_t srchval, theguid;
1938
1939                         verify(nvpair_value_uint64(pair, &srchval) == 0);
1940                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
1941                             &theguid) == 0);
1942                         if (theguid == srchval)
1943                                 return (nv);
1944                 }
1945                 break;
1946
1947         case DATA_TYPE_STRING: {
1948                 char *srchval, *val;
1949
1950                 verify(nvpair_value_string(pair, &srchval) == 0);
1951                 if (nvlist_lookup_string(nv, srchkey, &val) != 0)
1952                         break;
1953
1954                 /*
1955                  * Search for the requested value. Special cases:
1956                  *
1957                  * - ZPOOL_CONFIG_PATH for whole disk entries.  These end in
1958                  *   "s0" or "s0/old".  The "s0" part is hidden from the user,
1959                  *   but included in the string, so this matches around it.
1960                  * - looking for a top-level vdev name (i.e. ZPOOL_CONFIG_TYPE).
1961                  *
1962                  * Otherwise, all other searches are simple string compares.
1963                  */
1964 #ifdef illumos
1965                 if (strcmp(srchkey, ZPOOL_CONFIG_PATH) == 0 &&
1966                     ctd_check_path(val)) {
1967                         uint64_t wholedisk = 0;
1968
1969                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
1970                             &wholedisk);
1971                         if (wholedisk) {
1972                                 int slen = strlen(srchval);
1973                                 int vlen = strlen(val);
1974
1975                                 if (slen != vlen - 2)
1976                                         break;
1977
1978                                 /*
1979                                  * make_leaf_vdev() should only set
1980                                  * wholedisk for ZPOOL_CONFIG_PATHs which
1981                                  * will include "/dev/dsk/", giving plenty of
1982                                  * room for the indices used next.
1983                                  */
1984                                 ASSERT(vlen >= 6);
1985
1986                                 /*
1987                                  * strings identical except trailing "s0"
1988                                  */
1989                                 if (strcmp(&val[vlen - 2], "s0") == 0 &&
1990                                     strncmp(srchval, val, slen) == 0)
1991                                         return (nv);
1992
1993                                 /*
1994                                  * strings identical except trailing "s0/old"
1995                                  */
1996                                 if (strcmp(&val[vlen - 6], "s0/old") == 0 &&
1997                                     strcmp(&srchval[slen - 4], "/old") == 0 &&
1998                                     strncmp(srchval, val, slen - 4) == 0)
1999                                         return (nv);
2000
2001                                 break;
2002                         }
2003                 } else if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2004 #else
2005                 if (strcmp(srchkey, ZPOOL_CONFIG_TYPE) == 0 && val) {
2006 #endif
2007                         char *type, *idx, *end, *p;
2008                         uint64_t id, vdev_id;
2009
2010                         /*
2011                          * Determine our vdev type, keeping in mind
2012                          * that the srchval is composed of a type and
2013                          * vdev id pair (i.e. mirror-4).
2014                          */
2015                         if ((type = strdup(srchval)) == NULL)
2016                                 return (NULL);
2017
2018                         if ((p = strrchr(type, '-')) == NULL) {
2019                                 free(type);
2020                                 break;
2021                         }
2022                         idx = p + 1;
2023                         *p = '\0';
2024
2025                         /*
2026                          * If the types don't match then keep looking.
2027                          */
2028                         if (strncmp(val, type, strlen(val)) != 0) {
2029                                 free(type);
2030                                 break;
2031                         }
2032
2033                         verify(strncmp(type, VDEV_TYPE_RAIDZ,
2034                             strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2035                             strncmp(type, VDEV_TYPE_MIRROR,
2036                             strlen(VDEV_TYPE_MIRROR)) == 0);
2037                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
2038                             &id) == 0);
2039
2040                         errno = 0;
2041                         vdev_id = strtoull(idx, &end, 10);
2042
2043                         free(type);
2044                         if (errno != 0)
2045                                 return (NULL);
2046
2047                         /*
2048                          * Now verify that we have the correct vdev id.
2049                          */
2050                         if (vdev_id == id)
2051                                 return (nv);
2052                 }
2053
2054                 /*
2055                  * Common case
2056                  */
2057                 if (strcmp(srchval, val) == 0)
2058                         return (nv);
2059                 break;
2060         }
2061
2062         default:
2063                 break;
2064         }
2065
2066         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
2067             &child, &children) != 0)
2068                 return (NULL);
2069
2070         for (c = 0; c < children; c++) {
2071                 if ((ret = vdev_to_nvlist_iter(child[c], search,
2072                     avail_spare, l2cache, NULL)) != NULL) {
2073                         /*
2074                          * The 'is_log' value is only set for the toplevel
2075                          * vdev, not the leaf vdevs.  So we always lookup the
2076                          * log device from the root of the vdev tree (where
2077                          * 'log' is non-NULL).
2078                          */
2079                         if (log != NULL &&
2080                             nvlist_lookup_uint64(child[c],
2081                             ZPOOL_CONFIG_IS_LOG, &is_log) == 0 &&
2082                             is_log) {
2083                                 *log = B_TRUE;
2084                         }
2085                         return (ret);
2086                 }
2087         }
2088
2089         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_SPARES,
2090             &child, &children) == 0) {
2091                 for (c = 0; c < children; c++) {
2092                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2093                             avail_spare, l2cache, NULL)) != NULL) {
2094                                 *avail_spare = B_TRUE;
2095                                 return (ret);
2096                         }
2097                 }
2098         }
2099
2100         if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_L2CACHE,
2101             &child, &children) == 0) {
2102                 for (c = 0; c < children; c++) {
2103                         if ((ret = vdev_to_nvlist_iter(child[c], search,
2104                             avail_spare, l2cache, NULL)) != NULL) {
2105                                 *l2cache = B_TRUE;
2106                                 return (ret);
2107                         }
2108                 }
2109         }
2110
2111         return (NULL);
2112 }
2113
2114 /*
2115  * Given a physical path (minus the "/devices" prefix), find the
2116  * associated vdev.
2117  */
2118 nvlist_t *
2119 zpool_find_vdev_by_physpath(zpool_handle_t *zhp, const char *ppath,
2120     boolean_t *avail_spare, boolean_t *l2cache, boolean_t *log)
2121 {
2122         nvlist_t *search, *nvroot, *ret;
2123
2124         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2125         verify(nvlist_add_string(search, ZPOOL_CONFIG_PHYS_PATH, ppath) == 0);
2126
2127         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2128             &nvroot) == 0);
2129
2130         *avail_spare = B_FALSE;
2131         *l2cache = B_FALSE;
2132         if (log != NULL)
2133                 *log = B_FALSE;
2134         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2135         nvlist_free(search);
2136
2137         return (ret);
2138 }
2139
2140 /*
2141  * Determine if we have an "interior" top-level vdev (i.e mirror/raidz).
2142  */
2143 boolean_t
2144 zpool_vdev_is_interior(const char *name)
2145 {
2146         if (strncmp(name, VDEV_TYPE_RAIDZ, strlen(VDEV_TYPE_RAIDZ)) == 0 ||
2147             strncmp(name, VDEV_TYPE_MIRROR, strlen(VDEV_TYPE_MIRROR)) == 0)
2148                 return (B_TRUE);
2149         return (B_FALSE);
2150 }
2151
2152 nvlist_t *
2153 zpool_find_vdev(zpool_handle_t *zhp, const char *path, boolean_t *avail_spare,
2154     boolean_t *l2cache, boolean_t *log)
2155 {
2156         char buf[MAXPATHLEN];
2157         char *end;
2158         nvlist_t *nvroot, *search, *ret;
2159         uint64_t guid;
2160
2161         verify(nvlist_alloc(&search, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2162
2163         guid = strtoull(path, &end, 10);
2164         if (guid != 0 && *end == '\0') {
2165                 verify(nvlist_add_uint64(search, ZPOOL_CONFIG_GUID, guid) == 0);
2166         } else if (zpool_vdev_is_interior(path)) {
2167                 verify(nvlist_add_string(search, ZPOOL_CONFIG_TYPE, path) == 0);
2168         } else if (path[0] != '/') {
2169                 (void) snprintf(buf, sizeof (buf), "%s%s", _PATH_DEV, path);
2170                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, buf) == 0);
2171         } else {
2172                 verify(nvlist_add_string(search, ZPOOL_CONFIG_PATH, path) == 0);
2173         }
2174
2175         verify(nvlist_lookup_nvlist(zhp->zpool_config, ZPOOL_CONFIG_VDEV_TREE,
2176             &nvroot) == 0);
2177
2178         *avail_spare = B_FALSE;
2179         *l2cache = B_FALSE;
2180         if (log != NULL)
2181                 *log = B_FALSE;
2182         ret = vdev_to_nvlist_iter(nvroot, search, avail_spare, l2cache, log);
2183         nvlist_free(search);
2184
2185         return (ret);
2186 }
2187
2188 static int
2189 vdev_online(nvlist_t *nv)
2190 {
2191         uint64_t ival;
2192
2193         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, &ival) == 0 ||
2194             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, &ival) == 0 ||
2195             nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, &ival) == 0)
2196                 return (0);
2197
2198         return (1);
2199 }
2200
2201 /*
2202  * Helper function for zpool_get_physpaths().
2203  */
2204 static int
2205 vdev_get_one_physpath(nvlist_t *config, char *physpath, size_t physpath_size,
2206     size_t *bytes_written)
2207 {
2208         size_t bytes_left, pos, rsz;
2209         char *tmppath;
2210         const char *format;
2211
2212         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PHYS_PATH,
2213             &tmppath) != 0)
2214                 return (EZFS_NODEVICE);
2215
2216         pos = *bytes_written;
2217         bytes_left = physpath_size - pos;
2218         format = (pos == 0) ? "%s" : " %s";
2219
2220         rsz = snprintf(physpath + pos, bytes_left, format, tmppath);
2221         *bytes_written += rsz;
2222
2223         if (rsz >= bytes_left) {
2224                 /* if physpath was not copied properly, clear it */
2225                 if (bytes_left != 0) {
2226                         physpath[pos] = 0;
2227                 }
2228                 return (EZFS_NOSPC);
2229         }
2230         return (0);
2231 }
2232
2233 static int
2234 vdev_get_physpaths(nvlist_t *nv, char *physpath, size_t phypath_size,
2235     size_t *rsz, boolean_t is_spare)
2236 {
2237         char *type;
2238         int ret;
2239
2240         if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0)
2241                 return (EZFS_INVALCONFIG);
2242
2243         if (strcmp(type, VDEV_TYPE_DISK) == 0) {
2244                 /*
2245                  * An active spare device has ZPOOL_CONFIG_IS_SPARE set.
2246                  * For a spare vdev, we only want to boot from the active
2247                  * spare device.
2248                  */
2249                 if (is_spare) {
2250                         uint64_t spare = 0;
2251                         (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE,
2252                             &spare);
2253                         if (!spare)
2254                                 return (EZFS_INVALCONFIG);
2255                 }
2256
2257                 if (vdev_online(nv)) {
2258                         if ((ret = vdev_get_one_physpath(nv, physpath,
2259                             phypath_size, rsz)) != 0)
2260                                 return (ret);
2261                 }
2262         } else if (strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
2263             strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
2264             (is_spare = (strcmp(type, VDEV_TYPE_SPARE) == 0))) {
2265                 nvlist_t **child;
2266                 uint_t count;
2267                 int i, ret;
2268
2269                 if (nvlist_lookup_nvlist_array(nv,
2270                     ZPOOL_CONFIG_CHILDREN, &child, &count) != 0)
2271                         return (EZFS_INVALCONFIG);
2272
2273                 for (i = 0; i < count; i++) {
2274                         ret = vdev_get_physpaths(child[i], physpath,
2275                             phypath_size, rsz, is_spare);
2276                         if (ret == EZFS_NOSPC)
2277                                 return (ret);
2278                 }
2279         }
2280
2281         return (EZFS_POOL_INVALARG);
2282 }
2283
2284 /*
2285  * Get phys_path for a root pool config.
2286  * Return 0 on success; non-zero on failure.
2287  */
2288 static int
2289 zpool_get_config_physpath(nvlist_t *config, char *physpath, size_t phypath_size)
2290 {
2291         size_t rsz;
2292         nvlist_t *vdev_root;
2293         nvlist_t **child;
2294         uint_t count;
2295         char *type;
2296
2297         rsz = 0;
2298
2299         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
2300             &vdev_root) != 0)
2301                 return (EZFS_INVALCONFIG);
2302
2303         if (nvlist_lookup_string(vdev_root, ZPOOL_CONFIG_TYPE, &type) != 0 ||
2304             nvlist_lookup_nvlist_array(vdev_root, ZPOOL_CONFIG_CHILDREN,
2305             &child, &count) != 0)
2306                 return (EZFS_INVALCONFIG);
2307
2308         /*
2309          * root pool can only have a single top-level vdev.
2310          */
2311         if (strcmp(type, VDEV_TYPE_ROOT) != 0 || count != 1)
2312                 return (EZFS_POOL_INVALARG);
2313
2314         (void) vdev_get_physpaths(child[0], physpath, phypath_size, &rsz,
2315             B_FALSE);
2316
2317         /* No online devices */
2318         if (rsz == 0)
2319                 return (EZFS_NODEVICE);
2320
2321         return (0);
2322 }
2323
2324 /*
2325  * Get phys_path for a root pool
2326  * Return 0 on success; non-zero on failure.
2327  */
2328 int
2329 zpool_get_physpath(zpool_handle_t *zhp, char *physpath, size_t phypath_size)
2330 {
2331         return (zpool_get_config_physpath(zhp->zpool_config, physpath,
2332             phypath_size));
2333 }
2334
2335 /*
2336  * If the device has being dynamically expanded then we need to relabel
2337  * the disk to use the new unallocated space.
2338  */
2339 static int
2340 zpool_relabel_disk(libzfs_handle_t *hdl, const char *name)
2341 {
2342 #ifdef sun
2343         char path[MAXPATHLEN];
2344         char errbuf[1024];
2345         int fd, error;
2346         int (*_efi_use_whole_disk)(int);
2347
2348         if ((_efi_use_whole_disk = (int (*)(int))dlsym(RTLD_DEFAULT,
2349             "efi_use_whole_disk")) == NULL)
2350                 return (-1);
2351
2352         (void) snprintf(path, sizeof (path), "%s/%s", RDISK_ROOT, name);
2353
2354         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
2355                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2356                     "relabel '%s': unable to open device"), name);
2357                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
2358         }
2359
2360         /*
2361          * It's possible that we might encounter an error if the device
2362          * does not have any unallocated space left. If so, we simply
2363          * ignore that error and continue on.
2364          */
2365         error = _efi_use_whole_disk(fd);
2366         (void) close(fd);
2367         if (error && error != VT_ENOSPC) {
2368                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "cannot "
2369                     "relabel '%s': unable to read disk capacity"), name);
2370                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
2371         }
2372 #endif  /* sun */
2373         return (0);
2374 }
2375
2376 /*
2377  * Bring the specified vdev online.   The 'flags' parameter is a set of the
2378  * ZFS_ONLINE_* flags.
2379  */
2380 int
2381 zpool_vdev_online(zpool_handle_t *zhp, const char *path, int flags,
2382     vdev_state_t *newstate)
2383 {
2384         zfs_cmd_t zc = { 0 };
2385         char msg[1024];
2386         nvlist_t *tgt;
2387         boolean_t avail_spare, l2cache, islog;
2388         libzfs_handle_t *hdl = zhp->zpool_hdl;
2389
2390         if (flags & ZFS_ONLINE_EXPAND) {
2391                 (void) snprintf(msg, sizeof (msg),
2392                     dgettext(TEXT_DOMAIN, "cannot expand %s"), path);
2393         } else {
2394                 (void) snprintf(msg, sizeof (msg),
2395                     dgettext(TEXT_DOMAIN, "cannot online %s"), path);
2396         }
2397
2398         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2399         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2400             &islog)) == NULL)
2401                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2402
2403         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2404
2405         if (avail_spare)
2406                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2407
2408         if (flags & ZFS_ONLINE_EXPAND ||
2409             zpool_get_prop_int(zhp, ZPOOL_PROP_AUTOEXPAND, NULL)) {
2410                 char *pathname = NULL;
2411                 uint64_t wholedisk = 0;
2412
2413                 (void) nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_WHOLE_DISK,
2414                     &wholedisk);
2415                 verify(nvlist_lookup_string(tgt, ZPOOL_CONFIG_PATH,
2416                     &pathname) == 0);
2417
2418                 /*
2419                  * XXX - L2ARC 1.0 devices can't support expansion.
2420                  */
2421                 if (l2cache) {
2422                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2423                             "cannot expand cache devices"));
2424                         return (zfs_error(hdl, EZFS_VDEVNOTSUP, msg));
2425                 }
2426
2427                 if (wholedisk) {
2428                         pathname += strlen(DISK_ROOT) + 1;
2429                         (void) zpool_relabel_disk(hdl, pathname);
2430                 }
2431         }
2432
2433         zc.zc_cookie = VDEV_STATE_ONLINE;
2434         zc.zc_obj = flags;
2435
2436         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) != 0) {
2437                 if (errno == EINVAL) {
2438                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "was split "
2439                             "from this pool into a new one.  Use '%s' "
2440                             "instead"), "zpool detach");
2441                         return (zfs_error(hdl, EZFS_POSTSPLIT_ONLINE, msg));
2442                 }
2443                 return (zpool_standard_error(hdl, errno, msg));
2444         }
2445
2446         *newstate = zc.zc_cookie;
2447         return (0);
2448 }
2449
2450 /*
2451  * Take the specified vdev offline
2452  */
2453 int
2454 zpool_vdev_offline(zpool_handle_t *zhp, const char *path, boolean_t istmp)
2455 {
2456         zfs_cmd_t zc = { 0 };
2457         char msg[1024];
2458         nvlist_t *tgt;
2459         boolean_t avail_spare, l2cache;
2460         libzfs_handle_t *hdl = zhp->zpool_hdl;
2461
2462         (void) snprintf(msg, sizeof (msg),
2463             dgettext(TEXT_DOMAIN, "cannot offline %s"), path);
2464
2465         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2466         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2467             NULL)) == NULL)
2468                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2469
2470         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2471
2472         if (avail_spare)
2473                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2474
2475         zc.zc_cookie = VDEV_STATE_OFFLINE;
2476         zc.zc_obj = istmp ? ZFS_OFFLINE_TEMPORARY : 0;
2477
2478         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2479                 return (0);
2480
2481         switch (errno) {
2482         case EBUSY:
2483
2484                 /*
2485                  * There are no other replicas of this device.
2486                  */
2487                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2488
2489         case EEXIST:
2490                 /*
2491                  * The log device has unplayed logs
2492                  */
2493                 return (zfs_error(hdl, EZFS_UNPLAYED_LOGS, msg));
2494
2495         default:
2496                 return (zpool_standard_error(hdl, errno, msg));
2497         }
2498 }
2499
2500 /*
2501  * Mark the given vdev faulted.
2502  */
2503 int
2504 zpool_vdev_fault(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2505 {
2506         zfs_cmd_t zc = { 0 };
2507         char msg[1024];
2508         libzfs_handle_t *hdl = zhp->zpool_hdl;
2509
2510         (void) snprintf(msg, sizeof (msg),
2511             dgettext(TEXT_DOMAIN, "cannot fault %llu"), guid);
2512
2513         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2514         zc.zc_guid = guid;
2515         zc.zc_cookie = VDEV_STATE_FAULTED;
2516         zc.zc_obj = aux;
2517
2518         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2519                 return (0);
2520
2521         switch (errno) {
2522         case EBUSY:
2523
2524                 /*
2525                  * There are no other replicas of this device.
2526                  */
2527                 return (zfs_error(hdl, EZFS_NOREPLICAS, msg));
2528
2529         default:
2530                 return (zpool_standard_error(hdl, errno, msg));
2531         }
2532
2533 }
2534
2535 /*
2536  * Mark the given vdev degraded.
2537  */
2538 int
2539 zpool_vdev_degrade(zpool_handle_t *zhp, uint64_t guid, vdev_aux_t aux)
2540 {
2541         zfs_cmd_t zc = { 0 };
2542         char msg[1024];
2543         libzfs_handle_t *hdl = zhp->zpool_hdl;
2544
2545         (void) snprintf(msg, sizeof (msg),
2546             dgettext(TEXT_DOMAIN, "cannot degrade %llu"), guid);
2547
2548         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2549         zc.zc_guid = guid;
2550         zc.zc_cookie = VDEV_STATE_DEGRADED;
2551         zc.zc_obj = aux;
2552
2553         if (ioctl(hdl->libzfs_fd, ZFS_IOC_VDEV_SET_STATE, &zc) == 0)
2554                 return (0);
2555
2556         return (zpool_standard_error(hdl, errno, msg));
2557 }
2558
2559 /*
2560  * Returns TRUE if the given nvlist is a vdev that was originally swapped in as
2561  * a hot spare.
2562  */
2563 static boolean_t
2564 is_replacing_spare(nvlist_t *search, nvlist_t *tgt, int which)
2565 {
2566         nvlist_t **child;
2567         uint_t c, children;
2568         char *type;
2569
2570         if (nvlist_lookup_nvlist_array(search, ZPOOL_CONFIG_CHILDREN, &child,
2571             &children) == 0) {
2572                 verify(nvlist_lookup_string(search, ZPOOL_CONFIG_TYPE,
2573                     &type) == 0);
2574
2575                 if (strcmp(type, VDEV_TYPE_SPARE) == 0 &&
2576                     children == 2 && child[which] == tgt)
2577                         return (B_TRUE);
2578
2579                 for (c = 0; c < children; c++)
2580                         if (is_replacing_spare(child[c], tgt, which))
2581                                 return (B_TRUE);
2582         }
2583
2584         return (B_FALSE);
2585 }
2586
2587 /*
2588  * Attach new_disk (fully described by nvroot) to old_disk.
2589  * If 'replacing' is specified, the new disk will replace the old one.
2590  */
2591 int
2592 zpool_vdev_attach(zpool_handle_t *zhp,
2593     const char *old_disk, const char *new_disk, nvlist_t *nvroot, int replacing)
2594 {
2595         zfs_cmd_t zc = { 0 };
2596         char msg[1024];
2597         int ret;
2598         nvlist_t *tgt;
2599         boolean_t avail_spare, l2cache, islog;
2600         uint64_t val;
2601         char *newname;
2602         nvlist_t **child;
2603         uint_t children;
2604         nvlist_t *config_root;
2605         libzfs_handle_t *hdl = zhp->zpool_hdl;
2606         boolean_t rootpool = zpool_is_bootable(zhp);
2607
2608         if (replacing)
2609                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2610                     "cannot replace %s with %s"), old_disk, new_disk);
2611         else
2612                 (void) snprintf(msg, sizeof (msg), dgettext(TEXT_DOMAIN,
2613                     "cannot attach %s to %s"), new_disk, old_disk);
2614
2615         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2616         if ((tgt = zpool_find_vdev(zhp, old_disk, &avail_spare, &l2cache,
2617             &islog)) == 0)
2618                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2619
2620         if (avail_spare)
2621                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2622
2623         if (l2cache)
2624                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2625
2626         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2627         zc.zc_cookie = replacing;
2628
2629         if (nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
2630             &child, &children) != 0 || children != 1) {
2631                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2632                     "new device must be a single disk"));
2633                 return (zfs_error(hdl, EZFS_INVALCONFIG, msg));
2634         }
2635
2636         verify(nvlist_lookup_nvlist(zpool_get_config(zhp, NULL),
2637             ZPOOL_CONFIG_VDEV_TREE, &config_root) == 0);
2638
2639         if ((newname = zpool_vdev_name(NULL, NULL, child[0], B_FALSE)) == NULL)
2640                 return (-1);
2641
2642         /*
2643          * If the target is a hot spare that has been swapped in, we can only
2644          * replace it with another hot spare.
2645          */
2646         if (replacing &&
2647             nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_IS_SPARE, &val) == 0 &&
2648             (zpool_find_vdev(zhp, newname, &avail_spare, &l2cache,
2649             NULL) == NULL || !avail_spare) &&
2650             is_replacing_spare(config_root, tgt, 1)) {
2651                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2652                     "can only be replaced by another hot spare"));
2653                 free(newname);
2654                 return (zfs_error(hdl, EZFS_BADTARGET, msg));
2655         }
2656
2657         free(newname);
2658
2659         if (zcmd_write_conf_nvlist(hdl, &zc, nvroot) != 0)
2660                 return (-1);
2661
2662         ret = zfs_ioctl(hdl, ZFS_IOC_VDEV_ATTACH, &zc);
2663
2664         zcmd_free_nvlists(&zc);
2665
2666         if (ret == 0) {
2667                 if (rootpool) {
2668                         /*
2669                          * XXX need a better way to prevent user from
2670                          * booting up a half-baked vdev.
2671                          */
2672                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "Make "
2673                             "sure to wait until resilver is done "
2674                             "before rebooting.\n"));
2675                         (void) fprintf(stderr, "\n");
2676                         (void) fprintf(stderr, dgettext(TEXT_DOMAIN, "If "
2677                             "you boot from pool '%s', you may need to update\n"
2678                             "boot code on newly attached disk '%s'.\n\n"
2679                             "Assuming you use GPT partitioning and 'da0' is "
2680                             "your new boot disk\n"
2681                             "you may use the following command:\n\n"
2682                             "\tgpart bootcode -b /boot/pmbr -p "
2683                             "/boot/gptzfsboot -i 1 da0\n\n"),
2684                             zhp->zpool_name, new_disk);
2685                 }
2686                 return (0);
2687         }
2688
2689         switch (errno) {
2690         case ENOTSUP:
2691                 /*
2692                  * Can't attach to or replace this type of vdev.
2693                  */
2694                 if (replacing) {
2695                         uint64_t version = zpool_get_prop_int(zhp,
2696                             ZPOOL_PROP_VERSION, NULL);
2697
2698                         if (islog)
2699                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2700                                     "cannot replace a log with a spare"));
2701                         else if (version >= SPA_VERSION_MULTI_REPLACE)
2702                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2703                                     "already in replacing/spare config; wait "
2704                                     "for completion or use 'zpool detach'"));
2705                         else
2706                                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2707                                     "cannot replace a replacing device"));
2708                 } else {
2709                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2710                             "can only attach to mirrors and top-level "
2711                             "disks"));
2712                 }
2713                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2714                 break;
2715
2716         case EINVAL:
2717                 /*
2718                  * The new device must be a single disk.
2719                  */
2720                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2721                     "new device must be a single disk"));
2722                 (void) zfs_error(hdl, EZFS_INVALCONFIG, msg);
2723                 break;
2724
2725         case EBUSY:
2726                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "%s is busy"),
2727                     new_disk);
2728                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2729                 break;
2730
2731         case EOVERFLOW:
2732                 /*
2733                  * The new device is too small.
2734                  */
2735                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2736                     "device is too small"));
2737                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2738                 break;
2739
2740         case EDOM:
2741                 /*
2742                  * The new device has a different alignment requirement.
2743                  */
2744                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2745                     "devices have different sector alignment"));
2746                 (void) zfs_error(hdl, EZFS_BADDEV, msg);
2747                 break;
2748
2749         case ENAMETOOLONG:
2750                 /*
2751                  * The resulting top-level vdev spec won't fit in the label.
2752                  */
2753                 (void) zfs_error(hdl, EZFS_DEVOVERFLOW, msg);
2754                 break;
2755
2756         default:
2757                 (void) zpool_standard_error(hdl, errno, msg);
2758         }
2759
2760         return (-1);
2761 }
2762
2763 /*
2764  * Detach the specified device.
2765  */
2766 int
2767 zpool_vdev_detach(zpool_handle_t *zhp, const char *path)
2768 {
2769         zfs_cmd_t zc = { 0 };
2770         char msg[1024];
2771         nvlist_t *tgt;
2772         boolean_t avail_spare, l2cache;
2773         libzfs_handle_t *hdl = zhp->zpool_hdl;
2774
2775         (void) snprintf(msg, sizeof (msg),
2776             dgettext(TEXT_DOMAIN, "cannot detach %s"), path);
2777
2778         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
2779         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
2780             NULL)) == 0)
2781                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
2782
2783         if (avail_spare)
2784                 return (zfs_error(hdl, EZFS_ISSPARE, msg));
2785
2786         if (l2cache)
2787                 return (zfs_error(hdl, EZFS_ISL2CACHE, msg));
2788
2789         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
2790
2791         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_DETACH, &zc) == 0)
2792                 return (0);
2793
2794         switch (errno) {
2795
2796         case ENOTSUP:
2797                 /*
2798                  * Can't detach from this type of vdev.
2799                  */
2800                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "only "
2801                     "applicable to mirror and replacing vdevs"));
2802                 (void) zfs_error(hdl, EZFS_BADTARGET, msg);
2803                 break;
2804
2805         case EBUSY:
2806                 /*
2807                  * There are no other replicas of this device.
2808                  */
2809                 (void) zfs_error(hdl, EZFS_NOREPLICAS, msg);
2810                 break;
2811
2812         default:
2813                 (void) zpool_standard_error(hdl, errno, msg);
2814         }
2815
2816         return (-1);
2817 }
2818
2819 /*
2820  * Find a mirror vdev in the source nvlist.
2821  *
2822  * The mchild array contains a list of disks in one of the top-level mirrors
2823  * of the source pool.  The schild array contains a list of disks that the
2824  * user specified on the command line.  We loop over the mchild array to
2825  * see if any entry in the schild array matches.
2826  *
2827  * If a disk in the mchild array is found in the schild array, we return
2828  * the index of that entry.  Otherwise we return -1.
2829  */
2830 static int
2831 find_vdev_entry(zpool_handle_t *zhp, nvlist_t **mchild, uint_t mchildren,
2832     nvlist_t **schild, uint_t schildren)
2833 {
2834         uint_t mc;
2835
2836         for (mc = 0; mc < mchildren; mc++) {
2837                 uint_t sc;
2838                 char *mpath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2839                     mchild[mc], B_FALSE);
2840
2841                 for (sc = 0; sc < schildren; sc++) {
2842                         char *spath = zpool_vdev_name(zhp->zpool_hdl, zhp,
2843                             schild[sc], B_FALSE);
2844                         boolean_t result = (strcmp(mpath, spath) == 0);
2845
2846                         free(spath);
2847                         if (result) {
2848                                 free(mpath);
2849                                 return (mc);
2850                         }
2851                 }
2852
2853                 free(mpath);
2854         }
2855
2856         return (-1);
2857 }
2858
2859 /*
2860  * Split a mirror pool.  If newroot points to null, then a new nvlist
2861  * is generated and it is the responsibility of the caller to free it.
2862  */
2863 int
2864 zpool_vdev_split(zpool_handle_t *zhp, char *newname, nvlist_t **newroot,
2865     nvlist_t *props, splitflags_t flags)
2866 {
2867         zfs_cmd_t zc = { 0 };
2868         char msg[1024];
2869         nvlist_t *tree, *config, **child, **newchild, *newconfig = NULL;
2870         nvlist_t **varray = NULL, *zc_props = NULL;
2871         uint_t c, children, newchildren, lastlog = 0, vcount, found = 0;
2872         libzfs_handle_t *hdl = zhp->zpool_hdl;
2873         uint64_t vers;
2874         boolean_t freelist = B_FALSE, memory_err = B_TRUE;
2875         int retval = 0;
2876
2877         (void) snprintf(msg, sizeof (msg),
2878             dgettext(TEXT_DOMAIN, "Unable to split %s"), zhp->zpool_name);
2879
2880         if (!zpool_name_valid(hdl, B_FALSE, newname))
2881                 return (zfs_error(hdl, EZFS_INVALIDNAME, msg));
2882
2883         if ((config = zpool_get_config(zhp, NULL)) == NULL) {
2884                 (void) fprintf(stderr, gettext("Internal error: unable to "
2885                     "retrieve pool configuration\n"));
2886                 return (-1);
2887         }
2888
2889         verify(nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, &tree)
2890             == 0);
2891         verify(nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION, &vers) == 0);
2892
2893         if (props) {
2894                 prop_flags_t flags = { .create = B_FALSE, .import = B_TRUE };
2895                 if ((zc_props = zpool_valid_proplist(hdl, zhp->zpool_name,
2896                     props, vers, flags, msg)) == NULL)
2897                         return (-1);
2898         }
2899
2900         if (nvlist_lookup_nvlist_array(tree, ZPOOL_CONFIG_CHILDREN, &child,
2901             &children) != 0) {
2902                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2903                     "Source pool is missing vdev tree"));
2904                 if (zc_props)
2905                         nvlist_free(zc_props);
2906                 return (-1);
2907         }
2908
2909         varray = zfs_alloc(hdl, children * sizeof (nvlist_t *));
2910         vcount = 0;
2911
2912         if (*newroot == NULL ||
2913             nvlist_lookup_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN,
2914             &newchild, &newchildren) != 0)
2915                 newchildren = 0;
2916
2917         for (c = 0; c < children; c++) {
2918                 uint64_t is_log = B_FALSE, is_hole = B_FALSE;
2919                 char *type;
2920                 nvlist_t **mchild, *vdev;
2921                 uint_t mchildren;
2922                 int entry;
2923
2924                 /*
2925                  * Unlike cache & spares, slogs are stored in the
2926                  * ZPOOL_CONFIG_CHILDREN array.  We filter them out here.
2927                  */
2928                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_LOG,
2929                     &is_log);
2930                 (void) nvlist_lookup_uint64(child[c], ZPOOL_CONFIG_IS_HOLE,
2931                     &is_hole);
2932                 if (is_log || is_hole) {
2933                         /*
2934                          * Create a hole vdev and put it in the config.
2935                          */
2936                         if (nvlist_alloc(&vdev, NV_UNIQUE_NAME, 0) != 0)
2937                                 goto out;
2938                         if (nvlist_add_string(vdev, ZPOOL_CONFIG_TYPE,
2939                             VDEV_TYPE_HOLE) != 0)
2940                                 goto out;
2941                         if (nvlist_add_uint64(vdev, ZPOOL_CONFIG_IS_HOLE,
2942                             1) != 0)
2943                                 goto out;
2944                         if (lastlog == 0)
2945                                 lastlog = vcount;
2946                         varray[vcount++] = vdev;
2947                         continue;
2948                 }
2949                 lastlog = 0;
2950                 verify(nvlist_lookup_string(child[c], ZPOOL_CONFIG_TYPE, &type)
2951                     == 0);
2952                 if (strcmp(type, VDEV_TYPE_MIRROR) != 0) {
2953                         zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2954                             "Source pool must be composed only of mirrors\n"));
2955                         retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2956                         goto out;
2957                 }
2958
2959                 verify(nvlist_lookup_nvlist_array(child[c],
2960                     ZPOOL_CONFIG_CHILDREN, &mchild, &mchildren) == 0);
2961
2962                 /* find or add an entry for this top-level vdev */
2963                 if (newchildren > 0 &&
2964                     (entry = find_vdev_entry(zhp, mchild, mchildren,
2965                     newchild, newchildren)) >= 0) {
2966                         /* We found a disk that the user specified. */
2967                         vdev = mchild[entry];
2968                         ++found;
2969                 } else {
2970                         /* User didn't specify a disk for this vdev. */
2971                         vdev = mchild[mchildren - 1];
2972                 }
2973
2974                 if (nvlist_dup(vdev, &varray[vcount++], 0) != 0)
2975                         goto out;
2976         }
2977
2978         /* did we find every disk the user specified? */
2979         if (found != newchildren) {
2980                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, "Device list must "
2981                     "include at most one disk from each mirror"));
2982                 retval = zfs_error(hdl, EZFS_INVALCONFIG, msg);
2983                 goto out;
2984         }
2985
2986         /* Prepare the nvlist for populating. */
2987         if (*newroot == NULL) {
2988                 if (nvlist_alloc(newroot, NV_UNIQUE_NAME, 0) != 0)
2989                         goto out;
2990                 freelist = B_TRUE;
2991                 if (nvlist_add_string(*newroot, ZPOOL_CONFIG_TYPE,
2992                     VDEV_TYPE_ROOT) != 0)
2993                         goto out;
2994         } else {
2995                 verify(nvlist_remove_all(*newroot, ZPOOL_CONFIG_CHILDREN) == 0);
2996         }
2997
2998         /* Add all the children we found */
2999         if (nvlist_add_nvlist_array(*newroot, ZPOOL_CONFIG_CHILDREN, varray,
3000             lastlog == 0 ? vcount : lastlog) != 0)
3001                 goto out;
3002
3003         /*
3004          * If we're just doing a dry run, exit now with success.
3005          */
3006         if (flags.dryrun) {
3007                 memory_err = B_FALSE;
3008                 freelist = B_FALSE;
3009                 goto out;
3010         }
3011
3012         /* now build up the config list & call the ioctl */
3013         if (nvlist_alloc(&newconfig, NV_UNIQUE_NAME, 0) != 0)
3014                 goto out;
3015
3016         if (nvlist_add_nvlist(newconfig,
3017             ZPOOL_CONFIG_VDEV_TREE, *newroot) != 0 ||
3018             nvlist_add_string(newconfig,
3019             ZPOOL_CONFIG_POOL_NAME, newname) != 0 ||
3020             nvlist_add_uint64(newconfig, ZPOOL_CONFIG_VERSION, vers) != 0)
3021                 goto out;
3022
3023         /*
3024          * The new pool is automatically part of the namespace unless we
3025          * explicitly export it.
3026          */
3027         if (!flags.import)
3028                 zc.zc_cookie = ZPOOL_EXPORT_AFTER_SPLIT;
3029         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3030         (void) strlcpy(zc.zc_string, newname, sizeof (zc.zc_string));
3031         if (zcmd_write_conf_nvlist(hdl, &zc, newconfig) != 0)
3032                 goto out;
3033         if (zc_props != NULL && zcmd_write_src_nvlist(hdl, &zc, zc_props) != 0)
3034                 goto out;
3035
3036         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_SPLIT, &zc) != 0) {
3037                 retval = zpool_standard_error(hdl, errno, msg);
3038                 goto out;
3039         }
3040
3041         freelist = B_FALSE;
3042         memory_err = B_FALSE;
3043
3044 out:
3045         if (varray != NULL) {
3046                 int v;
3047
3048                 for (v = 0; v < vcount; v++)
3049                         nvlist_free(varray[v]);
3050                 free(varray);
3051         }
3052         zcmd_free_nvlists(&zc);
3053         if (zc_props)
3054                 nvlist_free(zc_props);
3055         if (newconfig)
3056                 nvlist_free(newconfig);
3057         if (freelist) {
3058                 nvlist_free(*newroot);
3059                 *newroot = NULL;
3060         }
3061
3062         if (retval != 0)
3063                 return (retval);
3064
3065         if (memory_err)
3066                 return (no_memory(hdl));
3067
3068         return (0);
3069 }
3070
3071 /*
3072  * Remove the given device.  Currently, this is supported only for hot spares
3073  * and level 2 cache devices.
3074  */
3075 int
3076 zpool_vdev_remove(zpool_handle_t *zhp, const char *path)
3077 {
3078         zfs_cmd_t zc = { 0 };
3079         char msg[1024];
3080         nvlist_t *tgt;
3081         boolean_t avail_spare, l2cache, islog;
3082         libzfs_handle_t *hdl = zhp->zpool_hdl;
3083         uint64_t version;
3084
3085         (void) snprintf(msg, sizeof (msg),
3086             dgettext(TEXT_DOMAIN, "cannot remove %s"), path);
3087
3088         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3089         if ((tgt = zpool_find_vdev(zhp, path, &avail_spare, &l2cache,
3090             &islog)) == 0)
3091                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3092         /*
3093          * XXX - this should just go away.
3094          */
3095         if (!avail_spare && !l2cache && !islog) {
3096                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3097                     "only inactive hot spares, cache, top-level, "
3098                     "or log devices can be removed"));
3099                 return (zfs_error(hdl, EZFS_NODEVICE, msg));
3100         }
3101
3102         version = zpool_get_prop_int(zhp, ZPOOL_PROP_VERSION, NULL);
3103         if (islog && version < SPA_VERSION_HOLES) {
3104                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3105                     "pool must be upgrade to support log removal"));
3106                 return (zfs_error(hdl, EZFS_BADVERSION, msg));
3107         }
3108
3109         verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID, &zc.zc_guid) == 0);
3110
3111         if (zfs_ioctl(hdl, ZFS_IOC_VDEV_REMOVE, &zc) == 0)
3112                 return (0);
3113
3114         return (zpool_standard_error(hdl, errno, msg));
3115 }
3116
3117 /*
3118  * Clear the errors for the pool, or the particular device if specified.
3119  */
3120 int
3121 zpool_clear(zpool_handle_t *zhp, const char *path, nvlist_t *rewindnvl)
3122 {
3123         zfs_cmd_t zc = { 0 };
3124         char msg[1024];
3125         nvlist_t *tgt;
3126         zpool_rewind_policy_t policy;
3127         boolean_t avail_spare, l2cache;
3128         libzfs_handle_t *hdl = zhp->zpool_hdl;
3129         nvlist_t *nvi = NULL;
3130         int error;
3131
3132         if (path)
3133                 (void) snprintf(msg, sizeof (msg),
3134                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3135                     path);
3136         else
3137                 (void) snprintf(msg, sizeof (msg),
3138                     dgettext(TEXT_DOMAIN, "cannot clear errors for %s"),
3139                     zhp->zpool_name);
3140
3141         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3142         if (path) {
3143                 if ((tgt = zpool_find_vdev(zhp, path, &avail_spare,
3144                     &l2cache, NULL)) == 0)
3145                         return (zfs_error(hdl, EZFS_NODEVICE, msg));
3146
3147                 /*
3148                  * Don't allow error clearing for hot spares.  Do allow
3149                  * error clearing for l2cache devices.
3150                  */
3151                 if (avail_spare)
3152                         return (zfs_error(hdl, EZFS_ISSPARE, msg));
3153
3154                 verify(nvlist_lookup_uint64(tgt, ZPOOL_CONFIG_GUID,
3155                     &zc.zc_guid) == 0);
3156         }
3157
3158         zpool_get_rewind_policy(rewindnvl, &policy);
3159         zc.zc_cookie = policy.zrp_request;
3160
3161         if (zcmd_alloc_dst_nvlist(hdl, &zc, zhp->zpool_config_size * 2) != 0)
3162                 return (-1);
3163
3164         if (zcmd_write_src_nvlist(hdl, &zc, rewindnvl) != 0)
3165                 return (-1);
3166
3167         while ((error = zfs_ioctl(hdl, ZFS_IOC_CLEAR, &zc)) != 0 &&
3168             errno == ENOMEM) {
3169                 if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
3170                         zcmd_free_nvlists(&zc);
3171                         return (-1);
3172                 }
3173         }
3174
3175         if (!error || ((policy.zrp_request & ZPOOL_TRY_REWIND) &&
3176             errno != EPERM && errno != EACCES)) {
3177                 if (policy.zrp_request &
3178                     (ZPOOL_DO_REWIND | ZPOOL_TRY_REWIND)) {
3179                         (void) zcmd_read_dst_nvlist(hdl, &zc, &nvi);
3180                         zpool_rewind_exclaim(hdl, zc.zc_name,
3181                             ((policy.zrp_request & ZPOOL_TRY_REWIND) != 0),
3182                             nvi);
3183                         nvlist_free(nvi);
3184                 }
3185                 zcmd_free_nvlists(&zc);
3186                 return (0);
3187         }
3188
3189         zcmd_free_nvlists(&zc);
3190         return (zpool_standard_error(hdl, errno, msg));
3191 }
3192
3193 /*
3194  * Similar to zpool_clear(), but takes a GUID (used by fmd).
3195  */
3196 int
3197 zpool_vdev_clear(zpool_handle_t *zhp, uint64_t guid)
3198 {
3199         zfs_cmd_t zc = { 0 };
3200         char msg[1024];
3201         libzfs_handle_t *hdl = zhp->zpool_hdl;
3202
3203         (void) snprintf(msg, sizeof (msg),
3204             dgettext(TEXT_DOMAIN, "cannot clear errors for %llx"),
3205             guid);
3206
3207         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3208         zc.zc_guid = guid;
3209         zc.zc_cookie = ZPOOL_NO_REWIND;
3210
3211         if (ioctl(hdl->libzfs_fd, ZFS_IOC_CLEAR, &zc) == 0)
3212                 return (0);
3213
3214         return (zpool_standard_error(hdl, errno, msg));
3215 }
3216
3217 /*
3218  * Change the GUID for a pool.
3219  */
3220 int
3221 zpool_reguid(zpool_handle_t *zhp)
3222 {
3223         char msg[1024];
3224         libzfs_handle_t *hdl = zhp->zpool_hdl;
3225         zfs_cmd_t zc = { 0 };
3226
3227         (void) snprintf(msg, sizeof (msg),
3228             dgettext(TEXT_DOMAIN, "cannot reguid '%s'"), zhp->zpool_name);
3229
3230         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3231         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REGUID, &zc) == 0)
3232                 return (0);
3233
3234         return (zpool_standard_error(hdl, errno, msg));
3235 }
3236
3237 /*
3238  * Reopen the pool.
3239  */
3240 int
3241 zpool_reopen(zpool_handle_t *zhp)
3242 {
3243         zfs_cmd_t zc = { 0 };
3244         char msg[1024];
3245         libzfs_handle_t *hdl = zhp->zpool_hdl;
3246
3247         (void) snprintf(msg, sizeof (msg),
3248             dgettext(TEXT_DOMAIN, "cannot reopen '%s'"),
3249             zhp->zpool_name);
3250
3251         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3252         if (zfs_ioctl(hdl, ZFS_IOC_POOL_REOPEN, &zc) == 0)
3253                 return (0);
3254         return (zpool_standard_error(hdl, errno, msg));
3255 }
3256
3257 /*
3258  * Convert from a devid string to a path.
3259  */
3260 static char *
3261 devid_to_path(char *devid_str)
3262 {
3263         ddi_devid_t devid;
3264         char *minor;
3265         char *path;
3266         devid_nmlist_t *list = NULL;
3267         int ret;
3268
3269         if (devid_str_decode(devid_str, &devid, &minor) != 0)
3270                 return (NULL);
3271
3272         ret = devid_deviceid_to_nmlist("/dev", devid, minor, &list);
3273
3274         devid_str_free(minor);
3275         devid_free(devid);
3276
3277         if (ret != 0)
3278                 return (NULL);
3279
3280         /*
3281          * In a case the strdup() fails, we will just return NULL below.
3282          */
3283         path = strdup(list[0].devname);
3284
3285         devid_free_nmlist(list);
3286
3287         return (path);
3288 }
3289
3290 /*
3291  * Convert from a path to a devid string.
3292  */
3293 static char *
3294 path_to_devid(const char *path)
3295 {
3296 #ifdef have_devid
3297         int fd;
3298         ddi_devid_t devid;
3299         char *minor, *ret;
3300
3301         if ((fd = open(path, O_RDONLY)) < 0)
3302                 return (NULL);
3303
3304         minor = NULL;
3305         ret = NULL;
3306         if (devid_get(fd, &devid) == 0) {
3307                 if (devid_get_minor_name(fd, &minor) == 0)
3308                         ret = devid_str_encode(devid, minor);
3309                 if (minor != NULL)
3310                         devid_str_free(minor);
3311                 devid_free(devid);
3312         }
3313         (void) close(fd);
3314
3315         return (ret);
3316 #else
3317         return (NULL);
3318 #endif
3319 }
3320
3321 /*
3322  * Issue the necessary ioctl() to update the stored path value for the vdev.  We
3323  * ignore any failure here, since a common case is for an unprivileged user to
3324  * type 'zpool status', and we'll display the correct information anyway.
3325  */
3326 static void
3327 set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
3328 {
3329         zfs_cmd_t zc = { 0 };
3330
3331         (void) strncpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3332         (void) strncpy(zc.zc_value, path, sizeof (zc.zc_value));
3333         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3334             &zc.zc_guid) == 0);
3335
3336         (void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
3337 }
3338
3339 /*
3340  * Given a vdev, return the name to display in iostat.  If the vdev has a path,
3341  * we use that, stripping off any leading "/dev/dsk/"; if not, we use the type.
3342  * We also check if this is a whole disk, in which case we strip off the
3343  * trailing 's0' slice name.
3344  *
3345  * This routine is also responsible for identifying when disks have been
3346  * reconfigured in a new location.  The kernel will have opened the device by
3347  * devid, but the path will still refer to the old location.  To catch this, we
3348  * first do a path -> devid translation (which is fast for the common case).  If
3349  * the devid matches, we're done.  If not, we do a reverse devid -> path
3350  * translation and issue the appropriate ioctl() to update the path of the vdev.
3351  * If 'zhp' is NULL, then this is an exported pool, and we don't need to do any
3352  * of these checks.
3353  */
3354 char *
3355 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
3356     boolean_t verbose)
3357 {
3358         char *path, *devid;
3359         uint64_t value;
3360         char buf[64];
3361         vdev_stat_t *vs;
3362         uint_t vsc;
3363         int have_stats;
3364         int have_path;
3365
3366         have_stats = nvlist_lookup_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
3367             (uint64_t **)&vs, &vsc) == 0;
3368         have_path = nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0;
3369
3370         /*
3371          * If the device is not currently present, assume it will not
3372          * come back at the same device path.  Display the device by GUID.
3373          */
3374         if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, &value) == 0 ||
3375             have_path && have_stats && vs->vs_state <= VDEV_STATE_CANT_OPEN) {
3376                 verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID,
3377                     &value) == 0);
3378                 (void) snprintf(buf, sizeof (buf), "%llu",
3379                     (u_longlong_t)value);
3380                 path = buf;
3381         } else if (have_path) {
3382
3383                 /*
3384                  * If the device is dead (faulted, offline, etc) then don't
3385                  * bother opening it.  Otherwise we may be forcing the user to
3386                  * open a misbehaving device, which can have undesirable
3387                  * effects.
3388                  */
3389                 if ((have_stats == 0 ||
3390                     vs->vs_state >= VDEV_STATE_DEGRADED) &&
3391                     zhp != NULL &&
3392                     nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &devid) == 0) {
3393                         /*
3394                          * Determine if the current path is correct.
3395                          */
3396                         char *newdevid = path_to_devid(path);
3397
3398                         if (newdevid == NULL ||
3399                             strcmp(devid, newdevid) != 0) {
3400                                 char *newpath;
3401
3402                                 if ((newpath = devid_to_path(devid)) != NULL) {
3403                                         /*
3404                                          * Update the path appropriately.
3405                                          */
3406                                         set_path(zhp, nv, newpath);
3407                                         if (nvlist_add_string(nv,
3408                                             ZPOOL_CONFIG_PATH, newpath) == 0)
3409                                                 verify(nvlist_lookup_string(nv,
3410                                                     ZPOOL_CONFIG_PATH,
3411                                                     &path) == 0);
3412                                         free(newpath);
3413                                 }
3414                         }
3415
3416                         if (newdevid)
3417                                 devid_str_free(newdevid);
3418                 }
3419
3420 #ifdef sun
3421                 if (strncmp(path, "/dev/dsk/", 9) == 0)
3422                         path += 9;
3423
3424                 if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
3425                     &value) == 0 && value) {
3426                         int pathlen = strlen(path);
3427                         char *tmp = zfs_strdup(hdl, path);
3428
3429                         /*
3430                          * If it starts with c#, and ends with "s0", chop
3431                          * the "s0" off, or if it ends with "s0/old", remove
3432                          * the "s0" from the middle.
3433                          */
3434                         if (CTD_CHECK(tmp)) {
3435                                 if (strcmp(&tmp[pathlen - 2], "s0") == 0) {
3436                                         tmp[pathlen - 2] = '\0';
3437                                 } else if (pathlen > 6 &&
3438                                     strcmp(&tmp[pathlen - 6], "s0/old") == 0) {
3439                                         (void) strcpy(&tmp[pathlen - 6],
3440                                             "/old");
3441                                 }
3442                         }
3443                         return (tmp);
3444                 }
3445 #else   /* !sun */
3446                 if (strncmp(path, _PATH_DEV, sizeof(_PATH_DEV) - 1) == 0)
3447                         path += sizeof(_PATH_DEV) - 1;
3448 #endif  /* !sun */
3449         } else {
3450                 verify(nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &path) == 0);
3451
3452                 /*
3453                  * If it's a raidz device, we need to stick in the parity level.
3454                  */
3455                 if (strcmp(path, VDEV_TYPE_RAIDZ) == 0) {
3456                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY,
3457                             &value) == 0);
3458                         (void) snprintf(buf, sizeof (buf), "%s%llu", path,
3459                             (u_longlong_t)value);
3460                         path = buf;
3461                 }
3462
3463                 /*
3464                  * We identify each top-level vdev by using a <type-id>
3465                  * naming convention.
3466                  */
3467                 if (verbose) {
3468                         uint64_t id;
3469
3470                         verify(nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID,
3471                             &id) == 0);
3472                         (void) snprintf(buf, sizeof (buf), "%s-%llu", path,
3473                             (u_longlong_t)id);
3474                         path = buf;
3475                 }
3476         }
3477
3478         return (zfs_strdup(hdl, path));
3479 }
3480
3481 static int
3482 zbookmark_mem_compare(const void *a, const void *b)
3483 {
3484         return (memcmp(a, b, sizeof (zbookmark_phys_t)));
3485 }
3486
3487 /*
3488  * Retrieve the persistent error log, uniquify the members, and return to the
3489  * caller.
3490  */
3491 int
3492 zpool_get_errlog(zpool_handle_t *zhp, nvlist_t **nverrlistp)
3493 {
3494         zfs_cmd_t zc = { 0 };
3495         uint64_t count;
3496         zbookmark_phys_t *zb = NULL;
3497         int i;
3498
3499         /*
3500          * Retrieve the raw error list from the kernel.  If the number of errors
3501          * has increased, allocate more space and continue until we get the
3502          * entire list.
3503          */
3504         verify(nvlist_lookup_uint64(zhp->zpool_config, ZPOOL_CONFIG_ERRCOUNT,
3505             &count) == 0);
3506         if (count == 0)
3507                 return (0);
3508         if ((zc.zc_nvlist_dst = (uintptr_t)zfs_alloc(zhp->zpool_hdl,
3509             count * sizeof (zbookmark_phys_t))) == (uintptr_t)NULL)
3510                 return (-1);
3511         zc.zc_nvlist_dst_size = count;
3512         (void) strcpy(zc.zc_name, zhp->zpool_name);
3513         for (;;) {
3514                 if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_ERROR_LOG,
3515                     &zc) != 0) {
3516                         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3517                         if (errno == ENOMEM) {
3518                                 void *dst;
3519
3520                                 count = zc.zc_nvlist_dst_size;
3521                                 dst = zfs_alloc(zhp->zpool_hdl, count *
3522                                     sizeof (zbookmark_phys_t));
3523                                 if (dst == NULL)
3524                                         return (-1);
3525                                 zc.zc_nvlist_dst = (uintptr_t)dst;
3526                         } else {
3527                                 return (-1);
3528                         }
3529                 } else {
3530                         break;
3531                 }
3532         }
3533
3534         /*
3535          * Sort the resulting bookmarks.  This is a little confusing due to the
3536          * implementation of ZFS_IOC_ERROR_LOG.  The bookmarks are copied last
3537          * to first, and 'zc_nvlist_dst_size' indicates the number of boomarks
3538          * _not_ copied as part of the process.  So we point the start of our
3539          * array appropriate and decrement the total number of elements.
3540          */
3541         zb = ((zbookmark_phys_t *)(uintptr_t)zc.zc_nvlist_dst) +
3542             zc.zc_nvlist_dst_size;
3543         count -= zc.zc_nvlist_dst_size;
3544
3545         qsort(zb, count, sizeof (zbookmark_phys_t), zbookmark_mem_compare);
3546
3547         verify(nvlist_alloc(nverrlistp, 0, KM_SLEEP) == 0);
3548
3549         /*
3550          * Fill in the nverrlistp with nvlist's of dataset and object numbers.
3551          */
3552         for (i = 0; i < count; i++) {
3553                 nvlist_t *nv;
3554
3555                 /* ignoring zb_blkid and zb_level for now */
3556                 if (i > 0 && zb[i-1].zb_objset == zb[i].zb_objset &&
3557                     zb[i-1].zb_object == zb[i].zb_object)
3558                         continue;
3559
3560                 if (nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) != 0)
3561                         goto nomem;
3562                 if (nvlist_add_uint64(nv, ZPOOL_ERR_DATASET,
3563                     zb[i].zb_objset) != 0) {
3564                         nvlist_free(nv);
3565                         goto nomem;
3566                 }
3567                 if (nvlist_add_uint64(nv, ZPOOL_ERR_OBJECT,
3568                     zb[i].zb_object) != 0) {
3569                         nvlist_free(nv);
3570                         goto nomem;
3571                 }
3572                 if (nvlist_add_nvlist(*nverrlistp, "ejk", nv) != 0) {
3573                         nvlist_free(nv);
3574                         goto nomem;
3575                 }
3576                 nvlist_free(nv);
3577         }
3578
3579         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3580         return (0);
3581
3582 nomem:
3583         free((void *)(uintptr_t)zc.zc_nvlist_dst);
3584         return (no_memory(zhp->zpool_hdl));
3585 }
3586
3587 /*
3588  * Upgrade a ZFS pool to the latest on-disk version.
3589  */
3590 int
3591 zpool_upgrade(zpool_handle_t *zhp, uint64_t new_version)
3592 {
3593         zfs_cmd_t zc = { 0 };
3594         libzfs_handle_t *hdl = zhp->zpool_hdl;
3595
3596         (void) strcpy(zc.zc_name, zhp->zpool_name);
3597         zc.zc_cookie = new_version;
3598
3599         if (zfs_ioctl(hdl, ZFS_IOC_POOL_UPGRADE, &zc) != 0)
3600                 return (zpool_standard_error_fmt(hdl, errno,
3601                     dgettext(TEXT_DOMAIN, "cannot upgrade '%s'"),
3602                     zhp->zpool_name));
3603         return (0);
3604 }
3605
3606 void
3607 zfs_save_arguments(int argc, char **argv, char *string, int len)
3608 {
3609         (void) strlcpy(string, basename(argv[0]), len);
3610         for (int i = 1; i < argc; i++) {
3611                 (void) strlcat(string, " ", len);
3612                 (void) strlcat(string, argv[i], len);
3613         }
3614 }
3615
3616 int
3617 zpool_log_history(libzfs_handle_t *hdl, const char *message)
3618 {
3619         zfs_cmd_t zc = { 0 };
3620         nvlist_t *args;
3621         int err;
3622
3623         args = fnvlist_alloc();
3624         fnvlist_add_string(args, "message", message);
3625         err = zcmd_write_src_nvlist(hdl, &zc, args);
3626         if (err == 0)
3627                 err = ioctl(hdl->libzfs_fd, ZFS_IOC_LOG_HISTORY, &zc);
3628         nvlist_free(args);
3629         zcmd_free_nvlists(&zc);
3630         return (err);
3631 }
3632
3633 /*
3634  * Perform ioctl to get some command history of a pool.
3635  *
3636  * 'buf' is the buffer to fill up to 'len' bytes.  'off' is the
3637  * logical offset of the history buffer to start reading from.
3638  *
3639  * Upon return, 'off' is the next logical offset to read from and
3640  * 'len' is the actual amount of bytes read into 'buf'.
3641  */
3642 static int
3643 get_history(zpool_handle_t *zhp, char *buf, uint64_t *off, uint64_t *len)
3644 {
3645         zfs_cmd_t zc = { 0 };
3646         libzfs_handle_t *hdl = zhp->zpool_hdl;
3647
3648         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3649
3650         zc.zc_history = (uint64_t)(uintptr_t)buf;
3651         zc.zc_history_len = *len;
3652         zc.zc_history_offset = *off;
3653
3654         if (ioctl(hdl->libzfs_fd, ZFS_IOC_POOL_GET_HISTORY, &zc) != 0) {
3655                 switch (errno) {
3656                 case EPERM:
3657                         return (zfs_error_fmt(hdl, EZFS_PERM,
3658                             dgettext(TEXT_DOMAIN,
3659                             "cannot show history for pool '%s'"),
3660                             zhp->zpool_name));
3661                 case ENOENT:
3662                         return (zfs_error_fmt(hdl, EZFS_NOHISTORY,
3663                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3664                             "'%s'"), zhp->zpool_name));
3665                 case ENOTSUP:
3666                         return (zfs_error_fmt(hdl, EZFS_BADVERSION,
3667                             dgettext(TEXT_DOMAIN, "cannot get history for pool "
3668                             "'%s', pool must be upgraded"), zhp->zpool_name));
3669                 default:
3670                         return (zpool_standard_error_fmt(hdl, errno,
3671                             dgettext(TEXT_DOMAIN,
3672                             "cannot get history for '%s'"), zhp->zpool_name));
3673                 }
3674         }
3675
3676         *len = zc.zc_history_len;
3677         *off = zc.zc_history_offset;
3678
3679         return (0);
3680 }
3681
3682 /*
3683  * Process the buffer of nvlists, unpacking and storing each nvlist record
3684  * into 'records'.  'leftover' is set to the number of bytes that weren't
3685  * processed as there wasn't a complete record.
3686  */
3687 int
3688 zpool_history_unpack(char *buf, uint64_t bytes_read, uint64_t *leftover,
3689     nvlist_t ***records, uint_t *numrecords)
3690 {
3691         uint64_t reclen;
3692         nvlist_t *nv;
3693         int i;
3694
3695         while (bytes_read > sizeof (reclen)) {
3696
3697                 /* get length of packed record (stored as little endian) */
3698                 for (i = 0, reclen = 0; i < sizeof (reclen); i++)
3699                         reclen += (uint64_t)(((uchar_t *)buf)[i]) << (8*i);
3700
3701                 if (bytes_read < sizeof (reclen) + reclen)
3702                         break;
3703
3704                 /* unpack record */
3705                 if (nvlist_unpack(buf + sizeof (reclen), reclen, &nv, 0) != 0)
3706                         return (ENOMEM);
3707                 bytes_read -= sizeof (reclen) + reclen;
3708                 buf += sizeof (reclen) + reclen;
3709
3710                 /* add record to nvlist array */
3711                 (*numrecords)++;
3712                 if (ISP2(*numrecords + 1)) {
3713                         *records = realloc(*records,
3714                             *numrecords * 2 * sizeof (nvlist_t *));
3715                 }
3716                 (*records)[*numrecords - 1] = nv;
3717         }
3718
3719         *leftover = bytes_read;
3720         return (0);
3721 }
3722
3723 /* from spa_history.c: spa_history_create_obj() */
3724 #define HIS_BUF_LEN_DEF (128 << 10)
3725 #define HIS_BUF_LEN_MAX (1 << 30)
3726
3727 /*
3728  * Retrieve the command history of a pool.
3729  */
3730 int
3731 zpool_get_history(zpool_handle_t *zhp, nvlist_t **nvhisp)
3732 {
3733         char *buf;
3734         uint64_t buflen = HIS_BUF_LEN_DEF;
3735         uint64_t off = 0;
3736         nvlist_t **records = NULL;
3737         uint_t numrecords = 0;
3738         int err, i;
3739
3740         buf = malloc(buflen);
3741         if (buf == NULL)
3742                 return (ENOMEM);
3743         do {
3744                 uint64_t bytes_read = buflen;
3745                 uint64_t leftover;
3746
3747                 if ((err = get_history(zhp, buf, &off, &bytes_read)) != 0)
3748                         break;
3749
3750                 /* if nothing else was read in, we're at EOF, just return */
3751                 if (bytes_read == 0)
3752                         break;
3753
3754                 if ((err = zpool_history_unpack(buf, bytes_read,
3755                     &leftover, &records, &numrecords)) != 0)
3756                         break;
3757                 off -= leftover;
3758                 if (leftover == bytes_read) {
3759                         /*
3760                          * no progress made, because buffer is not big enough
3761                          * to hold this record; resize and retry.
3762                          */
3763                         buflen *= 2;
3764                         free(buf);
3765                         buf = NULL;
3766                         if ((buflen >= HIS_BUF_LEN_MAX) ||
3767                             ((buf = malloc(buflen)) == NULL)) {
3768                                 err = ENOMEM;
3769                                 break;
3770                         }
3771                 }
3772
3773                 /* CONSTCOND */
3774         } while (1);
3775
3776         free(buf);
3777
3778         if (!err) {
3779                 verify(nvlist_alloc(nvhisp, NV_UNIQUE_NAME, 0) == 0);
3780                 verify(nvlist_add_nvlist_array(*nvhisp, ZPOOL_HIST_RECORD,
3781                     records, numrecords) == 0);
3782         }
3783         for (i = 0; i < numrecords; i++)
3784                 nvlist_free(records[i]);
3785         free(records);
3786
3787         return (err);
3788 }
3789
3790 void
3791 zpool_obj_to_path(zpool_handle_t *zhp, uint64_t dsobj, uint64_t obj,
3792     char *pathname, size_t len)
3793 {
3794         zfs_cmd_t zc = { 0 };
3795         boolean_t mounted = B_FALSE;
3796         char *mntpnt = NULL;
3797         char dsname[MAXNAMELEN];
3798
3799         if (dsobj == 0) {
3800                 /* special case for the MOS */
3801                 (void) snprintf(pathname, len, "<metadata>:<0x%llx>", obj);
3802                 return;
3803         }
3804
3805         /* get the dataset's name */
3806         (void) strlcpy(zc.zc_name, zhp->zpool_name, sizeof (zc.zc_name));
3807         zc.zc_obj = dsobj;
3808         if (ioctl(zhp->zpool_hdl->libzfs_fd,
3809             ZFS_IOC_DSOBJ_TO_DSNAME, &zc) != 0) {
3810                 /* just write out a path of two object numbers */
3811                 (void) snprintf(pathname, len, "<0x%llx>:<0x%llx>",
3812                     dsobj, obj);
3813                 return;
3814         }
3815         (void) strlcpy(dsname, zc.zc_value, sizeof (dsname));
3816
3817         /* find out if the dataset is mounted */
3818         mounted = is_mounted(zhp->zpool_hdl, dsname, &mntpnt);
3819
3820         /* get the corrupted object's path */
3821         (void) strlcpy(zc.zc_name, dsname, sizeof (zc.zc_name));
3822         zc.zc_obj = obj;
3823         if (ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_OBJ_TO_PATH,
3824             &zc) == 0) {
3825                 if (mounted) {
3826                         (void) snprintf(pathname, len, "%s%s", mntpnt,
3827                             zc.zc_value);
3828                 } else {
3829                         (void) snprintf(pathname, len, "%s:%s",
3830                             dsname, zc.zc_value);
3831                 }
3832         } else {
3833                 (void) snprintf(pathname, len, "%s:<0x%llx>", dsname, obj);
3834         }
3835         free(mntpnt);
3836 }
3837
3838 #ifdef sun
3839 /*
3840  * Read the EFI label from the config, if a label does not exist then
3841  * pass back the error to the caller. If the caller has passed a non-NULL
3842  * diskaddr argument then we set it to the starting address of the EFI
3843  * partition.
3844  */
3845 static int
3846 read_efi_label(nvlist_t *config, diskaddr_t *sb)
3847 {
3848         char *path;
3849         int fd;
3850         char diskname[MAXPATHLEN];
3851         int err = -1;
3852
3853         if (nvlist_lookup_string(config, ZPOOL_CONFIG_PATH, &path) != 0)
3854                 return (err);
3855
3856         (void) snprintf(diskname, sizeof (diskname), "%s%s", RDISK_ROOT,
3857             strrchr(path, '/'));
3858         if ((fd = open(diskname, O_RDONLY|O_NDELAY)) >= 0) {
3859                 struct dk_gpt *vtoc;
3860
3861                 if ((err = efi_alloc_and_read(fd, &vtoc)) >= 0) {
3862                         if (sb != NULL)
3863                                 *sb = vtoc->efi_parts[0].p_start;
3864                         efi_free(vtoc);
3865                 }
3866                 (void) close(fd);
3867         }
3868         return (err);
3869 }
3870
3871 /*
3872  * determine where a partition starts on a disk in the current
3873  * configuration
3874  */
3875 static diskaddr_t
3876 find_start_block(nvlist_t *config)
3877 {
3878         nvlist_t **child;
3879         uint_t c, children;
3880         diskaddr_t sb = MAXOFFSET_T;
3881         uint64_t wholedisk;
3882
3883         if (nvlist_lookup_nvlist_array(config,
3884             ZPOOL_CONFIG_CHILDREN, &child, &children) != 0) {
3885                 if (nvlist_lookup_uint64(config,
3886                     ZPOOL_CONFIG_WHOLE_DISK,
3887                     &wholedisk) != 0 || !wholedisk) {
3888                         return (MAXOFFSET_T);
3889                 }
3890                 if (read_efi_label(config, &sb) < 0)
3891                         sb = MAXOFFSET_T;
3892                 return (sb);
3893         }
3894
3895         for (c = 0; c < children; c++) {
3896                 sb = find_start_block(child[c]);
3897                 if (sb != MAXOFFSET_T) {
3898                         return (sb);
3899                 }
3900         }
3901         return (MAXOFFSET_T);
3902 }
3903 #endif /* sun */
3904
3905 /*
3906  * Label an individual disk.  The name provided is the short name,
3907  * stripped of any leading /dev path.
3908  */
3909 int
3910 zpool_label_disk(libzfs_handle_t *hdl, zpool_handle_t *zhp, const char *name)
3911 {
3912 #ifdef sun
3913         char path[MAXPATHLEN];
3914         struct dk_gpt *vtoc;
3915         int fd;
3916         size_t resv = EFI_MIN_RESV_SIZE;
3917         uint64_t slice_size;
3918         diskaddr_t start_block;
3919         char errbuf[1024];
3920
3921         /* prepare an error message just in case */
3922         (void) snprintf(errbuf, sizeof (errbuf),
3923             dgettext(TEXT_DOMAIN, "cannot label '%s'"), name);
3924
3925         if (zhp) {
3926                 nvlist_t *nvroot;
3927
3928                 verify(nvlist_lookup_nvlist(zhp->zpool_config,
3929                     ZPOOL_CONFIG_VDEV_TREE, &nvroot) == 0);
3930
3931                 if (zhp->zpool_start_block == 0)
3932                         start_block = find_start_block(nvroot);
3933                 else
3934                         start_block = zhp->zpool_start_block;
3935                 zhp->zpool_start_block = start_block;
3936         } else {
3937                 /* new pool */
3938                 start_block = NEW_START_BLOCK;
3939         }
3940
3941         (void) snprintf(path, sizeof (path), "%s/%s%s", RDISK_ROOT, name,
3942             BACKUP_SLICE);
3943
3944         if ((fd = open(path, O_RDWR | O_NDELAY)) < 0) {
3945                 /*
3946                  * This shouldn't happen.  We've long since verified that this
3947                  * is a valid device.
3948                  */
3949                 zfs_error_aux(hdl,
3950                     dgettext(TEXT_DOMAIN, "unable to open device"));
3951                 return (zfs_error(hdl, EZFS_OPENFAILED, errbuf));
3952         }
3953
3954         if (efi_alloc_and_init(fd, EFI_NUMPAR, &vtoc) != 0) {
3955                 /*
3956                  * The only way this can fail is if we run out of memory, or we
3957                  * were unable to read the disk's capacity
3958                  */
3959                 if (errno == ENOMEM)
3960                         (void) no_memory(hdl);
3961
3962                 (void) close(fd);
3963                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3964                     "unable to read disk capacity"), name);
3965
3966                 return (zfs_error(hdl, EZFS_NOCAP, errbuf));
3967         }
3968
3969         slice_size = vtoc->efi_last_u_lba + 1;
3970         slice_size -= EFI_MIN_RESV_SIZE;
3971         if (start_block == MAXOFFSET_T)
3972                 start_block = NEW_START_BLOCK;
3973         slice_size -= start_block;
3974
3975         vtoc->efi_parts[0].p_start = start_block;
3976         vtoc->efi_parts[0].p_size = slice_size;
3977
3978         /*
3979          * Why we use V_USR: V_BACKUP confuses users, and is considered
3980          * disposable by some EFI utilities (since EFI doesn't have a backup
3981          * slice).  V_UNASSIGNED is supposed to be used only for zero size
3982          * partitions, and efi_write() will fail if we use it.  V_ROOT, V_BOOT,
3983          * etc. were all pretty specific.  V_USR is as close to reality as we
3984          * can get, in the absence of V_OTHER.
3985          */
3986         vtoc->efi_parts[0].p_tag = V_USR;
3987         (void) strcpy(vtoc->efi_parts[0].p_name, "zfs");
3988
3989         vtoc->efi_parts[8].p_start = slice_size + start_block;
3990         vtoc->efi_parts[8].p_size = resv;
3991         vtoc->efi_parts[8].p_tag = V_RESERVED;
3992
3993         if (efi_write(fd, vtoc) != 0) {
3994                 /*
3995                  * Some block drivers (like pcata) may not support EFI
3996                  * GPT labels.  Print out a helpful error message dir-
3997                  * ecting the user to manually label the disk and give
3998                  * a specific slice.
3999                  */
4000                 (void) close(fd);
4001                 efi_free(vtoc);
4002
4003                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4004                     "try using fdisk(1M) and then provide a specific slice"));
4005                 return (zfs_error(hdl, EZFS_LABELFAILED, errbuf));
4006         }
4007
4008         (void) close(fd);
4009         efi_free(vtoc);
4010 #endif /* sun */
4011         return (0);
4012 }
4013
4014 static boolean_t
4015 supported_dump_vdev_type(libzfs_handle_t *hdl, nvlist_t *config, char *errbuf)
4016 {
4017         char *type;
4018         nvlist_t **child;
4019         uint_t children, c;
4020
4021         verify(nvlist_lookup_string(config, ZPOOL_CONFIG_TYPE, &type) == 0);
4022         if (strcmp(type, VDEV_TYPE_FILE) == 0 ||
4023             strcmp(type, VDEV_TYPE_HOLE) == 0 ||
4024             strcmp(type, VDEV_TYPE_MISSING) == 0) {
4025                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4026                     "vdev type '%s' is not supported"), type);
4027                 (void) zfs_error(hdl, EZFS_VDEVNOTSUP, errbuf);
4028                 return (B_FALSE);
4029         }
4030         if (nvlist_lookup_nvlist_array(config, ZPOOL_CONFIG_CHILDREN,
4031             &child, &children) == 0) {
4032                 for (c = 0; c < children; c++) {
4033                         if (!supported_dump_vdev_type(hdl, child[c], errbuf))
4034                                 return (B_FALSE);
4035                 }
4036         }
4037         return (B_TRUE);
4038 }
4039
4040 /*
4041  * Check if this zvol is allowable for use as a dump device; zero if
4042  * it is, > 0 if it isn't, < 0 if it isn't a zvol.
4043  *
4044  * Allowable storage configurations include mirrors, all raidz variants, and
4045  * pools with log, cache, and spare devices.  Pools which are backed by files or
4046  * have missing/hole vdevs are not suitable.
4047  */
4048 int
4049 zvol_check_dump_config(char *arg)
4050 {
4051         zpool_handle_t *zhp = NULL;
4052         nvlist_t *config, *nvroot;
4053         char *p, *volname;
4054         nvlist_t **top;
4055         uint_t toplevels;
4056         libzfs_handle_t *hdl;
4057         char errbuf[1024];
4058         char poolname[ZPOOL_MAXNAMELEN];
4059         int pathlen = strlen(ZVOL_FULL_DEV_DIR);
4060         int ret = 1;
4061
4062         if (strncmp(arg, ZVOL_FULL_DEV_DIR, pathlen)) {
4063                 return (-1);
4064         }
4065
4066         (void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
4067             "dump is not supported on device '%s'"), arg);
4068
4069         if ((hdl = libzfs_init()) == NULL)
4070                 return (1);
4071         libzfs_print_on_error(hdl, B_TRUE);
4072
4073         volname = arg + pathlen;
4074
4075         /* check the configuration of the pool */
4076         if ((p = strchr(volname, '/')) == NULL) {
4077                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4078                     "malformed dataset name"));
4079                 (void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
4080                 return (1);
4081         } else if (p - volname >= ZFS_MAXNAMELEN) {
4082                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4083                     "dataset name is too long"));
4084                 (void) zfs_error(hdl, EZFS_NAMETOOLONG, errbuf);
4085                 return (1);
4086         } else {
4087                 (void) strncpy(poolname, volname, p - volname);
4088                 poolname[p - volname] = '\0';
4089         }
4090
4091         if ((zhp = zpool_open(hdl, poolname)) == NULL) {
4092                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4093                     "could not open pool '%s'"), poolname);
4094                 (void) zfs_error(hdl, EZFS_OPENFAILED, errbuf);
4095                 goto out;
4096         }
4097         config = zpool_get_config(zhp, NULL);
4098         if (nvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE,
4099             &nvroot) != 0) {
4100                 zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
4101                     "could not obtain vdev configuration for  '%s'"), poolname);
4102                 (void) zfs_error(hdl, EZFS_INVALCONFIG, errbuf);
4103                 goto out;
4104         }
4105
4106         verify(nvlist_lookup_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
4107             &top, &toplevels) == 0);
4108
4109         if (!supported_dump_vdev_type(hdl, top[0], errbuf)) {
4110                 goto out;
4111         }
4112         ret = 0;
4113
4114 out:
4115         if (zhp)
4116                 zpool_close(zhp);
4117         libzfs_fini(hdl);
4118         return (ret);
4119 }