1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 2013 by Delphix. All rights reserved.
  25  */
  26 
  27 /*
  28  * Virtual Device Labels
  29  * ---------------------
  30  *
  31  * The vdev label serves several distinct purposes:
  32  *
  33  *      1. Uniquely identify this device as part of a ZFS pool and confirm its
  34  *         identity within the pool.
  35  *
  36  *      2. Verify that all the devices given in a configuration are present
  37  *         within the pool.
  38  *
  39  *      3. Determine the uberblock for the pool.
  40  *
  41  *      4. In case of an import operation, determine the configuration of the
  42  *         toplevel vdev of which it is a part.
  43  *
  44  *      5. If an import operation cannot find all the devices in the pool,
  45  *         provide enough information to the administrator to determine which
  46  *         devices are missing.
  47  *
  48  * It is important to note that while the kernel is responsible for writing the
  49  * label, it only consumes the information in the first three cases.  The
  50  * latter information is only consumed in userland when determining the
  51  * configuration to import a pool.
  52  *
  53  *
  54  * Label Organization
  55  * ------------------
  56  *
  57  * Before describing the contents of the label, it's important to understand how
  58  * the labels are written and updated with respect to the uberblock.
  59  *
  60  * When the pool configuration is altered, either because it was newly created
  61  * or a device was added, we want to update all the labels such that we can deal
  62  * with fatal failure at any point.  To this end, each disk has two labels which
  63  * are updated before and after the uberblock is synced.  Assuming we have
  64  * labels and an uberblock with the following transaction groups:
  65  *
  66  *              L1          UB          L2
  67  *           +------+    +------+    +------+
  68  *           |      |    |      |    |      |
  69  *           | t10  |    | t10  |    | t10  |
  70  *           |      |    |      |    |      |
  71  *           +------+    +------+    +------+
  72  *
  73  * In this stable state, the labels and the uberblock were all updated within
  74  * the same transaction group (10).  Each label is mirrored and checksummed, so
  75  * that we can detect when we fail partway through writing the label.
  76  *
  77  * In order to identify which labels are valid, the labels are written in the
  78  * following manner:
  79  *
  80  *      1. For each vdev, update 'L1' to the new label
  81  *      2. Update the uberblock
  82  *      3. For each vdev, update 'L2' to the new label
  83  *
  84  * Given arbitrary failure, we can determine the correct label to use based on
  85  * the transaction group.  If we fail after updating L1 but before updating the
  86  * UB, we will notice that L1's transaction group is greater than the uberblock,
  87  * so L2 must be valid.  If we fail after writing the uberblock but before
  88  * writing L2, we will notice that L2's transaction group is less than L1, and
  89  * therefore L1 is valid.
  90  *
  91  * Another added complexity is that not every label is updated when the config
  92  * is synced.  If we add a single device, we do not want to have to re-write
  93  * every label for every device in the pool.  This means that both L1 and L2 may
  94  * be older than the pool uberblock, because the necessary information is stored
  95  * on another vdev.
  96  *
  97  *
  98  * On-disk Format
  99  * --------------
 100  *
 101  * The vdev label consists of two distinct parts, and is wrapped within the
 102  * vdev_label_t structure.  The label includes 8k of padding to permit legacy
 103  * VTOC disk labels, but is otherwise ignored.
 104  *
 105  * The first half of the label is a packed nvlist which contains pool wide
 106  * properties, per-vdev properties, and configuration information.  It is
 107  * described in more detail below.
 108  *
 109  * The latter half of the label consists of a redundant array of uberblocks.
 110  * These uberblocks are updated whenever a transaction group is committed,
 111  * or when the configuration is updated.  When a pool is loaded, we scan each
 112  * vdev for the 'best' uberblock.
 113  *
 114  *
 115  * Configuration Information
 116  * -------------------------
 117  *
 118  * The nvlist describing the pool and vdev contains the following elements:
 119  *
 120  *      version         ZFS on-disk version
 121  *      name            Pool name
 122  *      state           Pool state
 123  *      txg             Transaction group in which this label was written
 124  *      pool_guid       Unique identifier for this pool
 125  *      vdev_tree       An nvlist describing vdev tree.
 126  *      features_for_read
 127  *                      An nvlist of the features necessary for reading the MOS.
 128  *
 129  * Each leaf device label also contains the following:
 130  *
 131  *      top_guid        Unique ID for top-level vdev in which this is contained
 132  *      guid            Unique ID for the leaf vdev
 133  *
 134  * The 'vs' configuration follows the format described in 'spa_config.c'.
 135  */
 136 
 137 #include <sys/zfs_context.h>
 138 #include <sys/spa.h>
 139 #include <sys/spa_impl.h>
 140 #include <sys/dmu.h>
 141 #include <sys/zap.h>
 142 #include <sys/vdev.h>
 143 #include <sys/vdev_impl.h>
 144 #include <sys/uberblock_impl.h>
 145 #include <sys/metaslab.h>
 146 #include <sys/zio.h>
 147 #include <sys/dsl_scan.h>
 148 #include <sys/fs/zfs.h>
 149 
 150 /*
 151  * Basic routines to read and write from a vdev label.
 152  * Used throughout the rest of this file.
 153  */
 154 uint64_t
 155 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
 156 {
 157         ASSERT(offset < sizeof (vdev_label_t));
 158         ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
 159 
 160         return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
 161             0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
 162 }
 163 
 164 /*
 165  * Returns back the vdev label associated with the passed in offset.
 166  */
 167 int
 168 vdev_label_number(uint64_t psize, uint64_t offset)
 169 {
 170         int l;
 171 
 172         if (offset >= psize - VDEV_LABEL_END_SIZE) {
 173                 offset -= psize - VDEV_LABEL_END_SIZE;
 174                 offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
 175         }
 176         l = offset / sizeof (vdev_label_t);
 177         return (l < VDEV_LABELS ? l : -1);
 178 }
 179 
 180 static void
 181 vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
 182         uint64_t size, zio_done_func_t *done, void *private, int flags)
 183 {
 184         ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
 185             SCL_STATE_ALL);
 186         ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
 187 
 188         zio_nowait(zio_read_phys(zio, vd,
 189             vdev_label_offset(vd->vdev_psize, l, offset),
 190             size, buf, ZIO_CHECKSUM_LABEL, done, private,
 191             ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
 192 }
 193 
 194 static void
 195 vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
 196         uint64_t size, zio_done_func_t *done, void *private, int flags)
 197 {
 198         ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL ||
 199             (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
 200             (SCL_CONFIG | SCL_STATE) &&
 201             dsl_pool_sync_context(spa_get_dsl(zio->io_spa))));
 202         ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
 203 
 204         zio_nowait(zio_write_phys(zio, vd,
 205             vdev_label_offset(vd->vdev_psize, l, offset),
 206             size, buf, ZIO_CHECKSUM_LABEL, done, private,
 207             ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
 208 }
 209 
 210 /*
 211  * Generate the nvlist representing this vdev's config.
 212  */
 213 nvlist_t *
 214 vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
 215     vdev_config_flag_t flags)
 216 {
 217         nvlist_t *nv = NULL;
 218 
 219         VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 220 
 221         VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
 222             vd->vdev_ops->vdev_op_type) == 0);
 223         if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
 224                 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id)
 225                     == 0);
 226         VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0);
 227 
 228         if (vd->vdev_path != NULL)
 229                 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH,
 230                     vd->vdev_path) == 0);
 231 
 232         if (vd->vdev_devid != NULL)
 233                 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID,
 234                     vd->vdev_devid) == 0);
 235 
 236         if (vd->vdev_physpath != NULL)
 237                 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
 238                     vd->vdev_physpath) == 0);
 239 
 240         if (vd->vdev_fru != NULL)
 241                 VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_FRU,
 242                     vd->vdev_fru) == 0);
 243 
 244         if (vd->vdev_nparity != 0) {
 245                 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
 246                     VDEV_TYPE_RAIDZ) == 0);
 247 
 248                 /*
 249                  * Make sure someone hasn't managed to sneak a fancy new vdev
 250                  * into a crufty old storage pool.
 251                  */
 252                 ASSERT(vd->vdev_nparity == 1 ||
 253                     (vd->vdev_nparity <= 2 &&
 254                     spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
 255                     (vd->vdev_nparity <= 3 &&
 256                     spa_version(spa) >= SPA_VERSION_RAIDZ3));
 257 
 258                 /*
 259                  * Note that we'll add the nparity tag even on storage pools
 260                  * that only support a single parity device -- older software
 261                  * will just ignore it.
 262                  */
 263                 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY,
 264                     vd->vdev_nparity) == 0);
 265         }
 266 
 267         if (vd->vdev_wholedisk != -1ULL)
 268                 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
 269                     vd->vdev_wholedisk) == 0);
 270 
 271         if (vd->vdev_not_present)
 272                 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0);
 273 
 274         if (vd->vdev_isspare)
 275                 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0);
 276 
 277         if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
 278             vd == vd->vdev_top) {
 279                 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
 280                     vd->vdev_ms_array) == 0);
 281                 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
 282                     vd->vdev_ms_shift) == 0);
 283                 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT,
 284                     vd->vdev_ashift) == 0);
 285                 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
 286                     vd->vdev_asize) == 0);
 287                 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG,
 288                     vd->vdev_islog) == 0);
 289                 if (vd->vdev_removing)
 290                         VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
 291                             vd->vdev_removing) == 0);
 292         }
 293 
 294         if (flags & VDEV_CONFIG_L2CACHE)
 295                 /* indicate that we support L2ARC persistency */
 296                 VERIFY(nvlist_add_boolean_value(nv,
 297                     ZPOOL_CONFIG_L2CACHE_PERSISTENT, B_TRUE) == 0);
 298 
 299         if (vd->vdev_dtl_smo.smo_object != 0)
 300                 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
 301                     vd->vdev_dtl_smo.smo_object) == 0);
 302 
 303         if (vd->vdev_crtxg)
 304                 VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG,
 305                     vd->vdev_crtxg) == 0);
 306 
 307         if (getstats) {
 308                 vdev_stat_t vs;
 309                 pool_scan_stat_t ps;
 310 
 311                 vdev_get_stats(vd, &vs);
 312                 VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
 313                     (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0);
 314 
 315                 /* provide either current or previous scan information */
 316                 if (spa_scan_get_stats(spa, &ps) == 0) {
 317                         VERIFY(nvlist_add_uint64_array(nv,
 318                             ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
 319                             sizeof (pool_scan_stat_t) / sizeof (uint64_t))
 320                             == 0);
 321                 }
 322         }
 323 
 324         if (!vd->vdev_ops->vdev_op_leaf) {
 325                 nvlist_t **child;
 326                 int c, idx;
 327 
 328                 ASSERT(!vd->vdev_ishole);
 329 
 330                 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
 331                     KM_SLEEP);
 332 
 333                 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
 334                         vdev_t *cvd = vd->vdev_child[c];
 335 
 336                         /*
 337                          * If we're generating an nvlist of removing
 338                          * vdevs then skip over any device which is
 339                          * not being removed.
 340                          */
 341                         if ((flags & VDEV_CONFIG_REMOVING) &&
 342                             !cvd->vdev_removing)
 343                                 continue;
 344 
 345                         child[idx++] = vdev_config_generate(spa, cvd,
 346                             getstats, flags);
 347                 }
 348 
 349                 if (idx) {
 350                         VERIFY(nvlist_add_nvlist_array(nv,
 351                             ZPOOL_CONFIG_CHILDREN, child, idx) == 0);
 352                 }
 353 
 354                 for (c = 0; c < idx; c++)
 355                         nvlist_free(child[c]);
 356 
 357                 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
 358 
 359         } else {
 360                 const char *aux = NULL;
 361 
 362                 if (vd->vdev_offline && !vd->vdev_tmpoffline)
 363                         VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE,
 364                             B_TRUE) == 0);
 365                 if (vd->vdev_resilvering)
 366                         VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVERING,
 367                             B_TRUE) == 0);
 368                 if (vd->vdev_faulted)
 369                         VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED,
 370                             B_TRUE) == 0);
 371                 if (vd->vdev_degraded)
 372                         VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED,
 373                             B_TRUE) == 0);
 374                 if (vd->vdev_removed)
 375                         VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED,
 376                             B_TRUE) == 0);
 377                 if (vd->vdev_unspare)
 378                         VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE,
 379                             B_TRUE) == 0);
 380                 if (vd->vdev_ishole)
 381                         VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE,
 382                             B_TRUE) == 0);
 383 
 384                 switch (vd->vdev_stat.vs_aux) {
 385                 case VDEV_AUX_ERR_EXCEEDED:
 386                         aux = "err_exceeded";
 387                         break;
 388 
 389                 case VDEV_AUX_EXTERNAL:
 390                         aux = "external";
 391                         break;
 392                 }
 393 
 394                 if (aux != NULL)
 395                         VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE,
 396                             aux) == 0);
 397 
 398                 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
 399                         VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
 400                             vd->vdev_orig_guid) == 0);
 401                 }
 402         }
 403 
 404         return (nv);
 405 }
 406 
 407 /*
 408  * Generate a view of the top-level vdevs.  If we currently have holes
 409  * in the namespace, then generate an array which contains a list of holey
 410  * vdevs.  Additionally, add the number of top-level children that currently
 411  * exist.
 412  */
 413 void
 414 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
 415 {
 416         vdev_t *rvd = spa->spa_root_vdev;
 417         uint64_t *array;
 418         uint_t c, idx;
 419 
 420         array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
 421 
 422         for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
 423                 vdev_t *tvd = rvd->vdev_child[c];
 424 
 425                 if (tvd->vdev_ishole)
 426                         array[idx++] = c;
 427         }
 428 
 429         if (idx) {
 430                 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
 431                     array, idx) == 0);
 432         }
 433 
 434         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
 435             rvd->vdev_children) == 0);
 436 
 437         kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
 438 }
 439 
 440 /*
 441  * Returns the configuration from the label of the given vdev. For vdevs
 442  * which don't have a txg value stored on their label (i.e. spares/cache)
 443  * or have not been completely initialized (txg = 0) just return
 444  * the configuration from the first valid label we find. Otherwise,
 445  * find the most up-to-date label that does not exceed the specified
 446  * 'txg' value.
 447  */
 448 nvlist_t *
 449 vdev_label_read_config(vdev_t *vd, uint64_t txg)
 450 {
 451         spa_t *spa = vd->vdev_spa;
 452         nvlist_t *config = NULL;
 453         vdev_phys_t *vp;
 454         zio_t *zio;
 455         uint64_t best_txg = 0;
 456         int error = 0;
 457         int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
 458             ZIO_FLAG_SPECULATIVE;
 459 
 460         ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
 461 
 462         if (!vdev_readable(vd))
 463                 return (NULL);
 464 
 465         vp = zio_buf_alloc(sizeof (vdev_phys_t));
 466 
 467 retry:
 468         for (int l = 0; l < VDEV_LABELS; l++) {
 469                 nvlist_t *label = NULL;
 470 
 471                 zio = zio_root(spa, NULL, NULL, flags);
 472 
 473                 vdev_label_read(zio, vd, l, vp,
 474                     offsetof(vdev_label_t, vl_vdev_phys),
 475                     sizeof (vdev_phys_t), NULL, NULL, flags);
 476 
 477                 if (zio_wait(zio) == 0 &&
 478                     nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
 479                     &label, 0) == 0) {
 480                         uint64_t label_txg = 0;
 481 
 482                         /*
 483                          * Auxiliary vdevs won't have txg values in their
 484                          * labels and newly added vdevs may not have been
 485                          * completely initialized so just return the
 486                          * configuration from the first valid label we
 487                          * encounter.
 488                          */
 489                         error = nvlist_lookup_uint64(label,
 490                             ZPOOL_CONFIG_POOL_TXG, &label_txg);
 491                         if ((error || label_txg == 0) && !config) {
 492                                 config = label;
 493                                 break;
 494                         } else if (label_txg <= txg && label_txg > best_txg) {
 495                                 best_txg = label_txg;
 496                                 nvlist_free(config);
 497                                 config = fnvlist_dup(label);
 498                         }
 499                 }
 500 
 501                 if (label != NULL) {
 502                         nvlist_free(label);
 503                         label = NULL;
 504                 }
 505         }
 506 
 507         if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
 508                 flags |= ZIO_FLAG_TRYHARD;
 509                 goto retry;
 510         }
 511 
 512         zio_buf_free(vp, sizeof (vdev_phys_t));
 513 
 514         return (config);
 515 }
 516 
 517 /*
 518  * Determine if a device is in use.  The 'spare_guid' parameter will be filled
 519  * in with the device guid if this spare is active elsewhere on the system.
 520  */
 521 static boolean_t
 522 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
 523     uint64_t *spare_guid, uint64_t *l2cache_guid)
 524 {
 525         spa_t *spa = vd->vdev_spa;
 526         uint64_t state, pool_guid, device_guid, txg, spare_pool;
 527         uint64_t vdtxg = 0;
 528         nvlist_t *label;
 529 
 530         if (spare_guid)
 531                 *spare_guid = 0ULL;
 532         if (l2cache_guid)
 533                 *l2cache_guid = 0ULL;
 534 
 535         /*
 536          * Read the label, if any, and perform some basic sanity checks.
 537          */
 538         if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
 539                 return (B_FALSE);
 540 
 541         (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
 542             &vdtxg);
 543 
 544         if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
 545             &state) != 0 ||
 546             nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
 547             &device_guid) != 0) {
 548                 nvlist_free(label);
 549                 return (B_FALSE);
 550         }
 551 
 552         if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 553             (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
 554             &pool_guid) != 0 ||
 555             nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
 556             &txg) != 0)) {
 557                 nvlist_free(label);
 558                 return (B_FALSE);
 559         }
 560 
 561         nvlist_free(label);
 562 
 563         /*
 564          * Check to see if this device indeed belongs to the pool it claims to
 565          * be a part of.  The only way this is allowed is if the device is a hot
 566          * spare (which we check for later on).
 567          */
 568         if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 569             !spa_guid_exists(pool_guid, device_guid) &&
 570             !spa_spare_exists(device_guid, NULL, NULL) &&
 571             !spa_l2cache_exists(device_guid, NULL))
 572                 return (B_FALSE);
 573 
 574         /*
 575          * If the transaction group is zero, then this an initialized (but
 576          * unused) label.  This is only an error if the create transaction
 577          * on-disk is the same as the one we're using now, in which case the
 578          * user has attempted to add the same vdev multiple times in the same
 579          * transaction.
 580          */
 581         if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 582             txg == 0 && vdtxg == crtxg)
 583                 return (B_TRUE);
 584 
 585         /*
 586          * Check to see if this is a spare device.  We do an explicit check for
 587          * spa_has_spare() here because it may be on our pending list of spares
 588          * to add.  We also check if it is an l2cache device.
 589          */
 590         if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
 591             spa_has_spare(spa, device_guid)) {
 592                 if (spare_guid)
 593                         *spare_guid = device_guid;
 594 
 595                 switch (reason) {
 596                 case VDEV_LABEL_CREATE:
 597                 case VDEV_LABEL_L2CACHE:
 598                         return (B_TRUE);
 599 
 600                 case VDEV_LABEL_REPLACE:
 601                         return (!spa_has_spare(spa, device_guid) ||
 602                             spare_pool != 0ULL);
 603 
 604                 case VDEV_LABEL_SPARE:
 605                         return (spa_has_spare(spa, device_guid));
 606                 }
 607         }
 608 
 609         /*
 610          * Check to see if this is an l2cache device.
 611          */
 612         if (spa_l2cache_exists(device_guid, NULL))
 613                 return (B_TRUE);
 614 
 615         /*
 616          * We can't rely on a pool's state if it's been imported
 617          * read-only.  Instead we look to see if the pools is marked
 618          * read-only in the namespace and set the state to active.
 619          */
 620         if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
 621             spa_mode(spa) == FREAD)
 622                 state = POOL_STATE_ACTIVE;
 623 
 624         /*
 625          * If the device is marked ACTIVE, then this device is in use by another
 626          * pool on the system.
 627          */
 628         return (state == POOL_STATE_ACTIVE);
 629 }
 630 
 631 /*
 632  * Initialize a vdev label.  We check to make sure each leaf device is not in
 633  * use, and writable.  We put down an initial label which we will later
 634  * overwrite with a complete label.  Note that it's important to do this
 635  * sequentially, not in parallel, so that we catch cases of multiple use of the
 636  * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
 637  * itself.
 638  */
 639 int
 640 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
 641 {
 642         spa_t *spa = vd->vdev_spa;
 643         nvlist_t *label;
 644         vdev_phys_t *vp;
 645         char *pad2;
 646         uberblock_t *ub;
 647         zio_t *zio;
 648         char *buf;
 649         size_t buflen;
 650         int error;
 651         uint64_t spare_guid, l2cache_guid;
 652         int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
 653 
 654         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
 655 
 656         for (int c = 0; c < vd->vdev_children; c++)
 657                 if ((error = vdev_label_init(vd->vdev_child[c],
 658                     crtxg, reason)) != 0)
 659                         return (error);
 660 
 661         /* Track the creation time for this vdev */
 662         vd->vdev_crtxg = crtxg;
 663 
 664         if (!vd->vdev_ops->vdev_op_leaf)
 665                 return (0);
 666 
 667         /*
 668          * Dead vdevs cannot be initialized.
 669          */
 670         if (vdev_is_dead(vd))
 671                 return (SET_ERROR(EIO));
 672 
 673         /*
 674          * Determine if the vdev is in use.
 675          */
 676         if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
 677             vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
 678                 return (SET_ERROR(EBUSY));
 679 
 680         /*
 681          * If this is a request to add or replace a spare or l2cache device
 682          * that is in use elsewhere on the system, then we must update the
 683          * guid (which was initialized to a random value) to reflect the
 684          * actual GUID (which is shared between multiple pools).
 685          */
 686         if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
 687             spare_guid != 0ULL) {
 688                 uint64_t guid_delta = spare_guid - vd->vdev_guid;
 689 
 690                 vd->vdev_guid += guid_delta;
 691 
 692                 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
 693                         pvd->vdev_guid_sum += guid_delta;
 694 
 695                 /*
 696                  * If this is a replacement, then we want to fallthrough to the
 697                  * rest of the code.  If we're adding a spare, then it's already
 698                  * labeled appropriately and we can just return.
 699                  */
 700                 if (reason == VDEV_LABEL_SPARE)
 701                         return (0);
 702                 ASSERT(reason == VDEV_LABEL_REPLACE ||
 703                     reason == VDEV_LABEL_SPLIT);
 704         }
 705 
 706         if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
 707             l2cache_guid != 0ULL) {
 708                 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
 709 
 710                 vd->vdev_guid += guid_delta;
 711 
 712                 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
 713                         pvd->vdev_guid_sum += guid_delta;
 714 
 715                 /*
 716                  * If this is a replacement, then we want to fallthrough to the
 717                  * rest of the code.  If we're adding an l2cache, then it's
 718                  * already labeled appropriately and we can just return.
 719                  */
 720                 if (reason == VDEV_LABEL_L2CACHE)
 721                         return (0);
 722                 ASSERT(reason == VDEV_LABEL_REPLACE);
 723         }
 724 
 725         /*
 726          * Initialize its label.
 727          */
 728         vp = zio_buf_alloc(sizeof (vdev_phys_t));
 729         bzero(vp, sizeof (vdev_phys_t));
 730 
 731         /*
 732          * Generate a label describing the pool and our top-level vdev.
 733          * We mark it as being from txg 0 to indicate that it's not
 734          * really part of an active pool just yet.  The labels will
 735          * be written again with a meaningful txg by spa_sync().
 736          */
 737         if (reason == VDEV_LABEL_SPARE ||
 738             (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
 739                 /*
 740                  * For inactive hot spares, we generate a special label that
 741                  * identifies as a mutually shared hot spare.  We write the
 742                  * label if we are adding a hot spare, or if we are removing an
 743                  * active hot spare (in which case we want to revert the
 744                  * labels).
 745                  */
 746                 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 747 
 748                 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
 749                     spa_version(spa)) == 0);
 750                 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
 751                     POOL_STATE_SPARE) == 0);
 752                 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
 753                     vd->vdev_guid) == 0);
 754         } else if (reason == VDEV_LABEL_L2CACHE ||
 755             (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
 756                 /*
 757                  * For level 2 ARC devices, add a special label.
 758                  */
 759                 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 760 
 761                 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
 762                     spa_version(spa)) == 0);
 763                 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
 764                     POOL_STATE_L2CACHE) == 0);
 765                 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
 766                     vd->vdev_guid) == 0);
 767         } else {
 768                 uint64_t txg = 0ULL;
 769 
 770                 if (reason == VDEV_LABEL_SPLIT)
 771                         txg = spa->spa_uberblock.ub_txg;
 772                 label = spa_config_generate(spa, vd, txg, B_FALSE);
 773 
 774                 /*
 775                  * Add our creation time.  This allows us to detect multiple
 776                  * vdev uses as described above, and automatically expires if we
 777                  * fail.
 778                  */
 779                 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
 780                     crtxg) == 0);
 781         }
 782 
 783         buf = vp->vp_nvlist;
 784         buflen = sizeof (vp->vp_nvlist);
 785 
 786         error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
 787         if (error != 0) {
 788                 nvlist_free(label);
 789                 zio_buf_free(vp, sizeof (vdev_phys_t));
 790                 /* EFAULT means nvlist_pack ran out of room */
 791                 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
 792         }
 793 
 794         /*
 795          * Initialize uberblock template.
 796          */
 797         ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
 798         bzero(ub, VDEV_UBERBLOCK_RING);
 799         *ub = spa->spa_uberblock;
 800         ub->ub_txg = 0;
 801 
 802         /* Initialize the 2nd padding area. */
 803         pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
 804         bzero(pad2, VDEV_PAD_SIZE);
 805 
 806         /*
 807          * Write everything in parallel.
 808          */
 809 retry:
 810         zio = zio_root(spa, NULL, NULL, flags);
 811 
 812         for (int l = 0; l < VDEV_LABELS; l++) {
 813 
 814                 vdev_label_write(zio, vd, l, vp,
 815                     offsetof(vdev_label_t, vl_vdev_phys),
 816                     sizeof (vdev_phys_t), NULL, NULL, flags);
 817 
 818                 /*
 819                  * Skip the 1st padding area.
 820                  * Zero out the 2nd padding area where it might have
 821                  * left over data from previous filesystem format.
 822                  */
 823                 vdev_label_write(zio, vd, l, pad2,
 824                     offsetof(vdev_label_t, vl_pad2),
 825                     VDEV_PAD_SIZE, NULL, NULL, flags);
 826 
 827                 vdev_label_write(zio, vd, l, ub,
 828                     offsetof(vdev_label_t, vl_uberblock),
 829                     VDEV_UBERBLOCK_RING, NULL, NULL, flags);
 830         }
 831 
 832         error = zio_wait(zio);
 833 
 834         if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
 835                 flags |= ZIO_FLAG_TRYHARD;
 836                 goto retry;
 837         }
 838 
 839         nvlist_free(label);
 840         zio_buf_free(pad2, VDEV_PAD_SIZE);
 841         zio_buf_free(ub, VDEV_UBERBLOCK_RING);
 842         zio_buf_free(vp, sizeof (vdev_phys_t));
 843 
 844         /*
 845          * If this vdev hasn't been previously identified as a spare, then we
 846          * mark it as such only if a) we are labeling it as a spare, or b) it
 847          * exists as a spare elsewhere in the system.  Do the same for
 848          * level 2 ARC devices.
 849          */
 850         if (error == 0 && !vd->vdev_isspare &&
 851             (reason == VDEV_LABEL_SPARE ||
 852             spa_spare_exists(vd->vdev_guid, NULL, NULL)))
 853                 spa_spare_add(vd);
 854 
 855         if (error == 0 && !vd->vdev_isl2cache &&
 856             (reason == VDEV_LABEL_L2CACHE ||
 857             spa_l2cache_exists(vd->vdev_guid, NULL)))
 858                 spa_l2cache_add(vd);
 859 
 860         return (error);
 861 }
 862 
 863 /*
 864  * ==========================================================================
 865  * uberblock load/sync
 866  * ==========================================================================
 867  */
 868 
 869 /*
 870  * Consider the following situation: txg is safely synced to disk.  We've
 871  * written the first uberblock for txg + 1, and then we lose power.  When we
 872  * come back up, we fail to see the uberblock for txg + 1 because, say,
 873  * it was on a mirrored device and the replica to which we wrote txg + 1
 874  * is now offline.  If we then make some changes and sync txg + 1, and then
 875  * the missing replica comes back, then for a few seconds we'll have two
 876  * conflicting uberblocks on disk with the same txg.  The solution is simple:
 877  * among uberblocks with equal txg, choose the one with the latest timestamp.
 878  */
 879 static int
 880 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
 881 {
 882         if (ub1->ub_txg < ub2->ub_txg)
 883                 return (-1);
 884         if (ub1->ub_txg > ub2->ub_txg)
 885                 return (1);
 886 
 887         if (ub1->ub_timestamp < ub2->ub_timestamp)
 888                 return (-1);
 889         if (ub1->ub_timestamp > ub2->ub_timestamp)
 890                 return (1);
 891 
 892         return (0);
 893 }
 894 
 895 struct ubl_cbdata {
 896         uberblock_t     *ubl_ubbest;    /* Best uberblock */
 897         vdev_t          *ubl_vd;        /* vdev associated with the above */
 898 };
 899 
 900 static void
 901 vdev_uberblock_load_done(zio_t *zio)
 902 {
 903         vdev_t *vd = zio->io_vd;
 904         spa_t *spa = zio->io_spa;
 905         zio_t *rio = zio->io_private;
 906         uberblock_t *ub = zio->io_data;
 907         struct ubl_cbdata *cbp = rio->io_private;
 908 
 909         ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
 910 
 911         if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
 912                 mutex_enter(&rio->io_lock);
 913                 if (ub->ub_txg <= spa->spa_load_max_txg &&
 914                     vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
 915                         /*
 916                          * Keep track of the vdev in which this uberblock
 917                          * was found. We will use this information later
 918                          * to obtain the config nvlist associated with
 919                          * this uberblock.
 920                          */
 921                         *cbp->ubl_ubbest = *ub;
 922                         cbp->ubl_vd = vd;
 923                 }
 924                 mutex_exit(&rio->io_lock);
 925         }
 926 
 927         zio_buf_free(zio->io_data, zio->io_size);
 928 }
 929 
 930 static void
 931 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
 932     struct ubl_cbdata *cbp)
 933 {
 934         for (int c = 0; c < vd->vdev_children; c++)
 935                 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
 936 
 937         if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
 938                 for (int l = 0; l < VDEV_LABELS; l++) {
 939                         for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
 940                                 vdev_label_read(zio, vd, l,
 941                                     zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
 942                                     VDEV_UBERBLOCK_OFFSET(vd, n),
 943                                     VDEV_UBERBLOCK_SIZE(vd),
 944                                     vdev_uberblock_load_done, zio, flags);
 945                         }
 946                 }
 947         }
 948 }
 949 
 950 /*
 951  * Reads the 'best' uberblock from disk along with its associated
 952  * configuration. First, we read the uberblock array of each label of each
 953  * vdev, keeping track of the uberblock with the highest txg in each array.
 954  * Then, we read the configuration from the same vdev as the best uberblock.
 955  */
 956 void
 957 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
 958 {
 959         zio_t *zio;
 960         spa_t *spa = rvd->vdev_spa;
 961         struct ubl_cbdata cb;
 962         int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
 963             ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
 964 
 965         ASSERT(ub);
 966         ASSERT(config);
 967 
 968         bzero(ub, sizeof (uberblock_t));
 969         *config = NULL;
 970 
 971         cb.ubl_ubbest = ub;
 972         cb.ubl_vd = NULL;
 973 
 974         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
 975         zio = zio_root(spa, NULL, &cb, flags);
 976         vdev_uberblock_load_impl(zio, rvd, flags, &cb);
 977         (void) zio_wait(zio);
 978 
 979         /*
 980          * It's possible that the best uberblock was discovered on a label
 981          * that has a configuration which was written in a future txg.
 982          * Search all labels on this vdev to find the configuration that
 983          * matches the txg for our uberblock.
 984          */
 985         if (cb.ubl_vd != NULL)
 986                 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
 987         spa_config_exit(spa, SCL_ALL, FTAG);
 988 }
 989 
 990 /*
 991  * On success, increment root zio's count of good writes.
 992  * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
 993  */
 994 static void
 995 vdev_uberblock_sync_done(zio_t *zio)
 996 {
 997         uint64_t *good_writes = zio->io_private;
 998 
 999         if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
1000                 atomic_add_64(good_writes, 1);
1001 }
1002 
1003 /*
1004  * Write the uberblock to all labels of all leaves of the specified vdev.
1005  */
1006 static void
1007 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
1008 {
1009         uberblock_t *ubbuf;
1010         int n;
1011 
1012         for (int c = 0; c < vd->vdev_children; c++)
1013                 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
1014 
1015         if (!vd->vdev_ops->vdev_op_leaf)
1016                 return;
1017 
1018         if (!vdev_writeable(vd))
1019                 return;
1020 
1021         n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
1022 
1023         ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
1024         bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1025         *ubbuf = *ub;
1026 
1027         for (int l = 0; l < VDEV_LABELS; l++)
1028                 vdev_label_write(zio, vd, l, ubbuf,
1029                     VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1030                     vdev_uberblock_sync_done, zio->io_private,
1031                     flags | ZIO_FLAG_DONT_PROPAGATE);
1032 
1033         zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1034 }
1035 
1036 /* Sync the uberblocks to all vdevs in svd[] */
1037 int
1038 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1039 {
1040         spa_t *spa = svd[0]->vdev_spa;
1041         zio_t *zio;
1042         uint64_t good_writes = 0;
1043 
1044         zio = zio_root(spa, NULL, &good_writes, flags);
1045 
1046         for (int v = 0; v < svdcount; v++)
1047                 vdev_uberblock_sync(zio, ub, svd[v], flags);
1048 
1049         (void) zio_wait(zio);
1050 
1051         /*
1052          * Flush the uberblocks to disk.  This ensures that the odd labels
1053          * are no longer needed (because the new uberblocks and the even
1054          * labels are safely on disk), so it is safe to overwrite them.
1055          */
1056         zio = zio_root(spa, NULL, NULL, flags);
1057 
1058         for (int v = 0; v < svdcount; v++)
1059                 zio_flush(zio, svd[v]);
1060 
1061         (void) zio_wait(zio);
1062 
1063         return (good_writes >= 1 ? 0 : EIO);
1064 }
1065 
1066 /*
1067  * On success, increment the count of good writes for our top-level vdev.
1068  */
1069 static void
1070 vdev_label_sync_done(zio_t *zio)
1071 {
1072         uint64_t *good_writes = zio->io_private;
1073 
1074         if (zio->io_error == 0)
1075                 atomic_add_64(good_writes, 1);
1076 }
1077 
1078 /*
1079  * If there weren't enough good writes, indicate failure to the parent.
1080  */
1081 static void
1082 vdev_label_sync_top_done(zio_t *zio)
1083 {
1084         uint64_t *good_writes = zio->io_private;
1085 
1086         if (*good_writes == 0)
1087                 zio->io_error = SET_ERROR(EIO);
1088 
1089         kmem_free(good_writes, sizeof (uint64_t));
1090 }
1091 
1092 /*
1093  * We ignore errors for log and cache devices, simply free the private data.
1094  */
1095 static void
1096 vdev_label_sync_ignore_done(zio_t *zio)
1097 {
1098         kmem_free(zio->io_private, sizeof (uint64_t));
1099 }
1100 
1101 /*
1102  * Write all even or odd labels to all leaves of the specified vdev.
1103  */
1104 static void
1105 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1106 {
1107         nvlist_t *label;
1108         vdev_phys_t *vp;
1109         char *buf;
1110         size_t buflen;
1111 
1112         for (int c = 0; c < vd->vdev_children; c++)
1113                 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1114 
1115         if (!vd->vdev_ops->vdev_op_leaf)
1116                 return;
1117 
1118         if (!vdev_writeable(vd))
1119                 return;
1120 
1121         /*
1122          * Generate a label describing the top-level config to which we belong.
1123          */
1124         label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1125 
1126         vp = zio_buf_alloc(sizeof (vdev_phys_t));
1127         bzero(vp, sizeof (vdev_phys_t));
1128 
1129         buf = vp->vp_nvlist;
1130         buflen = sizeof (vp->vp_nvlist);
1131 
1132         if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1133                 for (; l < VDEV_LABELS; l += 2) {
1134                         vdev_label_write(zio, vd, l, vp,
1135                             offsetof(vdev_label_t, vl_vdev_phys),
1136                             sizeof (vdev_phys_t),
1137                             vdev_label_sync_done, zio->io_private,
1138                             flags | ZIO_FLAG_DONT_PROPAGATE);
1139                 }
1140         }
1141 
1142         zio_buf_free(vp, sizeof (vdev_phys_t));
1143         nvlist_free(label);
1144 }
1145 
1146 int
1147 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1148 {
1149         list_t *dl = &spa->spa_config_dirty_list;
1150         vdev_t *vd;
1151         zio_t *zio;
1152         int error;
1153 
1154         /*
1155          * Write the new labels to disk.
1156          */
1157         zio = zio_root(spa, NULL, NULL, flags);
1158 
1159         for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1160                 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1161                     KM_SLEEP);
1162 
1163                 ASSERT(!vd->vdev_ishole);
1164 
1165                 zio_t *vio = zio_null(zio, spa, NULL,
1166                     (vd->vdev_islog || vd->vdev_aux != NULL) ?
1167                     vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1168                     good_writes, flags);
1169                 vdev_label_sync(vio, vd, l, txg, flags);
1170                 zio_nowait(vio);
1171         }
1172 
1173         error = zio_wait(zio);
1174 
1175         /*
1176          * Flush the new labels to disk.
1177          */
1178         zio = zio_root(spa, NULL, NULL, flags);
1179 
1180         for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1181                 zio_flush(zio, vd);
1182 
1183         (void) zio_wait(zio);
1184 
1185         return (error);
1186 }
1187 
1188 /*
1189  * Sync the uberblock and any changes to the vdev configuration.
1190  *
1191  * The order of operations is carefully crafted to ensure that
1192  * if the system panics or loses power at any time, the state on disk
1193  * is still transactionally consistent.  The in-line comments below
1194  * describe the failure semantics at each stage.
1195  *
1196  * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1197  * at any time, you can just call it again, and it will resume its work.
1198  */
1199 int
1200 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard)
1201 {
1202         spa_t *spa = svd[0]->vdev_spa;
1203         uberblock_t *ub = &spa->spa_uberblock;
1204         vdev_t *vd;
1205         zio_t *zio;
1206         int error;
1207         int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1208 
1209         /*
1210          * Normally, we don't want to try too hard to write every label and
1211          * uberblock.  If there is a flaky disk, we don't want the rest of the
1212          * sync process to block while we retry.  But if we can't write a
1213          * single label out, we should retry with ZIO_FLAG_TRYHARD before
1214          * bailing out and declaring the pool faulted.
1215          */
1216         if (tryhard)
1217                 flags |= ZIO_FLAG_TRYHARD;
1218 
1219         ASSERT(ub->ub_txg <= txg);
1220 
1221         /*
1222          * If this isn't a resync due to I/O errors,
1223          * and nothing changed in this transaction group,
1224          * and the vdev configuration hasn't changed,
1225          * then there's nothing to do.
1226          */
1227         if (ub->ub_txg < txg &&
1228             uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1229             list_is_empty(&spa->spa_config_dirty_list))
1230                 return (0);
1231 
1232         if (txg > spa_freeze_txg(spa))
1233                 return (0);
1234 
1235         ASSERT(txg <= spa->spa_final_txg);
1236 
1237         /*
1238          * Flush the write cache of every disk that's been written to
1239          * in this transaction group.  This ensures that all blocks
1240          * written in this txg will be committed to stable storage
1241          * before any uberblock that references them.
1242          */
1243         zio = zio_root(spa, NULL, NULL, flags);
1244 
1245         for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1246             vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1247                 zio_flush(zio, vd);
1248 
1249         (void) zio_wait(zio);
1250 
1251         /*
1252          * Sync out the even labels (L0, L2) for every dirty vdev.  If the
1253          * system dies in the middle of this process, that's OK: all of the
1254          * even labels that made it to disk will be newer than any uberblock,
1255          * and will therefore be considered invalid.  The odd labels (L1, L3),
1256          * which have not yet been touched, will still be valid.  We flush
1257          * the new labels to disk to ensure that all even-label updates
1258          * are committed to stable storage before the uberblock update.
1259          */
1260         if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1261                 return (error);
1262 
1263         /*
1264          * Sync the uberblocks to all vdevs in svd[].
1265          * If the system dies in the middle of this step, there are two cases
1266          * to consider, and the on-disk state is consistent either way:
1267          *
1268          * (1)  If none of the new uberblocks made it to disk, then the
1269          *      previous uberblock will be the newest, and the odd labels
1270          *      (which had not yet been touched) will be valid with respect
1271          *      to that uberblock.
1272          *
1273          * (2)  If one or more new uberblocks made it to disk, then they
1274          *      will be the newest, and the even labels (which had all
1275          *      been successfully committed) will be valid with respect
1276          *      to the new uberblocks.
1277          */
1278         if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1279                 return (error);
1280 
1281         /*
1282          * Sync out odd labels for every dirty vdev.  If the system dies
1283          * in the middle of this process, the even labels and the new
1284          * uberblocks will suffice to open the pool.  The next time
1285          * the pool is opened, the first thing we'll do -- before any
1286          * user data is modified -- is mark every vdev dirty so that
1287          * all labels will be brought up to date.  We flush the new labels
1288          * to disk to ensure that all odd-label updates are committed to
1289          * stable storage before the next transaction group begins.
1290          */
1291         return (vdev_label_sync_list(spa, 1, txg, flags));
1292 }