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         nv = fnvlist_alloc();
 220 
 221         fnvlist_add_string(nv, ZPOOL_CONFIG_TYPE, vd->vdev_ops->vdev_op_type);
 222         if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)))
 223                 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id);
 224         fnvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid);
 225 
 226         if (vd->vdev_path != NULL)
 227                 fnvlist_add_string(nv, ZPOOL_CONFIG_PATH, vd->vdev_path);
 228 
 229         if (vd->vdev_devid != NULL)
 230                 fnvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vd->vdev_devid);
 231 
 232         if (vd->vdev_physpath != NULL)
 233                 fnvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
 234                     vd->vdev_physpath);
 235 
 236         if (vd->vdev_fru != NULL)
 237                 fnvlist_add_string(nv, ZPOOL_CONFIG_FRU, vd->vdev_fru);
 238 
 239         if (vd->vdev_nparity != 0) {
 240                 ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
 241                     VDEV_TYPE_RAIDZ) == 0);
 242 
 243                 /*
 244                  * Make sure someone hasn't managed to sneak a fancy new vdev
 245                  * into a crufty old storage pool.
 246                  */
 247                 ASSERT(vd->vdev_nparity == 1 ||
 248                     (vd->vdev_nparity <= 2 &&
 249                     spa_version(spa) >= SPA_VERSION_RAIDZ2) ||
 250                     (vd->vdev_nparity <= 3 &&
 251                     spa_version(spa) >= SPA_VERSION_RAIDZ3));
 252 
 253                 /*
 254                  * Note that we'll add the nparity tag even on storage pools
 255                  * that only support a single parity device -- older software
 256                  * will just ignore it.
 257                  */
 258                 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY, vd->vdev_nparity);
 259         }
 260 
 261         if (vd->vdev_wholedisk != -1ULL)
 262                 fnvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
 263                     vd->vdev_wholedisk);
 264 
 265         if (vd->vdev_not_present)
 266                 fnvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1);
 267 
 268         if (vd->vdev_isspare)
 269                 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1);
 270 
 271         if (!(flags & (VDEV_CONFIG_SPARE | VDEV_CONFIG_L2CACHE)) &&
 272             vd == vd->vdev_top) {
 273                 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
 274                     vd->vdev_ms_array);
 275                 fnvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
 276                     vd->vdev_ms_shift);
 277                 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT, vd->vdev_ashift);
 278                 fnvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
 279                     vd->vdev_asize);
 280                 fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG, vd->vdev_islog);
 281                 if (vd->vdev_removing)
 282                         fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVING,
 283                             vd->vdev_removing);
 284         }
 285 
 286         if (flags & VDEV_CONFIG_L2CACHE)
 287                 /* indicate that we support L2ARC persistency */
 288                 VERIFY(nvlist_add_boolean_value(nv,
 289                     ZPOOL_CONFIG_L2CACHE_PERSISTENT, B_TRUE) == 0);
 290 
 291         if (vd->vdev_dtl_sm != NULL) {
 292                 fnvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
 293                     space_map_object(vd->vdev_dtl_sm));
 294         }
 295 
 296         if (vd->vdev_crtxg)
 297                 fnvlist_add_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, vd->vdev_crtxg);
 298 
 299         if (getstats) {
 300                 vdev_stat_t vs;
 301                 pool_scan_stat_t ps;
 302 
 303                 vdev_get_stats(vd, &vs);
 304                 fnvlist_add_uint64_array(nv, ZPOOL_CONFIG_VDEV_STATS,
 305                     (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t));
 306 
 307                 /* provide either current or previous scan information */
 308                 if (spa_scan_get_stats(spa, &ps) == 0) {
 309                         fnvlist_add_uint64_array(nv,
 310                             ZPOOL_CONFIG_SCAN_STATS, (uint64_t *)&ps,
 311                             sizeof (pool_scan_stat_t) / sizeof (uint64_t));
 312                 }
 313         }
 314 
 315         if (!vd->vdev_ops->vdev_op_leaf) {
 316                 nvlist_t **child;
 317                 int c, idx;
 318 
 319                 ASSERT(!vd->vdev_ishole);
 320 
 321                 child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
 322                     KM_SLEEP);
 323 
 324                 for (c = 0, idx = 0; c < vd->vdev_children; c++) {
 325                         vdev_t *cvd = vd->vdev_child[c];
 326 
 327                         /*
 328                          * If we're generating an nvlist of removing
 329                          * vdevs then skip over any device which is
 330                          * not being removed.
 331                          */
 332                         if ((flags & VDEV_CONFIG_REMOVING) &&
 333                             !cvd->vdev_removing)
 334                                 continue;
 335 
 336                         child[idx++] = vdev_config_generate(spa, cvd,
 337                             getstats, flags);
 338                 }
 339 
 340                 if (idx) {
 341                         fnvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 342                             child, idx);
 343                 }
 344 
 345                 for (c = 0; c < idx; c++)
 346                         nvlist_free(child[c]);
 347 
 348                 kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
 349 
 350         } else {
 351                 const char *aux = NULL;
 352 
 353                 if (vd->vdev_offline && !vd->vdev_tmpoffline)
 354                         fnvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE, B_TRUE);
 355                 if (vd->vdev_resilver_txg != 0)
 356                         fnvlist_add_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG,
 357                             vd->vdev_resilver_txg);
 358                 if (vd->vdev_faulted)
 359                         fnvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED, B_TRUE);
 360                 if (vd->vdev_degraded)
 361                         fnvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED, B_TRUE);
 362                 if (vd->vdev_removed)
 363                         fnvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED, B_TRUE);
 364                 if (vd->vdev_unspare)
 365                         fnvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE, B_TRUE);
 366                 if (vd->vdev_ishole)
 367                         fnvlist_add_uint64(nv, ZPOOL_CONFIG_IS_HOLE, B_TRUE);
 368 
 369                 switch (vd->vdev_stat.vs_aux) {
 370                 case VDEV_AUX_ERR_EXCEEDED:
 371                         aux = "err_exceeded";
 372                         break;
 373 
 374                 case VDEV_AUX_EXTERNAL:
 375                         aux = "external";
 376                         break;
 377                 }
 378 
 379                 if (aux != NULL)
 380                         fnvlist_add_string(nv, ZPOOL_CONFIG_AUX_STATE, aux);
 381 
 382                 if (vd->vdev_splitting && vd->vdev_orig_guid != 0LL) {
 383                         fnvlist_add_uint64(nv, ZPOOL_CONFIG_ORIG_GUID,
 384                             vd->vdev_orig_guid);
 385                 }
 386         }
 387 
 388         return (nv);
 389 }
 390 
 391 /*
 392  * Generate a view of the top-level vdevs.  If we currently have holes
 393  * in the namespace, then generate an array which contains a list of holey
 394  * vdevs.  Additionally, add the number of top-level children that currently
 395  * exist.
 396  */
 397 void
 398 vdev_top_config_generate(spa_t *spa, nvlist_t *config)
 399 {
 400         vdev_t *rvd = spa->spa_root_vdev;
 401         uint64_t *array;
 402         uint_t c, idx;
 403 
 404         array = kmem_alloc(rvd->vdev_children * sizeof (uint64_t), KM_SLEEP);
 405 
 406         for (c = 0, idx = 0; c < rvd->vdev_children; c++) {
 407                 vdev_t *tvd = rvd->vdev_child[c];
 408 
 409                 if (tvd->vdev_ishole)
 410                         array[idx++] = c;
 411         }
 412 
 413         if (idx) {
 414                 VERIFY(nvlist_add_uint64_array(config, ZPOOL_CONFIG_HOLE_ARRAY,
 415                     array, idx) == 0);
 416         }
 417 
 418         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VDEV_CHILDREN,
 419             rvd->vdev_children) == 0);
 420 
 421         kmem_free(array, rvd->vdev_children * sizeof (uint64_t));
 422 }
 423 
 424 /*
 425  * Returns the configuration from the label of the given vdev. For vdevs
 426  * which don't have a txg value stored on their label (i.e. spares/cache)
 427  * or have not been completely initialized (txg = 0) just return
 428  * the configuration from the first valid label we find. Otherwise,
 429  * find the most up-to-date label that does not exceed the specified
 430  * 'txg' value.
 431  */
 432 nvlist_t *
 433 vdev_label_read_config(vdev_t *vd, uint64_t txg)
 434 {
 435         spa_t *spa = vd->vdev_spa;
 436         nvlist_t *config = NULL;
 437         vdev_phys_t *vp;
 438         zio_t *zio;
 439         uint64_t best_txg = 0;
 440         int error = 0;
 441         int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
 442             ZIO_FLAG_SPECULATIVE;
 443 
 444         ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
 445 
 446         if (!vdev_readable(vd))
 447                 return (NULL);
 448 
 449         vp = zio_buf_alloc(sizeof (vdev_phys_t));
 450 
 451 retry:
 452         for (int l = 0; l < VDEV_LABELS; l++) {
 453                 nvlist_t *label = NULL;
 454 
 455                 zio = zio_root(spa, NULL, NULL, flags);
 456 
 457                 vdev_label_read(zio, vd, l, vp,
 458                     offsetof(vdev_label_t, vl_vdev_phys),
 459                     sizeof (vdev_phys_t), NULL, NULL, flags);
 460 
 461                 if (zio_wait(zio) == 0 &&
 462                     nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
 463                     &label, 0) == 0) {
 464                         uint64_t label_txg = 0;
 465 
 466                         /*
 467                          * Auxiliary vdevs won't have txg values in their
 468                          * labels and newly added vdevs may not have been
 469                          * completely initialized so just return the
 470                          * configuration from the first valid label we
 471                          * encounter.
 472                          */
 473                         error = nvlist_lookup_uint64(label,
 474                             ZPOOL_CONFIG_POOL_TXG, &label_txg);
 475                         if ((error || label_txg == 0) && !config) {
 476                                 config = label;
 477                                 break;
 478                         } else if (label_txg <= txg && label_txg > best_txg) {
 479                                 best_txg = label_txg;
 480                                 nvlist_free(config);
 481                                 config = fnvlist_dup(label);
 482                         }
 483                 }
 484 
 485                 if (label != NULL) {
 486                         nvlist_free(label);
 487                         label = NULL;
 488                 }
 489         }
 490 
 491         if (config == NULL && !(flags & ZIO_FLAG_TRYHARD)) {
 492                 flags |= ZIO_FLAG_TRYHARD;
 493                 goto retry;
 494         }
 495 
 496         zio_buf_free(vp, sizeof (vdev_phys_t));
 497 
 498         return (config);
 499 }
 500 
 501 /*
 502  * Determine if a device is in use.  The 'spare_guid' parameter will be filled
 503  * in with the device guid if this spare is active elsewhere on the system.
 504  */
 505 static boolean_t
 506 vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
 507     uint64_t *spare_guid, uint64_t *l2cache_guid)
 508 {
 509         spa_t *spa = vd->vdev_spa;
 510         uint64_t state, pool_guid, device_guid, txg, spare_pool;
 511         uint64_t vdtxg = 0;
 512         nvlist_t *label;
 513 
 514         if (spare_guid)
 515                 *spare_guid = 0ULL;
 516         if (l2cache_guid)
 517                 *l2cache_guid = 0ULL;
 518 
 519         /*
 520          * Read the label, if any, and perform some basic sanity checks.
 521          */
 522         if ((label = vdev_label_read_config(vd, -1ULL)) == NULL)
 523                 return (B_FALSE);
 524 
 525         (void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
 526             &vdtxg);
 527 
 528         if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
 529             &state) != 0 ||
 530             nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
 531             &device_guid) != 0) {
 532                 nvlist_free(label);
 533                 return (B_FALSE);
 534         }
 535 
 536         if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 537             (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
 538             &pool_guid) != 0 ||
 539             nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
 540             &txg) != 0)) {
 541                 nvlist_free(label);
 542                 return (B_FALSE);
 543         }
 544 
 545         nvlist_free(label);
 546 
 547         /*
 548          * Check to see if this device indeed belongs to the pool it claims to
 549          * be a part of.  The only way this is allowed is if the device is a hot
 550          * spare (which we check for later on).
 551          */
 552         if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 553             !spa_guid_exists(pool_guid, device_guid) &&
 554             !spa_spare_exists(device_guid, NULL, NULL) &&
 555             !spa_l2cache_exists(device_guid, NULL))
 556                 return (B_FALSE);
 557 
 558         /*
 559          * If the transaction group is zero, then this an initialized (but
 560          * unused) label.  This is only an error if the create transaction
 561          * on-disk is the same as the one we're using now, in which case the
 562          * user has attempted to add the same vdev multiple times in the same
 563          * transaction.
 564          */
 565         if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
 566             txg == 0 && vdtxg == crtxg)
 567                 return (B_TRUE);
 568 
 569         /*
 570          * Check to see if this is a spare device.  We do an explicit check for
 571          * spa_has_spare() here because it may be on our pending list of spares
 572          * to add.  We also check if it is an l2cache device.
 573          */
 574         if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
 575             spa_has_spare(spa, device_guid)) {
 576                 if (spare_guid)
 577                         *spare_guid = device_guid;
 578 
 579                 switch (reason) {
 580                 case VDEV_LABEL_CREATE:
 581                 case VDEV_LABEL_L2CACHE:
 582                         return (B_TRUE);
 583 
 584                 case VDEV_LABEL_REPLACE:
 585                         return (!spa_has_spare(spa, device_guid) ||
 586                             spare_pool != 0ULL);
 587 
 588                 case VDEV_LABEL_SPARE:
 589                         return (spa_has_spare(spa, device_guid));
 590                 }
 591         }
 592 
 593         /*
 594          * Check to see if this is an l2cache device.
 595          */
 596         if (spa_l2cache_exists(device_guid, NULL))
 597                 return (B_TRUE);
 598 
 599         /*
 600          * We can't rely on a pool's state if it's been imported
 601          * read-only.  Instead we look to see if the pools is marked
 602          * read-only in the namespace and set the state to active.
 603          */
 604         if ((spa = spa_by_guid(pool_guid, device_guid)) != NULL &&
 605             spa_mode(spa) == FREAD)
 606                 state = POOL_STATE_ACTIVE;
 607 
 608         /*
 609          * If the device is marked ACTIVE, then this device is in use by another
 610          * pool on the system.
 611          */
 612         return (state == POOL_STATE_ACTIVE);
 613 }
 614 
 615 /*
 616  * Initialize a vdev label.  We check to make sure each leaf device is not in
 617  * use, and writable.  We put down an initial label which we will later
 618  * overwrite with a complete label.  Note that it's important to do this
 619  * sequentially, not in parallel, so that we catch cases of multiple use of the
 620  * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
 621  * itself.
 622  */
 623 int
 624 vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
 625 {
 626         spa_t *spa = vd->vdev_spa;
 627         nvlist_t *label;
 628         vdev_phys_t *vp;
 629         char *pad2;
 630         uberblock_t *ub;
 631         zio_t *zio;
 632         char *buf;
 633         size_t buflen;
 634         int error;
 635         uint64_t spare_guid, l2cache_guid;
 636         int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
 637 
 638         ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
 639 
 640         for (int c = 0; c < vd->vdev_children; c++)
 641                 if ((error = vdev_label_init(vd->vdev_child[c],
 642                     crtxg, reason)) != 0)
 643                         return (error);
 644 
 645         /* Track the creation time for this vdev */
 646         vd->vdev_crtxg = crtxg;
 647 
 648         if (!vd->vdev_ops->vdev_op_leaf)
 649                 return (0);
 650 
 651         /*
 652          * Dead vdevs cannot be initialized.
 653          */
 654         if (vdev_is_dead(vd))
 655                 return (SET_ERROR(EIO));
 656 
 657         /*
 658          * Determine if the vdev is in use.
 659          */
 660         if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPLIT &&
 661             vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
 662                 return (SET_ERROR(EBUSY));
 663 
 664         /*
 665          * If this is a request to add or replace a spare or l2cache device
 666          * that is in use elsewhere on the system, then we must update the
 667          * guid (which was initialized to a random value) to reflect the
 668          * actual GUID (which is shared between multiple pools).
 669          */
 670         if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
 671             spare_guid != 0ULL) {
 672                 uint64_t guid_delta = spare_guid - vd->vdev_guid;
 673 
 674                 vd->vdev_guid += guid_delta;
 675 
 676                 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
 677                         pvd->vdev_guid_sum += guid_delta;
 678 
 679                 /*
 680                  * If this is a replacement, then we want to fallthrough to the
 681                  * rest of the code.  If we're adding a spare, then it's already
 682                  * labeled appropriately and we can just return.
 683                  */
 684                 if (reason == VDEV_LABEL_SPARE)
 685                         return (0);
 686                 ASSERT(reason == VDEV_LABEL_REPLACE ||
 687                     reason == VDEV_LABEL_SPLIT);
 688         }
 689 
 690         if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
 691             l2cache_guid != 0ULL) {
 692                 uint64_t guid_delta = l2cache_guid - vd->vdev_guid;
 693 
 694                 vd->vdev_guid += guid_delta;
 695 
 696                 for (vdev_t *pvd = vd; pvd != NULL; pvd = pvd->vdev_parent)
 697                         pvd->vdev_guid_sum += guid_delta;
 698 
 699                 /*
 700                  * If this is a replacement, then we want to fallthrough to the
 701                  * rest of the code.  If we're adding an l2cache, then it's
 702                  * already labeled appropriately and we can just return.
 703                  */
 704                 if (reason == VDEV_LABEL_L2CACHE)
 705                         return (0);
 706                 ASSERT(reason == VDEV_LABEL_REPLACE);
 707         }
 708 
 709         /*
 710          * Initialize its label.
 711          */
 712         vp = zio_buf_alloc(sizeof (vdev_phys_t));
 713         bzero(vp, sizeof (vdev_phys_t));
 714 
 715         /*
 716          * Generate a label describing the pool and our top-level vdev.
 717          * We mark it as being from txg 0 to indicate that it's not
 718          * really part of an active pool just yet.  The labels will
 719          * be written again with a meaningful txg by spa_sync().
 720          */
 721         if (reason == VDEV_LABEL_SPARE ||
 722             (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
 723                 /*
 724                  * For inactive hot spares, we generate a special label that
 725                  * identifies as a mutually shared hot spare.  We write the
 726                  * label if we are adding a hot spare, or if we are removing an
 727                  * active hot spare (in which case we want to revert the
 728                  * labels).
 729                  */
 730                 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 731 
 732                 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
 733                     spa_version(spa)) == 0);
 734                 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
 735                     POOL_STATE_SPARE) == 0);
 736                 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
 737                     vd->vdev_guid) == 0);
 738         } else if (reason == VDEV_LABEL_L2CACHE ||
 739             (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
 740                 /*
 741                  * For level 2 ARC devices, add a special label.
 742                  */
 743                 VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 744 
 745                 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
 746                     spa_version(spa)) == 0);
 747                 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
 748                     POOL_STATE_L2CACHE) == 0);
 749                 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
 750                     vd->vdev_guid) == 0);
 751         } else {
 752                 uint64_t txg = 0ULL;
 753 
 754                 if (reason == VDEV_LABEL_SPLIT)
 755                         txg = spa->spa_uberblock.ub_txg;
 756                 label = spa_config_generate(spa, vd, txg, B_FALSE);
 757 
 758                 /*
 759                  * Add our creation time.  This allows us to detect multiple
 760                  * vdev uses as described above, and automatically expires if we
 761                  * fail.
 762                  */
 763                 VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
 764                     crtxg) == 0);
 765         }
 766 
 767         buf = vp->vp_nvlist;
 768         buflen = sizeof (vp->vp_nvlist);
 769 
 770         error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
 771         if (error != 0) {
 772                 nvlist_free(label);
 773                 zio_buf_free(vp, sizeof (vdev_phys_t));
 774                 /* EFAULT means nvlist_pack ran out of room */
 775                 return (error == EFAULT ? ENAMETOOLONG : EINVAL);
 776         }
 777 
 778         /*
 779          * Initialize uberblock template.
 780          */
 781         ub = zio_buf_alloc(VDEV_UBERBLOCK_RING);
 782         bzero(ub, VDEV_UBERBLOCK_RING);
 783         *ub = spa->spa_uberblock;
 784         ub->ub_txg = 0;
 785 
 786         /* Initialize the 2nd padding area. */
 787         pad2 = zio_buf_alloc(VDEV_PAD_SIZE);
 788         bzero(pad2, VDEV_PAD_SIZE);
 789 
 790         /*
 791          * Write everything in parallel.
 792          */
 793 retry:
 794         zio = zio_root(spa, NULL, NULL, flags);
 795 
 796         for (int l = 0; l < VDEV_LABELS; l++) {
 797 
 798                 vdev_label_write(zio, vd, l, vp,
 799                     offsetof(vdev_label_t, vl_vdev_phys),
 800                     sizeof (vdev_phys_t), NULL, NULL, flags);
 801 
 802                 /*
 803                  * Skip the 1st padding area.
 804                  * Zero out the 2nd padding area where it might have
 805                  * left over data from previous filesystem format.
 806                  */
 807                 vdev_label_write(zio, vd, l, pad2,
 808                     offsetof(vdev_label_t, vl_pad2),
 809                     VDEV_PAD_SIZE, NULL, NULL, flags);
 810 
 811                 vdev_label_write(zio, vd, l, ub,
 812                     offsetof(vdev_label_t, vl_uberblock),
 813                     VDEV_UBERBLOCK_RING, NULL, NULL, flags);
 814         }
 815 
 816         error = zio_wait(zio);
 817 
 818         if (error != 0 && !(flags & ZIO_FLAG_TRYHARD)) {
 819                 flags |= ZIO_FLAG_TRYHARD;
 820                 goto retry;
 821         }
 822 
 823         nvlist_free(label);
 824         zio_buf_free(pad2, VDEV_PAD_SIZE);
 825         zio_buf_free(ub, VDEV_UBERBLOCK_RING);
 826         zio_buf_free(vp, sizeof (vdev_phys_t));
 827 
 828         /*
 829          * If this vdev hasn't been previously identified as a spare, then we
 830          * mark it as such only if a) we are labeling it as a spare, or b) it
 831          * exists as a spare elsewhere in the system.  Do the same for
 832          * level 2 ARC devices.
 833          */
 834         if (error == 0 && !vd->vdev_isspare &&
 835             (reason == VDEV_LABEL_SPARE ||
 836             spa_spare_exists(vd->vdev_guid, NULL, NULL)))
 837                 spa_spare_add(vd);
 838 
 839         if (error == 0 && !vd->vdev_isl2cache &&
 840             (reason == VDEV_LABEL_L2CACHE ||
 841             spa_l2cache_exists(vd->vdev_guid, NULL)))
 842                 spa_l2cache_add(vd);
 843 
 844         return (error);
 845 }
 846 
 847 /*
 848  * ==========================================================================
 849  * uberblock load/sync
 850  * ==========================================================================
 851  */
 852 
 853 /*
 854  * Consider the following situation: txg is safely synced to disk.  We've
 855  * written the first uberblock for txg + 1, and then we lose power.  When we
 856  * come back up, we fail to see the uberblock for txg + 1 because, say,
 857  * it was on a mirrored device and the replica to which we wrote txg + 1
 858  * is now offline.  If we then make some changes and sync txg + 1, and then
 859  * the missing replica comes back, then for a few seconds we'll have two
 860  * conflicting uberblocks on disk with the same txg.  The solution is simple:
 861  * among uberblocks with equal txg, choose the one with the latest timestamp.
 862  */
 863 static int
 864 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
 865 {
 866         if (ub1->ub_txg < ub2->ub_txg)
 867                 return (-1);
 868         if (ub1->ub_txg > ub2->ub_txg)
 869                 return (1);
 870 
 871         if (ub1->ub_timestamp < ub2->ub_timestamp)
 872                 return (-1);
 873         if (ub1->ub_timestamp > ub2->ub_timestamp)
 874                 return (1);
 875 
 876         return (0);
 877 }
 878 
 879 struct ubl_cbdata {
 880         uberblock_t     *ubl_ubbest;    /* Best uberblock */
 881         vdev_t          *ubl_vd;        /* vdev associated with the above */
 882 };
 883 
 884 static void
 885 vdev_uberblock_load_done(zio_t *zio)
 886 {
 887         vdev_t *vd = zio->io_vd;
 888         spa_t *spa = zio->io_spa;
 889         zio_t *rio = zio->io_private;
 890         uberblock_t *ub = zio->io_data;
 891         struct ubl_cbdata *cbp = rio->io_private;
 892 
 893         ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(vd));
 894 
 895         if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
 896                 mutex_enter(&rio->io_lock);
 897                 if (ub->ub_txg <= spa->spa_load_max_txg &&
 898                     vdev_uberblock_compare(ub, cbp->ubl_ubbest) > 0) {
 899                         /*
 900                          * Keep track of the vdev in which this uberblock
 901                          * was found. We will use this information later
 902                          * to obtain the config nvlist associated with
 903                          * this uberblock.
 904                          */
 905                         *cbp->ubl_ubbest = *ub;
 906                         cbp->ubl_vd = vd;
 907                 }
 908                 mutex_exit(&rio->io_lock);
 909         }
 910 
 911         zio_buf_free(zio->io_data, zio->io_size);
 912 }
 913 
 914 static void
 915 vdev_uberblock_load_impl(zio_t *zio, vdev_t *vd, int flags,
 916     struct ubl_cbdata *cbp)
 917 {
 918         for (int c = 0; c < vd->vdev_children; c++)
 919                 vdev_uberblock_load_impl(zio, vd->vdev_child[c], flags, cbp);
 920 
 921         if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
 922                 for (int l = 0; l < VDEV_LABELS; l++) {
 923                         for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
 924                                 vdev_label_read(zio, vd, l,
 925                                     zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
 926                                     VDEV_UBERBLOCK_OFFSET(vd, n),
 927                                     VDEV_UBERBLOCK_SIZE(vd),
 928                                     vdev_uberblock_load_done, zio, flags);
 929                         }
 930                 }
 931         }
 932 }
 933 
 934 /*
 935  * Reads the 'best' uberblock from disk along with its associated
 936  * configuration. First, we read the uberblock array of each label of each
 937  * vdev, keeping track of the uberblock with the highest txg in each array.
 938  * Then, we read the configuration from the same vdev as the best uberblock.
 939  */
 940 void
 941 vdev_uberblock_load(vdev_t *rvd, uberblock_t *ub, nvlist_t **config)
 942 {
 943         zio_t *zio;
 944         spa_t *spa = rvd->vdev_spa;
 945         struct ubl_cbdata cb;
 946         int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL |
 947             ZIO_FLAG_SPECULATIVE | ZIO_FLAG_TRYHARD;
 948 
 949         ASSERT(ub);
 950         ASSERT(config);
 951 
 952         bzero(ub, sizeof (uberblock_t));
 953         *config = NULL;
 954 
 955         cb.ubl_ubbest = ub;
 956         cb.ubl_vd = NULL;
 957 
 958         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
 959         zio = zio_root(spa, NULL, &cb, flags);
 960         vdev_uberblock_load_impl(zio, rvd, flags, &cb);
 961         (void) zio_wait(zio);
 962 
 963         /*
 964          * It's possible that the best uberblock was discovered on a label
 965          * that has a configuration which was written in a future txg.
 966          * Search all labels on this vdev to find the configuration that
 967          * matches the txg for our uberblock.
 968          */
 969         if (cb.ubl_vd != NULL)
 970                 *config = vdev_label_read_config(cb.ubl_vd, ub->ub_txg);
 971         spa_config_exit(spa, SCL_ALL, FTAG);
 972 }
 973 
 974 /*
 975  * On success, increment root zio's count of good writes.
 976  * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
 977  */
 978 static void
 979 vdev_uberblock_sync_done(zio_t *zio)
 980 {
 981         uint64_t *good_writes = zio->io_private;
 982 
 983         if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
 984                 atomic_add_64(good_writes, 1);
 985 }
 986 
 987 /*
 988  * Write the uberblock to all labels of all leaves of the specified vdev.
 989  */
 990 static void
 991 vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
 992 {
 993         uberblock_t *ubbuf;
 994         int n;
 995 
 996         for (int c = 0; c < vd->vdev_children; c++)
 997                 vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
 998 
 999         if (!vd->vdev_ops->vdev_op_leaf)
1000                 return;
1001 
1002         if (!vdev_writeable(vd))
1003                 return;
1004 
1005         n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
1006 
1007         ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
1008         bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1009         *ubbuf = *ub;
1010 
1011         for (int l = 0; l < VDEV_LABELS; l++)
1012                 vdev_label_write(zio, vd, l, ubbuf,
1013                     VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
1014                     vdev_uberblock_sync_done, zio->io_private,
1015                     flags | ZIO_FLAG_DONT_PROPAGATE);
1016 
1017         zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
1018 }
1019 
1020 /* Sync the uberblocks to all vdevs in svd[] */
1021 int
1022 vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
1023 {
1024         spa_t *spa = svd[0]->vdev_spa;
1025         zio_t *zio;
1026         uint64_t good_writes = 0;
1027 
1028         zio = zio_root(spa, NULL, &good_writes, flags);
1029 
1030         for (int v = 0; v < svdcount; v++)
1031                 vdev_uberblock_sync(zio, ub, svd[v], flags);
1032 
1033         (void) zio_wait(zio);
1034 
1035         /*
1036          * Flush the uberblocks to disk.  This ensures that the odd labels
1037          * are no longer needed (because the new uberblocks and the even
1038          * labels are safely on disk), so it is safe to overwrite them.
1039          */
1040         zio = zio_root(spa, NULL, NULL, flags);
1041 
1042         for (int v = 0; v < svdcount; v++)
1043                 zio_flush(zio, svd[v]);
1044 
1045         (void) zio_wait(zio);
1046 
1047         return (good_writes >= 1 ? 0 : EIO);
1048 }
1049 
1050 /*
1051  * On success, increment the count of good writes for our top-level vdev.
1052  */
1053 static void
1054 vdev_label_sync_done(zio_t *zio)
1055 {
1056         uint64_t *good_writes = zio->io_private;
1057 
1058         if (zio->io_error == 0)
1059                 atomic_add_64(good_writes, 1);
1060 }
1061 
1062 /*
1063  * If there weren't enough good writes, indicate failure to the parent.
1064  */
1065 static void
1066 vdev_label_sync_top_done(zio_t *zio)
1067 {
1068         uint64_t *good_writes = zio->io_private;
1069 
1070         if (*good_writes == 0)
1071                 zio->io_error = SET_ERROR(EIO);
1072 
1073         kmem_free(good_writes, sizeof (uint64_t));
1074 }
1075 
1076 /*
1077  * We ignore errors for log and cache devices, simply free the private data.
1078  */
1079 static void
1080 vdev_label_sync_ignore_done(zio_t *zio)
1081 {
1082         kmem_free(zio->io_private, sizeof (uint64_t));
1083 }
1084 
1085 /*
1086  * Write all even or odd labels to all leaves of the specified vdev.
1087  */
1088 static void
1089 vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
1090 {
1091         nvlist_t *label;
1092         vdev_phys_t *vp;
1093         char *buf;
1094         size_t buflen;
1095 
1096         for (int c = 0; c < vd->vdev_children; c++)
1097                 vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
1098 
1099         if (!vd->vdev_ops->vdev_op_leaf)
1100                 return;
1101 
1102         if (!vdev_writeable(vd))
1103                 return;
1104 
1105         /*
1106          * Generate a label describing the top-level config to which we belong.
1107          */
1108         label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
1109 
1110         vp = zio_buf_alloc(sizeof (vdev_phys_t));
1111         bzero(vp, sizeof (vdev_phys_t));
1112 
1113         buf = vp->vp_nvlist;
1114         buflen = sizeof (vp->vp_nvlist);
1115 
1116         if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
1117                 for (; l < VDEV_LABELS; l += 2) {
1118                         vdev_label_write(zio, vd, l, vp,
1119                             offsetof(vdev_label_t, vl_vdev_phys),
1120                             sizeof (vdev_phys_t),
1121                             vdev_label_sync_done, zio->io_private,
1122                             flags | ZIO_FLAG_DONT_PROPAGATE);
1123                 }
1124         }
1125 
1126         zio_buf_free(vp, sizeof (vdev_phys_t));
1127         nvlist_free(label);
1128 }
1129 
1130 int
1131 vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
1132 {
1133         list_t *dl = &spa->spa_config_dirty_list;
1134         vdev_t *vd;
1135         zio_t *zio;
1136         int error;
1137 
1138         /*
1139          * Write the new labels to disk.
1140          */
1141         zio = zio_root(spa, NULL, NULL, flags);
1142 
1143         for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
1144                 uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
1145                     KM_SLEEP);
1146 
1147                 ASSERT(!vd->vdev_ishole);
1148 
1149                 zio_t *vio = zio_null(zio, spa, NULL,
1150                     (vd->vdev_islog || vd->vdev_aux != NULL) ?
1151                     vdev_label_sync_ignore_done : vdev_label_sync_top_done,
1152                     good_writes, flags);
1153                 vdev_label_sync(vio, vd, l, txg, flags);
1154                 zio_nowait(vio);
1155         }
1156 
1157         error = zio_wait(zio);
1158 
1159         /*
1160          * Flush the new labels to disk.
1161          */
1162         zio = zio_root(spa, NULL, NULL, flags);
1163 
1164         for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
1165                 zio_flush(zio, vd);
1166 
1167         (void) zio_wait(zio);
1168 
1169         return (error);
1170 }
1171 
1172 /*
1173  * Sync the uberblock and any changes to the vdev configuration.
1174  *
1175  * The order of operations is carefully crafted to ensure that
1176  * if the system panics or loses power at any time, the state on disk
1177  * is still transactionally consistent.  The in-line comments below
1178  * describe the failure semantics at each stage.
1179  *
1180  * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
1181  * at any time, you can just call it again, and it will resume its work.
1182  */
1183 int
1184 vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg, boolean_t tryhard)
1185 {
1186         spa_t *spa = svd[0]->vdev_spa;
1187         uberblock_t *ub = &spa->spa_uberblock;
1188         vdev_t *vd;
1189         zio_t *zio;
1190         int error;
1191         int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
1192 
1193         /*
1194          * Normally, we don't want to try too hard to write every label and
1195          * uberblock.  If there is a flaky disk, we don't want the rest of the
1196          * sync process to block while we retry.  But if we can't write a
1197          * single label out, we should retry with ZIO_FLAG_TRYHARD before
1198          * bailing out and declaring the pool faulted.
1199          */
1200         if (tryhard)
1201                 flags |= ZIO_FLAG_TRYHARD;
1202 
1203         ASSERT(ub->ub_txg <= txg);
1204 
1205         /*
1206          * If this isn't a resync due to I/O errors,
1207          * and nothing changed in this transaction group,
1208          * and the vdev configuration hasn't changed,
1209          * then there's nothing to do.
1210          */
1211         if (ub->ub_txg < txg &&
1212             uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1213             list_is_empty(&spa->spa_config_dirty_list))
1214                 return (0);
1215 
1216         if (txg > spa_freeze_txg(spa))
1217                 return (0);
1218 
1219         ASSERT(txg <= spa->spa_final_txg);
1220 
1221         /*
1222          * Flush the write cache of every disk that's been written to
1223          * in this transaction group.  This ensures that all blocks
1224          * written in this txg will be committed to stable storage
1225          * before any uberblock that references them.
1226          */
1227         zio = zio_root(spa, NULL, NULL, flags);
1228 
1229         for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
1230             vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
1231                 zio_flush(zio, vd);
1232 
1233         (void) zio_wait(zio);
1234 
1235         /*
1236          * Sync out the even labels (L0, L2) for every dirty vdev.  If the
1237          * system dies in the middle of this process, that's OK: all of the
1238          * even labels that made it to disk will be newer than any uberblock,
1239          * and will therefore be considered invalid.  The odd labels (L1, L3),
1240          * which have not yet been touched, will still be valid.  We flush
1241          * the new labels to disk to ensure that all even-label updates
1242          * are committed to stable storage before the uberblock update.
1243          */
1244         if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1245                 return (error);
1246 
1247         /*
1248          * Sync the uberblocks to all vdevs in svd[].
1249          * If the system dies in the middle of this step, there are two cases
1250          * to consider, and the on-disk state is consistent either way:
1251          *
1252          * (1)  If none of the new uberblocks made it to disk, then the
1253          *      previous uberblock will be the newest, and the odd labels
1254          *      (which had not yet been touched) will be valid with respect
1255          *      to that uberblock.
1256          *
1257          * (2)  If one or more new uberblocks made it to disk, then they
1258          *      will be the newest, and the even labels (which had all
1259          *      been successfully committed) will be valid with respect
1260          *      to the new uberblocks.
1261          */
1262         if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1263                 return (error);
1264 
1265         /*
1266          * Sync out odd labels for every dirty vdev.  If the system dies
1267          * in the middle of this process, the even labels and the new
1268          * uberblocks will suffice to open the pool.  The next time
1269          * the pool is opened, the first thing we'll do -- before any
1270          * user data is modified -- is mark every vdev dirty so that
1271          * all labels will be brought up to date.  We flush the new labels
1272          * to disk to ensure that all odd-label updates are committed to
1273          * stable storage before the next transaction group begins.
1274          */
1275         return (vdev_label_sync_list(spa, 1, txg, flags));
1276 }