Print this page
    
5882 Temporary pool names
Reviewed by: Matt Ahrens <matt@delphix.com>
Reviewed by: Igor Kozhukhov <igor@dilos.org>
Reviewed by: John Kennedy <john.kennedy@delphix.com>
Approved by: Dan McDonald <danmcd@joyent.com>
    
      
        | Split | 
	Close | 
      
      | Expand all | 
      | Collapse all | 
    
    
          --- old/usr/src/uts/common/fs/zfs/spa_config.c
          +++ new/usr/src/uts/common/fs/zfs/spa_config.c
   1    1  /*
   2    2   * CDDL HEADER START
   3    3   *
   4    4   * The contents of this file are subject to the terms of the
   5    5   * Common Development and Distribution License (the "License").
   6    6   * You may not use this file except in compliance with the License.
   7    7   *
   8    8   * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9    9   * or http://www.opensolaris.org/os/licensing.
  10   10   * See the License for the specific language governing permissions
  11   11   * and limitations under the License.
  12   12   *
  13   13   * When distributing Covered Code, include this CDDL HEADER in each
  14   14   * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15   15   * If applicable, add the following below this CDDL HEADER, with the
  16   16   * fields enclosed by brackets "[]" replaced with your own identifying
  17   17   * information: Portions Copyright [yyyy] [name of copyright owner]
  18   18   *
  19   19   * CDDL HEADER END
  20   20   */
  21   21  
  22   22  /*
  23   23   * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24   24   * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  25   25   * Copyright (c) 2011, 2018 by Delphix. All rights reserved.
  26   26   * Copyright 2017 Joyent, Inc.
  27   27   */
  28   28  
  29   29  #include <sys/spa.h>
  30   30  #include <sys/fm/fs/zfs.h>
  31   31  #include <sys/spa_impl.h>
  32   32  #include <sys/nvpair.h>
  33   33  #include <sys/uio.h>
  34   34  #include <sys/fs/zfs.h>
  35   35  #include <sys/vdev_impl.h>
  36   36  #include <sys/zfs_ioctl.h>
  37   37  #include <sys/utsname.h>
  38   38  #include <sys/systeminfo.h>
  39   39  #include <sys/sunddi.h>
  40   40  #include <sys/zfeature.h>
  41   41  #ifdef _KERNEL
  42   42  #include <sys/kobj.h>
  43   43  #include <sys/zone.h>
  44   44  #endif
  45   45  
  46   46  /*
  47   47   * Pool configuration repository.
  48   48   *
  49   49   * Pool configuration is stored as a packed nvlist on the filesystem.  By
  50   50   * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
  51   51   * (when the ZFS module is loaded).  Pools can also have the 'cachefile'
  52   52   * property set that allows them to be stored in an alternate location until
  53   53   * the control of external software.
  54   54   *
  55   55   * For each cache file, we have a single nvlist which holds all the
  56   56   * configuration information.  When the module loads, we read this information
  57   57   * from /etc/zfs/zpool.cache and populate the SPA namespace.  This namespace is
  58   58   * maintained independently in spa.c.  Whenever the namespace is modified, or
  59   59   * the configuration of a pool is changed, we call spa_write_cachefile(), which
  60   60   * walks through all the active pools and writes the configuration to disk.
  61   61   */
  62   62  
  63   63  static uint64_t spa_config_generation = 1;
  64   64  
  65   65  /*
  66   66   * This can be overridden in userland to preserve an alternate namespace for
  67   67   * userland pools when doing testing.
  68   68   */
  69   69  const char *spa_config_path = ZPOOL_CACHE;
  70   70  
  71   71  /*
  72   72   * Called when the module is first loaded, this routine loads the configuration
  73   73   * file into the SPA namespace.  It does not actually open or load the pools; it
  74   74   * only populates the namespace.
  75   75   */
  76   76  void
  77   77  spa_config_load(void)
  78   78  {
  79   79          void *buf = NULL;
  80   80          nvlist_t *nvlist, *child;
  81   81          nvpair_t *nvpair;
  82   82          char *pathname;
  83   83          struct _buf *file;
  84   84          uint64_t fsize;
  85   85  
  86   86          /*
  87   87           * Open the configuration file.
  88   88           */
  89   89          pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
  90   90  
  91   91          (void) snprintf(pathname, MAXPATHLEN, "%s%s",
  92   92              (rootdir != NULL) ? "./" : "", spa_config_path);
  93   93  
  94   94          file = kobj_open_file(pathname);
  95   95  
  96   96          kmem_free(pathname, MAXPATHLEN);
  97   97  
  98   98          if (file == (struct _buf *)-1)
  99   99                  return;
 100  100  
 101  101          if (kobj_get_filesize(file, &fsize) != 0)
 102  102                  goto out;
 103  103  
 104  104          buf = kmem_alloc(fsize, KM_SLEEP);
 105  105  
 106  106          /*
 107  107           * Read the nvlist from the file.
 108  108           */
 109  109          if (kobj_read_file(file, buf, fsize, 0) < 0)
 110  110                  goto out;
 111  111  
 112  112          /*
 113  113           * Unpack the nvlist.
 114  114           */
 115  115          if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
 116  116                  goto out;
 117  117  
 118  118          /*
 119  119           * Iterate over all elements in the nvlist, creating a new spa_t for
 120  120           * each one with the specified configuration.
 121  121           */
 122  122          mutex_enter(&spa_namespace_lock);
 123  123          nvpair = NULL;
 124  124          while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
 125  125                  if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
 126  126                          continue;
 127  127  
 128  128                  child = fnvpair_value_nvlist(nvpair);
 129  129  
 130  130                  if (spa_lookup(nvpair_name(nvpair)) != NULL)
 131  131                          continue;
 132  132                  (void) spa_add(nvpair_name(nvpair), child, NULL);
 133  133          }
 134  134          mutex_exit(&spa_namespace_lock);
 135  135  
 136  136          nvlist_free(nvlist);
 137  137  
 138  138  out:
 139  139          if (buf != NULL)
 140  140                  kmem_free(buf, fsize);
 141  141  
 142  142          kobj_close_file(file);
 143  143  }
 144  144  
 145  145  static int
 146  146  spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
 147  147  {
 148  148          size_t buflen;
 149  149          char *buf;
 150  150          vnode_t *vp;
 151  151          int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
 152  152          char *temp;
 153  153          int err;
 154  154  
 155  155          /*
 156  156           * If the nvlist is empty (NULL), then remove the old cachefile.
 157  157           */
 158  158          if (nvl == NULL) {
 159  159                  err = vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
 160  160                  return (err);
 161  161          }
 162  162  
 163  163          /*
 164  164           * Pack the configuration into a buffer.
 165  165           */
 166  166          buf = fnvlist_pack(nvl, &buflen);
 167  167          temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
 168  168  
 169  169          /*
 170  170           * Write the configuration to disk.  We need to do the traditional
 171  171           * 'write to temporary file, sync, move over original' to make sure we
 172  172           * always have a consistent view of the data.
 173  173           */
 174  174          (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
 175  175  
 176  176          err = vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0);
 177  177          if (err == 0) {
 178  178                  err = vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
 179  179                      0, RLIM64_INFINITY, kcred, NULL);
 180  180                  if (err == 0)
 181  181                          err = VOP_FSYNC(vp, FSYNC, kcred, NULL);
 182  182                  if (err == 0)
 183  183                          err = vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
 184  184                  (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
 185  185                  VN_RELE(vp);
 186  186          }
 187  187  
 188  188          (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
 189  189  
 190  190          fnvlist_pack_free(buf, buflen);
 191  191          kmem_free(temp, MAXPATHLEN);
 192  192          return (err);
 193  193  }
 194  194  
 195  195  /*
 196  196   * Synchronize pool configuration to disk.  This must be called with the
 197  197   * namespace lock held. Synchronizing the pool cache is typically done after
 198  198   * the configuration has been synced to the MOS. This exposes a window where
 199  199   * the MOS config will have been updated but the cache file has not. If
 200  200   * the system were to crash at that instant then the cached config may not
  
    | 
      ↓ open down ↓ | 
    200 lines elided | 
    
      ↑ open up ↑ | 
  
 201  201   * contain the correct information to open the pool and an explicit import
 202  202   * would be required.
 203  203   */
 204  204  void
 205  205  spa_write_cachefile(spa_t *target, boolean_t removing, boolean_t postsysevent)
 206  206  {
 207  207          spa_config_dirent_t *dp, *tdp;
 208  208          nvlist_t *nvl;
 209  209          boolean_t ccw_failure;
 210  210          int error;
      211 +        char *pool_name;
 211  212  
 212  213          ASSERT(MUTEX_HELD(&spa_namespace_lock));
 213  214  
 214  215          if (rootdir == NULL || !(spa_mode_global & FWRITE))
 215  216                  return;
 216  217  
 217  218          /*
 218  219           * Iterate over all cachefiles for the pool, past or present.  When the
 219  220           * cachefile is changed, the new one is pushed onto this list, allowing
 220  221           * us to update previous cachefiles that no longer contain this pool.
 221  222           */
 222  223          ccw_failure = B_FALSE;
 223  224          for (dp = list_head(&target->spa_config_list); dp != NULL;
 224  225              dp = list_next(&target->spa_config_list, dp)) {
 225  226                  spa_t *spa = NULL;
 226  227                  if (dp->scd_path == NULL)
 227  228                          continue;
 228  229  
 229  230                  /*
 230  231                   * Iterate over all pools, adding any matching pools to 'nvl'.
 231  232                   */
 232  233                  nvl = NULL;
 233  234                  while ((spa = spa_next(spa)) != NULL) {
 234  235                          /*
 235  236                           * Skip over our own pool if we're about to remove
 236  237                           * ourselves from the spa namespace or any pool that
 237  238                           * is readonly. Since we cannot guarantee that a
 238  239                           * readonly pool would successfully import upon reboot,
 239  240                           * we don't allow them to be written to the cache file.
 240  241                           */
 241  242                          if ((spa == target && removing) ||
 242  243                              !spa_writeable(spa))
 243  244                                  continue;
 244  245  
 245  246                          mutex_enter(&spa->spa_props_lock);
 246  247                          tdp = list_head(&spa->spa_config_list);
  
    | 
      ↓ open down ↓ | 
    26 lines elided | 
    
      ↑ open up ↑ | 
  
 247  248                          if (spa->spa_config == NULL ||
 248  249                              tdp->scd_path == NULL ||
 249  250                              strcmp(tdp->scd_path, dp->scd_path) != 0) {
 250  251                                  mutex_exit(&spa->spa_props_lock);
 251  252                                  continue;
 252  253                          }
 253  254  
 254  255                          if (nvl == NULL)
 255  256                                  nvl = fnvlist_alloc();
 256  257  
 257      -                        fnvlist_add_nvlist(nvl, spa->spa_name,
      258 +                        if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
      259 +                                pool_name = fnvlist_lookup_string(
      260 +                                    spa->spa_config, ZPOOL_CONFIG_POOL_NAME);
      261 +                        } else {
      262 +                                pool_name = spa_name(spa);
      263 +                        }
      264 +
      265 +                        fnvlist_add_nvlist(nvl, pool_name,
 258  266                              spa->spa_config);
 259  267                          mutex_exit(&spa->spa_props_lock);
 260  268                  }
 261  269  
 262  270                  error = spa_config_write(dp, nvl);
 263  271                  if (error != 0)
 264  272                          ccw_failure = B_TRUE;
 265  273                  nvlist_free(nvl);
 266  274          }
 267  275  
 268  276          if (ccw_failure) {
 269  277                  /*
 270  278                   * Keep trying so that configuration data is
 271  279                   * written if/when any temporary filesystem
 272  280                   * resource issues are resolved.
 273  281                   */
 274  282                  if (target->spa_ccw_fail_time == 0) {
 275  283                          zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
 276  284                              target, NULL, NULL, 0, 0);
 277  285                  }
 278  286                  target->spa_ccw_fail_time = gethrtime();
 279  287                  spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
 280  288          } else {
 281  289                  /*
 282  290                   * Do not rate limit future attempts to update
 283  291                   * the config cache.
 284  292                   */
 285  293                  target->spa_ccw_fail_time = 0;
 286  294          }
 287  295  
 288  296          /*
 289  297           * Remove any config entries older than the current one.
 290  298           */
 291  299          dp = list_head(&target->spa_config_list);
 292  300          while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
 293  301                  list_remove(&target->spa_config_list, tdp);
 294  302                  if (tdp->scd_path != NULL)
 295  303                          spa_strfree(tdp->scd_path);
 296  304                  kmem_free(tdp, sizeof (spa_config_dirent_t));
 297  305          }
 298  306  
 299  307          spa_config_generation++;
 300  308  
 301  309          if (postsysevent)
 302  310                  spa_event_notify(target, NULL, NULL, ESC_ZFS_CONFIG_SYNC);
 303  311  }
 304  312  
 305  313  /*
 306  314   * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
 307  315   * and we don't want to allow the local zone to see all the pools anyway.
 308  316   * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
 309  317   * information for all pool visible within the zone.
 310  318   */
 311  319  nvlist_t *
 312  320  spa_all_configs(uint64_t *generation)
 313  321  {
 314  322          nvlist_t *pools;
 315  323          spa_t *spa = NULL;
 316  324  
 317  325          if (*generation == spa_config_generation)
 318  326                  return (NULL);
 319  327  
 320  328          pools = fnvlist_alloc();
 321  329  
 322  330          mutex_enter(&spa_namespace_lock);
 323  331          while ((spa = spa_next(spa)) != NULL) {
 324  332                  if (INGLOBALZONE(curproc) ||
 325  333                      zone_dataset_visible(spa_name(spa), NULL)) {
 326  334                          mutex_enter(&spa->spa_props_lock);
 327  335                          fnvlist_add_nvlist(pools, spa_name(spa),
 328  336                              spa->spa_config);
 329  337                          mutex_exit(&spa->spa_props_lock);
 330  338                  }
 331  339          }
 332  340          *generation = spa_config_generation;
 333  341          mutex_exit(&spa_namespace_lock);
 334  342  
 335  343          return (pools);
 336  344  }
 337  345  
 338  346  void
 339  347  spa_config_set(spa_t *spa, nvlist_t *config)
 340  348  {
 341  349          mutex_enter(&spa->spa_props_lock);
 342  350          if (spa->spa_config != NULL && spa->spa_config != config)
 343  351                  nvlist_free(spa->spa_config);
 344  352          spa->spa_config = config;
 345  353          mutex_exit(&spa->spa_props_lock);
 346  354  }
 347  355  
 348  356  /*
 349  357   * Generate the pool's configuration based on the current in-core state.
 350  358   *
 351  359   * We infer whether to generate a complete config or just one top-level config
  
    | 
      ↓ open down ↓ | 
    84 lines elided | 
    
      ↑ open up ↑ | 
  
 352  360   * based on whether vd is the root vdev.
 353  361   */
 354  362  nvlist_t *
 355  363  spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
 356  364  {
 357  365          nvlist_t *config, *nvroot;
 358  366          vdev_t *rvd = spa->spa_root_vdev;
 359  367          unsigned long hostid = 0;
 360  368          boolean_t locked = B_FALSE;
 361  369          uint64_t split_guid;
      370 +        char *pool_name;
 362  371  
 363  372          if (vd == NULL) {
 364  373                  vd = rvd;
 365  374                  locked = B_TRUE;
 366  375                  spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
 367  376          }
 368  377  
 369  378          ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
 370  379              (SCL_CONFIG | SCL_STATE));
 371  380  
 372  381          /*
 373  382           * If txg is -1, report the current value of spa->spa_config_txg.
 374  383           */
 375  384          if (txg == -1ULL)
 376  385                  txg = spa->spa_config_txg;
 377  386  
      387 +        /*
      388 +         * Originally, users had to handle spa namespace collisions by either
      389 +         * exporting the already imported pool or by specifying a new name for
      390 +         * the pool with a conflicting name. In the case of root pools from
      391 +         * virtual guests, neither approach to collision resolution is
      392 +         * reasonable. This is addressed by extending the new name syntax with
      393 +         * an option to specify that the new name is temporary. When specified,
      394 +         * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
      395 +         * to use the previous name, which we do below.
      396 +         */
      397 +        if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
      398 +                pool_name = fnvlist_lookup_string(spa->spa_config,
      399 +                    ZPOOL_CONFIG_POOL_NAME);
      400 +        } else {
      401 +                pool_name = spa_name(spa);
      402 +        }
      403 +
 378  404          config = fnvlist_alloc();
 379  405  
 380  406          fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa));
 381      -        fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, spa_name(spa));
      407 +        fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, pool_name);
 382  408          fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa));
 383  409          fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg);
 384  410          fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa));
 385  411          if (spa->spa_comment != NULL) {
 386  412                  fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT,
 387  413                      spa->spa_comment);
 388  414          }
 389  415  
 390  416          hostid = zone_get_hostid(NULL);
 391  417  
 392  418          if (hostid != 0) {
 393  419                  fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid);
 394  420          }
 395  421          fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname.nodename);
 396  422  
 397  423          int config_gen_flags = 0;
 398  424          if (vd != rvd) {
 399  425                  fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
 400  426                      vd->vdev_top->vdev_guid);
 401  427                  fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
 402  428                      vd->vdev_guid);
 403  429                  if (vd->vdev_isspare) {
 404  430                          fnvlist_add_uint64(config,
 405  431                              ZPOOL_CONFIG_IS_SPARE, 1ULL);
 406  432                  }
 407  433                  if (vd->vdev_islog) {
 408  434                          fnvlist_add_uint64(config,
 409  435                              ZPOOL_CONFIG_IS_LOG, 1ULL);
 410  436                  }
 411  437                  vd = vd->vdev_top;              /* label contains top config */
 412  438          } else {
 413  439                  /*
 414  440                   * Only add the (potentially large) split information
 415  441                   * in the mos config, and not in the vdev labels
 416  442                   */
 417  443                  if (spa->spa_config_splitting != NULL)
 418  444                          fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
 419  445                              spa->spa_config_splitting);
 420  446                  fnvlist_add_boolean(config,
 421  447                      ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS);
 422  448  
 423  449                  config_gen_flags |= VDEV_CONFIG_MOS;
 424  450          }
 425  451  
 426  452          /*
 427  453           * Add the top-level config.  We even add this on pools which
 428  454           * don't support holes in the namespace.
 429  455           */
 430  456          vdev_top_config_generate(spa, config);
 431  457  
 432  458          /*
 433  459           * If we're splitting, record the original pool's guid.
 434  460           */
 435  461          if (spa->spa_config_splitting != NULL &&
 436  462              nvlist_lookup_uint64(spa->spa_config_splitting,
 437  463              ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
 438  464                  fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
 439  465                      split_guid);
 440  466          }
 441  467  
 442  468          nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags);
 443  469          fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot);
 444  470          nvlist_free(nvroot);
 445  471  
 446  472          /*
 447  473           * Store what's necessary for reading the MOS in the label.
 448  474           */
 449  475          fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
 450  476              spa->spa_label_features);
 451  477  
 452  478          if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
 453  479                  ddt_histogram_t *ddh;
 454  480                  ddt_stat_t *dds;
 455  481                  ddt_object_t *ddo;
 456  482  
 457  483                  ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
 458  484                  ddt_get_dedup_histogram(spa, ddh);
 459  485                  fnvlist_add_uint64_array(config,
 460  486                      ZPOOL_CONFIG_DDT_HISTOGRAM,
 461  487                      (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t));
 462  488                  kmem_free(ddh, sizeof (ddt_histogram_t));
 463  489  
 464  490                  ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
 465  491                  ddt_get_dedup_object_stats(spa, ddo);
 466  492                  fnvlist_add_uint64_array(config,
 467  493                      ZPOOL_CONFIG_DDT_OBJ_STATS,
 468  494                      (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t));
 469  495                  kmem_free(ddo, sizeof (ddt_object_t));
 470  496  
 471  497                  dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
 472  498                  ddt_get_dedup_stats(spa, dds);
 473  499                  fnvlist_add_uint64_array(config,
 474  500                      ZPOOL_CONFIG_DDT_STATS,
 475  501                      (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t));
 476  502                  kmem_free(dds, sizeof (ddt_stat_t));
 477  503          }
 478  504  
 479  505          if (locked)
 480  506                  spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
 481  507  
 482  508          return (config);
 483  509  }
 484  510  
 485  511  /*
 486  512   * Update all disk labels, generate a fresh config based on the current
 487  513   * in-core state, and sync the global config cache (do not sync the config
 488  514   * cache if this is a booting rootpool).
 489  515   */
 490  516  void
 491  517  spa_config_update(spa_t *spa, int what)
 492  518  {
 493  519          vdev_t *rvd = spa->spa_root_vdev;
 494  520          uint64_t txg;
 495  521          int c;
 496  522  
 497  523          ASSERT(MUTEX_HELD(&spa_namespace_lock));
 498  524  
 499  525          spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
 500  526          txg = spa_last_synced_txg(spa) + 1;
 501  527          if (what == SPA_CONFIG_UPDATE_POOL) {
 502  528                  vdev_config_dirty(rvd);
 503  529          } else {
 504  530                  /*
 505  531                   * If we have top-level vdevs that were added but have
 506  532                   * not yet been prepared for allocation, do that now.
 507  533                   * (It's safe now because the config cache is up to date,
 508  534                   * so it will be able to translate the new DVAs.)
 509  535                   * See comments in spa_vdev_add() for full details.
 510  536                   */
 511  537                  for (c = 0; c < rvd->vdev_children; c++) {
 512  538                          vdev_t *tvd = rvd->vdev_child[c];
 513  539  
 514  540                          /*
 515  541                           * Explicitly skip vdevs that are indirect or
 516  542                           * log vdevs that are being removed. The reason
 517  543                           * is that both of those can have vdev_ms_array
 518  544                           * set to 0 and we wouldn't want to change their
 519  545                           * metaslab size nor call vdev_expand() on them.
 520  546                           */
 521  547                          if (!vdev_is_concrete(tvd) ||
 522  548                              (tvd->vdev_islog && tvd->vdev_removing))
 523  549                                  continue;
 524  550  
 525  551                          if (tvd->vdev_ms_array == 0)
 526  552                                  vdev_metaslab_set_size(tvd);
 527  553                          vdev_expand(tvd, txg);
 528  554                  }
 529  555          }
 530  556          spa_config_exit(spa, SCL_ALL, FTAG);
 531  557  
 532  558          /*
 533  559           * Wait for the mosconfig to be regenerated and synced.
 534  560           */
 535  561          txg_wait_synced(spa->spa_dsl_pool, txg);
 536  562  
 537  563          /*
 538  564           * Update the global config cache to reflect the new mosconfig.
 539  565           */
 540  566          if (!spa->spa_is_root) {
 541  567                  spa_write_cachefile(spa, B_FALSE,
 542  568                      what != SPA_CONFIG_UPDATE_POOL);
 543  569          }
 544  570  
 545  571          if (what == SPA_CONFIG_UPDATE_POOL)
 546  572                  spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
 547  573  }
  
    | 
      ↓ open down ↓ | 
    156 lines elided | 
    
      ↑ open up ↑ | 
  
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX