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 2011 Nexenta Systems, Inc. All rights reserved.
  25  * Copyright (c) 2013 by Delphix. All rights reserved.
  26  */
  27 
  28 #include <sys/spa.h>
  29 #include <sys/fm/fs/zfs.h>
  30 #include <sys/spa_impl.h>
  31 #include <sys/nvpair.h>
  32 #include <sys/uio.h>
  33 #include <sys/fs/zfs.h>
  34 #include <sys/vdev_impl.h>
  35 #include <sys/zfs_ioctl.h>
  36 #include <sys/utsname.h>
  37 #include <sys/systeminfo.h>
  38 #include <sys/sunddi.h>
  39 #include <sys/zfeature.h>
  40 #ifdef _KERNEL
  41 #include <sys/kobj.h>
  42 #include <sys/zone.h>
  43 #endif
  44 
  45 /*
  46  * Pool configuration repository.
  47  *
  48  * Pool configuration is stored as a packed nvlist on the filesystem.  By
  49  * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
  50  * (when the ZFS module is loaded).  Pools can also have the 'cachefile'
  51  * property set that allows them to be stored in an alternate location until
  52  * the control of external software.
  53  *
  54  * For each cache file, we have a single nvlist which holds all the
  55  * configuration information.  When the module loads, we read this information
  56  * from /etc/zfs/zpool.cache and populate the SPA namespace.  This namespace is
  57  * maintained independently in spa.c.  Whenever the namespace is modified, or
  58  * the configuration of a pool is changed, we call spa_config_sync(), which
  59  * walks through all the active pools and writes the configuration to disk.
  60  */
  61 
  62 static uint64_t spa_config_generation = 1;
  63 
  64 /*
  65  * This can be overridden in userland to preserve an alternate namespace for
  66  * userland pools when doing testing.
  67  */
  68 const char *spa_config_path = ZPOOL_CACHE;
  69 
  70 /*
  71  * Called when the module is first loaded, this routine loads the configuration
  72  * file into the SPA namespace.  It does not actually open or load the pools; it
  73  * only populates the namespace.
  74  */
  75 void
  76 spa_config_load(void)
  77 {
  78         void *buf = NULL;
  79         nvlist_t *nvlist, *child;
  80         nvpair_t *nvpair;
  81         char *pathname;
  82         struct _buf *file;
  83         uint64_t fsize;
  84 
  85         /*
  86          * Open the configuration file.
  87          */
  88         pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
  89 
  90         (void) snprintf(pathname, MAXPATHLEN, "%s%s",
  91             (rootdir != NULL) ? "./" : "", spa_config_path);
  92 
  93         file = kobj_open_file(pathname);
  94 
  95         kmem_free(pathname, MAXPATHLEN);
  96 
  97         if (file == (struct _buf *)-1)
  98                 return;
  99 
 100         if (kobj_get_filesize(file, &fsize) != 0)
 101                 goto out;
 102 
 103         buf = kmem_alloc(fsize, KM_SLEEP);
 104 
 105         /*
 106          * Read the nvlist from the file.
 107          */
 108         if (kobj_read_file(file, buf, fsize, 0) < 0)
 109                 goto out;
 110 
 111         /*
 112          * Unpack the nvlist.
 113          */
 114         if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
 115                 goto out;
 116 
 117         /*
 118          * Iterate over all elements in the nvlist, creating a new spa_t for
 119          * each one with the specified configuration.
 120          */
 121         mutex_enter(&spa_namespace_lock);
 122         nvpair = NULL;
 123         while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
 124                 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
 125                         continue;
 126 
 127                 VERIFY(nvpair_value_nvlist(nvpair, &child) == 0);
 128 
 129                 if (spa_lookup(nvpair_name(nvpair)) != NULL)
 130                         continue;
 131                 (void) spa_add(nvpair_name(nvpair), child, NULL);
 132         }
 133         mutex_exit(&spa_namespace_lock);
 134 
 135         nvlist_free(nvlist);
 136 
 137 out:
 138         if (buf != NULL)
 139                 kmem_free(buf, fsize);
 140 
 141         kobj_close_file(file);
 142 }
 143 
 144 static int
 145 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
 146 {
 147         size_t buflen;
 148         char *buf;
 149         vnode_t *vp;
 150         int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
 151         char *temp;
 152         int err;
 153 
 154         /*
 155          * If the nvlist is empty (NULL), then remove the old cachefile.
 156          */
 157         if (nvl == NULL) {
 158                 err = vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
 159                 return (err);
 160         }
 161 
 162         /*
 163          * Pack the configuration into a buffer.
 164          */
 165         VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0);
 166 
 167         buf = kmem_alloc(buflen, KM_SLEEP);
 168         temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
 169 
 170         VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR,
 171             KM_SLEEP) == 0);
 172 
 173         /*
 174          * Write the configuration to disk.  We need to do the traditional
 175          * 'write to temporary file, sync, move over original' to make sure we
 176          * always have a consistent view of the data.
 177          */
 178         (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
 179 
 180         err = vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0);
 181         if (err == 0) {
 182                 err = vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
 183                     0, RLIM64_INFINITY, kcred, NULL);
 184                 if (err == 0)
 185                         err = VOP_FSYNC(vp, FSYNC, kcred, NULL);
 186                 if (err == 0)
 187                         err = vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
 188                 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
 189                 VN_RELE(vp);
 190         }
 191 
 192         (void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
 193 
 194         kmem_free(buf, buflen);
 195         kmem_free(temp, MAXPATHLEN);
 196         return (err);
 197 }
 198 
 199 /*
 200  * Synchronize pool configuration to disk.  This must be called with the
 201  * namespace lock held. Synchronizing the pool cache is typically done after
 202  * the configuration has been synced to the MOS. This exposes a window where
 203  * the MOS config will have been updated but the cache file has not. If
 204  * the system were to crash at that instant then the cached config may not
 205  * contain the correct information to open the pool and an explicity import
 206  * would be required.
 207  */
 208 void
 209 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
 210 {
 211         spa_config_dirent_t *dp, *tdp;
 212         nvlist_t *nvl;
 213         boolean_t ccw_failure;
 214         int error;
 215 
 216         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 217 
 218         if (rootdir == NULL || !(spa_mode_global & FWRITE))
 219                 return;
 220 
 221         /*
 222          * Iterate over all cachefiles for the pool, past or present.  When the
 223          * cachefile is changed, the new one is pushed onto this list, allowing
 224          * us to update previous cachefiles that no longer contain this pool.
 225          */
 226         ccw_failure = B_FALSE;
 227         for (dp = list_head(&target->spa_config_list); dp != NULL;
 228             dp = list_next(&target->spa_config_list, dp)) {
 229                 spa_t *spa = NULL;
 230                 if (dp->scd_path == NULL)
 231                         continue;
 232 
 233                 /*
 234                  * Iterate over all pools, adding any matching pools to 'nvl'.
 235                  */
 236                 nvl = NULL;
 237                 while ((spa = spa_next(spa)) != NULL) {
 238                         /*
 239                          * Skip over our own pool if we're about to remove
 240                          * ourselves from the spa namespace or any pool that
 241                          * is readonly. Since we cannot guarantee that a
 242                          * readonly pool would successfully import upon reboot,
 243                          * we don't allow them to be written to the cache file.
 244                          */
 245                         if ((spa == target && removing) ||
 246                             !spa_writeable(spa))
 247                                 continue;
 248 
 249                         mutex_enter(&spa->spa_props_lock);
 250                         tdp = list_head(&spa->spa_config_list);
 251                         if (spa->spa_config == NULL ||
 252                             tdp->scd_path == NULL ||
 253                             strcmp(tdp->scd_path, dp->scd_path) != 0) {
 254                                 mutex_exit(&spa->spa_props_lock);
 255                                 continue;
 256                         }
 257 
 258                         if (nvl == NULL)
 259                                 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
 260                                     KM_SLEEP) == 0);
 261 
 262                         VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
 263                             spa->spa_config) == 0);
 264                         mutex_exit(&spa->spa_props_lock);
 265                 }
 266 
 267                 error = spa_config_write(dp, nvl);
 268                 if (error != 0)
 269                         ccw_failure = B_TRUE;
 270                 nvlist_free(nvl);
 271         }
 272 
 273         if (ccw_failure) {
 274                 /*
 275                  * Keep trying so that configuration data is
 276                  * written if/when any temporary filesystem
 277                  * resource issues are resolved.
 278                  */
 279                 if (target->spa_ccw_fail_time == 0) {
 280                         zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
 281                             target, NULL, NULL, 0, 0);
 282                 }
 283                 target->spa_ccw_fail_time = gethrtime();
 284                 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
 285         } else {
 286                 /*
 287                  * Do not rate limit future attempts to update
 288                  * the config cache.
 289                  */
 290                 target->spa_ccw_fail_time = 0;
 291         }
 292 
 293         /*
 294          * Remove any config entries older than the current one.
 295          */
 296         dp = list_head(&target->spa_config_list);
 297         while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
 298                 list_remove(&target->spa_config_list, tdp);
 299                 if (tdp->scd_path != NULL)
 300                         spa_strfree(tdp->scd_path);
 301                 kmem_free(tdp, sizeof (spa_config_dirent_t));
 302         }
 303 
 304         spa_config_generation++;
 305 
 306         if (postsysevent)
 307                 spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
 308 }
 309 
 310 /*
 311  * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
 312  * and we don't want to allow the local zone to see all the pools anyway.
 313  * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
 314  * information for all pool visible within the zone.
 315  */
 316 nvlist_t *
 317 spa_all_configs(uint64_t *generation)
 318 {
 319         nvlist_t *pools;
 320         spa_t *spa = NULL;
 321 
 322         if (*generation == spa_config_generation)
 323                 return (NULL);
 324 
 325         VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 326 
 327         mutex_enter(&spa_namespace_lock);
 328         while ((spa = spa_next(spa)) != NULL) {
 329                 if (INGLOBALZONE(curproc) ||
 330                     zone_dataset_visible(spa_name(spa), NULL)) {
 331                         mutex_enter(&spa->spa_props_lock);
 332                         VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
 333                             spa->spa_config) == 0);
 334                         mutex_exit(&spa->spa_props_lock);
 335                 }
 336         }
 337         *generation = spa_config_generation;
 338         mutex_exit(&spa_namespace_lock);
 339 
 340         return (pools);
 341 }
 342 
 343 void
 344 spa_config_set(spa_t *spa, nvlist_t *config)
 345 {
 346         mutex_enter(&spa->spa_props_lock);
 347         nvlist_free(spa->spa_config);
 348         spa->spa_config = config;
 349         mutex_exit(&spa->spa_props_lock);
 350 }
 351 
 352 /*
 353  * Generate the pool's configuration based on the current in-core state.
 354  *
 355  * We infer whether to generate a complete config or just one top-level config
 356  * based on whether vd is the root vdev.
 357  */
 358 nvlist_t *
 359 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
 360 {
 361         nvlist_t *config, *nvroot;
 362         vdev_t *rvd = spa->spa_root_vdev;
 363         unsigned long hostid = 0;
 364         boolean_t locked = B_FALSE;
 365         uint64_t split_guid;
 366 
 367         if (vd == NULL) {
 368                 vd = rvd;
 369                 locked = B_TRUE;
 370                 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
 371         }
 372 
 373         ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
 374             (SCL_CONFIG | SCL_STATE));
 375 
 376         /*
 377          * If txg is -1, report the current value of spa->spa_config_txg.
 378          */
 379         if (txg == -1ULL)
 380                 txg = spa->spa_config_txg;
 381 
 382         VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 383 
 384         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
 385             spa_version(spa)) == 0);
 386         VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
 387             spa_name(spa)) == 0);
 388         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
 389             spa_state(spa)) == 0);
 390         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
 391             txg) == 0);
 392         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 393             spa_guid(spa)) == 0);
 394         VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
 395             ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
 396 
 397 
 398 #ifdef  _KERNEL
 399         hostid = zone_get_hostid(NULL);
 400 #else   /* _KERNEL */
 401         /*
 402          * We're emulating the system's hostid in userland, so we can't use
 403          * zone_get_hostid().
 404          */
 405         (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
 406 #endif  /* _KERNEL */
 407         if (hostid != 0) {
 408                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 409                     hostid) == 0);
 410         }
 411         VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 412             utsname.nodename) == 0);
 413 
 414         if (vd != rvd) {
 415                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
 416                     vd->vdev_top->vdev_guid) == 0);
 417                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
 418                     vd->vdev_guid) == 0);
 419                 if (vd->vdev_isspare)
 420                         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
 421                             1ULL) == 0);
 422                 if (vd->vdev_islog)
 423                         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
 424                             1ULL) == 0);
 425                 vd = vd->vdev_top;           /* label contains top config */
 426         } else {
 427                 /*
 428                  * Only add the (potentially large) split information
 429                  * in the mos config, and not in the vdev labels
 430                  */
 431                 if (spa->spa_config_splitting != NULL)
 432                         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
 433                             spa->spa_config_splitting) == 0);
 434         }
 435 
 436         /*
 437          * Add the top-level config.  We even add this on pools which
 438          * don't support holes in the namespace.
 439          */
 440         vdev_top_config_generate(spa, config);
 441 
 442         /*
 443          * If we're splitting, record the original pool's guid.
 444          */
 445         if (spa->spa_config_splitting != NULL &&
 446             nvlist_lookup_uint64(spa->spa_config_splitting,
 447             ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
 448                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
 449                     split_guid) == 0);
 450         }
 451 
 452         nvroot = vdev_config_generate(spa, vd, getstats, 0);
 453         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
 454         nvlist_free(nvroot);
 455 
 456         /*
 457          * Store what's necessary for reading the MOS in the label.
 458          */
 459         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
 460             spa->spa_label_features) == 0);
 461 
 462         if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
 463                 ddt_histogram_t *ddh;
 464                 ddt_stat_t *dds;
 465                 ddt_object_t *ddo;
 466 
 467                 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
 468                 ddt_get_dedup_histogram(spa, ddh);
 469                 VERIFY(nvlist_add_uint64_array(config,
 470                     ZPOOL_CONFIG_DDT_HISTOGRAM,
 471                     (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
 472                 kmem_free(ddh, sizeof (ddt_histogram_t));
 473 
 474                 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
 475                 ddt_get_dedup_object_stats(spa, ddo);
 476                 VERIFY(nvlist_add_uint64_array(config,
 477                     ZPOOL_CONFIG_DDT_OBJ_STATS,
 478                     (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
 479                 kmem_free(ddo, sizeof (ddt_object_t));
 480 
 481                 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
 482                 ddt_get_dedup_stats(spa, dds);
 483                 VERIFY(nvlist_add_uint64_array(config,
 484                     ZPOOL_CONFIG_DDT_STATS,
 485                     (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
 486                 kmem_free(dds, sizeof (ddt_stat_t));
 487         }
 488 
 489         if (locked)
 490                 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
 491 
 492         return (config);
 493 }
 494 
 495 /*
 496  * Update all disk labels, generate a fresh config based on the current
 497  * in-core state, and sync the global config cache (do not sync the config
 498  * cache if this is a booting rootpool).
 499  */
 500 void
 501 spa_config_update(spa_t *spa, int what)
 502 {
 503         vdev_t *rvd = spa->spa_root_vdev;
 504         uint64_t txg;
 505         int c;
 506 
 507         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 508 
 509         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
 510         txg = spa_last_synced_txg(spa) + 1;
 511         if (what == SPA_CONFIG_UPDATE_POOL) {
 512                 vdev_config_dirty(rvd);
 513         } else {
 514                 /*
 515                  * If we have top-level vdevs that were added but have
 516                  * not yet been prepared for allocation, do that now.
 517                  * (It's safe now because the config cache is up to date,
 518                  * so it will be able to translate the new DVAs.)
 519                  * See comments in spa_vdev_add() for full details.
 520                  */
 521                 for (c = 0; c < rvd->vdev_children; c++) {
 522                         vdev_t *tvd = rvd->vdev_child[c];
 523                         if (tvd->vdev_ms_array == 0)
 524                                 vdev_metaslab_set_size(tvd);
 525                         vdev_expand(tvd, txg);
 526                 }
 527         }
 528         spa_config_exit(spa, SCL_ALL, FTAG);
 529 
 530         /*
 531          * Wait for the mosconfig to be regenerated and synced.
 532          */
 533         txg_wait_synced(spa->spa_dsl_pool, txg);
 534 
 535         /*
 536          * Update the global config cache to reflect the new mosconfig.
 537          */
 538         if (!spa->spa_is_root)
 539                 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
 540 
 541         if (what == SPA_CONFIG_UPDATE_POOL)
 542                 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
 543 }