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) 2012 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.
 202  */
 203 void
 204 spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent)
 205 {
 206         spa_config_dirent_t *dp, *tdp;
 207         nvlist_t *nvl;
 208         boolean_t ccw_failure;
 209         int error;
 210 
 211         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 212 
 213         if (rootdir == NULL || !(spa_mode_global & FWRITE))
 214                 return;
 215 
 216         /*
 217          * Iterate over all cachefiles for the pool, past or present.  When the
 218          * cachefile is changed, the new one is pushed onto this list, allowing
 219          * us to update previous cachefiles that no longer contain this pool.
 220          */
 221         ccw_failure = B_FALSE;
 222         for (dp = list_head(&target->spa_config_list); dp != NULL;
 223             dp = list_next(&target->spa_config_list, dp)) {
 224                 spa_t *spa = NULL;
 225                 if (dp->scd_path == NULL)
 226                         continue;
 227 
 228                 /*
 229                  * Iterate over all pools, adding any matching pools to 'nvl'.
 230                  */
 231                 nvl = NULL;
 232                 while ((spa = spa_next(spa)) != NULL) {
 233                         /*
 234                          * Skip over our own pool if we're about to remove
 235                          * ourselves from the spa namespace or any pool that
 236                          * is readonly. Since we cannot guarantee that a
 237                          * readonly pool would successfully import upon reboot,
 238                          * we don't allow them to be written to the cache file.
 239                          */
 240                         if ((spa == target && removing) ||
 241                             !spa_writeable(spa))
 242                                 continue;
 243 
 244                         mutex_enter(&spa->spa_props_lock);
 245                         tdp = list_head(&spa->spa_config_list);
 246                         if (spa->spa_config == NULL ||
 247                             tdp->scd_path == NULL ||
 248                             strcmp(tdp->scd_path, dp->scd_path) != 0) {
 249                                 mutex_exit(&spa->spa_props_lock);
 250                                 continue;
 251                         }
 252 
 253                         if (nvl == NULL)
 254                                 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME,
 255                                     KM_SLEEP) == 0);
 256 
 257                         VERIFY(nvlist_add_nvlist(nvl, spa->spa_name,
 258                             spa->spa_config) == 0);
 259                         mutex_exit(&spa->spa_props_lock);
 260                 }
 261 
 262                 error = spa_config_write(dp, nvl);
 263                 if (error != 0)
 264                         ccw_failure = B_TRUE;
 265                 nvlist_free(nvl);
 266         }
 267 
 268         if (ccw_failure) {
 269                 /*
 270                  * Keep trying so that configuration data is
 271                  * written if/when any temporary filesystem
 272                  * resource issues are resolved.
 273                  */
 274                 target->spa_ccw_fail_time = ddi_get_lbolt64();
 275                 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
 276                 zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
 277                     target, NULL, NULL, 0, 0);
 278         } else {
 279                 /*
 280                  * Do not rate limit future attempts to update
 281                  * the config cache.
 282                  */
 283                 target->spa_ccw_fail_time = 0;
 284         }
 285 
 286         /*
 287          * Remove any config entries older than the current one.
 288          */
 289         dp = list_head(&target->spa_config_list);
 290         while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
 291                 list_remove(&target->spa_config_list, tdp);
 292                 if (tdp->scd_path != NULL)
 293                         spa_strfree(tdp->scd_path);
 294                 kmem_free(tdp, sizeof (spa_config_dirent_t));
 295         }
 296 
 297         spa_config_generation++;
 298 
 299         if (postsysevent)
 300                 spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC);
 301 }
 302 
 303 /*
 304  * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
 305  * and we don't want to allow the local zone to see all the pools anyway.
 306  * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
 307  * information for all pool visible within the zone.
 308  */
 309 nvlist_t *
 310 spa_all_configs(uint64_t *generation)
 311 {
 312         nvlist_t *pools;
 313         spa_t *spa = NULL;
 314 
 315         if (*generation == spa_config_generation)
 316                 return (NULL);
 317 
 318         VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 319 
 320         mutex_enter(&spa_namespace_lock);
 321         while ((spa = spa_next(spa)) != NULL) {
 322                 if (INGLOBALZONE(curproc) ||
 323                     zone_dataset_visible(spa_name(spa), NULL)) {
 324                         mutex_enter(&spa->spa_props_lock);
 325                         VERIFY(nvlist_add_nvlist(pools, spa_name(spa),
 326                             spa->spa_config) == 0);
 327                         mutex_exit(&spa->spa_props_lock);
 328                 }
 329         }
 330         *generation = spa_config_generation;
 331         mutex_exit(&spa_namespace_lock);
 332 
 333         return (pools);
 334 }
 335 
 336 void
 337 spa_config_set(spa_t *spa, nvlist_t *config)
 338 {
 339         mutex_enter(&spa->spa_props_lock);
 340         if (spa->spa_config != NULL)
 341                 nvlist_free(spa->spa_config);
 342         spa->spa_config = config;
 343         mutex_exit(&spa->spa_props_lock);
 344 }
 345 
 346 /*
 347  * Generate the pool's configuration based on the current in-core state.
 348  * We infer whether to generate a complete config or just one top-level config
 349  * based on whether vd is the root vdev.
 350  */
 351 nvlist_t *
 352 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
 353 {
 354         nvlist_t *config, *nvroot;
 355         vdev_t *rvd = spa->spa_root_vdev;
 356         unsigned long hostid = 0;
 357         boolean_t locked = B_FALSE;
 358         uint64_t split_guid;
 359 
 360         if (vd == NULL) {
 361                 vd = rvd;
 362                 locked = B_TRUE;
 363                 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
 364         }
 365 
 366         ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
 367             (SCL_CONFIG | SCL_STATE));
 368 
 369         /*
 370          * If txg is -1, report the current value of spa->spa_config_txg.
 371          */
 372         if (txg == -1ULL)
 373                 txg = spa->spa_config_txg;
 374 
 375         VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0);
 376 
 377         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION,
 378             spa_version(spa)) == 0);
 379         VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME,
 380             spa_name(spa)) == 0);
 381         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE,
 382             spa_state(spa)) == 0);
 383         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG,
 384             txg) == 0);
 385         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID,
 386             spa_guid(spa)) == 0);
 387         VERIFY(spa->spa_comment == NULL || nvlist_add_string(config,
 388             ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0);
 389 
 390 
 391 #ifdef  _KERNEL
 392         hostid = zone_get_hostid(NULL);
 393 #else   /* _KERNEL */
 394         /*
 395          * We're emulating the system's hostid in userland, so we can't use
 396          * zone_get_hostid().
 397          */
 398         (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
 399 #endif  /* _KERNEL */
 400         if (hostid != 0) {
 401                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID,
 402                     hostid) == 0);
 403         }
 404         VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME,
 405             utsname.nodename) == 0);
 406 
 407         if (vd != rvd) {
 408                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
 409                     vd->vdev_top->vdev_guid) == 0);
 410                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
 411                     vd->vdev_guid) == 0);
 412                 if (vd->vdev_isspare)
 413                         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE,
 414                             1ULL) == 0);
 415                 if (vd->vdev_islog)
 416                         VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG,
 417                             1ULL) == 0);
 418                 vd = vd->vdev_top;           /* label contains top config */
 419         } else {
 420                 /*
 421                  * Only add the (potentially large) split information
 422                  * in the mos config, and not in the vdev labels
 423                  */
 424                 if (spa->spa_config_splitting != NULL)
 425                         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
 426                             spa->spa_config_splitting) == 0);
 427         }
 428 
 429         /*
 430          * Add the top-level config.  We even add this on pools which
 431          * don't support holes in the namespace.
 432          */
 433         vdev_top_config_generate(spa, config);
 434 
 435         /*
 436          * If we're splitting, record the original pool's guid.
 437          */
 438         if (spa->spa_config_splitting != NULL &&
 439             nvlist_lookup_uint64(spa->spa_config_splitting,
 440             ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
 441                 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
 442                     split_guid) == 0);
 443         }
 444 
 445         nvroot = vdev_config_generate(spa, vd, getstats, 0);
 446         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0);
 447         nvlist_free(nvroot);
 448 
 449         /*
 450          * Store what's necessary for reading the MOS in the label.
 451          */
 452         VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
 453             spa->spa_label_features) == 0);
 454 
 455         if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
 456                 ddt_histogram_t *ddh;
 457                 ddt_stat_t *dds;
 458                 ddt_object_t *ddo;
 459 
 460                 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
 461                 ddt_get_dedup_histogram(spa, ddh);
 462                 VERIFY(nvlist_add_uint64_array(config,
 463                     ZPOOL_CONFIG_DDT_HISTOGRAM,
 464                     (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0);
 465                 kmem_free(ddh, sizeof (ddt_histogram_t));
 466 
 467                 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
 468                 ddt_get_dedup_object_stats(spa, ddo);
 469                 VERIFY(nvlist_add_uint64_array(config,
 470                     ZPOOL_CONFIG_DDT_OBJ_STATS,
 471                     (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0);
 472                 kmem_free(ddo, sizeof (ddt_object_t));
 473 
 474                 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
 475                 ddt_get_dedup_stats(spa, dds);
 476                 VERIFY(nvlist_add_uint64_array(config,
 477                     ZPOOL_CONFIG_DDT_STATS,
 478                     (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0);
 479                 kmem_free(dds, sizeof (ddt_stat_t));
 480         }
 481 
 482         if (locked)
 483                 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
 484 
 485         return (config);
 486 }
 487 
 488 /*
 489  * Update all disk labels, generate a fresh config based on the current
 490  * in-core state, and sync the global config cache (do not sync the config
 491  * cache if this is a booting rootpool).
 492  */
 493 void
 494 spa_config_update(spa_t *spa, int what)
 495 {
 496         vdev_t *rvd = spa->spa_root_vdev;
 497         uint64_t txg;
 498         int c;
 499 
 500         ASSERT(MUTEX_HELD(&spa_namespace_lock));
 501 
 502         spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
 503         txg = spa_last_synced_txg(spa) + 1;
 504         if (what == SPA_CONFIG_UPDATE_POOL) {
 505                 vdev_config_dirty(rvd);
 506         } else {
 507                 /*
 508                  * If we have top-level vdevs that were added but have
 509                  * not yet been prepared for allocation, do that now.
 510                  * (It's safe now because the config cache is up to date,
 511                  * so it will be able to translate the new DVAs.)
 512                  * See comments in spa_vdev_add() for full details.
 513                  */
 514                 for (c = 0; c < rvd->vdev_children; c++) {
 515                         vdev_t *tvd = rvd->vdev_child[c];
 516                         if (tvd->vdev_ms_array == 0)
 517                                 vdev_metaslab_set_size(tvd);
 518                         vdev_expand(tvd, txg);
 519                 }
 520         }
 521         spa_config_exit(spa, SCL_ALL, FTAG);
 522 
 523         /*
 524          * Wait for the mosconfig to be regenerated and synced.
 525          */
 526         txg_wait_synced(spa->spa_dsl_pool, txg);
 527 
 528         /*
 529          * Update the global config cache to reflect the new mosconfig.
 530          */
 531         if (!spa->spa_is_root)
 532                 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL);
 533 
 534         if (what == SPA_CONFIG_UPDATE_POOL)
 535                 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
 536 }