spa_config.c revision 249209
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/zfs_context.h> 29#include <sys/spa.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/sunddi.h> 38#include <sys/zfeature.h> 39#ifdef _KERNEL 40#include <sys/kobj.h> 41#include <sys/zone.h> 42#endif 43 44/* 45 * Pool configuration repository. 46 * 47 * Pool configuration is stored as a packed nvlist on the filesystem. By 48 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot 49 * (when the ZFS module is loaded). Pools can also have the 'cachefile' 50 * property set that allows them to be stored in an alternate location until 51 * the control of external software. 52 * 53 * For each cache file, we have a single nvlist which holds all the 54 * configuration information. When the module loads, we read this information 55 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is 56 * maintained independently in spa.c. Whenever the namespace is modified, or 57 * the configuration of a pool is changed, we call spa_config_sync(), which 58 * walks through all the active pools and writes the configuration to disk. 59 */ 60 61static uint64_t spa_config_generation = 1; 62 63/* 64 * This can be overridden in userland to preserve an alternate namespace for 65 * userland pools when doing testing. 66 */ 67const char *spa_config_path = ZPOOL_CACHE; 68 69/* 70 * Called when the module is first loaded, this routine loads the configuration 71 * file into the SPA namespace. It does not actually open or load the pools; it 72 * only populates the namespace. 73 */ 74void 75spa_config_load(void) 76{ 77 void *buf = NULL; 78 nvlist_t *nvlist, *child; 79 nvpair_t *nvpair; 80 char *pathname; 81 struct _buf *file; 82 uint64_t fsize; 83 84 /* 85 * Open the configuration file. 86 */ 87 pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 88 89 (void) snprintf(pathname, MAXPATHLEN, "%s", spa_config_path); 90 91 file = kobj_open_file(pathname); 92 93 kmem_free(pathname, MAXPATHLEN); 94 95 if (file == (struct _buf *)-1) 96 return; 97 98 if (kobj_get_filesize(file, &fsize) != 0) 99 goto out; 100 101 buf = kmem_alloc(fsize, KM_SLEEP); 102 103 /* 104 * Read the nvlist from the file. 105 */ 106 if (kobj_read_file(file, buf, fsize, 0) < 0) 107 goto out; 108 109 /* 110 * Unpack the nvlist. 111 */ 112 if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0) 113 goto out; 114 115 /* 116 * Iterate over all elements in the nvlist, creating a new spa_t for 117 * each one with the specified configuration. 118 */ 119 mutex_enter(&spa_namespace_lock); 120 nvpair = NULL; 121 while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) { 122 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST) 123 continue; 124 125 VERIFY(nvpair_value_nvlist(nvpair, &child) == 0); 126 127 if (spa_lookup(nvpair_name(nvpair)) != NULL) 128 continue; 129 (void) spa_add(nvpair_name(nvpair), child, NULL); 130 } 131 mutex_exit(&spa_namespace_lock); 132 133 nvlist_free(nvlist); 134 135out: 136 if (buf != NULL) 137 kmem_free(buf, fsize); 138 139 kobj_close_file(file); 140} 141 142static void 143spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl) 144{ 145 size_t buflen; 146 char *buf; 147 vnode_t *vp; 148 int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX; 149 char *temp; 150 151 /* 152 * If the nvlist is empty (NULL), then remove the old cachefile. 153 */ 154 if (nvl == NULL) { 155 (void) vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE); 156 return; 157 } 158 159 /* 160 * Pack the configuration into a buffer. 161 */ 162 VERIFY(nvlist_size(nvl, &buflen, NV_ENCODE_XDR) == 0); 163 164 buf = kmem_alloc(buflen, KM_SLEEP); 165 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 166 167 VERIFY(nvlist_pack(nvl, &buf, &buflen, NV_ENCODE_XDR, 168 KM_SLEEP) == 0); 169 170 /* 171 * Write the configuration to disk. We need to do the traditional 172 * 'write to temporary file, sync, move over original' to make sure we 173 * always have a consistent view of the data. 174 */ 175 (void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path); 176 177 if (vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0) == 0) { 178 if (vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE, 179 0, RLIM64_INFINITY, kcred, NULL) == 0 && 180 VOP_FSYNC(vp, FSYNC, kcred, NULL) == 0) { 181 (void) vn_rename(temp, dp->scd_path, UIO_SYSSPACE); 182 } 183 (void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL); 184 } 185 186 (void) vn_remove(temp, UIO_SYSSPACE, RMFILE); 187 188 kmem_free(buf, buflen); 189 kmem_free(temp, MAXPATHLEN); 190} 191 192/* 193 * Synchronize pool configuration to disk. This must be called with the 194 * namespace lock held. 195 */ 196void 197spa_config_sync(spa_t *target, boolean_t removing, boolean_t postsysevent) 198{ 199 spa_config_dirent_t *dp, *tdp; 200 nvlist_t *nvl; 201 202 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 203 204 if (rootdir == NULL || !(spa_mode_global & FWRITE)) 205 return; 206 207 /* 208 * Iterate over all cachefiles for the pool, past or present. When the 209 * cachefile is changed, the new one is pushed onto this list, allowing 210 * us to update previous cachefiles that no longer contain this pool. 211 */ 212 for (dp = list_head(&target->spa_config_list); dp != NULL; 213 dp = list_next(&target->spa_config_list, dp)) { 214 spa_t *spa = NULL; 215 if (dp->scd_path == NULL) 216 continue; 217 218 /* 219 * Iterate over all pools, adding any matching pools to 'nvl'. 220 */ 221 nvl = NULL; 222 while ((spa = spa_next(spa)) != NULL) { 223 /* 224 * Skip over our own pool if we're about to remove 225 * ourselves from the spa namespace or any pool that 226 * is readonly. Since we cannot guarantee that a 227 * readonly pool would successfully import upon reboot, 228 * we don't allow them to be written to the cache file. 229 */ 230 if ((spa == target && removing) || 231 !spa_writeable(spa)) 232 continue; 233 234 mutex_enter(&spa->spa_props_lock); 235 tdp = list_head(&spa->spa_config_list); 236 if (spa->spa_config == NULL || 237 tdp->scd_path == NULL || 238 strcmp(tdp->scd_path, dp->scd_path) != 0) { 239 mutex_exit(&spa->spa_props_lock); 240 continue; 241 } 242 243 if (nvl == NULL) 244 VERIFY(nvlist_alloc(&nvl, NV_UNIQUE_NAME, 245 KM_SLEEP) == 0); 246 247 VERIFY(nvlist_add_nvlist(nvl, spa->spa_name, 248 spa->spa_config) == 0); 249 mutex_exit(&spa->spa_props_lock); 250 } 251 252 spa_config_write(dp, nvl); 253 nvlist_free(nvl); 254 } 255 256 /* 257 * Remove any config entries older than the current one. 258 */ 259 dp = list_head(&target->spa_config_list); 260 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) { 261 list_remove(&target->spa_config_list, tdp); 262 if (tdp->scd_path != NULL) 263 spa_strfree(tdp->scd_path); 264 kmem_free(tdp, sizeof (spa_config_dirent_t)); 265 } 266 267 spa_config_generation++; 268 269 if (postsysevent) 270 spa_event_notify(target, NULL, ESC_ZFS_CONFIG_SYNC); 271} 272 273/* 274 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache, 275 * and we don't want to allow the local zone to see all the pools anyway. 276 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration 277 * information for all pool visible within the zone. 278 */ 279nvlist_t * 280spa_all_configs(uint64_t *generation) 281{ 282 nvlist_t *pools; 283 spa_t *spa = NULL; 284 285 if (*generation == spa_config_generation) 286 return (NULL); 287 288 VERIFY(nvlist_alloc(&pools, NV_UNIQUE_NAME, KM_SLEEP) == 0); 289 290 mutex_enter(&spa_namespace_lock); 291 while ((spa = spa_next(spa)) != NULL) { 292 if (INGLOBALZONE(curthread) || 293 zone_dataset_visible(spa_name(spa), NULL)) { 294 mutex_enter(&spa->spa_props_lock); 295 VERIFY(nvlist_add_nvlist(pools, spa_name(spa), 296 spa->spa_config) == 0); 297 mutex_exit(&spa->spa_props_lock); 298 } 299 } 300 *generation = spa_config_generation; 301 mutex_exit(&spa_namespace_lock); 302 303 return (pools); 304} 305 306void 307spa_config_set(spa_t *spa, nvlist_t *config) 308{ 309 mutex_enter(&spa->spa_props_lock); 310 if (spa->spa_config != NULL) 311 nvlist_free(spa->spa_config); 312 spa->spa_config = config; 313 mutex_exit(&spa->spa_props_lock); 314} 315 316/* 317 * Generate the pool's configuration based on the current in-core state. 318 * We infer whether to generate a complete config or just one top-level config 319 * based on whether vd is the root vdev. 320 */ 321nvlist_t * 322spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats) 323{ 324 nvlist_t *config, *nvroot; 325 vdev_t *rvd = spa->spa_root_vdev; 326 unsigned long hostid = 0; 327 boolean_t locked = B_FALSE; 328 uint64_t split_guid; 329 330 if (vd == NULL) { 331 vd = rvd; 332 locked = B_TRUE; 333 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 334 } 335 336 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) == 337 (SCL_CONFIG | SCL_STATE)); 338 339 /* 340 * If txg is -1, report the current value of spa->spa_config_txg. 341 */ 342 if (txg == -1ULL) 343 txg = spa->spa_config_txg; 344 345 VERIFY(nvlist_alloc(&config, NV_UNIQUE_NAME, KM_SLEEP) == 0); 346 347 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, 348 spa_version(spa)) == 0); 349 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, 350 spa_name(spa)) == 0); 351 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, 352 spa_state(spa)) == 0); 353 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, 354 txg) == 0); 355 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, 356 spa_guid(spa)) == 0); 357 VERIFY(spa->spa_comment == NULL || nvlist_add_string(config, 358 ZPOOL_CONFIG_COMMENT, spa->spa_comment) == 0); 359 360 361#ifdef _KERNEL 362 hostid = zone_get_hostid(NULL); 363#else /* _KERNEL */ 364 /* 365 * We're emulating the system's hostid in userland, so we can't use 366 * zone_get_hostid(). 367 */ 368 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid); 369#endif /* _KERNEL */ 370 if (hostid != 0) { 371 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, 372 hostid) == 0); 373 } 374 VERIFY(nvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, 375 utsname.nodename) == 0); 376 377 if (vd != rvd) { 378 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID, 379 vd->vdev_top->vdev_guid) == 0); 380 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_GUID, 381 vd->vdev_guid) == 0); 382 if (vd->vdev_isspare) 383 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_SPARE, 384 1ULL) == 0); 385 if (vd->vdev_islog) 386 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_IS_LOG, 387 1ULL) == 0); 388 vd = vd->vdev_top; /* label contains top config */ 389 } else { 390 /* 391 * Only add the (potentially large) split information 392 * in the mos config, and not in the vdev labels 393 */ 394 if (spa->spa_config_splitting != NULL) 395 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT, 396 spa->spa_config_splitting) == 0); 397 } 398 399 /* 400 * Add the top-level config. We even add this on pools which 401 * don't support holes in the namespace. 402 */ 403 vdev_top_config_generate(spa, config); 404 405 /* 406 * If we're splitting, record the original pool's guid. 407 */ 408 if (spa->spa_config_splitting != NULL && 409 nvlist_lookup_uint64(spa->spa_config_splitting, 410 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) { 411 VERIFY(nvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID, 412 split_guid) == 0); 413 } 414 415 nvroot = vdev_config_generate(spa, vd, getstats, 0); 416 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot) == 0); 417 nvlist_free(nvroot); 418 419 /* 420 * Store what's necessary for reading the MOS in the label. 421 */ 422 VERIFY(nvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, 423 spa->spa_label_features) == 0); 424 425 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) { 426 ddt_histogram_t *ddh; 427 ddt_stat_t *dds; 428 ddt_object_t *ddo; 429 430 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP); 431 ddt_get_dedup_histogram(spa, ddh); 432 VERIFY(nvlist_add_uint64_array(config, 433 ZPOOL_CONFIG_DDT_HISTOGRAM, 434 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)) == 0); 435 kmem_free(ddh, sizeof (ddt_histogram_t)); 436 437 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP); 438 ddt_get_dedup_object_stats(spa, ddo); 439 VERIFY(nvlist_add_uint64_array(config, 440 ZPOOL_CONFIG_DDT_OBJ_STATS, 441 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)) == 0); 442 kmem_free(ddo, sizeof (ddt_object_t)); 443 444 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP); 445 ddt_get_dedup_stats(spa, dds); 446 VERIFY(nvlist_add_uint64_array(config, 447 ZPOOL_CONFIG_DDT_STATS, 448 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)) == 0); 449 kmem_free(dds, sizeof (ddt_stat_t)); 450 } 451 452 if (locked) 453 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 454 455 return (config); 456} 457 458/* 459 * Update all disk labels, generate a fresh config based on the current 460 * in-core state, and sync the global config cache (do not sync the config 461 * cache if this is a booting rootpool). 462 */ 463void 464spa_config_update(spa_t *spa, int what) 465{ 466 vdev_t *rvd = spa->spa_root_vdev; 467 uint64_t txg; 468 int c; 469 470 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 471 472 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 473 txg = spa_last_synced_txg(spa) + 1; 474 if (what == SPA_CONFIG_UPDATE_POOL) { 475 vdev_config_dirty(rvd); 476 } else { 477 /* 478 * If we have top-level vdevs that were added but have 479 * not yet been prepared for allocation, do that now. 480 * (It's safe now because the config cache is up to date, 481 * so it will be able to translate the new DVAs.) 482 * See comments in spa_vdev_add() for full details. 483 */ 484 for (c = 0; c < rvd->vdev_children; c++) { 485 vdev_t *tvd = rvd->vdev_child[c]; 486 if (tvd->vdev_ms_array == 0) 487 vdev_metaslab_set_size(tvd); 488 vdev_expand(tvd, txg); 489 } 490 } 491 spa_config_exit(spa, SCL_ALL, FTAG); 492 493 /* 494 * Wait for the mosconfig to be regenerated and synced. 495 */ 496 txg_wait_synced(spa->spa_dsl_pool, txg); 497 498 /* 499 * Update the global config cache to reflect the new mosconfig. 500 */ 501 if (!spa->spa_is_root) 502 spa_config_sync(spa, B_FALSE, what != SPA_CONFIG_UPDATE_POOL); 503 504 if (what == SPA_CONFIG_UPDATE_POOL) 505 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS); 506} 507