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 https://opensource.org/licenses/CDDL-1.0. 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) 2011, 2020 by Delphix. All rights reserved. 26 * Copyright 2017 Joyent, Inc. 27 * Copyright (c) 2021, Colm Buckley <colm@tuatha.org> 28 */ 29 30 #include <sys/spa.h> 31 #include <sys/file.h> 32 #include <sys/fm/fs/zfs.h> 33 #include <sys/spa_impl.h> 34 #include <sys/nvpair.h> 35 #include <sys/fs/zfs.h> 36 #include <sys/vdev_impl.h> 37 #include <sys/zfs_ioctl.h> 38 #include <sys/systeminfo.h> 39 #include <sys/sunddi.h> 40 #include <sys/zfeature.h> 41 #include <sys/zfs_file.h> 42 #include <sys/zfs_context.h> 43 #ifdef _KERNEL 44 #include <sys/zone.h> 45 #endif 46 47 /* 48 * Pool configuration repository. 49 * 50 * Pool configuration is stored as a packed nvlist on the filesystem. By 51 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot 52 * (when the ZFS module is loaded). Pools can also have the 'cachefile' 53 * property set that allows them to be stored in an alternate location until 54 * the control of external software. 55 * 56 * For each cache file, we have a single nvlist which holds all the 57 * configuration information. When the module loads, we read this information 58 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is 59 * maintained independently in spa.c. Whenever the namespace is modified, or 60 * the configuration of a pool is changed, we call spa_write_cachefile(), which 61 * walks through all the active pools and writes the configuration to disk. 62 */ 63 64 static uint64_t spa_config_generation = 1; 65 66 /* 67 * This can be overridden in userland to preserve an alternate namespace for 68 * userland pools when doing testing. 69 */ 70 char *spa_config_path = (char *)ZPOOL_CACHE; 71 #ifdef _KERNEL 72 static int zfs_autoimport_disable = B_TRUE; 73 #endif 74 75 /* 76 * Called when the module is first loaded, this routine loads the configuration 77 * file into the SPA namespace. It does not actually open or load the pools; it 78 * only populates the namespace. 79 */ 80 void 81 spa_config_load(void) 82 { 83 void *buf = NULL; 84 nvlist_t *nvlist, *child; 85 nvpair_t *nvpair; 86 char *pathname; 87 zfs_file_t *fp; 88 zfs_file_attr_t zfa; 89 uint64_t fsize; 90 int err; 91 92 #ifdef _KERNEL 93 if (zfs_autoimport_disable) 94 return; 95 #endif 96 97 /* 98 * Open the configuration file. 99 */ 100 pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 101 102 (void) snprintf(pathname, MAXPATHLEN, "%s", spa_config_path); 103 104 err = zfs_file_open(pathname, O_RDONLY, 0, &fp); 105 106 #ifdef __FreeBSD__ 107 if (err) 108 err = zfs_file_open(ZPOOL_CACHE_BOOT, O_RDONLY, 0, &fp); 109 #endif 110 kmem_free(pathname, MAXPATHLEN); 111 112 if (err) 113 return; 114 115 if (zfs_file_getattr(fp, &zfa)) 116 goto out; 117 118 fsize = zfa.zfa_size; 119 buf = kmem_alloc(fsize, KM_SLEEP); 120 121 /* 122 * Read the nvlist from the file. 123 */ 124 if (zfs_file_read(fp, buf, fsize, NULL) < 0) 125 goto out; 126 127 /* 128 * Unpack the nvlist. 129 */ 130 if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0) 131 goto out; 132 133 /* 134 * Iterate over all elements in the nvlist, creating a new spa_t for 135 * each one with the specified configuration. 136 */ 137 mutex_enter(&spa_namespace_lock); 138 nvpair = NULL; 139 while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) { 140 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST) 141 continue; 142 143 child = fnvpair_value_nvlist(nvpair); 144 145 if (spa_lookup(nvpair_name(nvpair)) != NULL) 146 continue; 147 (void) spa_add(nvpair_name(nvpair), child, NULL); 148 } 149 mutex_exit(&spa_namespace_lock); 150 151 nvlist_free(nvlist); 152 153 out: 154 if (buf != NULL) 155 kmem_free(buf, fsize); 156 157 zfs_file_close(fp); 158 } 159 160 static int 161 spa_config_remove(spa_config_dirent_t *dp) 162 { 163 int error = 0; 164 165 /* 166 * Remove the cache file. If zfs_file_unlink() in not supported by the 167 * platform fallback to truncating the file which is functionally 168 * equivalent. 169 */ 170 error = zfs_file_unlink(dp->scd_path); 171 if (error == EOPNOTSUPP) { 172 int flags = O_RDWR | O_TRUNC; 173 zfs_file_t *fp; 174 175 error = zfs_file_open(dp->scd_path, flags, 0644, &fp); 176 if (error == 0) { 177 (void) zfs_file_fsync(fp, O_SYNC); 178 (void) zfs_file_close(fp); 179 } 180 } 181 182 return (error); 183 } 184 185 static int 186 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl) 187 { 188 size_t buflen; 189 char *buf; 190 int oflags = O_RDWR | O_TRUNC | O_CREAT | O_LARGEFILE; 191 char *temp; 192 int err; 193 zfs_file_t *fp; 194 195 /* 196 * If the nvlist is empty (NULL), then remove the old cachefile. 197 */ 198 if (nvl == NULL) { 199 err = spa_config_remove(dp); 200 if (err == ENOENT) 201 err = 0; 202 203 return (err); 204 } 205 206 /* 207 * Pack the configuration into a buffer. 208 */ 209 buf = fnvlist_pack(nvl, &buflen); 210 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 211 212 /* 213 * Write the configuration to disk. Due to the complexity involved 214 * in performing a rename and remove from within the kernel the file 215 * is instead truncated and overwritten in place. This way we always 216 * have a consistent view of the data or a zero length file. 217 */ 218 err = zfs_file_open(dp->scd_path, oflags, 0644, &fp); 219 if (err == 0) { 220 err = zfs_file_write(fp, buf, buflen, NULL); 221 if (err == 0) 222 err = zfs_file_fsync(fp, O_SYNC); 223 224 zfs_file_close(fp); 225 if (err) 226 (void) spa_config_remove(dp); 227 } 228 fnvlist_pack_free(buf, buflen); 229 kmem_free(temp, MAXPATHLEN); 230 return (err); 231 } 232 233 /* 234 * Synchronize pool configuration to disk. This must be called with the 235 * namespace lock held. Synchronizing the pool cache is typically done after 236 * the configuration has been synced to the MOS. This exposes a window where 237 * the MOS config will have been updated but the cache file has not. If 238 * the system were to crash at that instant then the cached config may not 239 * contain the correct information to open the pool and an explicit import 240 * would be required. 241 */ 242 void 243 spa_write_cachefile(spa_t *target, boolean_t removing, boolean_t postsysevent) 244 { 245 spa_config_dirent_t *dp, *tdp; 246 nvlist_t *nvl; 247 char *pool_name; 248 boolean_t ccw_failure; 249 int error = 0; 250 251 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 252 253 if (!(spa_mode_global & SPA_MODE_WRITE)) 254 return; 255 256 /* 257 * Iterate over all cachefiles for the pool, past or present. When the 258 * cachefile is changed, the new one is pushed onto this list, allowing 259 * us to update previous cachefiles that no longer contain this pool. 260 */ 261 ccw_failure = B_FALSE; 262 for (dp = list_head(&target->spa_config_list); dp != NULL; 263 dp = list_next(&target->spa_config_list, dp)) { 264 spa_t *spa = NULL; 265 if (dp->scd_path == NULL) 266 continue; 267 268 /* 269 * Iterate over all pools, adding any matching pools to 'nvl'. 270 */ 271 nvl = NULL; 272 while ((spa = spa_next(spa)) != NULL) { 273 /* 274 * Skip over our own pool if we're about to remove 275 * ourselves from the spa namespace or any pool that 276 * is readonly. Since we cannot guarantee that a 277 * readonly pool would successfully import upon reboot, 278 * we don't allow them to be written to the cache file. 279 */ 280 if ((spa == target && removing) || 281 !spa_writeable(spa)) 282 continue; 283 284 mutex_enter(&spa->spa_props_lock); 285 tdp = list_head(&spa->spa_config_list); 286 if (spa->spa_config == NULL || 287 tdp == NULL || 288 tdp->scd_path == NULL || 289 strcmp(tdp->scd_path, dp->scd_path) != 0) { 290 mutex_exit(&spa->spa_props_lock); 291 continue; 292 } 293 294 if (nvl == NULL) 295 nvl = fnvlist_alloc(); 296 297 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) 298 pool_name = fnvlist_lookup_string( 299 spa->spa_config, ZPOOL_CONFIG_POOL_NAME); 300 else 301 pool_name = spa_name(spa); 302 303 fnvlist_add_nvlist(nvl, pool_name, spa->spa_config); 304 mutex_exit(&spa->spa_props_lock); 305 } 306 307 error = spa_config_write(dp, nvl); 308 if (error != 0) 309 ccw_failure = B_TRUE; 310 nvlist_free(nvl); 311 } 312 313 if (ccw_failure) { 314 /* 315 * Keep trying so that configuration data is 316 * written if/when any temporary filesystem 317 * resource issues are resolved. 318 */ 319 if (target->spa_ccw_fail_time == 0) { 320 (void) zfs_ereport_post( 321 FM_EREPORT_ZFS_CONFIG_CACHE_WRITE, 322 target, NULL, NULL, NULL, 0); 323 } 324 target->spa_ccw_fail_time = gethrtime(); 325 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE); 326 } else { 327 /* 328 * Do not rate limit future attempts to update 329 * the config cache. 330 */ 331 target->spa_ccw_fail_time = 0; 332 } 333 334 /* 335 * Remove any config entries older than the current one. 336 */ 337 dp = list_head(&target->spa_config_list); 338 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) { 339 list_remove(&target->spa_config_list, tdp); 340 if (tdp->scd_path != NULL) 341 spa_strfree(tdp->scd_path); 342 kmem_free(tdp, sizeof (spa_config_dirent_t)); 343 } 344 345 spa_config_generation++; 346 347 if (postsysevent) 348 spa_event_notify(target, NULL, NULL, ESC_ZFS_CONFIG_SYNC); 349 } 350 351 /* 352 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache, 353 * and we don't want to allow the local zone to see all the pools anyway. 354 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration 355 * information for all pool visible within the zone. 356 */ 357 nvlist_t * 358 spa_all_configs(uint64_t *generation) 359 { 360 nvlist_t *pools; 361 spa_t *spa = NULL; 362 363 if (*generation == spa_config_generation) 364 return (NULL); 365 366 pools = fnvlist_alloc(); 367 368 mutex_enter(&spa_namespace_lock); 369 while ((spa = spa_next(spa)) != NULL) { 370 if (INGLOBALZONE(curproc) || 371 zone_dataset_visible(spa_name(spa), NULL)) { 372 mutex_enter(&spa->spa_props_lock); 373 fnvlist_add_nvlist(pools, spa_name(spa), 374 spa->spa_config); 375 mutex_exit(&spa->spa_props_lock); 376 } 377 } 378 *generation = spa_config_generation; 379 mutex_exit(&spa_namespace_lock); 380 381 return (pools); 382 } 383 384 void 385 spa_config_set(spa_t *spa, nvlist_t *config) 386 { 387 mutex_enter(&spa->spa_props_lock); 388 if (spa->spa_config != NULL && spa->spa_config != config) 389 nvlist_free(spa->spa_config); 390 spa->spa_config = config; 391 mutex_exit(&spa->spa_props_lock); 392 } 393 394 /* 395 * Generate the pool's configuration based on the current in-core state. 396 * 397 * We infer whether to generate a complete config or just one top-level config 398 * based on whether vd is the root vdev. 399 */ 400 nvlist_t * 401 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats) 402 { 403 nvlist_t *config, *nvroot; 404 vdev_t *rvd = spa->spa_root_vdev; 405 unsigned long hostid = 0; 406 boolean_t locked = B_FALSE; 407 uint64_t split_guid; 408 char *pool_name; 409 410 if (vd == NULL) { 411 vd = rvd; 412 locked = B_TRUE; 413 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 414 } 415 416 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) == 417 (SCL_CONFIG | SCL_STATE)); 418 419 /* 420 * If txg is -1, report the current value of spa->spa_config_txg. 421 */ 422 if (txg == -1ULL) 423 txg = spa->spa_config_txg; 424 425 /* 426 * Originally, users had to handle spa namespace collisions by either 427 * exporting the already imported pool or by specifying a new name for 428 * the pool with a conflicting name. In the case of root pools from 429 * virtual guests, neither approach to collision resolution is 430 * reasonable. This is addressed by extending the new name syntax with 431 * an option to specify that the new name is temporary. When specified, 432 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us 433 * to use the previous name, which we do below. 434 */ 435 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) { 436 VERIFY0(nvlist_lookup_string(spa->spa_config, 437 ZPOOL_CONFIG_POOL_NAME, &pool_name)); 438 } else 439 pool_name = spa_name(spa); 440 441 config = fnvlist_alloc(); 442 443 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa)); 444 fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, pool_name); 445 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa)); 446 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg); 447 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa)); 448 fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, spa->spa_errata); 449 if (spa->spa_comment != NULL) 450 fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT, 451 spa->spa_comment); 452 if (spa->spa_compatibility != NULL) 453 fnvlist_add_string(config, ZPOOL_CONFIG_COMPATIBILITY, 454 spa->spa_compatibility); 455 456 hostid = spa_get_hostid(spa); 457 if (hostid != 0) 458 fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid); 459 fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname()->nodename); 460 461 int config_gen_flags = 0; 462 if (vd != rvd) { 463 fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID, 464 vd->vdev_top->vdev_guid); 465 fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID, 466 vd->vdev_guid); 467 if (vd->vdev_isspare) 468 fnvlist_add_uint64(config, 469 ZPOOL_CONFIG_IS_SPARE, 1ULL); 470 if (vd->vdev_islog) 471 fnvlist_add_uint64(config, 472 ZPOOL_CONFIG_IS_LOG, 1ULL); 473 vd = vd->vdev_top; /* label contains top config */ 474 } else { 475 /* 476 * Only add the (potentially large) split information 477 * in the mos config, and not in the vdev labels 478 */ 479 if (spa->spa_config_splitting != NULL) 480 fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT, 481 spa->spa_config_splitting); 482 483 fnvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS); 484 485 config_gen_flags |= VDEV_CONFIG_MOS; 486 } 487 488 /* 489 * Add the top-level config. We even add this on pools which 490 * don't support holes in the namespace. 491 */ 492 vdev_top_config_generate(spa, config); 493 494 /* 495 * If we're splitting, record the original pool's guid. 496 */ 497 if (spa->spa_config_splitting != NULL && 498 nvlist_lookup_uint64(spa->spa_config_splitting, 499 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) { 500 fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID, split_guid); 501 } 502 503 nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags); 504 fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot); 505 nvlist_free(nvroot); 506 507 /* 508 * Store what's necessary for reading the MOS in the label. 509 */ 510 fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, 511 spa->spa_label_features); 512 513 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) { 514 ddt_histogram_t *ddh; 515 ddt_stat_t *dds; 516 ddt_object_t *ddo; 517 518 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP); 519 ddt_get_dedup_histogram(spa, ddh); 520 fnvlist_add_uint64_array(config, 521 ZPOOL_CONFIG_DDT_HISTOGRAM, 522 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)); 523 kmem_free(ddh, sizeof (ddt_histogram_t)); 524 525 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP); 526 ddt_get_dedup_object_stats(spa, ddo); 527 fnvlist_add_uint64_array(config, 528 ZPOOL_CONFIG_DDT_OBJ_STATS, 529 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)); 530 kmem_free(ddo, sizeof (ddt_object_t)); 531 532 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP); 533 ddt_get_dedup_stats(spa, dds); 534 fnvlist_add_uint64_array(config, 535 ZPOOL_CONFIG_DDT_STATS, 536 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)); 537 kmem_free(dds, sizeof (ddt_stat_t)); 538 } 539 540 if (locked) 541 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 542 543 return (config); 544 } 545 546 /* 547 * Update all disk labels, generate a fresh config based on the current 548 * in-core state, and sync the global config cache (do not sync the config 549 * cache if this is a booting rootpool). 550 */ 551 void 552 spa_config_update(spa_t *spa, int what) 553 { 554 vdev_t *rvd = spa->spa_root_vdev; 555 uint64_t txg; 556 int c; 557 558 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 559 560 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 561 txg = spa_last_synced_txg(spa) + 1; 562 if (what == SPA_CONFIG_UPDATE_POOL) { 563 vdev_config_dirty(rvd); 564 } else { 565 /* 566 * If we have top-level vdevs that were added but have 567 * not yet been prepared for allocation, do that now. 568 * (It's safe now because the config cache is up to date, 569 * so it will be able to translate the new DVAs.) 570 * See comments in spa_vdev_add() for full details. 571 */ 572 for (c = 0; c < rvd->vdev_children; c++) { 573 vdev_t *tvd = rvd->vdev_child[c]; 574 575 /* 576 * Explicitly skip vdevs that are indirect or 577 * log vdevs that are being removed. The reason 578 * is that both of those can have vdev_ms_array 579 * set to 0 and we wouldn't want to change their 580 * metaslab size nor call vdev_expand() on them. 581 */ 582 if (!vdev_is_concrete(tvd) || 583 (tvd->vdev_islog && tvd->vdev_removing)) 584 continue; 585 586 if (tvd->vdev_ms_array == 0) 587 vdev_metaslab_set_size(tvd); 588 vdev_expand(tvd, txg); 589 } 590 } 591 spa_config_exit(spa, SCL_ALL, FTAG); 592 593 /* 594 * Wait for the mosconfig to be regenerated and synced. 595 */ 596 txg_wait_synced(spa->spa_dsl_pool, txg); 597 598 /* 599 * Update the global config cache to reflect the new mosconfig. 600 */ 601 if (!spa->spa_is_root) { 602 spa_write_cachefile(spa, B_FALSE, 603 what != SPA_CONFIG_UPDATE_POOL); 604 } 605 606 if (what == SPA_CONFIG_UPDATE_POOL) 607 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS); 608 } 609 610 EXPORT_SYMBOL(spa_config_load); 611 EXPORT_SYMBOL(spa_all_configs); 612 EXPORT_SYMBOL(spa_config_set); 613 EXPORT_SYMBOL(spa_config_generate); 614 EXPORT_SYMBOL(spa_config_update); 615 616 #ifdef __linux__ 617 /* string sysctls require a char array on FreeBSD */ 618 ZFS_MODULE_PARAM(zfs_spa, spa_, config_path, STRING, ZMOD_RD, 619 "SPA config file (/etc/zfs/zpool.cache)"); 620 #endif 621 622 ZFS_MODULE_PARAM(zfs, zfs_, autoimport_disable, INT, ZMOD_RW, 623 "Disable pool import at module load"); 624