1 // SPDX-License-Identifier: CDDL-1.0 2 /* 3 * CDDL HEADER START 4 * 5 * The contents of this file are subject to the terms of the 6 * Common Development and Distribution License (the "License"). 7 * You may not use this file except in compliance with the License. 8 * 9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 10 * or https://opensource.org/licenses/CDDL-1.0. 11 * See the License for the specific language governing permissions 12 * and limitations under the License. 13 * 14 * When distributing Covered Code, include this CDDL HEADER in each 15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 16 * If applicable, add the following below this CDDL HEADER, with the 17 * fields enclosed by brackets "[]" replaced with your own identifying 18 * information: Portions Copyright [yyyy] [name of copyright owner] 19 * 20 * CDDL HEADER END 21 */ 22 23 /* 24 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 25 * Copyright 2011 Nexenta Systems, Inc. All rights reserved. 26 * Copyright (c) 2011, 2020 by Delphix. All rights reserved. 27 * Copyright 2017 Joyent, Inc. 28 * Copyright (c) 2021, Colm Buckley <colm@tuatha.org> 29 */ 30 31 #include <sys/spa.h> 32 #include <sys/file.h> 33 #include <sys/fm/fs/zfs.h> 34 #include <sys/spa_impl.h> 35 #include <sys/nvpair.h> 36 #include <sys/fs/zfs.h> 37 #include <sys/vdev_impl.h> 38 #include <sys/zfs_ioctl.h> 39 #include <sys/systeminfo.h> 40 #include <sys/sunddi.h> 41 #include <sys/zfeature.h> 42 #include <sys/zfs_file.h> 43 #include <sys/zfs_context.h> 44 #ifdef _KERNEL 45 #include <sys/zone.h> 46 #endif 47 48 /* 49 * Pool configuration repository. 50 * 51 * Pool configuration is stored as a packed nvlist on the filesystem. By 52 * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot 53 * (when the ZFS module is loaded). Pools can also have the 'cachefile' 54 * property set that allows them to be stored in an alternate location until 55 * the control of external software. 56 * 57 * For each cache file, we have a single nvlist which holds all the 58 * configuration information. When the module loads, we read this information 59 * from /etc/zfs/zpool.cache and populate the SPA namespace. This namespace is 60 * maintained independently in spa.c. Whenever the namespace is modified, or 61 * the configuration of a pool is changed, we call spa_write_cachefile(), which 62 * walks through all the active pools and writes the configuration to disk. 63 */ 64 65 static uint64_t spa_config_generation = 1; 66 67 /* 68 * This can be overridden in userland to preserve an alternate namespace for 69 * userland pools when doing testing. 70 */ 71 char *spa_config_path = (char *)ZPOOL_CACHE; 72 #ifdef _KERNEL 73 static int zfs_autoimport_disable = B_TRUE; 74 #endif 75 76 /* 77 * Called when the module is first loaded, this routine loads the configuration 78 * file into the SPA namespace. It does not actually open or load the pools; it 79 * only populates the namespace. 80 */ 81 void 82 spa_config_load(void) 83 { 84 void *buf = NULL; 85 nvlist_t *nvlist, *child; 86 nvpair_t *nvpair; 87 char *pathname; 88 zfs_file_t *fp; 89 zfs_file_attr_t zfa; 90 uint64_t fsize; 91 int err; 92 93 #ifdef _KERNEL 94 if (zfs_autoimport_disable) 95 return; 96 #endif 97 98 /* 99 * Open the configuration file. 100 */ 101 pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP); 102 103 (void) snprintf(pathname, MAXPATHLEN, "%s", spa_config_path); 104 105 err = zfs_file_open(pathname, O_RDONLY, 0, &fp); 106 107 #ifdef __FreeBSD__ 108 if (err) 109 err = zfs_file_open(ZPOOL_CACHE_BOOT, O_RDONLY, 0, &fp); 110 #endif 111 kmem_free(pathname, MAXPATHLEN); 112 113 if (err) 114 return; 115 116 if (zfs_file_getattr(fp, &zfa)) 117 goto out; 118 119 fsize = zfa.zfa_size; 120 buf = kmem_alloc(fsize, KM_SLEEP); 121 122 /* 123 * Read the nvlist from the file. 124 */ 125 if (zfs_file_read(fp, buf, fsize, NULL) < 0) 126 goto out; 127 128 /* 129 * Unpack the nvlist. 130 */ 131 if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0) 132 goto out; 133 134 /* 135 * Iterate over all elements in the nvlist, creating a new spa_t for 136 * each one with the specified configuration. 137 */ 138 mutex_enter(&spa_namespace_lock); 139 nvpair = NULL; 140 while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) { 141 if (nvpair_type(nvpair) != DATA_TYPE_NVLIST) 142 continue; 143 144 child = fnvpair_value_nvlist(nvpair); 145 146 if (spa_lookup(nvpair_name(nvpair)) != NULL) 147 continue; 148 (void) spa_add(nvpair_name(nvpair), child, NULL); 149 } 150 mutex_exit(&spa_namespace_lock); 151 152 nvlist_free(nvlist); 153 154 out: 155 if (buf != NULL) 156 kmem_free(buf, fsize); 157 158 zfs_file_close(fp); 159 } 160 161 static int 162 spa_config_remove(spa_config_dirent_t *dp) 163 { 164 int error = 0; 165 166 /* 167 * Remove the cache file. If zfs_file_unlink() in not supported by the 168 * platform fallback to truncating the file which is functionally 169 * equivalent. 170 */ 171 error = zfs_file_unlink(dp->scd_path); 172 if (error == EOPNOTSUPP) { 173 int flags = O_RDWR | O_TRUNC; 174 zfs_file_t *fp; 175 176 error = zfs_file_open(dp->scd_path, flags, 0644, &fp); 177 if (error == 0) { 178 (void) zfs_file_fsync(fp, O_SYNC); 179 (void) zfs_file_close(fp); 180 } 181 } 182 183 return (error); 184 } 185 186 static int 187 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl) 188 { 189 size_t buflen; 190 char *buf; 191 int oflags = O_RDWR | O_TRUNC | O_CREAT | O_LARGEFILE; 192 char *temp; 193 int err; 194 zfs_file_t *fp; 195 196 /* 197 * If the nvlist is empty (NULL), then remove the old cachefile. 198 */ 199 if (nvl == NULL) { 200 err = spa_config_remove(dp); 201 if (err == ENOENT) 202 err = 0; 203 204 return (err); 205 } 206 207 /* 208 * Pack the configuration into a buffer. 209 */ 210 buf = fnvlist_pack(nvl, &buflen); 211 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP); 212 213 /* 214 * Write the configuration to disk. Due to the complexity involved 215 * in performing a rename and remove from within the kernel the file 216 * is instead truncated and overwritten in place. This way we always 217 * have a consistent view of the data or a zero length file. 218 */ 219 err = zfs_file_open(dp->scd_path, oflags, 0644, &fp); 220 if (err == 0) { 221 err = zfs_file_write(fp, buf, buflen, NULL); 222 if (err == 0) 223 err = zfs_file_fsync(fp, O_SYNC); 224 225 zfs_file_close(fp); 226 if (err) 227 (void) spa_config_remove(dp); 228 } 229 fnvlist_pack_free(buf, buflen); 230 kmem_free(temp, MAXPATHLEN); 231 return (err); 232 } 233 234 /* 235 * Synchronize pool configuration to disk. This must be called with the 236 * namespace lock held. Synchronizing the pool cache is typically done after 237 * the configuration has been synced to the MOS. This exposes a window where 238 * the MOS config will have been updated but the cache file has not. If 239 * the system were to crash at that instant then the cached config may not 240 * contain the correct information to open the pool and an explicit import 241 * would be required. 242 */ 243 void 244 spa_write_cachefile(spa_t *target, boolean_t removing, boolean_t postsysevent, 245 boolean_t postblkidevent) 246 { 247 spa_config_dirent_t *dp, *tdp; 248 nvlist_t *nvl; 249 const char *pool_name; 250 boolean_t ccw_failure; 251 int error = 0; 252 253 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 254 255 if (!(spa_mode_global & SPA_MODE_WRITE)) 256 return; 257 258 /* 259 * Iterate over all cachefiles for the pool, past or present. When the 260 * cachefile is changed, the new one is pushed onto this list, allowing 261 * us to update previous cachefiles that no longer contain this pool. 262 */ 263 ccw_failure = B_FALSE; 264 for (dp = list_head(&target->spa_config_list); dp != NULL; 265 dp = list_next(&target->spa_config_list, dp)) { 266 spa_t *spa = NULL; 267 if (dp->scd_path == NULL) 268 continue; 269 270 /* 271 * Iterate over all pools, adding any matching pools to 'nvl'. 272 */ 273 nvl = NULL; 274 while ((spa = spa_next(spa)) != NULL) { 275 /* 276 * Skip over our own pool if we're about to remove 277 * ourselves from the spa namespace or any pool that 278 * is readonly. Since we cannot guarantee that a 279 * readonly pool would successfully import upon reboot, 280 * we don't allow them to be written to the cache file. 281 */ 282 if ((spa == target && removing) || 283 !spa_writeable(spa)) 284 continue; 285 286 mutex_enter(&spa->spa_props_lock); 287 tdp = list_head(&spa->spa_config_list); 288 if (spa->spa_config == NULL || 289 tdp == NULL || 290 tdp->scd_path == NULL || 291 strcmp(tdp->scd_path, dp->scd_path) != 0) { 292 mutex_exit(&spa->spa_props_lock); 293 continue; 294 } 295 296 if (nvl == NULL) 297 nvl = fnvlist_alloc(); 298 299 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) 300 pool_name = fnvlist_lookup_string( 301 spa->spa_config, ZPOOL_CONFIG_POOL_NAME); 302 else 303 pool_name = spa_name(spa); 304 305 fnvlist_add_nvlist(nvl, pool_name, spa->spa_config); 306 mutex_exit(&spa->spa_props_lock); 307 } 308 309 error = spa_config_write(dp, nvl); 310 if (error != 0) 311 ccw_failure = B_TRUE; 312 nvlist_free(nvl); 313 } 314 315 if (ccw_failure) { 316 /* 317 * Keep trying so that configuration data is 318 * written if/when any temporary filesystem 319 * resource issues are resolved. 320 */ 321 if (target->spa_ccw_fail_time == 0) { 322 (void) zfs_ereport_post( 323 FM_EREPORT_ZFS_CONFIG_CACHE_WRITE, 324 target, NULL, NULL, NULL, 0); 325 } 326 target->spa_ccw_fail_time = gethrtime(); 327 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE); 328 } else { 329 /* 330 * Do not rate limit future attempts to update 331 * the config cache. 332 */ 333 target->spa_ccw_fail_time = 0; 334 } 335 336 /* 337 * Remove any config entries older than the current one. 338 */ 339 dp = list_head(&target->spa_config_list); 340 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) { 341 list_remove(&target->spa_config_list, tdp); 342 if (tdp->scd_path != NULL) 343 spa_strfree(tdp->scd_path); 344 kmem_free(tdp, sizeof (spa_config_dirent_t)); 345 } 346 347 spa_config_generation++; 348 349 if (postsysevent) 350 spa_event_notify(target, NULL, NULL, ESC_ZFS_CONFIG_SYNC); 351 352 /* 353 * Post udev event to sync blkid information if the pool is created 354 * or a new vdev is added to the pool. 355 */ 356 if ((target->spa_root_vdev) && postblkidevent) { 357 vdev_post_kobj_evt(target->spa_root_vdev); 358 for (int i = 0; i < target->spa_l2cache.sav_count; i++) 359 vdev_post_kobj_evt(target->spa_l2cache.sav_vdevs[i]); 360 for (int i = 0; i < target->spa_spares.sav_count; i++) 361 vdev_post_kobj_evt(target->spa_spares.sav_vdevs[i]); 362 } 363 } 364 365 /* 366 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache, 367 * and we don't want to allow the local zone to see all the pools anyway. 368 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration 369 * information for all pool visible within the zone. 370 */ 371 int 372 spa_all_configs(uint64_t *generation, nvlist_t **pools) 373 { 374 spa_t *spa = NULL; 375 376 if (*generation == spa_config_generation) 377 return (SET_ERROR(EEXIST)); 378 379 int error = mutex_enter_interruptible(&spa_namespace_lock); 380 if (error) 381 return (SET_ERROR(EINTR)); 382 383 *pools = fnvlist_alloc(); 384 while ((spa = spa_next(spa)) != NULL) { 385 if (INGLOBALZONE(curproc) || 386 zone_dataset_visible(spa_name(spa), NULL)) { 387 mutex_enter(&spa->spa_props_lock); 388 fnvlist_add_nvlist(*pools, spa_name(spa), 389 spa->spa_config); 390 mutex_exit(&spa->spa_props_lock); 391 } 392 } 393 *generation = spa_config_generation; 394 mutex_exit(&spa_namespace_lock); 395 396 return (0); 397 } 398 399 void 400 spa_config_set(spa_t *spa, nvlist_t *config) 401 { 402 mutex_enter(&spa->spa_props_lock); 403 if (spa->spa_config != NULL && spa->spa_config != config) 404 nvlist_free(spa->spa_config); 405 spa->spa_config = config; 406 mutex_exit(&spa->spa_props_lock); 407 } 408 409 /* 410 * Generate the pool's configuration based on the current in-core state. 411 * 412 * We infer whether to generate a complete config or just one top-level config 413 * based on whether vd is the root vdev. 414 */ 415 nvlist_t * 416 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats) 417 { 418 nvlist_t *config, *nvroot; 419 vdev_t *rvd = spa->spa_root_vdev; 420 unsigned long hostid = 0; 421 boolean_t locked = B_FALSE; 422 uint64_t split_guid; 423 const char *pool_name; 424 425 if (vd == NULL) { 426 vd = rvd; 427 locked = B_TRUE; 428 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 429 } 430 431 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) == 432 (SCL_CONFIG | SCL_STATE)); 433 434 /* 435 * If txg is -1, report the current value of spa->spa_config_txg. 436 */ 437 if (txg == -1ULL) 438 txg = spa->spa_config_txg; 439 440 /* 441 * Originally, users had to handle spa namespace collisions by either 442 * exporting the already imported pool or by specifying a new name for 443 * the pool with a conflicting name. In the case of root pools from 444 * virtual guests, neither approach to collision resolution is 445 * reasonable. This is addressed by extending the new name syntax with 446 * an option to specify that the new name is temporary. When specified, 447 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us 448 * to use the previous name, which we do below. 449 */ 450 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) { 451 VERIFY0(nvlist_lookup_string(spa->spa_config, 452 ZPOOL_CONFIG_POOL_NAME, &pool_name)); 453 } else 454 pool_name = spa_name(spa); 455 456 config = fnvlist_alloc(); 457 458 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa)); 459 fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, pool_name); 460 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa)); 461 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg); 462 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa)); 463 fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, spa->spa_errata); 464 if (spa->spa_comment != NULL) 465 fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT, 466 spa->spa_comment); 467 if (spa->spa_compatibility != NULL) 468 fnvlist_add_string(config, ZPOOL_CONFIG_COMPATIBILITY, 469 spa->spa_compatibility); 470 471 hostid = spa_get_hostid(spa); 472 if (hostid != 0) 473 fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid); 474 fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname()->nodename); 475 476 int config_gen_flags = 0; 477 if (vd != rvd) { 478 fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID, 479 vd->vdev_top->vdev_guid); 480 fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID, 481 vd->vdev_guid); 482 if (vd->vdev_isspare) 483 fnvlist_add_uint64(config, 484 ZPOOL_CONFIG_IS_SPARE, 1ULL); 485 if (vd->vdev_islog) 486 fnvlist_add_uint64(config, 487 ZPOOL_CONFIG_IS_LOG, 1ULL); 488 vd = vd->vdev_top; /* label contains top config */ 489 } else { 490 /* 491 * Only add the (potentially large) split information 492 * in the mos config, and not in the vdev labels 493 */ 494 if (spa->spa_config_splitting != NULL) 495 fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT, 496 spa->spa_config_splitting); 497 498 fnvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS); 499 500 config_gen_flags |= VDEV_CONFIG_MOS; 501 } 502 503 /* 504 * Add the top-level config. We even add this on pools which 505 * don't support holes in the namespace. 506 */ 507 vdev_top_config_generate(spa, config); 508 509 /* 510 * If we're splitting, record the original pool's guid. 511 */ 512 if (spa->spa_config_splitting != NULL && 513 nvlist_lookup_uint64(spa->spa_config_splitting, 514 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) { 515 fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID, split_guid); 516 } 517 518 nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags); 519 fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot); 520 nvlist_free(nvroot); 521 522 /* 523 * Store what's necessary for reading the MOS in the label. 524 */ 525 fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, 526 spa->spa_label_features); 527 528 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) { 529 ddt_histogram_t *ddh; 530 ddt_stat_t *dds; 531 ddt_object_t *ddo; 532 533 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP); 534 ddt_get_dedup_histogram(spa, ddh); 535 fnvlist_add_uint64_array(config, 536 ZPOOL_CONFIG_DDT_HISTOGRAM, 537 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)); 538 kmem_free(ddh, sizeof (ddt_histogram_t)); 539 540 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP); 541 ddt_get_dedup_object_stats(spa, ddo); 542 fnvlist_add_uint64_array(config, 543 ZPOOL_CONFIG_DDT_OBJ_STATS, 544 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)); 545 kmem_free(ddo, sizeof (ddt_object_t)); 546 547 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP); 548 ddt_get_dedup_stats(spa, dds); 549 fnvlist_add_uint64_array(config, 550 ZPOOL_CONFIG_DDT_STATS, 551 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)); 552 kmem_free(dds, sizeof (ddt_stat_t)); 553 } 554 555 if (locked) 556 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 557 558 return (config); 559 } 560 561 /* 562 * Update all disk labels, generate a fresh config based on the current 563 * in-core state, and sync the global config cache (do not sync the config 564 * cache if this is a booting rootpool). 565 */ 566 void 567 spa_config_update(spa_t *spa, int what) 568 { 569 vdev_t *rvd = spa->spa_root_vdev; 570 uint64_t txg; 571 int c; 572 573 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 574 575 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 576 txg = spa_last_synced_txg(spa) + 1; 577 if (what == SPA_CONFIG_UPDATE_POOL) { 578 vdev_config_dirty(rvd); 579 } else { 580 /* 581 * If we have top-level vdevs that were added but have 582 * not yet been prepared for allocation, do that now. 583 * (It's safe now because the config cache is up to date, 584 * so it will be able to translate the new DVAs.) 585 * See comments in spa_vdev_add() for full details. 586 */ 587 for (c = 0; c < rvd->vdev_children; c++) { 588 vdev_t *tvd = rvd->vdev_child[c]; 589 590 /* 591 * Explicitly skip vdevs that are indirect or 592 * log vdevs that are being removed. The reason 593 * is that both of those can have vdev_ms_array 594 * set to 0 and we wouldn't want to change their 595 * metaslab size nor call vdev_expand() on them. 596 */ 597 if (!vdev_is_concrete(tvd) || 598 (tvd->vdev_islog && tvd->vdev_removing)) 599 continue; 600 601 if (tvd->vdev_ms_array == 0) 602 vdev_metaslab_set_size(tvd); 603 vdev_expand(tvd, txg); 604 } 605 } 606 spa_config_exit(spa, SCL_ALL, FTAG); 607 608 /* 609 * Wait for the mosconfig to be regenerated and synced. 610 */ 611 txg_wait_synced(spa->spa_dsl_pool, txg); 612 613 /* 614 * Update the global config cache to reflect the new mosconfig. 615 */ 616 if (!spa->spa_is_root) { 617 spa_write_cachefile(spa, B_FALSE, 618 what != SPA_CONFIG_UPDATE_POOL, 619 what != SPA_CONFIG_UPDATE_POOL); 620 } 621 622 if (what == SPA_CONFIG_UPDATE_POOL) 623 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS); 624 } 625 626 EXPORT_SYMBOL(spa_config_load); 627 EXPORT_SYMBOL(spa_all_configs); 628 EXPORT_SYMBOL(spa_config_set); 629 EXPORT_SYMBOL(spa_config_generate); 630 EXPORT_SYMBOL(spa_config_update); 631 632 #ifdef __linux__ 633 /* string sysctls require a char array on FreeBSD */ 634 ZFS_MODULE_PARAM(zfs_spa, spa_, config_path, STRING, ZMOD_RD, 635 "SPA config file (/etc/zfs/zpool.cache)"); 636 #endif 637 638 #ifdef _KERNEL 639 ZFS_MODULE_PARAM(zfs, zfs_, autoimport_disable, INT, ZMOD_RW, 640 "Disable pool import at module load"); 641 #endif 642