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 boolean_t postblkidevent) 245 { 246 spa_config_dirent_t *dp, *tdp; 247 nvlist_t *nvl; 248 char *pool_name; 249 boolean_t ccw_failure; 250 int error = 0; 251 252 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 253 254 if (!(spa_mode_global & SPA_MODE_WRITE)) 255 return; 256 257 /* 258 * Iterate over all cachefiles for the pool, past or present. When the 259 * cachefile is changed, the new one is pushed onto this list, allowing 260 * us to update previous cachefiles that no longer contain this pool. 261 */ 262 ccw_failure = B_FALSE; 263 for (dp = list_head(&target->spa_config_list); dp != NULL; 264 dp = list_next(&target->spa_config_list, dp)) { 265 spa_t *spa = NULL; 266 if (dp->scd_path == NULL) 267 continue; 268 269 /* 270 * Iterate over all pools, adding any matching pools to 'nvl'. 271 */ 272 nvl = NULL; 273 while ((spa = spa_next(spa)) != NULL) { 274 /* 275 * Skip over our own pool if we're about to remove 276 * ourselves from the spa namespace or any pool that 277 * is readonly. Since we cannot guarantee that a 278 * readonly pool would successfully import upon reboot, 279 * we don't allow them to be written to the cache file. 280 */ 281 if ((spa == target && removing) || 282 !spa_writeable(spa)) 283 continue; 284 285 mutex_enter(&spa->spa_props_lock); 286 tdp = list_head(&spa->spa_config_list); 287 if (spa->spa_config == NULL || 288 tdp == NULL || 289 tdp->scd_path == NULL || 290 strcmp(tdp->scd_path, dp->scd_path) != 0) { 291 mutex_exit(&spa->spa_props_lock); 292 continue; 293 } 294 295 if (nvl == NULL) 296 nvl = fnvlist_alloc(); 297 298 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) 299 pool_name = fnvlist_lookup_string( 300 spa->spa_config, ZPOOL_CONFIG_POOL_NAME); 301 else 302 pool_name = spa_name(spa); 303 304 fnvlist_add_nvlist(nvl, pool_name, spa->spa_config); 305 mutex_exit(&spa->spa_props_lock); 306 } 307 308 error = spa_config_write(dp, nvl); 309 if (error != 0) 310 ccw_failure = B_TRUE; 311 nvlist_free(nvl); 312 } 313 314 if (ccw_failure) { 315 /* 316 * Keep trying so that configuration data is 317 * written if/when any temporary filesystem 318 * resource issues are resolved. 319 */ 320 if (target->spa_ccw_fail_time == 0) { 321 (void) zfs_ereport_post( 322 FM_EREPORT_ZFS_CONFIG_CACHE_WRITE, 323 target, NULL, NULL, NULL, 0); 324 } 325 target->spa_ccw_fail_time = gethrtime(); 326 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE); 327 } else { 328 /* 329 * Do not rate limit future attempts to update 330 * the config cache. 331 */ 332 target->spa_ccw_fail_time = 0; 333 } 334 335 /* 336 * Remove any config entries older than the current one. 337 */ 338 dp = list_head(&target->spa_config_list); 339 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) { 340 list_remove(&target->spa_config_list, tdp); 341 if (tdp->scd_path != NULL) 342 spa_strfree(tdp->scd_path); 343 kmem_free(tdp, sizeof (spa_config_dirent_t)); 344 } 345 346 spa_config_generation++; 347 348 if (postsysevent) 349 spa_event_notify(target, NULL, NULL, ESC_ZFS_CONFIG_SYNC); 350 351 /* 352 * Post udev event to sync blkid information if the pool is created 353 * or a new vdev is added to the pool. 354 */ 355 if ((target->spa_root_vdev) && postblkidevent) { 356 vdev_post_kobj_evt(target->spa_root_vdev); 357 for (int i = 0; i < target->spa_l2cache.sav_count; i++) 358 vdev_post_kobj_evt(target->spa_l2cache.sav_vdevs[i]); 359 for (int i = 0; i < target->spa_spares.sav_count; i++) 360 vdev_post_kobj_evt(target->spa_spares.sav_vdevs[i]); 361 } 362 } 363 364 /* 365 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache, 366 * and we don't want to allow the local zone to see all the pools anyway. 367 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration 368 * information for all pool visible within the zone. 369 */ 370 nvlist_t * 371 spa_all_configs(uint64_t *generation) 372 { 373 nvlist_t *pools; 374 spa_t *spa = NULL; 375 376 if (*generation == spa_config_generation) 377 return (NULL); 378 379 pools = fnvlist_alloc(); 380 381 mutex_enter(&spa_namespace_lock); 382 while ((spa = spa_next(spa)) != NULL) { 383 if (INGLOBALZONE(curproc) || 384 zone_dataset_visible(spa_name(spa), NULL)) { 385 mutex_enter(&spa->spa_props_lock); 386 fnvlist_add_nvlist(pools, spa_name(spa), 387 spa->spa_config); 388 mutex_exit(&spa->spa_props_lock); 389 } 390 } 391 *generation = spa_config_generation; 392 mutex_exit(&spa_namespace_lock); 393 394 return (pools); 395 } 396 397 void 398 spa_config_set(spa_t *spa, nvlist_t *config) 399 { 400 mutex_enter(&spa->spa_props_lock); 401 if (spa->spa_config != NULL && spa->spa_config != config) 402 nvlist_free(spa->spa_config); 403 spa->spa_config = config; 404 mutex_exit(&spa->spa_props_lock); 405 } 406 407 /* 408 * Generate the pool's configuration based on the current in-core state. 409 * 410 * We infer whether to generate a complete config or just one top-level config 411 * based on whether vd is the root vdev. 412 */ 413 nvlist_t * 414 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats) 415 { 416 nvlist_t *config, *nvroot; 417 vdev_t *rvd = spa->spa_root_vdev; 418 unsigned long hostid = 0; 419 boolean_t locked = B_FALSE; 420 uint64_t split_guid; 421 char *pool_name; 422 423 if (vd == NULL) { 424 vd = rvd; 425 locked = B_TRUE; 426 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER); 427 } 428 429 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) == 430 (SCL_CONFIG | SCL_STATE)); 431 432 /* 433 * If txg is -1, report the current value of spa->spa_config_txg. 434 */ 435 if (txg == -1ULL) 436 txg = spa->spa_config_txg; 437 438 /* 439 * Originally, users had to handle spa namespace collisions by either 440 * exporting the already imported pool or by specifying a new name for 441 * the pool with a conflicting name. In the case of root pools from 442 * virtual guests, neither approach to collision resolution is 443 * reasonable. This is addressed by extending the new name syntax with 444 * an option to specify that the new name is temporary. When specified, 445 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us 446 * to use the previous name, which we do below. 447 */ 448 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) { 449 VERIFY0(nvlist_lookup_string(spa->spa_config, 450 ZPOOL_CONFIG_POOL_NAME, &pool_name)); 451 } else 452 pool_name = spa_name(spa); 453 454 config = fnvlist_alloc(); 455 456 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa)); 457 fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, pool_name); 458 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa)); 459 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg); 460 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa)); 461 fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, spa->spa_errata); 462 if (spa->spa_comment != NULL) 463 fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT, 464 spa->spa_comment); 465 if (spa->spa_compatibility != NULL) 466 fnvlist_add_string(config, ZPOOL_CONFIG_COMPATIBILITY, 467 spa->spa_compatibility); 468 469 hostid = spa_get_hostid(spa); 470 if (hostid != 0) 471 fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid); 472 fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname()->nodename); 473 474 int config_gen_flags = 0; 475 if (vd != rvd) { 476 fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID, 477 vd->vdev_top->vdev_guid); 478 fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID, 479 vd->vdev_guid); 480 if (vd->vdev_isspare) 481 fnvlist_add_uint64(config, 482 ZPOOL_CONFIG_IS_SPARE, 1ULL); 483 if (vd->vdev_islog) 484 fnvlist_add_uint64(config, 485 ZPOOL_CONFIG_IS_LOG, 1ULL); 486 vd = vd->vdev_top; /* label contains top config */ 487 } else { 488 /* 489 * Only add the (potentially large) split information 490 * in the mos config, and not in the vdev labels 491 */ 492 if (spa->spa_config_splitting != NULL) 493 fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT, 494 spa->spa_config_splitting); 495 496 fnvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS); 497 498 config_gen_flags |= VDEV_CONFIG_MOS; 499 } 500 501 /* 502 * Add the top-level config. We even add this on pools which 503 * don't support holes in the namespace. 504 */ 505 vdev_top_config_generate(spa, config); 506 507 /* 508 * If we're splitting, record the original pool's guid. 509 */ 510 if (spa->spa_config_splitting != NULL && 511 nvlist_lookup_uint64(spa->spa_config_splitting, 512 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) { 513 fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID, split_guid); 514 } 515 516 nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags); 517 fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot); 518 nvlist_free(nvroot); 519 520 /* 521 * Store what's necessary for reading the MOS in the label. 522 */ 523 fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ, 524 spa->spa_label_features); 525 526 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) { 527 ddt_histogram_t *ddh; 528 ddt_stat_t *dds; 529 ddt_object_t *ddo; 530 531 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP); 532 ddt_get_dedup_histogram(spa, ddh); 533 fnvlist_add_uint64_array(config, 534 ZPOOL_CONFIG_DDT_HISTOGRAM, 535 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t)); 536 kmem_free(ddh, sizeof (ddt_histogram_t)); 537 538 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP); 539 ddt_get_dedup_object_stats(spa, ddo); 540 fnvlist_add_uint64_array(config, 541 ZPOOL_CONFIG_DDT_OBJ_STATS, 542 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t)); 543 kmem_free(ddo, sizeof (ddt_object_t)); 544 545 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP); 546 ddt_get_dedup_stats(spa, dds); 547 fnvlist_add_uint64_array(config, 548 ZPOOL_CONFIG_DDT_STATS, 549 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t)); 550 kmem_free(dds, sizeof (ddt_stat_t)); 551 } 552 553 if (locked) 554 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG); 555 556 return (config); 557 } 558 559 /* 560 * Update all disk labels, generate a fresh config based on the current 561 * in-core state, and sync the global config cache (do not sync the config 562 * cache if this is a booting rootpool). 563 */ 564 void 565 spa_config_update(spa_t *spa, int what) 566 { 567 vdev_t *rvd = spa->spa_root_vdev; 568 uint64_t txg; 569 int c; 570 571 ASSERT(MUTEX_HELD(&spa_namespace_lock)); 572 573 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER); 574 txg = spa_last_synced_txg(spa) + 1; 575 if (what == SPA_CONFIG_UPDATE_POOL) { 576 vdev_config_dirty(rvd); 577 } else { 578 /* 579 * If we have top-level vdevs that were added but have 580 * not yet been prepared for allocation, do that now. 581 * (It's safe now because the config cache is up to date, 582 * so it will be able to translate the new DVAs.) 583 * See comments in spa_vdev_add() for full details. 584 */ 585 for (c = 0; c < rvd->vdev_children; c++) { 586 vdev_t *tvd = rvd->vdev_child[c]; 587 588 /* 589 * Explicitly skip vdevs that are indirect or 590 * log vdevs that are being removed. The reason 591 * is that both of those can have vdev_ms_array 592 * set to 0 and we wouldn't want to change their 593 * metaslab size nor call vdev_expand() on them. 594 */ 595 if (!vdev_is_concrete(tvd) || 596 (tvd->vdev_islog && tvd->vdev_removing)) 597 continue; 598 599 if (tvd->vdev_ms_array == 0) 600 vdev_metaslab_set_size(tvd); 601 vdev_expand(tvd, txg); 602 } 603 } 604 spa_config_exit(spa, SCL_ALL, FTAG); 605 606 /* 607 * Wait for the mosconfig to be regenerated and synced. 608 */ 609 txg_wait_synced(spa->spa_dsl_pool, txg); 610 611 /* 612 * Update the global config cache to reflect the new mosconfig. 613 */ 614 if (!spa->spa_is_root) { 615 spa_write_cachefile(spa, B_FALSE, 616 what != SPA_CONFIG_UPDATE_POOL, 617 what != SPA_CONFIG_UPDATE_POOL); 618 } 619 620 if (what == SPA_CONFIG_UPDATE_POOL) 621 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS); 622 } 623 624 EXPORT_SYMBOL(spa_config_load); 625 EXPORT_SYMBOL(spa_all_configs); 626 EXPORT_SYMBOL(spa_config_set); 627 EXPORT_SYMBOL(spa_config_generate); 628 EXPORT_SYMBOL(spa_config_update); 629 630 #ifdef __linux__ 631 /* string sysctls require a char array on FreeBSD */ 632 ZFS_MODULE_PARAM(zfs_spa, spa_, config_path, STRING, ZMOD_RD, 633 "SPA config file (/etc/zfs/zpool.cache)"); 634 #endif 635 636 ZFS_MODULE_PARAM(zfs, zfs_, autoimport_disable, INT, ZMOD_RW, 637 "Disable pool import at module load"); 638