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. When
52 * pools are imported they are added to the /etc/zfs/zpool.cache file and
53 * removed from it when exported. For each cache file, we have a single nvlist
54 * which holds all the configuration information. Pools can also have the
55 * 'cachefile' property set which allows this config to be stored in an
56 * alternate location under the control of external software.
57 *
58 * The kernel independantly maintains an AVL tree of imported pools. See the
59 * "SPA locking" comment in spa.c. Whenever a pool configuration is modified
60 * we call spa_write_cachefile() which walks through all the active pools and
61 * writes the updated configuration to to /etc/zfs/zpool.cache file.
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
72 static int
spa_config_remove(spa_config_dirent_t * dp)73 spa_config_remove(spa_config_dirent_t *dp)
74 {
75 int error = 0;
76
77 /*
78 * Remove the cache file. If zfs_file_unlink() in not supported by the
79 * platform fallback to truncating the file which is functionally
80 * equivalent.
81 */
82 error = zfs_file_unlink(dp->scd_path);
83 if (error == EOPNOTSUPP) {
84 int flags = O_RDWR | O_TRUNC;
85 zfs_file_t *fp;
86
87 error = zfs_file_open(dp->scd_path, flags, 0644, &fp);
88 if (error == 0) {
89 (void) zfs_file_fsync(fp, O_SYNC);
90 (void) zfs_file_close(fp);
91 }
92 }
93
94 return (error);
95 }
96
97 static int
spa_config_write(spa_config_dirent_t * dp,nvlist_t * nvl)98 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
99 {
100 size_t buflen;
101 char *buf;
102 int oflags = O_RDWR | O_TRUNC | O_CREAT | O_LARGEFILE;
103 char *temp;
104 int err;
105 zfs_file_t *fp;
106
107 /*
108 * If the nvlist is empty (NULL), then remove the old cachefile.
109 */
110 if (nvl == NULL) {
111 err = spa_config_remove(dp);
112 if (err == ENOENT)
113 err = 0;
114
115 return (err);
116 }
117
118 /*
119 * Pack the configuration into a buffer.
120 */
121 buf = fnvlist_pack(nvl, &buflen);
122 temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
123
124 /*
125 * Write the configuration to disk. Due to the complexity involved
126 * in performing a rename and remove from within the kernel the file
127 * is instead truncated and overwritten in place. This way we always
128 * have a consistent view of the data or a zero length file.
129 */
130 err = zfs_file_open(dp->scd_path, oflags, 0644, &fp);
131 if (err == 0) {
132 err = zfs_file_write(fp, buf, buflen, NULL);
133 if (err == 0)
134 err = zfs_file_fsync(fp, O_SYNC);
135
136 zfs_file_close(fp);
137 if (err)
138 (void) spa_config_remove(dp);
139 }
140 fnvlist_pack_free(buf, buflen);
141 kmem_free(temp, MAXPATHLEN);
142 return (err);
143 }
144
145 /*
146 * Synchronize pool configuration to disk. This must be called with the
147 * namespace lock held. Synchronizing the pool cache is typically done after
148 * the configuration has been synced to the MOS. This exposes a window where
149 * the MOS config will have been updated but the cache file has not. If
150 * the system were to crash at that instant then the cached config may not
151 * contain the correct information to open the pool and an explicit import
152 * would be required.
153 */
154 void
spa_write_cachefile(spa_t * target,boolean_t removing,boolean_t postsysevent,boolean_t postblkidevent)155 spa_write_cachefile(spa_t *target, boolean_t removing, boolean_t postsysevent,
156 boolean_t postblkidevent)
157 {
158 spa_config_dirent_t *dp, *tdp;
159 nvlist_t *nvl;
160 const char *pool_name;
161 boolean_t ccw_failure;
162 int error = 0;
163
164 ASSERT(MUTEX_HELD(&spa_namespace_lock));
165
166 if (!(spa_mode_global & SPA_MODE_WRITE))
167 return;
168
169 /*
170 * Iterate over all cachefiles for the pool, past or present. When the
171 * cachefile is changed, the new one is pushed onto this list, allowing
172 * us to update previous cachefiles that no longer contain this pool.
173 */
174 ccw_failure = B_FALSE;
175 for (dp = list_head(&target->spa_config_list); dp != NULL;
176 dp = list_next(&target->spa_config_list, dp)) {
177 spa_t *spa = NULL;
178 if (dp->scd_path == NULL)
179 continue;
180
181 /*
182 * Iterate over all pools, adding any matching pools to 'nvl'.
183 */
184 nvl = NULL;
185 while ((spa = spa_next(spa)) != NULL) {
186 /*
187 * Skip over our own pool if we're about to remove
188 * ourselves from the spa namespace or any pool that
189 * is readonly. Since we cannot guarantee that a
190 * readonly pool would successfully import upon reboot,
191 * we don't allow them to be written to the cache file.
192 */
193 if ((spa == target && removing) ||
194 !spa_writeable(spa))
195 continue;
196
197 mutex_enter(&spa->spa_props_lock);
198 tdp = list_head(&spa->spa_config_list);
199 if (spa->spa_config == NULL ||
200 tdp == NULL ||
201 tdp->scd_path == NULL ||
202 strcmp(tdp->scd_path, dp->scd_path) != 0) {
203 mutex_exit(&spa->spa_props_lock);
204 continue;
205 }
206
207 if (nvl == NULL)
208 nvl = fnvlist_alloc();
209
210 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME)
211 pool_name = fnvlist_lookup_string(
212 spa->spa_config, ZPOOL_CONFIG_POOL_NAME);
213 else
214 pool_name = spa_name(spa);
215
216 fnvlist_add_nvlist(nvl, pool_name, spa->spa_config);
217 mutex_exit(&spa->spa_props_lock);
218 }
219
220 error = spa_config_write(dp, nvl);
221 if (error != 0)
222 ccw_failure = B_TRUE;
223 nvlist_free(nvl);
224 }
225
226 if (ccw_failure) {
227 /*
228 * Keep trying so that configuration data is
229 * written if/when any temporary filesystem
230 * resource issues are resolved.
231 */
232 if (target->spa_ccw_fail_time == 0) {
233 (void) zfs_ereport_post(
234 FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
235 target, NULL, NULL, NULL, 0);
236 }
237 target->spa_ccw_fail_time = gethrtime();
238 spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
239 } else {
240 /*
241 * Do not rate limit future attempts to update
242 * the config cache.
243 */
244 target->spa_ccw_fail_time = 0;
245 }
246
247 /*
248 * Remove any config entries older than the current one.
249 */
250 dp = list_head(&target->spa_config_list);
251 while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
252 list_remove(&target->spa_config_list, tdp);
253 if (tdp->scd_path != NULL)
254 spa_strfree(tdp->scd_path);
255 kmem_free(tdp, sizeof (spa_config_dirent_t));
256 }
257
258 spa_config_generation++;
259
260 if (postsysevent)
261 spa_event_notify(target, NULL, NULL, ESC_ZFS_CONFIG_SYNC);
262
263 /*
264 * Post udev event to sync blkid information if the pool is created
265 * or a new vdev is added to the pool.
266 */
267 if ((target->spa_root_vdev) && postblkidevent) {
268 vdev_post_kobj_evt(target->spa_root_vdev);
269 for (int i = 0; i < target->spa_l2cache.sav_count; i++)
270 vdev_post_kobj_evt(target->spa_l2cache.sav_vdevs[i]);
271 for (int i = 0; i < target->spa_spares.sav_count; i++)
272 vdev_post_kobj_evt(target->spa_spares.sav_vdevs[i]);
273 }
274 }
275
276 /*
277 * Sigh. Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
278 * and we don't want to allow the local zone to see all the pools anyway.
279 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
280 * information for all pool visible within the zone.
281 */
282 int
spa_all_configs(uint64_t * generation,nvlist_t ** pools)283 spa_all_configs(uint64_t *generation, nvlist_t **pools)
284 {
285 spa_t *spa = NULL;
286
287 if (*generation == spa_config_generation)
288 return (SET_ERROR(EEXIST));
289
290 int error = mutex_enter_interruptible(&spa_namespace_lock);
291 if (error)
292 return (SET_ERROR(EINTR));
293
294 *pools = fnvlist_alloc();
295 while ((spa = spa_next(spa)) != NULL) {
296 if (INGLOBALZONE(curproc) ||
297 zone_dataset_visible(spa_name(spa), NULL)) {
298 mutex_enter(&spa->spa_props_lock);
299 fnvlist_add_nvlist(*pools, spa_name(spa),
300 spa->spa_config);
301 mutex_exit(&spa->spa_props_lock);
302 }
303 }
304 *generation = spa_config_generation;
305 mutex_exit(&spa_namespace_lock);
306
307 return (0);
308 }
309
310 void
spa_config_set(spa_t * spa,nvlist_t * config)311 spa_config_set(spa_t *spa, nvlist_t *config)
312 {
313 mutex_enter(&spa->spa_props_lock);
314 if (spa->spa_config != NULL && spa->spa_config != config)
315 nvlist_free(spa->spa_config);
316 spa->spa_config = config;
317 mutex_exit(&spa->spa_props_lock);
318 }
319
320 /*
321 * Generate the pool's configuration based on the current in-core state.
322 *
323 * We infer whether to generate a complete config or just one top-level config
324 * based on whether vd is the root vdev.
325 */
326 nvlist_t *
spa_config_generate(spa_t * spa,vdev_t * vd,uint64_t txg,int getstats)327 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
328 {
329 nvlist_t *config, *nvroot;
330 vdev_t *rvd = spa->spa_root_vdev;
331 unsigned long hostid = 0;
332 boolean_t locked = B_FALSE;
333 uint64_t split_guid;
334 const char *pool_name;
335
336 if (vd == NULL) {
337 vd = rvd;
338 locked = B_TRUE;
339 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
340 }
341
342 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
343 (SCL_CONFIG | SCL_STATE));
344
345 /*
346 * If txg is -1, report the current value of spa->spa_config_txg.
347 */
348 if (txg == -1ULL)
349 txg = spa->spa_config_txg;
350
351 /*
352 * Originally, users had to handle spa namespace collisions by either
353 * exporting the already imported pool or by specifying a new name for
354 * the pool with a conflicting name. In the case of root pools from
355 * virtual guests, neither approach to collision resolution is
356 * reasonable. This is addressed by extending the new name syntax with
357 * an option to specify that the new name is temporary. When specified,
358 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
359 * to use the previous name, which we do below.
360 */
361 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
362 VERIFY0(nvlist_lookup_string(spa->spa_config,
363 ZPOOL_CONFIG_POOL_NAME, &pool_name));
364 } else
365 pool_name = spa_name(spa);
366
367 config = fnvlist_alloc();
368
369 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa));
370 fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, pool_name);
371 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa));
372 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg);
373 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa));
374 fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, spa->spa_errata);
375 if (spa->spa_comment != NULL)
376 fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT,
377 spa->spa_comment);
378 if (spa->spa_compatibility != NULL)
379 fnvlist_add_string(config, ZPOOL_CONFIG_COMPATIBILITY,
380 spa->spa_compatibility);
381
382 hostid = spa_get_hostid(spa);
383 if (hostid != 0)
384 fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid);
385 fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname()->nodename);
386
387 int config_gen_flags = 0;
388 if (vd != rvd) {
389 fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
390 vd->vdev_top->vdev_guid);
391 fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
392 vd->vdev_guid);
393 if (vd->vdev_isspare)
394 fnvlist_add_uint64(config,
395 ZPOOL_CONFIG_IS_SPARE, 1ULL);
396 if (vd->vdev_islog)
397 fnvlist_add_uint64(config,
398 ZPOOL_CONFIG_IS_LOG, 1ULL);
399 vd = vd->vdev_top; /* label contains top config */
400 } else {
401 /*
402 * Only add the (potentially large) split information
403 * in the mos config, and not in the vdev labels
404 */
405 if (spa->spa_config_splitting != NULL)
406 fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
407 spa->spa_config_splitting);
408
409 fnvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS);
410
411 config_gen_flags |= VDEV_CONFIG_MOS;
412 }
413
414 /*
415 * Add the top-level config. We even add this on pools which
416 * don't support holes in the namespace.
417 */
418 vdev_top_config_generate(spa, config);
419
420 /*
421 * If we're splitting, record the original pool's guid.
422 */
423 if (spa->spa_config_splitting != NULL &&
424 nvlist_lookup_uint64(spa->spa_config_splitting,
425 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
426 fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID, split_guid);
427 }
428
429 nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags);
430 fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot);
431 nvlist_free(nvroot);
432
433 /*
434 * Store what's necessary for reading the MOS in the label.
435 */
436 fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
437 spa->spa_label_features);
438
439 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
440 ddt_histogram_t *ddh;
441 ddt_stat_t *dds;
442 ddt_object_t *ddo;
443
444 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
445 ddt_get_dedup_histogram(spa, ddh);
446 fnvlist_add_uint64_array(config,
447 ZPOOL_CONFIG_DDT_HISTOGRAM,
448 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t));
449 kmem_free(ddh, sizeof (ddt_histogram_t));
450
451 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
452 ddt_get_dedup_object_stats(spa, ddo);
453 fnvlist_add_uint64_array(config,
454 ZPOOL_CONFIG_DDT_OBJ_STATS,
455 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t));
456 kmem_free(ddo, sizeof (ddt_object_t));
457
458 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
459 ddt_get_dedup_stats(spa, dds);
460 fnvlist_add_uint64_array(config,
461 ZPOOL_CONFIG_DDT_STATS,
462 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t));
463 kmem_free(dds, sizeof (ddt_stat_t));
464 }
465
466 if (locked)
467 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
468
469 return (config);
470 }
471
472 /*
473 * Update all disk labels, generate a fresh config based on the current
474 * in-core state, and sync the global config cache (do not sync the config
475 * cache if this is a booting rootpool).
476 */
477 void
spa_config_update(spa_t * spa,int what)478 spa_config_update(spa_t *spa, int what)
479 {
480 vdev_t *rvd = spa->spa_root_vdev;
481 uint64_t txg;
482 int c;
483
484 ASSERT(MUTEX_HELD(&spa_namespace_lock));
485
486 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
487 txg = spa_last_synced_txg(spa) + 1;
488 if (what == SPA_CONFIG_UPDATE_POOL) {
489 vdev_config_dirty(rvd);
490 } else {
491 /*
492 * If we have top-level vdevs that were added but have
493 * not yet been prepared for allocation, do that now.
494 * (It's safe now because the config cache is up to date,
495 * so it will be able to translate the new DVAs.)
496 * See comments in spa_vdev_add() for full details.
497 */
498 for (c = 0; c < rvd->vdev_children; c++) {
499 vdev_t *tvd = rvd->vdev_child[c];
500
501 /*
502 * Explicitly skip vdevs that are indirect or
503 * log vdevs that are being removed. The reason
504 * is that both of those can have vdev_ms_array
505 * set to 0 and we wouldn't want to change their
506 * metaslab size nor call vdev_expand() on them.
507 */
508 if (!vdev_is_concrete(tvd) ||
509 (tvd->vdev_islog && tvd->vdev_removing))
510 continue;
511
512 if (tvd->vdev_ms_array == 0)
513 vdev_metaslab_set_size(tvd);
514 vdev_expand(tvd, txg);
515 }
516 }
517 spa_config_exit(spa, SCL_ALL, FTAG);
518
519 /*
520 * Wait for the mosconfig to be regenerated and synced.
521 */
522 txg_wait_synced(spa->spa_dsl_pool, txg);
523
524 /*
525 * Update the global config cache to reflect the new mosconfig.
526 */
527 if (!spa->spa_is_root) {
528 spa_write_cachefile(spa, B_FALSE,
529 what != SPA_CONFIG_UPDATE_POOL,
530 what != SPA_CONFIG_UPDATE_POOL);
531 }
532
533 if (what == SPA_CONFIG_UPDATE_POOL)
534 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
535 }
536
537 EXPORT_SYMBOL(spa_all_configs);
538 EXPORT_SYMBOL(spa_config_set);
539 EXPORT_SYMBOL(spa_config_generate);
540 EXPORT_SYMBOL(spa_config_update);
541
542 #ifdef __linux__
543 /* string sysctls require a char array on FreeBSD */
544 ZFS_MODULE_PARAM(zfs_spa, spa_, config_path, STRING, ZMOD_RD,
545 "SPA config file (/etc/zfs/zpool.cache)");
546 #endif
547