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
spa_config_load(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
spa_config_remove(spa_config_dirent_t * dp)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
spa_config_write(spa_config_dirent_t * dp,nvlist_t * nvl)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
spa_write_cachefile(spa_t * target,boolean_t removing,boolean_t postsysevent,boolean_t postblkidevent)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 const 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 int
spa_all_configs(uint64_t * generation,nvlist_t ** pools)371 spa_all_configs(uint64_t *generation, nvlist_t **pools)
372 {
373 spa_t *spa = NULL;
374
375 if (*generation == spa_config_generation)
376 return (SET_ERROR(EEXIST));
377
378 int error = mutex_enter_interruptible(&spa_namespace_lock);
379 if (error)
380 return (SET_ERROR(EINTR));
381
382 *pools = fnvlist_alloc();
383 while ((spa = spa_next(spa)) != NULL) {
384 if (INGLOBALZONE(curproc) ||
385 zone_dataset_visible(spa_name(spa), NULL)) {
386 mutex_enter(&spa->spa_props_lock);
387 fnvlist_add_nvlist(*pools, spa_name(spa),
388 spa->spa_config);
389 mutex_exit(&spa->spa_props_lock);
390 }
391 }
392 *generation = spa_config_generation;
393 mutex_exit(&spa_namespace_lock);
394
395 return (0);
396 }
397
398 void
spa_config_set(spa_t * spa,nvlist_t * config)399 spa_config_set(spa_t *spa, nvlist_t *config)
400 {
401 mutex_enter(&spa->spa_props_lock);
402 if (spa->spa_config != NULL && spa->spa_config != config)
403 nvlist_free(spa->spa_config);
404 spa->spa_config = config;
405 mutex_exit(&spa->spa_props_lock);
406 }
407
408 /*
409 * Generate the pool's configuration based on the current in-core state.
410 *
411 * We infer whether to generate a complete config or just one top-level config
412 * based on whether vd is the root vdev.
413 */
414 nvlist_t *
spa_config_generate(spa_t * spa,vdev_t * vd,uint64_t txg,int getstats)415 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
416 {
417 nvlist_t *config, *nvroot;
418 vdev_t *rvd = spa->spa_root_vdev;
419 unsigned long hostid = 0;
420 boolean_t locked = B_FALSE;
421 uint64_t split_guid;
422 const char *pool_name;
423
424 if (vd == NULL) {
425 vd = rvd;
426 locked = B_TRUE;
427 spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
428 }
429
430 ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
431 (SCL_CONFIG | SCL_STATE));
432
433 /*
434 * If txg is -1, report the current value of spa->spa_config_txg.
435 */
436 if (txg == -1ULL)
437 txg = spa->spa_config_txg;
438
439 /*
440 * Originally, users had to handle spa namespace collisions by either
441 * exporting the already imported pool or by specifying a new name for
442 * the pool with a conflicting name. In the case of root pools from
443 * virtual guests, neither approach to collision resolution is
444 * reasonable. This is addressed by extending the new name syntax with
445 * an option to specify that the new name is temporary. When specified,
446 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
447 * to use the previous name, which we do below.
448 */
449 if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
450 VERIFY0(nvlist_lookup_string(spa->spa_config,
451 ZPOOL_CONFIG_POOL_NAME, &pool_name));
452 } else
453 pool_name = spa_name(spa);
454
455 config = fnvlist_alloc();
456
457 fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa));
458 fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, pool_name);
459 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa));
460 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg);
461 fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa));
462 fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, spa->spa_errata);
463 if (spa->spa_comment != NULL)
464 fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT,
465 spa->spa_comment);
466 if (spa->spa_compatibility != NULL)
467 fnvlist_add_string(config, ZPOOL_CONFIG_COMPATIBILITY,
468 spa->spa_compatibility);
469
470 hostid = spa_get_hostid(spa);
471 if (hostid != 0)
472 fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid);
473 fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname()->nodename);
474
475 int config_gen_flags = 0;
476 if (vd != rvd) {
477 fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
478 vd->vdev_top->vdev_guid);
479 fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
480 vd->vdev_guid);
481 if (vd->vdev_isspare)
482 fnvlist_add_uint64(config,
483 ZPOOL_CONFIG_IS_SPARE, 1ULL);
484 if (vd->vdev_islog)
485 fnvlist_add_uint64(config,
486 ZPOOL_CONFIG_IS_LOG, 1ULL);
487 vd = vd->vdev_top; /* label contains top config */
488 } else {
489 /*
490 * Only add the (potentially large) split information
491 * in the mos config, and not in the vdev labels
492 */
493 if (spa->spa_config_splitting != NULL)
494 fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
495 spa->spa_config_splitting);
496
497 fnvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS);
498
499 config_gen_flags |= VDEV_CONFIG_MOS;
500 }
501
502 /*
503 * Add the top-level config. We even add this on pools which
504 * don't support holes in the namespace.
505 */
506 vdev_top_config_generate(spa, config);
507
508 /*
509 * If we're splitting, record the original pool's guid.
510 */
511 if (spa->spa_config_splitting != NULL &&
512 nvlist_lookup_uint64(spa->spa_config_splitting,
513 ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
514 fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID, split_guid);
515 }
516
517 nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags);
518 fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot);
519 nvlist_free(nvroot);
520
521 /*
522 * Store what's necessary for reading the MOS in the label.
523 */
524 fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
525 spa->spa_label_features);
526
527 if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
528 ddt_histogram_t *ddh;
529 ddt_stat_t *dds;
530 ddt_object_t *ddo;
531
532 ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
533 ddt_get_dedup_histogram(spa, ddh);
534 fnvlist_add_uint64_array(config,
535 ZPOOL_CONFIG_DDT_HISTOGRAM,
536 (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t));
537 kmem_free(ddh, sizeof (ddt_histogram_t));
538
539 ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
540 ddt_get_dedup_object_stats(spa, ddo);
541 fnvlist_add_uint64_array(config,
542 ZPOOL_CONFIG_DDT_OBJ_STATS,
543 (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t));
544 kmem_free(ddo, sizeof (ddt_object_t));
545
546 dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
547 ddt_get_dedup_stats(spa, dds);
548 fnvlist_add_uint64_array(config,
549 ZPOOL_CONFIG_DDT_STATS,
550 (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t));
551 kmem_free(dds, sizeof (ddt_stat_t));
552 }
553
554 if (locked)
555 spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
556
557 return (config);
558 }
559
560 /*
561 * Update all disk labels, generate a fresh config based on the current
562 * in-core state, and sync the global config cache (do not sync the config
563 * cache if this is a booting rootpool).
564 */
565 void
spa_config_update(spa_t * spa,int what)566 spa_config_update(spa_t *spa, int what)
567 {
568 vdev_t *rvd = spa->spa_root_vdev;
569 uint64_t txg;
570 int c;
571
572 ASSERT(MUTEX_HELD(&spa_namespace_lock));
573
574 spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
575 txg = spa_last_synced_txg(spa) + 1;
576 if (what == SPA_CONFIG_UPDATE_POOL) {
577 vdev_config_dirty(rvd);
578 } else {
579 /*
580 * If we have top-level vdevs that were added but have
581 * not yet been prepared for allocation, do that now.
582 * (It's safe now because the config cache is up to date,
583 * so it will be able to translate the new DVAs.)
584 * See comments in spa_vdev_add() for full details.
585 */
586 for (c = 0; c < rvd->vdev_children; c++) {
587 vdev_t *tvd = rvd->vdev_child[c];
588
589 /*
590 * Explicitly skip vdevs that are indirect or
591 * log vdevs that are being removed. The reason
592 * is that both of those can have vdev_ms_array
593 * set to 0 and we wouldn't want to change their
594 * metaslab size nor call vdev_expand() on them.
595 */
596 if (!vdev_is_concrete(tvd) ||
597 (tvd->vdev_islog && tvd->vdev_removing))
598 continue;
599
600 if (tvd->vdev_ms_array == 0)
601 vdev_metaslab_set_size(tvd);
602 vdev_expand(tvd, txg);
603 }
604 }
605 spa_config_exit(spa, SCL_ALL, FTAG);
606
607 /*
608 * Wait for the mosconfig to be regenerated and synced.
609 */
610 txg_wait_synced(spa->spa_dsl_pool, txg);
611
612 /*
613 * Update the global config cache to reflect the new mosconfig.
614 */
615 if (!spa->spa_is_root) {
616 spa_write_cachefile(spa, B_FALSE,
617 what != SPA_CONFIG_UPDATE_POOL,
618 what != SPA_CONFIG_UPDATE_POOL);
619 }
620
621 if (what == SPA_CONFIG_UPDATE_POOL)
622 spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
623 }
624
625 EXPORT_SYMBOL(spa_config_load);
626 EXPORT_SYMBOL(spa_all_configs);
627 EXPORT_SYMBOL(spa_config_set);
628 EXPORT_SYMBOL(spa_config_generate);
629 EXPORT_SYMBOL(spa_config_update);
630
631 #ifdef __linux__
632 /* string sysctls require a char array on FreeBSD */
633 ZFS_MODULE_PARAM(zfs_spa, spa_, config_path, STRING, ZMOD_RD,
634 "SPA config file (/etc/zfs/zpool.cache)");
635 #endif
636
637 ZFS_MODULE_PARAM(zfs, zfs_, autoimport_disable, INT, ZMOD_RW,
638 "Disable pool import at module load");
639