xref: /freebsd/sys/contrib/openzfs/module/zfs/spa_config.c (revision 2f10ffc003be396f3fc23cd2888023896560252b)
1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3  * This file and its contents are supplied under the terms of the
4  * Common Development and Distribution License ("CDDL"), version 1.0.
5  * You may only use this file in accordance with the terms of version
6  * 1.0 of the CDDL.
7  *
8  * A full copy of the text of the CDDL should have accompanied this
9  * source.  A copy of the CDDL is also available via the Internet at
10  * https://opensource.org/license/CDDL-1.0.
11  */
12 
13 /*
14  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
15  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
16  * Copyright (c) 2011, 2020 by Delphix. All rights reserved.
17  * Copyright 2017 Joyent, Inc.
18  * Copyright (c) 2021, Colm Buckley <colm@tuatha.org>
19  */
20 
21 #include <sys/spa.h>
22 #include <sys/file.h>
23 #include <sys/fm/fs/zfs.h>
24 #include <sys/spa_impl.h>
25 #include <sys/nvpair.h>
26 #include <sys/fs/zfs.h>
27 #include <sys/vdev_impl.h>
28 #include <sys/zfs_ioctl.h>
29 #include <sys/systeminfo.h>
30 #include <sys/sunddi.h>
31 #include <sys/zfeature.h>
32 #include <sys/zfs_file.h>
33 #include <sys/zfs_context.h>
34 #ifdef _KERNEL
35 #include <sys/zone.h>
36 #endif
37 
38 /*
39  * Pool configuration repository.
40  *
41  * Pool configuration is stored as a packed nvlist on the filesystem.  When
42  * pools are imported they are added to the /etc/zfs/zpool.cache file and
43  * removed from it when exported.  For each cache file, we have a single nvlist
44  * which holds all the configuration information.  Pools can also have the
45  * 'cachefile' property set which allows this config to be stored in an
46  * alternate location under the control of external software.
47  *
48  * The kernel independently maintains an AVL tree of imported pools.  See the
49  * "SPA locking" comment in spa.c.  Whenever a pool configuration is modified
50  * we call spa_write_cachefile() which walks through all the active pools and
51  * writes the updated configuration to to /etc/zfs/zpool.cache file.
52  */
53 
54 static uint64_t spa_config_generation = 1;
55 
56 /*
57  * This can be overridden in userland to preserve an alternate namespace for
58  * userland pools when doing testing.
59  */
60 char *spa_config_path = (char *)ZPOOL_CACHE;
61 
62 static int
63 spa_config_remove(spa_config_dirent_t *dp)
64 {
65 	int error = 0;
66 
67 	/*
68 	 * Remove the cache file.  If zfs_file_unlink() in not supported by the
69 	 * platform fallback to truncating the file which is functionally
70 	 * equivalent.
71 	 */
72 	error = zfs_file_unlink(dp->scd_path);
73 	if (error == EOPNOTSUPP) {
74 		int flags = O_RDWR | O_TRUNC;
75 		zfs_file_t *fp;
76 
77 		error = zfs_file_open(dp->scd_path, flags, 0644, kcred, &fp);
78 		if (error == 0) {
79 			(void) zfs_file_fsync(fp, O_SYNC);
80 			(void) zfs_file_close(fp);
81 		}
82 	}
83 
84 	return (error);
85 }
86 
87 static int
88 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
89 {
90 	size_t buflen;
91 	char *buf;
92 	int oflags = O_RDWR | O_TRUNC | O_CREAT | O_LARGEFILE;
93 	char *temp;
94 	int err;
95 	zfs_file_t *fp;
96 
97 	/*
98 	 * If the nvlist is empty (NULL), then remove the old cachefile.
99 	 */
100 	if (nvl == NULL) {
101 		err = spa_config_remove(dp);
102 		if (err == ENOENT)
103 			err = 0;
104 
105 		return (err);
106 	}
107 
108 	/*
109 	 * Pack the configuration into a buffer.
110 	 */
111 	buf = fnvlist_pack(nvl, &buflen);
112 	temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
113 
114 	/*
115 	 * Write the configuration to disk.  Due to the complexity involved
116 	 * in performing a rename and remove from within the kernel the file
117 	 * is instead truncated and overwritten in place.  This way we always
118 	 * have a consistent view of the data or a zero length file.
119 	 */
120 	err = zfs_file_open(dp->scd_path, oflags, 0644, kcred, &fp);
121 	if (err == 0) {
122 		err = zfs_file_write(fp, buf, buflen, NULL);
123 		if (err == 0)
124 			err = zfs_file_fsync(fp, O_SYNC);
125 
126 		zfs_file_close(fp);
127 		if (err)
128 			(void) spa_config_remove(dp);
129 	}
130 	fnvlist_pack_free(buf, buflen);
131 	kmem_free(temp, MAXPATHLEN);
132 	return (err);
133 }
134 
135 /*
136  * Synchronize pool configuration to disk.  This must be called with the
137  * namespace lock held. Synchronizing the pool cache is typically done after
138  * the configuration has been synced to the MOS. This exposes a window where
139  * the MOS config will have been updated but the cache file has not. If
140  * the system were to crash at that instant then the cached config may not
141  * contain the correct information to open the pool and an explicit import
142  * would be required.
143  */
144 void
145 spa_write_cachefile(spa_t *target, boolean_t removing, boolean_t postsysevent,
146     boolean_t postblkidevent)
147 {
148 	spa_config_dirent_t *dp, *tdp;
149 	nvlist_t *nvl;
150 	const char *pool_name;
151 	boolean_t ccw_failure;
152 	int error = 0;
153 
154 	ASSERT(spa_namespace_held());
155 
156 	if (!(spa_mode_global & SPA_MODE_WRITE))
157 		return;
158 
159 	/*
160 	 * Iterate over all cachefiles for the pool, past or present.  When the
161 	 * cachefile is changed, the new one is pushed onto this list, allowing
162 	 * us to update previous cachefiles that no longer contain this pool.
163 	 */
164 	ccw_failure = B_FALSE;
165 	for (dp = list_head(&target->spa_config_list); dp != NULL;
166 	    dp = list_next(&target->spa_config_list, dp)) {
167 		spa_t *spa = NULL;
168 		if (dp->scd_path == NULL)
169 			continue;
170 
171 		/*
172 		 * Iterate over all pools, adding any matching pools to 'nvl'.
173 		 */
174 		nvl = NULL;
175 		while ((spa = spa_next(spa)) != NULL) {
176 			/*
177 			 * Skip over our own pool if we're about to remove
178 			 * ourselves from the spa namespace or any pool that
179 			 * is readonly. Since we cannot guarantee that a
180 			 * readonly pool would successfully import upon reboot,
181 			 * we don't allow them to be written to the cache file.
182 			 */
183 			if ((spa == target && removing) ||
184 			    !spa_writeable(spa))
185 				continue;
186 
187 			mutex_enter(&spa->spa_props_lock);
188 			tdp = list_head(&spa->spa_config_list);
189 			if (spa->spa_config == NULL ||
190 			    tdp == NULL ||
191 			    tdp->scd_path == NULL ||
192 			    strcmp(tdp->scd_path, dp->scd_path) != 0) {
193 				mutex_exit(&spa->spa_props_lock);
194 				continue;
195 			}
196 
197 			if (nvl == NULL)
198 				nvl = fnvlist_alloc();
199 
200 			if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME)
201 				pool_name = fnvlist_lookup_string(
202 				    spa->spa_config, ZPOOL_CONFIG_POOL_NAME);
203 			else
204 				pool_name = spa_name(spa);
205 
206 			fnvlist_add_nvlist(nvl, pool_name, spa->spa_config);
207 			mutex_exit(&spa->spa_props_lock);
208 		}
209 
210 		error = spa_config_write(dp, nvl);
211 		if (error != 0)
212 			ccw_failure = B_TRUE;
213 		nvlist_free(nvl);
214 	}
215 
216 	if (ccw_failure) {
217 		/*
218 		 * Keep trying so that configuration data is
219 		 * written if/when any temporary filesystem
220 		 * resource issues are resolved.
221 		 */
222 		if (target->spa_ccw_fail_time == 0) {
223 			(void) zfs_ereport_post(
224 			    FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
225 			    target, NULL, NULL, NULL, 0);
226 		}
227 		target->spa_ccw_fail_time = gethrtime();
228 		spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
229 	} else {
230 		/*
231 		 * Do not rate limit future attempts to update
232 		 * the config cache.
233 		 */
234 		target->spa_ccw_fail_time = 0;
235 	}
236 
237 	/*
238 	 * Remove any config entries older than the current one.
239 	 */
240 	dp = list_head(&target->spa_config_list);
241 	while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
242 		list_remove(&target->spa_config_list, tdp);
243 		if (tdp->scd_path != NULL)
244 			spa_strfree(tdp->scd_path);
245 		kmem_free(tdp, sizeof (spa_config_dirent_t));
246 	}
247 
248 	spa_config_generation++;
249 
250 	if (postsysevent)
251 		spa_event_notify(target, NULL, NULL, ESC_ZFS_CONFIG_SYNC);
252 
253 	/*
254 	 * Post udev event to sync blkid information if the pool is created
255 	 * or a new vdev is added to the pool.
256 	 */
257 	if ((target->spa_root_vdev) && postblkidevent) {
258 		vdev_post_kobj_evt(target->spa_root_vdev);
259 		for (int i = 0; i < target->spa_l2cache.sav_count; i++)
260 			vdev_post_kobj_evt(target->spa_l2cache.sav_vdevs[i]);
261 		for (int i = 0; i < target->spa_spares.sav_count; i++)
262 			vdev_post_kobj_evt(target->spa_spares.sav_vdevs[i]);
263 	}
264 }
265 
266 /*
267  * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
268  * and we don't want to allow the local zone to see all the pools anyway.
269  * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
270  * information for all pool visible within the zone.
271  */
272 int
273 spa_all_configs(uint64_t *generation, nvlist_t **pools)
274 {
275 	spa_t *spa = NULL;
276 
277 	if (*generation == spa_config_generation)
278 		return (SET_ERROR(EEXIST));
279 
280 	int error = spa_namespace_enter_interruptible(FTAG);
281 	if (error)
282 		return (SET_ERROR(EINTR));
283 
284 	*pools = fnvlist_alloc();
285 	while ((spa = spa_next(spa)) != NULL) {
286 		if (INGLOBALZONE(curproc) ||
287 		    zone_dataset_visible(spa_name(spa), NULL)) {
288 			mutex_enter(&spa->spa_props_lock);
289 			fnvlist_add_nvlist(*pools, spa_name(spa),
290 			    spa->spa_config);
291 			mutex_exit(&spa->spa_props_lock);
292 		}
293 	}
294 	*generation = spa_config_generation;
295 	spa_namespace_exit(FTAG);
296 
297 	return (0);
298 }
299 
300 void
301 spa_config_set(spa_t *spa, nvlist_t *config)
302 {
303 	mutex_enter(&spa->spa_props_lock);
304 	if (spa->spa_config != NULL && spa->spa_config != config)
305 		nvlist_free(spa->spa_config);
306 	spa->spa_config = config;
307 	mutex_exit(&spa->spa_props_lock);
308 }
309 
310 /*
311  * Generate the pool's configuration based on the current in-core state.
312  *
313  * We infer whether to generate a complete config or just one top-level config
314  * based on whether vd is the root vdev.
315  */
316 nvlist_t *
317 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
318 {
319 	nvlist_t *config, *nvroot;
320 	vdev_t *rvd = spa->spa_root_vdev;
321 	unsigned long hostid = 0;
322 	boolean_t locked = B_FALSE;
323 	uint64_t split_guid;
324 	const char *pool_name;
325 
326 	if (vd == NULL) {
327 		vd = rvd;
328 		locked = B_TRUE;
329 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
330 	}
331 
332 	ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
333 	    (SCL_CONFIG | SCL_STATE));
334 
335 	/*
336 	 * If txg is -1, report the current value of spa->spa_config_txg.
337 	 */
338 	if (txg == -1ULL)
339 		txg = spa->spa_config_txg;
340 
341 	/*
342 	 * Originally, users had to handle spa namespace collisions by either
343 	 * exporting the already imported pool or by specifying a new name for
344 	 * the pool with a conflicting name. In the case of root pools from
345 	 * virtual guests, neither approach to collision resolution is
346 	 * reasonable. This is addressed by extending the new name syntax with
347 	 * an option to specify that the new name is temporary. When specified,
348 	 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
349 	 * to use the previous name, which we do below.
350 	 */
351 	if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
352 		VERIFY0(nvlist_lookup_string(spa->spa_config,
353 		    ZPOOL_CONFIG_POOL_NAME, &pool_name));
354 	} else
355 		pool_name = spa_name(spa);
356 
357 	config = fnvlist_alloc();
358 
359 	fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa));
360 	fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, pool_name);
361 	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa));
362 	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg);
363 	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa));
364 	fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, spa->spa_errata);
365 	fnvlist_add_uint64(config, ZPOOL_CONFIG_MIN_ALLOC, spa->spa_min_alloc);
366 	fnvlist_add_uint64(config, ZPOOL_CONFIG_MAX_ALLOC, spa->spa_max_alloc);
367 	if (spa->spa_comment != NULL)
368 		fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT,
369 		    spa->spa_comment);
370 	if (spa->spa_compatibility != NULL)
371 		fnvlist_add_string(config, ZPOOL_CONFIG_COMPATIBILITY,
372 		    spa->spa_compatibility);
373 
374 	hostid = spa_get_hostid(spa);
375 	if (hostid != 0)
376 		fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid);
377 	fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname()->nodename);
378 
379 	int config_gen_flags = 0;
380 	if (vd != rvd) {
381 		fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
382 		    vd->vdev_top->vdev_guid);
383 		fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
384 		    vd->vdev_guid);
385 		if (vd->vdev_isspare)
386 			fnvlist_add_uint64(config,
387 			    ZPOOL_CONFIG_IS_SPARE, 1ULL);
388 		if (vd->vdev_islog)
389 			fnvlist_add_uint64(config,
390 			    ZPOOL_CONFIG_IS_LOG, 1ULL);
391 		vd = vd->vdev_top;		/* label contains top config */
392 	} else {
393 		/*
394 		 * Only add the (potentially large) split information
395 		 * in the mos config, and not in the vdev labels
396 		 */
397 		if (spa->spa_config_splitting != NULL)
398 			fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
399 			    spa->spa_config_splitting);
400 
401 		fnvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS);
402 
403 		config_gen_flags |= VDEV_CONFIG_MOS;
404 	}
405 
406 	/*
407 	 * Add the top-level config.  We even add this on pools which
408 	 * don't support holes in the namespace.
409 	 */
410 	vdev_top_config_generate(spa, config);
411 
412 	/*
413 	 * If we're splitting, record the original pool's guid.
414 	 */
415 	if (spa->spa_config_splitting != NULL &&
416 	    nvlist_lookup_uint64(spa->spa_config_splitting,
417 	    ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
418 		fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID, split_guid);
419 	}
420 
421 	nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags);
422 	fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot);
423 	nvlist_free(nvroot);
424 
425 	/*
426 	 * Store what's necessary for reading the MOS in the label.
427 	 */
428 	fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
429 	    spa->spa_label_features);
430 
431 	if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
432 		ddt_histogram_t *ddh;
433 		ddt_stat_t *dds;
434 		ddt_object_t *ddo;
435 
436 		ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
437 		ddt_get_dedup_histogram(spa, ddh);
438 		fnvlist_add_uint64_array(config,
439 		    ZPOOL_CONFIG_DDT_HISTOGRAM,
440 		    (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t));
441 		kmem_free(ddh, sizeof (ddt_histogram_t));
442 
443 		ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
444 		ddt_get_dedup_object_stats(spa, ddo);
445 		fnvlist_add_uint64_array(config,
446 		    ZPOOL_CONFIG_DDT_OBJ_STATS,
447 		    (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t));
448 		kmem_free(ddo, sizeof (ddt_object_t));
449 
450 		dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
451 		ddt_get_dedup_stats(spa, dds);
452 		fnvlist_add_uint64_array(config,
453 		    ZPOOL_CONFIG_DDT_STATS,
454 		    (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t));
455 		kmem_free(dds, sizeof (ddt_stat_t));
456 	}
457 
458 	if (locked)
459 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
460 
461 	return (config);
462 }
463 
464 /*
465  * Update all disk labels, generate a fresh config based on the current
466  * in-core state, and sync the global config cache (do not sync the config
467  * cache if this is a booting rootpool).
468  */
469 void
470 spa_config_update(spa_t *spa, int what)
471 {
472 	vdev_t *rvd = spa->spa_root_vdev;
473 	uint64_t txg;
474 	int c;
475 
476 	ASSERT(spa_namespace_held());
477 
478 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
479 	txg = spa_last_synced_txg(spa) + 1;
480 	if (what == SPA_CONFIG_UPDATE_POOL) {
481 		vdev_config_dirty(rvd);
482 	} else {
483 		/*
484 		 * If we have top-level vdevs that were added but have
485 		 * not yet been prepared for allocation, do that now.
486 		 * (It's safe now because the config cache is up to date,
487 		 * so it will be able to translate the new DVAs.)
488 		 * See comments in spa_vdev_add() for full details.
489 		 */
490 		for (c = 0; c < rvd->vdev_children; c++) {
491 			vdev_t *tvd = rvd->vdev_child[c];
492 
493 			/*
494 			 * Explicitly skip vdevs that are indirect or
495 			 * log vdevs that are being removed. The reason
496 			 * is that both of those can have vdev_ms_array
497 			 * set to 0 and we wouldn't want to change their
498 			 * metaslab size nor call vdev_expand() on them.
499 			 */
500 			if (!vdev_is_concrete(tvd) ||
501 			    (tvd->vdev_islog && tvd->vdev_removing))
502 				continue;
503 
504 			if (tvd->vdev_ms_array == 0)
505 				vdev_metaslab_set_size(tvd);
506 			vdev_expand(tvd, txg);
507 		}
508 	}
509 	spa_config_exit(spa, SCL_ALL, FTAG);
510 
511 	/*
512 	 * Wait for the mosconfig to be regenerated and synced.
513 	 */
514 	txg_wait_synced(spa->spa_dsl_pool, txg);
515 
516 	/*
517 	 * Update the global config cache to reflect the new mosconfig.
518 	 */
519 	if (!spa->spa_is_root) {
520 		spa_write_cachefile(spa, B_FALSE,
521 		    what != SPA_CONFIG_UPDATE_POOL,
522 		    what != SPA_CONFIG_UPDATE_POOL);
523 	}
524 
525 	if (what == SPA_CONFIG_UPDATE_POOL)
526 		spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
527 }
528 
529 EXPORT_SYMBOL(spa_all_configs);
530 EXPORT_SYMBOL(spa_config_set);
531 EXPORT_SYMBOL(spa_config_generate);
532 EXPORT_SYMBOL(spa_config_update);
533 
534 #ifdef __linux__
535 /* string sysctls require a char array on FreeBSD */
536 ZFS_MODULE_PARAM(zfs_spa, spa_, config_path, STRING, ZMOD_RD,
537 	"SPA config file (/etc/zfs/zpool.cache)");
538 #endif
539