xref: /freebsd/sys/contrib/openzfs/module/zfs/spa_config.c (revision 067bb8206d86526a72c7e7ba6285103b2cad3ad3)
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 http://www.opensolaris.org/os/licensing.
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  */
28 
29 #include <sys/spa.h>
30 #include <sys/file.h>
31 #include <sys/fm/fs/zfs.h>
32 #include <sys/spa_impl.h>
33 #include <sys/nvpair.h>
34 #include <sys/uio.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 #ifdef _KERNEL
43 #include <sys/zone.h>
44 #endif
45 
46 /*
47  * Pool configuration repository.
48  *
49  * Pool configuration is stored as a packed nvlist on the filesystem.  By
50  * default, all pools are stored in /etc/zfs/zpool.cache and loaded on boot
51  * (when the ZFS module is loaded).  Pools can also have the 'cachefile'
52  * property set that allows them to be stored in an alternate location until
53  * the control of external software.
54  *
55  * For each cache file, we have a single nvlist which holds all the
56  * configuration information.  When the module loads, we read this information
57  * from /etc/zfs/zpool.cache and populate the SPA namespace.  This namespace is
58  * maintained independently in spa.c.  Whenever the namespace is modified, or
59  * the configuration of a pool is changed, we call spa_write_cachefile(), which
60  * walks through all the active pools and writes the configuration to disk.
61  */
62 
63 static uint64_t spa_config_generation = 1;
64 
65 /*
66  * This can be overridden in userland to preserve an alternate namespace for
67  * userland pools when doing testing.
68  */
69 char *spa_config_path = ZPOOL_CACHE;
70 int zfs_autoimport_disable = 1;
71 
72 /*
73  * Called when the module is first loaded, this routine loads the configuration
74  * file into the SPA namespace.  It does not actually open or load the pools; it
75  * only populates the namespace.
76  */
77 void
78 spa_config_load(void)
79 {
80 	void *buf = NULL;
81 	nvlist_t *nvlist, *child;
82 	nvpair_t *nvpair;
83 	char *pathname;
84 	zfs_file_t *fp;
85 	zfs_file_attr_t zfa;
86 	uint64_t fsize;
87 	int err;
88 
89 #ifdef _KERNEL
90 	if (zfs_autoimport_disable)
91 		return;
92 #endif
93 
94 	/*
95 	 * Open the configuration file.
96 	 */
97 	pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
98 
99 	(void) snprintf(pathname, MAXPATHLEN, "%s", spa_config_path);
100 
101 	err = zfs_file_open(pathname, O_RDONLY, 0, &fp);
102 
103 #ifdef __FreeBSD__
104 	if (err)
105 		err = zfs_file_open(ZPOOL_CACHE_BOOT, O_RDONLY, 0, &fp);
106 #endif
107 	kmem_free(pathname, MAXPATHLEN);
108 
109 	if (err)
110 		return;
111 
112 	if (zfs_file_getattr(fp, &zfa))
113 		goto out;
114 
115 	fsize = zfa.zfa_size;
116 	buf = kmem_alloc(fsize, KM_SLEEP);
117 
118 	/*
119 	 * Read the nvlist from the file.
120 	 */
121 	if (zfs_file_read(fp, buf, fsize, NULL) < 0)
122 		goto out;
123 
124 	/*
125 	 * Unpack the nvlist.
126 	 */
127 	if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
128 		goto out;
129 
130 	/*
131 	 * Iterate over all elements in the nvlist, creating a new spa_t for
132 	 * each one with the specified configuration.
133 	 */
134 	mutex_enter(&spa_namespace_lock);
135 	nvpair = NULL;
136 	while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
137 		if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
138 			continue;
139 
140 		child = fnvpair_value_nvlist(nvpair);
141 
142 		if (spa_lookup(nvpair_name(nvpair)) != NULL)
143 			continue;
144 		(void) spa_add(nvpair_name(nvpair), child, NULL);
145 	}
146 	mutex_exit(&spa_namespace_lock);
147 
148 	nvlist_free(nvlist);
149 
150 out:
151 	if (buf != NULL)
152 		kmem_free(buf, fsize);
153 
154 	zfs_file_close(fp);
155 }
156 
157 static int
158 spa_config_remove(spa_config_dirent_t *dp)
159 {
160 	int error = 0;
161 
162 	/*
163 	 * Remove the cache file.  If zfs_file_unlink() in not supported by the
164 	 * platform fallback to truncating the file which is functionally
165 	 * equivalent.
166 	 */
167 	error = zfs_file_unlink(dp->scd_path);
168 	if (error == EOPNOTSUPP) {
169 		int flags = O_RDWR | O_TRUNC;
170 		zfs_file_t *fp;
171 
172 		error = zfs_file_open(dp->scd_path, flags, 0644, &fp);
173 		if (error == 0) {
174 			(void) zfs_file_fsync(fp, O_SYNC);
175 			(void) zfs_file_close(fp);
176 		}
177 	}
178 
179 	return (error);
180 }
181 
182 static int
183 spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
184 {
185 	size_t buflen;
186 	char *buf;
187 	int oflags = O_RDWR | O_TRUNC | O_CREAT | O_LARGEFILE;
188 	char *temp;
189 	int err;
190 	zfs_file_t *fp;
191 
192 	/*
193 	 * If the nvlist is empty (NULL), then remove the old cachefile.
194 	 */
195 	if (nvl == NULL) {
196 		err = spa_config_remove(dp);
197 		if (err == ENOENT)
198 			err = 0;
199 
200 		return (err);
201 	}
202 
203 	/*
204 	 * Pack the configuration into a buffer.
205 	 */
206 	buf = fnvlist_pack(nvl, &buflen);
207 	temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
208 
209 	/*
210 	 * Write the configuration to disk.  Due to the complexity involved
211 	 * in performing a rename and remove from within the kernel the file
212 	 * is instead truncated and overwritten in place.  This way we always
213 	 * have a consistent view of the data or a zero length file.
214 	 */
215 	err = zfs_file_open(dp->scd_path, oflags, 0644, &fp);
216 	if (err == 0) {
217 		err = zfs_file_write(fp, buf, buflen, NULL);
218 		if (err == 0)
219 			err = zfs_file_fsync(fp, O_SYNC);
220 
221 		zfs_file_close(fp);
222 		if (err)
223 			(void) spa_config_remove(dp);
224 	}
225 	fnvlist_pack_free(buf, buflen);
226 	kmem_free(temp, MAXPATHLEN);
227 	return (err);
228 }
229 
230 /*
231  * Synchronize pool configuration to disk.  This must be called with the
232  * namespace lock held. Synchronizing the pool cache is typically done after
233  * the configuration has been synced to the MOS. This exposes a window where
234  * the MOS config will have been updated but the cache file has not. If
235  * the system were to crash at that instant then the cached config may not
236  * contain the correct information to open the pool and an explicit import
237  * would be required.
238  */
239 void
240 spa_write_cachefile(spa_t *target, boolean_t removing, boolean_t postsysevent)
241 {
242 	spa_config_dirent_t *dp, *tdp;
243 	nvlist_t *nvl;
244 	char *pool_name;
245 	boolean_t ccw_failure;
246 	int error = 0;
247 
248 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
249 
250 	if (!(spa_mode_global & SPA_MODE_WRITE))
251 		return;
252 
253 	/*
254 	 * Iterate over all cachefiles for the pool, past or present.  When the
255 	 * cachefile is changed, the new one is pushed onto this list, allowing
256 	 * us to update previous cachefiles that no longer contain this pool.
257 	 */
258 	ccw_failure = B_FALSE;
259 	for (dp = list_head(&target->spa_config_list); dp != NULL;
260 	    dp = list_next(&target->spa_config_list, dp)) {
261 		spa_t *spa = NULL;
262 		if (dp->scd_path == NULL)
263 			continue;
264 
265 		/*
266 		 * Iterate over all pools, adding any matching pools to 'nvl'.
267 		 */
268 		nvl = NULL;
269 		while ((spa = spa_next(spa)) != NULL) {
270 			/*
271 			 * Skip over our own pool if we're about to remove
272 			 * ourselves from the spa namespace or any pool that
273 			 * is readonly. Since we cannot guarantee that a
274 			 * readonly pool would successfully import upon reboot,
275 			 * we don't allow them to be written to the cache file.
276 			 */
277 			if ((spa == target && removing) ||
278 			    !spa_writeable(spa))
279 				continue;
280 
281 			mutex_enter(&spa->spa_props_lock);
282 			tdp = list_head(&spa->spa_config_list);
283 			if (spa->spa_config == NULL ||
284 			    tdp == NULL ||
285 			    tdp->scd_path == NULL ||
286 			    strcmp(tdp->scd_path, dp->scd_path) != 0) {
287 				mutex_exit(&spa->spa_props_lock);
288 				continue;
289 			}
290 
291 			if (nvl == NULL)
292 				nvl = fnvlist_alloc();
293 
294 			if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME)
295 				pool_name = fnvlist_lookup_string(
296 				    spa->spa_config, ZPOOL_CONFIG_POOL_NAME);
297 			else
298 				pool_name = spa_name(spa);
299 
300 			fnvlist_add_nvlist(nvl, pool_name, spa->spa_config);
301 			mutex_exit(&spa->spa_props_lock);
302 		}
303 
304 		error = spa_config_write(dp, nvl);
305 		if (error != 0)
306 			ccw_failure = B_TRUE;
307 		nvlist_free(nvl);
308 	}
309 
310 	if (ccw_failure) {
311 		/*
312 		 * Keep trying so that configuration data is
313 		 * written if/when any temporary filesystem
314 		 * resource issues are resolved.
315 		 */
316 		if (target->spa_ccw_fail_time == 0) {
317 			(void) zfs_ereport_post(
318 			    FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
319 			    target, NULL, NULL, NULL, 0);
320 		}
321 		target->spa_ccw_fail_time = gethrtime();
322 		spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
323 	} else {
324 		/*
325 		 * Do not rate limit future attempts to update
326 		 * the config cache.
327 		 */
328 		target->spa_ccw_fail_time = 0;
329 	}
330 
331 	/*
332 	 * Remove any config entries older than the current one.
333 	 */
334 	dp = list_head(&target->spa_config_list);
335 	while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
336 		list_remove(&target->spa_config_list, tdp);
337 		if (tdp->scd_path != NULL)
338 			spa_strfree(tdp->scd_path);
339 		kmem_free(tdp, sizeof (spa_config_dirent_t));
340 	}
341 
342 	spa_config_generation++;
343 
344 	if (postsysevent)
345 		spa_event_notify(target, NULL, NULL, ESC_ZFS_CONFIG_SYNC);
346 }
347 
348 /*
349  * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
350  * and we don't want to allow the local zone to see all the pools anyway.
351  * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
352  * information for all pool visible within the zone.
353  */
354 nvlist_t *
355 spa_all_configs(uint64_t *generation)
356 {
357 	nvlist_t *pools;
358 	spa_t *spa = NULL;
359 
360 	if (*generation == spa_config_generation)
361 		return (NULL);
362 
363 	pools = fnvlist_alloc();
364 
365 	mutex_enter(&spa_namespace_lock);
366 	while ((spa = spa_next(spa)) != NULL) {
367 		if (INGLOBALZONE(curproc) ||
368 		    zone_dataset_visible(spa_name(spa), NULL)) {
369 			mutex_enter(&spa->spa_props_lock);
370 			fnvlist_add_nvlist(pools, spa_name(spa),
371 			    spa->spa_config);
372 			mutex_exit(&spa->spa_props_lock);
373 		}
374 	}
375 	*generation = spa_config_generation;
376 	mutex_exit(&spa_namespace_lock);
377 
378 	return (pools);
379 }
380 
381 void
382 spa_config_set(spa_t *spa, nvlist_t *config)
383 {
384 	mutex_enter(&spa->spa_props_lock);
385 	if (spa->spa_config != NULL && spa->spa_config != config)
386 		nvlist_free(spa->spa_config);
387 	spa->spa_config = config;
388 	mutex_exit(&spa->spa_props_lock);
389 }
390 
391 /*
392  * Generate the pool's configuration based on the current in-core state.
393  *
394  * We infer whether to generate a complete config or just one top-level config
395  * based on whether vd is the root vdev.
396  */
397 nvlist_t *
398 spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
399 {
400 	nvlist_t *config, *nvroot;
401 	vdev_t *rvd = spa->spa_root_vdev;
402 	unsigned long hostid = 0;
403 	boolean_t locked = B_FALSE;
404 	uint64_t split_guid;
405 	char *pool_name;
406 
407 	if (vd == NULL) {
408 		vd = rvd;
409 		locked = B_TRUE;
410 		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
411 	}
412 
413 	ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
414 	    (SCL_CONFIG | SCL_STATE));
415 
416 	/*
417 	 * If txg is -1, report the current value of spa->spa_config_txg.
418 	 */
419 	if (txg == -1ULL)
420 		txg = spa->spa_config_txg;
421 
422 	/*
423 	 * Originally, users had to handle spa namespace collisions by either
424 	 * exporting the already imported pool or by specifying a new name for
425 	 * the pool with a conflicting name. In the case of root pools from
426 	 * virtual guests, neither approach to collision resolution is
427 	 * reasonable. This is addressed by extending the new name syntax with
428 	 * an option to specify that the new name is temporary. When specified,
429 	 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
430 	 * to use the previous name, which we do below.
431 	 */
432 	if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
433 		VERIFY0(nvlist_lookup_string(spa->spa_config,
434 		    ZPOOL_CONFIG_POOL_NAME, &pool_name));
435 	} else
436 		pool_name = spa_name(spa);
437 
438 	config = fnvlist_alloc();
439 
440 	fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa));
441 	fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, pool_name);
442 	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa));
443 	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg);
444 	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa));
445 	fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, spa->spa_errata);
446 	if (spa->spa_comment != NULL)
447 		fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT,
448 		    spa->spa_comment);
449 
450 	hostid = spa_get_hostid(spa);
451 	if (hostid != 0)
452 		fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid);
453 	fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname()->nodename);
454 
455 	int config_gen_flags = 0;
456 	if (vd != rvd) {
457 		fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
458 		    vd->vdev_top->vdev_guid);
459 		fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
460 		    vd->vdev_guid);
461 		if (vd->vdev_isspare)
462 			fnvlist_add_uint64(config,
463 			    ZPOOL_CONFIG_IS_SPARE, 1ULL);
464 		if (vd->vdev_islog)
465 			fnvlist_add_uint64(config,
466 			    ZPOOL_CONFIG_IS_LOG, 1ULL);
467 		vd = vd->vdev_top;		/* label contains top config */
468 	} else {
469 		/*
470 		 * Only add the (potentially large) split information
471 		 * in the mos config, and not in the vdev labels
472 		 */
473 		if (spa->spa_config_splitting != NULL)
474 			fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
475 			    spa->spa_config_splitting);
476 
477 		fnvlist_add_boolean(config, ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS);
478 
479 		config_gen_flags |= VDEV_CONFIG_MOS;
480 	}
481 
482 	/*
483 	 * Add the top-level config.  We even add this on pools which
484 	 * don't support holes in the namespace.
485 	 */
486 	vdev_top_config_generate(spa, config);
487 
488 	/*
489 	 * If we're splitting, record the original pool's guid.
490 	 */
491 	if (spa->spa_config_splitting != NULL &&
492 	    nvlist_lookup_uint64(spa->spa_config_splitting,
493 	    ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
494 		fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID, split_guid);
495 	}
496 
497 	nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags);
498 	fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot);
499 	nvlist_free(nvroot);
500 
501 	/*
502 	 * Store what's necessary for reading the MOS in the label.
503 	 */
504 	fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
505 	    spa->spa_label_features);
506 
507 	if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
508 		ddt_histogram_t *ddh;
509 		ddt_stat_t *dds;
510 		ddt_object_t *ddo;
511 
512 		ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
513 		ddt_get_dedup_histogram(spa, ddh);
514 		fnvlist_add_uint64_array(config,
515 		    ZPOOL_CONFIG_DDT_HISTOGRAM,
516 		    (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t));
517 		kmem_free(ddh, sizeof (ddt_histogram_t));
518 
519 		ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
520 		ddt_get_dedup_object_stats(spa, ddo);
521 		fnvlist_add_uint64_array(config,
522 		    ZPOOL_CONFIG_DDT_OBJ_STATS,
523 		    (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t));
524 		kmem_free(ddo, sizeof (ddt_object_t));
525 
526 		dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
527 		ddt_get_dedup_stats(spa, dds);
528 		fnvlist_add_uint64_array(config,
529 		    ZPOOL_CONFIG_DDT_STATS,
530 		    (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t));
531 		kmem_free(dds, sizeof (ddt_stat_t));
532 	}
533 
534 	if (locked)
535 		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
536 
537 	return (config);
538 }
539 
540 /*
541  * Update all disk labels, generate a fresh config based on the current
542  * in-core state, and sync the global config cache (do not sync the config
543  * cache if this is a booting rootpool).
544  */
545 void
546 spa_config_update(spa_t *spa, int what)
547 {
548 	vdev_t *rvd = spa->spa_root_vdev;
549 	uint64_t txg;
550 	int c;
551 
552 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
553 
554 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
555 	txg = spa_last_synced_txg(spa) + 1;
556 	if (what == SPA_CONFIG_UPDATE_POOL) {
557 		vdev_config_dirty(rvd);
558 	} else {
559 		/*
560 		 * If we have top-level vdevs that were added but have
561 		 * not yet been prepared for allocation, do that now.
562 		 * (It's safe now because the config cache is up to date,
563 		 * so it will be able to translate the new DVAs.)
564 		 * See comments in spa_vdev_add() for full details.
565 		 */
566 		for (c = 0; c < rvd->vdev_children; c++) {
567 			vdev_t *tvd = rvd->vdev_child[c];
568 
569 			/*
570 			 * Explicitly skip vdevs that are indirect or
571 			 * log vdevs that are being removed. The reason
572 			 * is that both of those can have vdev_ms_array
573 			 * set to 0 and we wouldn't want to change their
574 			 * metaslab size nor call vdev_expand() on them.
575 			 */
576 			if (!vdev_is_concrete(tvd) ||
577 			    (tvd->vdev_islog && tvd->vdev_removing))
578 				continue;
579 
580 			if (tvd->vdev_ms_array == 0)
581 				vdev_metaslab_set_size(tvd);
582 			vdev_expand(tvd, txg);
583 		}
584 	}
585 	spa_config_exit(spa, SCL_ALL, FTAG);
586 
587 	/*
588 	 * Wait for the mosconfig to be regenerated and synced.
589 	 */
590 	txg_wait_synced(spa->spa_dsl_pool, txg);
591 
592 	/*
593 	 * Update the global config cache to reflect the new mosconfig.
594 	 */
595 	if (!spa->spa_is_root) {
596 		spa_write_cachefile(spa, B_FALSE,
597 		    what != SPA_CONFIG_UPDATE_POOL);
598 	}
599 
600 	if (what == SPA_CONFIG_UPDATE_POOL)
601 		spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
602 }
603 
604 EXPORT_SYMBOL(spa_config_load);
605 EXPORT_SYMBOL(spa_all_configs);
606 EXPORT_SYMBOL(spa_config_set);
607 EXPORT_SYMBOL(spa_config_generate);
608 EXPORT_SYMBOL(spa_config_update);
609 
610 /* BEGIN CSTYLED */
611 #ifdef __linux__
612 /* string sysctls require a char array on FreeBSD */
613 ZFS_MODULE_PARAM(zfs_spa, spa_, config_path, STRING, ZMOD_RD,
614 	"SPA config file (/etc/zfs/zpool.cache)");
615 #endif
616 
617 ZFS_MODULE_PARAM(zfs, zfs_, autoimport_disable, INT, ZMOD_RW,
618 	"Disable pool import at module load");
619 /* END CSTYLED */
620