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, 2018 by Delphix. All rights reserved.
26 * Copyright 2017 Joyent, Inc.
27 */
28
29#include <sys/spa.h>
30#include <sys/fm/fs/zfs.h>
31#include <sys/spa_impl.h>
32#include <sys/nvpair.h>
33#include <sys/uio.h>
34#include <sys/fs/zfs.h>
35#include <sys/vdev_impl.h>
36#include <sys/zfs_ioctl.h>
37#include <sys/utsname.h>
38#include <sys/systeminfo.h>
39#include <sys/sunddi.h>
40#include <sys/zfeature.h>
41#ifdef _KERNEL
42#include <sys/kobj.h>
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
63static 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 */
69const char *spa_config_path = ZPOOL_CACHE;
70
71/*
72 * Called when the module is first loaded, this routine loads the configuration
73 * file into the SPA namespace.  It does not actually open or load the pools; it
74 * only populates the namespace.
75 */
76void
77spa_config_load(void)
78{
79	void *buf = NULL;
80	nvlist_t *nvlist, *child;
81	nvpair_t *nvpair;
82	char *pathname;
83	struct _buf *file;
84	uint64_t fsize;
85
86	/*
87	 * Open the configuration file.
88	 */
89	pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
90
91	(void) snprintf(pathname, MAXPATHLEN, "%s%s",
92	    (rootdir != NULL) ? "./" : "", spa_config_path);
93
94	file = kobj_open_file(pathname);
95
96	kmem_free(pathname, MAXPATHLEN);
97
98	if (file == (struct _buf *)-1)
99		return;
100
101	if (kobj_get_filesize(file, &fsize) != 0)
102		goto out;
103
104	buf = kmem_alloc(fsize, KM_SLEEP);
105
106	/*
107	 * Read the nvlist from the file.
108	 */
109	if (kobj_read_file(file, buf, fsize, 0) < 0)
110		goto out;
111
112	/*
113	 * Unpack the nvlist.
114	 */
115	if (nvlist_unpack(buf, fsize, &nvlist, KM_SLEEP) != 0)
116		goto out;
117
118	/*
119	 * Iterate over all elements in the nvlist, creating a new spa_t for
120	 * each one with the specified configuration.
121	 */
122	mutex_enter(&spa_namespace_lock);
123	nvpair = NULL;
124	while ((nvpair = nvlist_next_nvpair(nvlist, nvpair)) != NULL) {
125		if (nvpair_type(nvpair) != DATA_TYPE_NVLIST)
126			continue;
127
128		child = fnvpair_value_nvlist(nvpair);
129
130		if (spa_lookup(nvpair_name(nvpair)) != NULL)
131			continue;
132		(void) spa_add(nvpair_name(nvpair), child, NULL);
133	}
134	mutex_exit(&spa_namespace_lock);
135
136	nvlist_free(nvlist);
137
138out:
139	if (buf != NULL)
140		kmem_free(buf, fsize);
141
142	kobj_close_file(file);
143}
144
145static int
146spa_config_write(spa_config_dirent_t *dp, nvlist_t *nvl)
147{
148	size_t buflen;
149	char *buf;
150	vnode_t *vp;
151	int oflags = FWRITE | FTRUNC | FCREAT | FOFFMAX;
152	char *temp;
153	int err;
154
155	/*
156	 * If the nvlist is empty (NULL), then remove the old cachefile.
157	 */
158	if (nvl == NULL) {
159		err = vn_remove(dp->scd_path, UIO_SYSSPACE, RMFILE);
160		return (err);
161	}
162
163	/*
164	 * Pack the configuration into a buffer.
165	 */
166	buf = fnvlist_pack(nvl, &buflen);
167	temp = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
168
169	/*
170	 * Write the configuration to disk.  We need to do the traditional
171	 * 'write to temporary file, sync, move over original' to make sure we
172	 * always have a consistent view of the data.
173	 */
174	(void) snprintf(temp, MAXPATHLEN, "%s.tmp", dp->scd_path);
175
176	err = vn_open(temp, UIO_SYSSPACE, oflags, 0644, &vp, CRCREAT, 0);
177	if (err == 0) {
178		err = vn_rdwr(UIO_WRITE, vp, buf, buflen, 0, UIO_SYSSPACE,
179		    0, RLIM64_INFINITY, kcred, NULL);
180		if (err == 0)
181			err = VOP_FSYNC(vp, FSYNC, kcred, NULL);
182		if (err == 0)
183			err = vn_rename(temp, dp->scd_path, UIO_SYSSPACE);
184		(void) VOP_CLOSE(vp, oflags, 1, 0, kcred, NULL);
185		VN_RELE(vp);
186	}
187
188	(void) vn_remove(temp, UIO_SYSSPACE, RMFILE);
189
190	fnvlist_pack_free(buf, buflen);
191	kmem_free(temp, MAXPATHLEN);
192	return (err);
193}
194
195/*
196 * Synchronize pool configuration to disk.  This must be called with the
197 * namespace lock held. Synchronizing the pool cache is typically done after
198 * the configuration has been synced to the MOS. This exposes a window where
199 * the MOS config will have been updated but the cache file has not. If
200 * the system were to crash at that instant then the cached config may not
201 * contain the correct information to open the pool and an explicit import
202 * would be required.
203 */
204void
205spa_write_cachefile(spa_t *target, boolean_t removing, boolean_t postsysevent)
206{
207	spa_config_dirent_t *dp, *tdp;
208	nvlist_t *nvl;
209	boolean_t ccw_failure;
210	int error;
211	char *pool_name;
212
213	ASSERT(MUTEX_HELD(&spa_namespace_lock));
214
215	if (rootdir == NULL || !(spa_mode_global & FWRITE))
216		return;
217
218	/*
219	 * Iterate over all cachefiles for the pool, past or present.  When the
220	 * cachefile is changed, the new one is pushed onto this list, allowing
221	 * us to update previous cachefiles that no longer contain this pool.
222	 */
223	ccw_failure = B_FALSE;
224	for (dp = list_head(&target->spa_config_list); dp != NULL;
225	    dp = list_next(&target->spa_config_list, dp)) {
226		spa_t *spa = NULL;
227		if (dp->scd_path == NULL)
228			continue;
229
230		/*
231		 * Iterate over all pools, adding any matching pools to 'nvl'.
232		 */
233		nvl = NULL;
234		while ((spa = spa_next(spa)) != NULL) {
235			/*
236			 * Skip over our own pool if we're about to remove
237			 * ourselves from the spa namespace or any pool that
238			 * is readonly. Since we cannot guarantee that a
239			 * readonly pool would successfully import upon reboot,
240			 * we don't allow them to be written to the cache file.
241			 */
242			if ((spa == target && removing) ||
243			    !spa_writeable(spa))
244				continue;
245
246			mutex_enter(&spa->spa_props_lock);
247			tdp = list_head(&spa->spa_config_list);
248			if (spa->spa_config == NULL ||
249			    tdp->scd_path == NULL ||
250			    strcmp(tdp->scd_path, dp->scd_path) != 0) {
251				mutex_exit(&spa->spa_props_lock);
252				continue;
253			}
254
255			if (nvl == NULL)
256				nvl = fnvlist_alloc();
257
258			if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
259				pool_name = fnvlist_lookup_string(
260				    spa->spa_config, ZPOOL_CONFIG_POOL_NAME);
261			} else {
262				pool_name = spa_name(spa);
263			}
264
265			fnvlist_add_nvlist(nvl, pool_name,
266			    spa->spa_config);
267			mutex_exit(&spa->spa_props_lock);
268		}
269
270		error = spa_config_write(dp, nvl);
271		if (error != 0)
272			ccw_failure = B_TRUE;
273		nvlist_free(nvl);
274	}
275
276	if (ccw_failure) {
277		/*
278		 * Keep trying so that configuration data is
279		 * written if/when any temporary filesystem
280		 * resource issues are resolved.
281		 */
282		if (target->spa_ccw_fail_time == 0) {
283			zfs_ereport_post(FM_EREPORT_ZFS_CONFIG_CACHE_WRITE,
284			    target, NULL, NULL, NULL, 0, 0);
285		}
286		target->spa_ccw_fail_time = gethrtime();
287		spa_async_request(target, SPA_ASYNC_CONFIG_UPDATE);
288	} else {
289		/*
290		 * Do not rate limit future attempts to update
291		 * the config cache.
292		 */
293		target->spa_ccw_fail_time = 0;
294	}
295
296	/*
297	 * Remove any config entries older than the current one.
298	 */
299	dp = list_head(&target->spa_config_list);
300	while ((tdp = list_next(&target->spa_config_list, dp)) != NULL) {
301		list_remove(&target->spa_config_list, tdp);
302		if (tdp->scd_path != NULL)
303			spa_strfree(tdp->scd_path);
304		kmem_free(tdp, sizeof (spa_config_dirent_t));
305	}
306
307	spa_config_generation++;
308
309	if (postsysevent)
310		spa_event_notify(target, NULL, NULL, ESC_ZFS_CONFIG_SYNC);
311}
312
313/*
314 * Sigh.  Inside a local zone, we don't have access to /etc/zfs/zpool.cache,
315 * and we don't want to allow the local zone to see all the pools anyway.
316 * So we have to invent the ZFS_IOC_CONFIG ioctl to grab the configuration
317 * information for all pool visible within the zone.
318 */
319nvlist_t *
320spa_all_configs(uint64_t *generation)
321{
322	nvlist_t *pools;
323	spa_t *spa = NULL;
324
325	if (*generation == spa_config_generation)
326		return (NULL);
327
328	pools = fnvlist_alloc();
329
330	mutex_enter(&spa_namespace_lock);
331	while ((spa = spa_next(spa)) != NULL) {
332		if (INGLOBALZONE(curproc) ||
333		    zone_dataset_visible(spa_name(spa), NULL)) {
334			mutex_enter(&spa->spa_props_lock);
335			fnvlist_add_nvlist(pools, spa_name(spa),
336			    spa->spa_config);
337			mutex_exit(&spa->spa_props_lock);
338		}
339	}
340	*generation = spa_config_generation;
341	mutex_exit(&spa_namespace_lock);
342
343	return (pools);
344}
345
346void
347spa_config_set(spa_t *spa, nvlist_t *config)
348{
349	mutex_enter(&spa->spa_props_lock);
350	if (spa->spa_config != NULL && spa->spa_config != config)
351		nvlist_free(spa->spa_config);
352	spa->spa_config = config;
353	mutex_exit(&spa->spa_props_lock);
354}
355
356/*
357 * Generate the pool's configuration based on the current in-core state.
358 *
359 * We infer whether to generate a complete config or just one top-level config
360 * based on whether vd is the root vdev.
361 */
362nvlist_t *
363spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg, int getstats)
364{
365	nvlist_t *config, *nvroot;
366	vdev_t *rvd = spa->spa_root_vdev;
367	unsigned long hostid = 0;
368	boolean_t locked = B_FALSE;
369	uint64_t split_guid;
370	char *pool_name;
371
372	if (vd == NULL) {
373		vd = rvd;
374		locked = B_TRUE;
375		spa_config_enter(spa, SCL_CONFIG | SCL_STATE, FTAG, RW_READER);
376	}
377
378	ASSERT(spa_config_held(spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
379	    (SCL_CONFIG | SCL_STATE));
380
381	/*
382	 * If txg is -1, report the current value of spa->spa_config_txg.
383	 */
384	if (txg == -1ULL)
385		txg = spa->spa_config_txg;
386
387	/*
388	 * Originally, users had to handle spa namespace collisions by either
389	 * exporting the already imported pool or by specifying a new name for
390	 * the pool with a conflicting name. In the case of root pools from
391	 * virtual guests, neither approach to collision resolution is
392	 * reasonable. This is addressed by extending the new name syntax with
393	 * an option to specify that the new name is temporary. When specified,
394	 * ZFS_IMPORT_TEMP_NAME will be set in spa->spa_import_flags to tell us
395	 * to use the previous name, which we do below.
396	 */
397	if (spa->spa_import_flags & ZFS_IMPORT_TEMP_NAME) {
398		pool_name = fnvlist_lookup_string(spa->spa_config,
399		    ZPOOL_CONFIG_POOL_NAME);
400	} else {
401		pool_name = spa_name(spa);
402	}
403
404	config = fnvlist_alloc();
405
406	fnvlist_add_uint64(config, ZPOOL_CONFIG_VERSION, spa_version(spa));
407	fnvlist_add_string(config, ZPOOL_CONFIG_POOL_NAME, pool_name);
408	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_STATE, spa_state(spa));
409	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_TXG, txg);
410	fnvlist_add_uint64(config, ZPOOL_CONFIG_POOL_GUID, spa_guid(spa));
411	fnvlist_add_uint64(config, ZPOOL_CONFIG_ERRATA, spa->spa_errata);
412	if (spa->spa_comment != NULL) {
413		fnvlist_add_string(config, ZPOOL_CONFIG_COMMENT,
414		    spa->spa_comment);
415	}
416
417	hostid = spa_get_hostid();
418	if (hostid != 0) {
419		fnvlist_add_uint64(config, ZPOOL_CONFIG_HOSTID, hostid);
420	}
421	fnvlist_add_string(config, ZPOOL_CONFIG_HOSTNAME, utsname.nodename);
422
423	int config_gen_flags = 0;
424	if (vd != rvd) {
425		fnvlist_add_uint64(config, ZPOOL_CONFIG_TOP_GUID,
426		    vd->vdev_top->vdev_guid);
427		fnvlist_add_uint64(config, ZPOOL_CONFIG_GUID,
428		    vd->vdev_guid);
429		if (vd->vdev_isspare) {
430			fnvlist_add_uint64(config,
431			    ZPOOL_CONFIG_IS_SPARE, 1ULL);
432		}
433		if (vd->vdev_islog) {
434			fnvlist_add_uint64(config,
435			    ZPOOL_CONFIG_IS_LOG, 1ULL);
436		}
437		vd = vd->vdev_top;		/* label contains top config */
438	} else {
439		/*
440		 * Only add the (potentially large) split information
441		 * in the mos config, and not in the vdev labels
442		 */
443		if (spa->spa_config_splitting != NULL)
444			fnvlist_add_nvlist(config, ZPOOL_CONFIG_SPLIT,
445			    spa->spa_config_splitting);
446		fnvlist_add_boolean(config,
447		    ZPOOL_CONFIG_HAS_PER_VDEV_ZAPS);
448
449		config_gen_flags |= VDEV_CONFIG_MOS;
450	}
451
452	/*
453	 * Add the top-level config.  We even add this on pools which
454	 * don't support holes in the namespace.
455	 */
456	vdev_top_config_generate(spa, config);
457
458	/*
459	 * If we're splitting, record the original pool's guid.
460	 */
461	if (spa->spa_config_splitting != NULL &&
462	    nvlist_lookup_uint64(spa->spa_config_splitting,
463	    ZPOOL_CONFIG_SPLIT_GUID, &split_guid) == 0) {
464		fnvlist_add_uint64(config, ZPOOL_CONFIG_SPLIT_GUID,
465		    split_guid);
466	}
467
468	nvroot = vdev_config_generate(spa, vd, getstats, config_gen_flags);
469	fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot);
470	nvlist_free(nvroot);
471
472	/*
473	 * Store what's necessary for reading the MOS in the label.
474	 */
475	fnvlist_add_nvlist(config, ZPOOL_CONFIG_FEATURES_FOR_READ,
476	    spa->spa_label_features);
477
478	if (getstats && spa_load_state(spa) == SPA_LOAD_NONE) {
479		ddt_histogram_t *ddh;
480		ddt_stat_t *dds;
481		ddt_object_t *ddo;
482
483		ddh = kmem_zalloc(sizeof (ddt_histogram_t), KM_SLEEP);
484		ddt_get_dedup_histogram(spa, ddh);
485		fnvlist_add_uint64_array(config,
486		    ZPOOL_CONFIG_DDT_HISTOGRAM,
487		    (uint64_t *)ddh, sizeof (*ddh) / sizeof (uint64_t));
488		kmem_free(ddh, sizeof (ddt_histogram_t));
489
490		ddo = kmem_zalloc(sizeof (ddt_object_t), KM_SLEEP);
491		ddt_get_dedup_object_stats(spa, ddo);
492		fnvlist_add_uint64_array(config,
493		    ZPOOL_CONFIG_DDT_OBJ_STATS,
494		    (uint64_t *)ddo, sizeof (*ddo) / sizeof (uint64_t));
495		kmem_free(ddo, sizeof (ddt_object_t));
496
497		dds = kmem_zalloc(sizeof (ddt_stat_t), KM_SLEEP);
498		ddt_get_dedup_stats(spa, dds);
499		fnvlist_add_uint64_array(config,
500		    ZPOOL_CONFIG_DDT_STATS,
501		    (uint64_t *)dds, sizeof (*dds) / sizeof (uint64_t));
502		kmem_free(dds, sizeof (ddt_stat_t));
503	}
504
505	if (locked)
506		spa_config_exit(spa, SCL_CONFIG | SCL_STATE, FTAG);
507
508	return (config);
509}
510
511/*
512 * Update all disk labels, generate a fresh config based on the current
513 * in-core state, and sync the global config cache (do not sync the config
514 * cache if this is a booting rootpool).
515 */
516void
517spa_config_update(spa_t *spa, int what)
518{
519	vdev_t *rvd = spa->spa_root_vdev;
520	uint64_t txg;
521	int c;
522
523	ASSERT(MUTEX_HELD(&spa_namespace_lock));
524
525	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
526	txg = spa_last_synced_txg(spa) + 1;
527	if (what == SPA_CONFIG_UPDATE_POOL) {
528		vdev_config_dirty(rvd);
529	} else {
530		/*
531		 * If we have top-level vdevs that were added but have
532		 * not yet been prepared for allocation, do that now.
533		 * (It's safe now because the config cache is up to date,
534		 * so it will be able to translate the new DVAs.)
535		 * See comments in spa_vdev_add() for full details.
536		 */
537		for (c = 0; c < rvd->vdev_children; c++) {
538			vdev_t *tvd = rvd->vdev_child[c];
539
540			/*
541			 * Explicitly skip vdevs that are indirect or
542			 * log vdevs that are being removed. The reason
543			 * is that both of those can have vdev_ms_array
544			 * set to 0 and we wouldn't want to change their
545			 * metaslab size nor call vdev_expand() on them.
546			 */
547			if (!vdev_is_concrete(tvd) ||
548			    (tvd->vdev_islog && tvd->vdev_removing))
549				continue;
550
551			if (tvd->vdev_ms_array == 0)
552				vdev_metaslab_set_size(tvd);
553			vdev_expand(tvd, txg);
554		}
555	}
556	spa_config_exit(spa, SCL_ALL, FTAG);
557
558	/*
559	 * Wait for the mosconfig to be regenerated and synced.
560	 */
561	txg_wait_synced(spa->spa_dsl_pool, txg);
562
563	/*
564	 * Update the global config cache to reflect the new mosconfig.
565	 */
566	if (!spa->spa_is_root) {
567		spa_write_cachefile(spa, B_FALSE,
568		    what != SPA_CONFIG_UPDATE_POOL);
569	}
570
571	if (what == SPA_CONFIG_UPDATE_POOL)
572		spa_config_update(spa, SPA_CONFIG_UPDATE_VDEVS);
573}
574