xref: /freebsd-src/sys/contrib/openzfs/module/os/freebsd/zfs/spa_os.c (revision 2a58b312b62f908ec92311d1bd8536dbaeb8e55b)
1eda14cbcSMatt Macy /*
2eda14cbcSMatt Macy  * CDDL HEADER START
3eda14cbcSMatt Macy  *
4eda14cbcSMatt Macy  * The contents of this file are subject to the terms of the
5eda14cbcSMatt Macy  * Common Development and Distribution License (the "License").
6eda14cbcSMatt Macy  * You may not use this file except in compliance with the License.
7eda14cbcSMatt Macy  *
8eda14cbcSMatt Macy  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9271171e0SMartin Matuska  * or https://opensource.org/licenses/CDDL-1.0.
10eda14cbcSMatt Macy  * See the License for the specific language governing permissions
11eda14cbcSMatt Macy  * and limitations under the License.
12eda14cbcSMatt Macy  *
13eda14cbcSMatt Macy  * When distributing Covered Code, include this CDDL HEADER in each
14eda14cbcSMatt Macy  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15eda14cbcSMatt Macy  * If applicable, add the following below this CDDL HEADER, with the
16eda14cbcSMatt Macy  * fields enclosed by brackets "[]" replaced with your own identifying
17eda14cbcSMatt Macy  * information: Portions Copyright [yyyy] [name of copyright owner]
18eda14cbcSMatt Macy  *
19eda14cbcSMatt Macy  * CDDL HEADER END
20eda14cbcSMatt Macy  */
21eda14cbcSMatt Macy 
22eda14cbcSMatt Macy /*
23eda14cbcSMatt Macy  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24eda14cbcSMatt Macy  * Copyright (c) 2011 by Delphix. All rights reserved.
25eda14cbcSMatt Macy  * Copyright (c) 2013 Martin Matuska <mm@FreeBSD.org>. All rights reserved.
26eda14cbcSMatt Macy  */
27eda14cbcSMatt Macy 
28eda14cbcSMatt Macy 
29eda14cbcSMatt Macy #include <sys/zfs_context.h>
30eda14cbcSMatt Macy #include <sys/fm/fs/zfs.h>
31eda14cbcSMatt Macy #include <sys/spa_impl.h>
32eda14cbcSMatt Macy #include <sys/zio.h>
33eda14cbcSMatt Macy #include <sys/zio_checksum.h>
34eda14cbcSMatt Macy #include <sys/dmu.h>
35eda14cbcSMatt Macy #include <sys/dmu_tx.h>
36eda14cbcSMatt Macy #include <sys/zap.h>
37eda14cbcSMatt Macy #include <sys/zil.h>
38eda14cbcSMatt Macy #include <sys/ddt.h>
39eda14cbcSMatt Macy #include <sys/vdev_impl.h>
40eda14cbcSMatt Macy #include <sys/vdev_os.h>
41eda14cbcSMatt Macy #include <sys/vdev_removal.h>
42eda14cbcSMatt Macy #include <sys/vdev_indirect_mapping.h>
43eda14cbcSMatt Macy #include <sys/vdev_indirect_births.h>
44eda14cbcSMatt Macy #include <sys/metaslab.h>
45eda14cbcSMatt Macy #include <sys/metaslab_impl.h>
46eda14cbcSMatt Macy #include <sys/uberblock_impl.h>
47eda14cbcSMatt Macy #include <sys/txg.h>
48eda14cbcSMatt Macy #include <sys/avl.h>
49eda14cbcSMatt Macy #include <sys/bpobj.h>
50eda14cbcSMatt Macy #include <sys/dmu_traverse.h>
51eda14cbcSMatt Macy #include <sys/dmu_objset.h>
52eda14cbcSMatt Macy #include <sys/unique.h>
53eda14cbcSMatt Macy #include <sys/dsl_pool.h>
54eda14cbcSMatt Macy #include <sys/dsl_dataset.h>
55eda14cbcSMatt Macy #include <sys/dsl_dir.h>
56eda14cbcSMatt Macy #include <sys/dsl_prop.h>
57eda14cbcSMatt Macy #include <sys/dsl_synctask.h>
58eda14cbcSMatt Macy #include <sys/fs/zfs.h>
59eda14cbcSMatt Macy #include <sys/arc.h>
60eda14cbcSMatt Macy #include <sys/callb.h>
61eda14cbcSMatt Macy #include <sys/zfs_ioctl.h>
62eda14cbcSMatt Macy #include <sys/dsl_scan.h>
63eda14cbcSMatt Macy #include <sys/dmu_send.h>
64eda14cbcSMatt Macy #include <sys/dsl_destroy.h>
65eda14cbcSMatt Macy #include <sys/dsl_userhold.h>
66eda14cbcSMatt Macy #include <sys/zfeature.h>
67eda14cbcSMatt Macy #include <sys/zvol.h>
68eda14cbcSMatt Macy #include <sys/abd.h>
69eda14cbcSMatt Macy #include <sys/callb.h>
70eda14cbcSMatt Macy #include <sys/zone.h>
71eda14cbcSMatt Macy 
72eda14cbcSMatt Macy #include "zfs_prop.h"
73eda14cbcSMatt Macy #include "zfs_comutil.h"
74eda14cbcSMatt Macy 
75eda14cbcSMatt Macy static nvlist_t *
spa_generate_rootconf(const char * name)76eda14cbcSMatt Macy spa_generate_rootconf(const char *name)
77eda14cbcSMatt Macy {
78eda14cbcSMatt Macy 	nvlist_t **configs, **tops;
79eda14cbcSMatt Macy 	nvlist_t *config;
80eda14cbcSMatt Macy 	nvlist_t *best_cfg, *nvtop, *nvroot;
81eda14cbcSMatt Macy 	uint64_t *holes;
82eda14cbcSMatt Macy 	uint64_t best_txg;
83eda14cbcSMatt Macy 	uint64_t nchildren;
84eda14cbcSMatt Macy 	uint64_t pgid;
85eda14cbcSMatt Macy 	uint64_t count;
86eda14cbcSMatt Macy 	uint64_t i;
87eda14cbcSMatt Macy 	uint_t   nholes;
88eda14cbcSMatt Macy 
89eda14cbcSMatt Macy 	if (vdev_geom_read_pool_label(name, &configs, &count) != 0)
90eda14cbcSMatt Macy 		return (NULL);
91eda14cbcSMatt Macy 
92eda14cbcSMatt Macy 	ASSERT3U(count, !=, 0);
93eda14cbcSMatt Macy 	best_txg = 0;
94eda14cbcSMatt Macy 	for (i = 0; i < count; i++) {
95eda14cbcSMatt Macy 		uint64_t txg;
96eda14cbcSMatt Macy 
9715f0b8c3SMartin Matuska 		if (configs[i] == NULL)
9815f0b8c3SMartin Matuska 			continue;
9916038816SMartin Matuska 		txg = fnvlist_lookup_uint64(configs[i], ZPOOL_CONFIG_POOL_TXG);
100eda14cbcSMatt Macy 		if (txg > best_txg) {
101eda14cbcSMatt Macy 			best_txg = txg;
102eda14cbcSMatt Macy 			best_cfg = configs[i];
103eda14cbcSMatt Macy 		}
104eda14cbcSMatt Macy 	}
105eda14cbcSMatt Macy 
106eda14cbcSMatt Macy 	nchildren = 1;
107eda14cbcSMatt Macy 	nvlist_lookup_uint64(best_cfg, ZPOOL_CONFIG_VDEV_CHILDREN, &nchildren);
108eda14cbcSMatt Macy 	holes = NULL;
109eda14cbcSMatt Macy 	nvlist_lookup_uint64_array(best_cfg, ZPOOL_CONFIG_HOLE_ARRAY,
110eda14cbcSMatt Macy 	    &holes, &nholes);
111eda14cbcSMatt Macy 
112eda14cbcSMatt Macy 	tops = kmem_zalloc(nchildren * sizeof (void *), KM_SLEEP);
113eda14cbcSMatt Macy 	for (i = 0; i < nchildren; i++) {
114eda14cbcSMatt Macy 		if (i >= count)
115eda14cbcSMatt Macy 			break;
116eda14cbcSMatt Macy 		if (configs[i] == NULL)
117eda14cbcSMatt Macy 			continue;
11816038816SMartin Matuska 		nvtop = fnvlist_lookup_nvlist(configs[i],
11916038816SMartin Matuska 		    ZPOOL_CONFIG_VDEV_TREE);
12016038816SMartin Matuska 		tops[i] = fnvlist_dup(nvtop);
121eda14cbcSMatt Macy 	}
122eda14cbcSMatt Macy 	for (i = 0; holes != NULL && i < nholes; i++) {
123eda14cbcSMatt Macy 		if (i >= nchildren)
124eda14cbcSMatt Macy 			continue;
125eda14cbcSMatt Macy 		if (tops[holes[i]] != NULL)
126eda14cbcSMatt Macy 			continue;
12716038816SMartin Matuska 		tops[holes[i]] = fnvlist_alloc();
12816038816SMartin Matuska 		fnvlist_add_string(tops[holes[i]], ZPOOL_CONFIG_TYPE,
12916038816SMartin Matuska 		    VDEV_TYPE_HOLE);
13016038816SMartin Matuska 		fnvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_ID, holes[i]);
13116038816SMartin Matuska 		fnvlist_add_uint64(tops[holes[i]], ZPOOL_CONFIG_GUID, 0);
132eda14cbcSMatt Macy 	}
133eda14cbcSMatt Macy 	for (i = 0; i < nchildren; i++) {
134eda14cbcSMatt Macy 		if (tops[i] != NULL)
135eda14cbcSMatt Macy 			continue;
13616038816SMartin Matuska 		tops[i] = fnvlist_alloc();
13716038816SMartin Matuska 		fnvlist_add_string(tops[i], ZPOOL_CONFIG_TYPE,
13816038816SMartin Matuska 		    VDEV_TYPE_MISSING);
13916038816SMartin Matuska 		fnvlist_add_uint64(tops[i], ZPOOL_CONFIG_ID, i);
14016038816SMartin Matuska 		fnvlist_add_uint64(tops[i], ZPOOL_CONFIG_GUID, 0);
141eda14cbcSMatt Macy 	}
142eda14cbcSMatt Macy 
143eda14cbcSMatt Macy 	/*
144eda14cbcSMatt Macy 	 * Create pool config based on the best vdev config.
145eda14cbcSMatt Macy 	 */
14616038816SMartin Matuska 	config = fnvlist_dup(best_cfg);
147eda14cbcSMatt Macy 
148eda14cbcSMatt Macy 	/*
149eda14cbcSMatt Macy 	 * Put this pool's top-level vdevs into a root vdev.
150eda14cbcSMatt Macy 	 */
15116038816SMartin Matuska 	pgid = fnvlist_lookup_uint64(config, ZPOOL_CONFIG_POOL_GUID);
15216038816SMartin Matuska 	nvroot = fnvlist_alloc();
15316038816SMartin Matuska 	fnvlist_add_string(nvroot, ZPOOL_CONFIG_TYPE, VDEV_TYPE_ROOT);
15416038816SMartin Matuska 	fnvlist_add_uint64(nvroot, ZPOOL_CONFIG_ID, 0ULL);
15516038816SMartin Matuska 	fnvlist_add_uint64(nvroot, ZPOOL_CONFIG_GUID, pgid);
156681ce946SMartin Matuska 	fnvlist_add_nvlist_array(nvroot, ZPOOL_CONFIG_CHILDREN,
157681ce946SMartin Matuska 	    (const nvlist_t * const *)tops, nchildren);
158eda14cbcSMatt Macy 
159eda14cbcSMatt Macy 	/*
160eda14cbcSMatt Macy 	 * Replace the existing vdev_tree with the new root vdev in
161eda14cbcSMatt Macy 	 * this pool's configuration (remove the old, add the new).
162eda14cbcSMatt Macy 	 */
16316038816SMartin Matuska 	fnvlist_add_nvlist(config, ZPOOL_CONFIG_VDEV_TREE, nvroot);
164eda14cbcSMatt Macy 
165eda14cbcSMatt Macy 	/*
166eda14cbcSMatt Macy 	 * Drop vdev config elements that should not be present at pool level.
167eda14cbcSMatt Macy 	 */
16816038816SMartin Matuska 	fnvlist_remove(config, ZPOOL_CONFIG_GUID);
16916038816SMartin Matuska 	fnvlist_remove(config, ZPOOL_CONFIG_TOP_GUID);
170eda14cbcSMatt Macy 
171eda14cbcSMatt Macy 	for (i = 0; i < count; i++)
17216038816SMartin Matuska 		fnvlist_free(configs[i]);
173eda14cbcSMatt Macy 	kmem_free(configs, count * sizeof (void *));
174eda14cbcSMatt Macy 	for (i = 0; i < nchildren; i++)
17516038816SMartin Matuska 		fnvlist_free(tops[i]);
176eda14cbcSMatt Macy 	kmem_free(tops, nchildren * sizeof (void *));
17716038816SMartin Matuska 	fnvlist_free(nvroot);
178eda14cbcSMatt Macy 	return (config);
179eda14cbcSMatt Macy }
180eda14cbcSMatt Macy 
181eda14cbcSMatt Macy int
spa_import_rootpool(const char * name,bool checkpointrewind)182eda14cbcSMatt Macy spa_import_rootpool(const char *name, bool checkpointrewind)
183eda14cbcSMatt Macy {
184eda14cbcSMatt Macy 	spa_t *spa;
185eda14cbcSMatt Macy 	vdev_t *rvd;
186eda14cbcSMatt Macy 	nvlist_t *config, *nvtop;
187*2a58b312SMartin Matuska 	const char *pname;
188eda14cbcSMatt Macy 	int error;
189eda14cbcSMatt Macy 
190eda14cbcSMatt Macy 	/*
191eda14cbcSMatt Macy 	 * Read the label from the boot device and generate a configuration.
192eda14cbcSMatt Macy 	 */
193eda14cbcSMatt Macy 	config = spa_generate_rootconf(name);
194eda14cbcSMatt Macy 
195eda14cbcSMatt Macy 	mutex_enter(&spa_namespace_lock);
196eda14cbcSMatt Macy 	if (config != NULL) {
19716038816SMartin Matuska 		pname = fnvlist_lookup_string(config, ZPOOL_CONFIG_POOL_NAME);
19816038816SMartin Matuska 		VERIFY0(strcmp(name, pname));
199eda14cbcSMatt Macy 
200eda14cbcSMatt Macy 		if ((spa = spa_lookup(pname)) != NULL) {
201eda14cbcSMatt Macy 			/*
202eda14cbcSMatt Macy 			 * The pool could already be imported,
203eda14cbcSMatt Macy 			 * e.g., after reboot -r.
204eda14cbcSMatt Macy 			 */
205eda14cbcSMatt Macy 			if (spa->spa_state == POOL_STATE_ACTIVE) {
206eda14cbcSMatt Macy 				mutex_exit(&spa_namespace_lock);
20716038816SMartin Matuska 				fnvlist_free(config);
208eda14cbcSMatt Macy 				return (0);
209eda14cbcSMatt Macy 			}
210eda14cbcSMatt Macy 
211eda14cbcSMatt Macy 			/*
212eda14cbcSMatt Macy 			 * Remove the existing root pool from the namespace so
213eda14cbcSMatt Macy 			 * that we can replace it with the correct config
214eda14cbcSMatt Macy 			 * we just read in.
215eda14cbcSMatt Macy 			 */
216eda14cbcSMatt Macy 			spa_remove(spa);
217eda14cbcSMatt Macy 		}
218eda14cbcSMatt Macy 		spa = spa_add(pname, config, NULL);
219eda14cbcSMatt Macy 
220eda14cbcSMatt Macy 		/*
221eda14cbcSMatt Macy 		 * Set spa_ubsync.ub_version as it can be used in vdev_alloc()
222eda14cbcSMatt Macy 		 * via spa_version().
223eda14cbcSMatt Macy 		 */
224eda14cbcSMatt Macy 		if (nvlist_lookup_uint64(config, ZPOOL_CONFIG_VERSION,
225eda14cbcSMatt Macy 		    &spa->spa_ubsync.ub_version) != 0)
226eda14cbcSMatt Macy 			spa->spa_ubsync.ub_version = SPA_VERSION_INITIAL;
227eda14cbcSMatt Macy 	} else if ((spa = spa_lookup(name)) == NULL) {
228eda14cbcSMatt Macy 		mutex_exit(&spa_namespace_lock);
22916038816SMartin Matuska 		fnvlist_free(config);
230eda14cbcSMatt Macy 		cmn_err(CE_NOTE, "Cannot find the pool label for '%s'",
231eda14cbcSMatt Macy 		    name);
232eda14cbcSMatt Macy 		return (EIO);
233eda14cbcSMatt Macy 	} else {
23416038816SMartin Matuska 		config = fnvlist_dup(spa->spa_config);
235eda14cbcSMatt Macy 	}
236eda14cbcSMatt Macy 	spa->spa_is_root = B_TRUE;
237eda14cbcSMatt Macy 	spa->spa_import_flags = ZFS_IMPORT_VERBATIM;
238eda14cbcSMatt Macy 	if (checkpointrewind) {
239eda14cbcSMatt Macy 		spa->spa_import_flags |= ZFS_IMPORT_CHECKPOINT;
240eda14cbcSMatt Macy 	}
241eda14cbcSMatt Macy 
242eda14cbcSMatt Macy 	/*
243eda14cbcSMatt Macy 	 * Build up a vdev tree based on the boot device's label config.
244eda14cbcSMatt Macy 	 */
24516038816SMartin Matuska 	nvtop = fnvlist_lookup_nvlist(config, ZPOOL_CONFIG_VDEV_TREE);
246eda14cbcSMatt Macy 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
247eda14cbcSMatt Macy 	error = spa_config_parse(spa, &rvd, nvtop, NULL, 0,
248eda14cbcSMatt Macy 	    VDEV_ALLOC_ROOTPOOL);
249eda14cbcSMatt Macy 	spa_config_exit(spa, SCL_ALL, FTAG);
250eda14cbcSMatt Macy 	if (error) {
251eda14cbcSMatt Macy 		mutex_exit(&spa_namespace_lock);
25216038816SMartin Matuska 		fnvlist_free(config);
253eda14cbcSMatt Macy 		cmn_err(CE_NOTE, "Can not parse the config for pool '%s'",
254c7046f76SMartin Matuska 		    name);
255eda14cbcSMatt Macy 		return (error);
256eda14cbcSMatt Macy 	}
257eda14cbcSMatt Macy 
258eda14cbcSMatt Macy 	spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
259eda14cbcSMatt Macy 	vdev_free(rvd);
260eda14cbcSMatt Macy 	spa_config_exit(spa, SCL_ALL, FTAG);
261eda14cbcSMatt Macy 	mutex_exit(&spa_namespace_lock);
262eda14cbcSMatt Macy 
26316038816SMartin Matuska 	fnvlist_free(config);
264eda14cbcSMatt Macy 	return (0);
265eda14cbcSMatt Macy }
266eda14cbcSMatt Macy 
267eda14cbcSMatt Macy const char *
spa_history_zone(void)268eda14cbcSMatt Macy spa_history_zone(void)
269eda14cbcSMatt Macy {
270eda14cbcSMatt Macy 	return ("freebsd");
271eda14cbcSMatt Macy }
272c03c5b1cSMartin Matuska 
273c03c5b1cSMartin Matuska void
spa_import_os(spa_t * spa)274c03c5b1cSMartin Matuska spa_import_os(spa_t *spa)
275c03c5b1cSMartin Matuska {
276c03c5b1cSMartin Matuska 	(void) spa;
277c03c5b1cSMartin Matuska }
278c03c5b1cSMartin Matuska 
279c03c5b1cSMartin Matuska void
spa_export_os(spa_t * spa)280c03c5b1cSMartin Matuska spa_export_os(spa_t *spa)
281c03c5b1cSMartin Matuska {
282c03c5b1cSMartin Matuska 	(void) spa;
283c03c5b1cSMartin Matuska }
284c03c5b1cSMartin Matuska 
285c03c5b1cSMartin Matuska void
spa_activate_os(spa_t * spa)286c03c5b1cSMartin Matuska spa_activate_os(spa_t *spa)
287c03c5b1cSMartin Matuska {
288c03c5b1cSMartin Matuska 	(void) spa;
289c03c5b1cSMartin Matuska }
290c03c5b1cSMartin Matuska 
291c03c5b1cSMartin Matuska void
spa_deactivate_os(spa_t * spa)292c03c5b1cSMartin Matuska spa_deactivate_os(spa_t *spa)
293c03c5b1cSMartin Matuska {
294c03c5b1cSMartin Matuska 	(void) spa;
295c03c5b1cSMartin Matuska }
296