xref: /illumos-gate/usr/src/uts/common/fs/zfs/vdev_label.c (revision e14bb3258d05c1b1077e2db7cf77088924e56919)
1fa9e4066Sahrens /*
2fa9e4066Sahrens  * CDDL HEADER START
3fa9e4066Sahrens  *
4fa9e4066Sahrens  * The contents of this file are subject to the terms of the
5441d80aaSlling  * Common Development and Distribution License (the "License").
6441d80aaSlling  * You may not use this file except in compliance with the License.
7fa9e4066Sahrens  *
8fa9e4066Sahrens  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9fa9e4066Sahrens  * or http://www.opensolaris.org/os/licensing.
10fa9e4066Sahrens  * See the License for the specific language governing permissions
11fa9e4066Sahrens  * and limitations under the License.
12fa9e4066Sahrens  *
13fa9e4066Sahrens  * When distributing Covered Code, include this CDDL HEADER in each
14fa9e4066Sahrens  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15fa9e4066Sahrens  * If applicable, add the following below this CDDL HEADER, with the
16fa9e4066Sahrens  * fields enclosed by brackets "[]" replaced with your own identifying
17fa9e4066Sahrens  * information: Portions Copyright [yyyy] [name of copyright owner]
18fa9e4066Sahrens  *
19fa9e4066Sahrens  * CDDL HEADER END
20fa9e4066Sahrens  */
21fa9e4066Sahrens /*
2221bf64a7Sgw  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
23fa9e4066Sahrens  * Use is subject to license terms.
24fa9e4066Sahrens  */
25fa9e4066Sahrens 
26fa9e4066Sahrens /*
27fa9e4066Sahrens  * Virtual Device Labels
28fa9e4066Sahrens  * ---------------------
29fa9e4066Sahrens  *
30fa9e4066Sahrens  * The vdev label serves several distinct purposes:
31fa9e4066Sahrens  *
32fa9e4066Sahrens  *	1. Uniquely identify this device as part of a ZFS pool and confirm its
33fa9e4066Sahrens  *	   identity within the pool.
34fa9e4066Sahrens  *
35fa9e4066Sahrens  * 	2. Verify that all the devices given in a configuration are present
36fa9e4066Sahrens  *         within the pool.
37fa9e4066Sahrens  *
38fa9e4066Sahrens  * 	3. Determine the uberblock for the pool.
39fa9e4066Sahrens  *
40fa9e4066Sahrens  * 	4. In case of an import operation, determine the configuration of the
41fa9e4066Sahrens  *         toplevel vdev of which it is a part.
42fa9e4066Sahrens  *
43fa9e4066Sahrens  * 	5. If an import operation cannot find all the devices in the pool,
44fa9e4066Sahrens  *         provide enough information to the administrator to determine which
45fa9e4066Sahrens  *         devices are missing.
46fa9e4066Sahrens  *
47fa9e4066Sahrens  * It is important to note that while the kernel is responsible for writing the
48fa9e4066Sahrens  * label, it only consumes the information in the first three cases.  The
49fa9e4066Sahrens  * latter information is only consumed in userland when determining the
50fa9e4066Sahrens  * configuration to import a pool.
51fa9e4066Sahrens  *
52fa9e4066Sahrens  *
53fa9e4066Sahrens  * Label Organization
54fa9e4066Sahrens  * ------------------
55fa9e4066Sahrens  *
56fa9e4066Sahrens  * Before describing the contents of the label, it's important to understand how
57fa9e4066Sahrens  * the labels are written and updated with respect to the uberblock.
58fa9e4066Sahrens  *
59fa9e4066Sahrens  * When the pool configuration is altered, either because it was newly created
60fa9e4066Sahrens  * or a device was added, we want to update all the labels such that we can deal
61fa9e4066Sahrens  * with fatal failure at any point.  To this end, each disk has two labels which
62fa9e4066Sahrens  * are updated before and after the uberblock is synced.  Assuming we have
633d7072f8Seschrock  * labels and an uberblock with the following transaction groups:
64fa9e4066Sahrens  *
65fa9e4066Sahrens  *              L1          UB          L2
66fa9e4066Sahrens  *           +------+    +------+    +------+
67fa9e4066Sahrens  *           |      |    |      |    |      |
68fa9e4066Sahrens  *           | t10  |    | t10  |    | t10  |
69fa9e4066Sahrens  *           |      |    |      |    |      |
70fa9e4066Sahrens  *           +------+    +------+    +------+
71fa9e4066Sahrens  *
72fa9e4066Sahrens  * In this stable state, the labels and the uberblock were all updated within
73fa9e4066Sahrens  * the same transaction group (10).  Each label is mirrored and checksummed, so
74fa9e4066Sahrens  * that we can detect when we fail partway through writing the label.
75fa9e4066Sahrens  *
76fa9e4066Sahrens  * In order to identify which labels are valid, the labels are written in the
77fa9e4066Sahrens  * following manner:
78fa9e4066Sahrens  *
79fa9e4066Sahrens  * 	1. For each vdev, update 'L1' to the new label
80fa9e4066Sahrens  * 	2. Update the uberblock
81fa9e4066Sahrens  * 	3. For each vdev, update 'L2' to the new label
82fa9e4066Sahrens  *
83fa9e4066Sahrens  * Given arbitrary failure, we can determine the correct label to use based on
84fa9e4066Sahrens  * the transaction group.  If we fail after updating L1 but before updating the
85fa9e4066Sahrens  * UB, we will notice that L1's transaction group is greater than the uberblock,
86fa9e4066Sahrens  * so L2 must be valid.  If we fail after writing the uberblock but before
87fa9e4066Sahrens  * writing L2, we will notice that L2's transaction group is less than L1, and
88fa9e4066Sahrens  * therefore L1 is valid.
89fa9e4066Sahrens  *
90fa9e4066Sahrens  * Another added complexity is that not every label is updated when the config
91fa9e4066Sahrens  * is synced.  If we add a single device, we do not want to have to re-write
92fa9e4066Sahrens  * every label for every device in the pool.  This means that both L1 and L2 may
93fa9e4066Sahrens  * be older than the pool uberblock, because the necessary information is stored
94fa9e4066Sahrens  * on another vdev.
95fa9e4066Sahrens  *
96fa9e4066Sahrens  *
97fa9e4066Sahrens  * On-disk Format
98fa9e4066Sahrens  * --------------
99fa9e4066Sahrens  *
100fa9e4066Sahrens  * The vdev label consists of two distinct parts, and is wrapped within the
101fa9e4066Sahrens  * vdev_label_t structure.  The label includes 8k of padding to permit legacy
102fa9e4066Sahrens  * VTOC disk labels, but is otherwise ignored.
103fa9e4066Sahrens  *
104fa9e4066Sahrens  * The first half of the label is a packed nvlist which contains pool wide
105fa9e4066Sahrens  * properties, per-vdev properties, and configuration information.  It is
106fa9e4066Sahrens  * described in more detail below.
107fa9e4066Sahrens  *
108fa9e4066Sahrens  * The latter half of the label consists of a redundant array of uberblocks.
109fa9e4066Sahrens  * These uberblocks are updated whenever a transaction group is committed,
110fa9e4066Sahrens  * or when the configuration is updated.  When a pool is loaded, we scan each
111fa9e4066Sahrens  * vdev for the 'best' uberblock.
112fa9e4066Sahrens  *
113fa9e4066Sahrens  *
114fa9e4066Sahrens  * Configuration Information
115fa9e4066Sahrens  * -------------------------
116fa9e4066Sahrens  *
117fa9e4066Sahrens  * The nvlist describing the pool and vdev contains the following elements:
118fa9e4066Sahrens  *
119fa9e4066Sahrens  * 	version		ZFS on-disk version
120fa9e4066Sahrens  * 	name		Pool name
121fa9e4066Sahrens  * 	state		Pool state
122fa9e4066Sahrens  * 	txg		Transaction group in which this label was written
123fa9e4066Sahrens  * 	pool_guid	Unique identifier for this pool
124fa9e4066Sahrens  * 	vdev_tree	An nvlist describing vdev tree.
125fa9e4066Sahrens  *
126fa9e4066Sahrens  * Each leaf device label also contains the following:
127fa9e4066Sahrens  *
128fa9e4066Sahrens  * 	top_guid	Unique ID for top-level vdev in which this is contained
129fa9e4066Sahrens  * 	guid		Unique ID for the leaf vdev
130fa9e4066Sahrens  *
131fa9e4066Sahrens  * The 'vs' configuration follows the format described in 'spa_config.c'.
132fa9e4066Sahrens  */
133fa9e4066Sahrens 
134fa9e4066Sahrens #include <sys/zfs_context.h>
135fa9e4066Sahrens #include <sys/spa.h>
136fa9e4066Sahrens #include <sys/spa_impl.h>
137fa9e4066Sahrens #include <sys/dmu.h>
138fa9e4066Sahrens #include <sys/zap.h>
139fa9e4066Sahrens #include <sys/vdev.h>
140fa9e4066Sahrens #include <sys/vdev_impl.h>
141fa9e4066Sahrens #include <sys/uberblock_impl.h>
142fa9e4066Sahrens #include <sys/metaslab.h>
143fa9e4066Sahrens #include <sys/zio.h>
144fa9e4066Sahrens #include <sys/fs/zfs.h>
145fa9e4066Sahrens 
146fa9e4066Sahrens /*
147fa9e4066Sahrens  * Basic routines to read and write from a vdev label.
148fa9e4066Sahrens  * Used throughout the rest of this file.
149fa9e4066Sahrens  */
150fa9e4066Sahrens uint64_t
151fa9e4066Sahrens vdev_label_offset(uint64_t psize, int l, uint64_t offset)
152fa9e4066Sahrens {
153ecc2d604Sbonwick 	ASSERT(offset < sizeof (vdev_label_t));
154e7437265Sahrens 	ASSERT(P2PHASE_TYPED(psize, sizeof (vdev_label_t), uint64_t) == 0);
155ecc2d604Sbonwick 
156fa9e4066Sahrens 	return (offset + l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
157fa9e4066Sahrens 	    0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
158fa9e4066Sahrens }
159fa9e4066Sahrens 
16021bf64a7Sgw /*
16121bf64a7Sgw  * Returns back the vdev label associated with the passed in offset.
16221bf64a7Sgw  */
16321bf64a7Sgw int
16421bf64a7Sgw vdev_label_number(uint64_t psize, uint64_t offset)
16521bf64a7Sgw {
16621bf64a7Sgw 	int l;
16721bf64a7Sgw 
16821bf64a7Sgw 	if (offset >= psize - VDEV_LABEL_END_SIZE) {
16921bf64a7Sgw 		offset -= psize - VDEV_LABEL_END_SIZE;
17021bf64a7Sgw 		offset += (VDEV_LABELS / 2) * sizeof (vdev_label_t);
17121bf64a7Sgw 	}
17221bf64a7Sgw 	l = offset / sizeof (vdev_label_t);
17321bf64a7Sgw 	return (l < VDEV_LABELS ? l : -1);
17421bf64a7Sgw }
17521bf64a7Sgw 
176fa9e4066Sahrens static void
177fa9e4066Sahrens vdev_label_read(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
178*e14bb325SJeff Bonwick 	uint64_t size, zio_done_func_t *done, void *private, int flags)
179fa9e4066Sahrens {
180*e14bb325SJeff Bonwick 	ASSERT(spa_config_held(zio->io_spa, SCL_STATE_ALL, RW_WRITER) ==
181*e14bb325SJeff Bonwick 	    SCL_STATE_ALL);
182*e14bb325SJeff Bonwick 	ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
183fa9e4066Sahrens 
184fa9e4066Sahrens 	zio_nowait(zio_read_phys(zio, vd,
185fa9e4066Sahrens 	    vdev_label_offset(vd->vdev_psize, l, offset),
186fa9e4066Sahrens 	    size, buf, ZIO_CHECKSUM_LABEL, done, private,
187*e14bb325SJeff Bonwick 	    ZIO_PRIORITY_SYNC_READ, flags, B_TRUE));
188fa9e4066Sahrens }
189fa9e4066Sahrens 
190fa9e4066Sahrens static void
191fa9e4066Sahrens vdev_label_write(zio_t *zio, vdev_t *vd, int l, void *buf, uint64_t offset,
19217f17c2dSbonwick 	uint64_t size, zio_done_func_t *done, void *private, int flags)
193fa9e4066Sahrens {
194*e14bb325SJeff Bonwick 	ASSERT(spa_config_held(zio->io_spa, SCL_ALL, RW_WRITER) == SCL_ALL ||
195*e14bb325SJeff Bonwick 	    (spa_config_held(zio->io_spa, SCL_CONFIG | SCL_STATE, RW_READER) ==
196*e14bb325SJeff Bonwick 	    (SCL_CONFIG | SCL_STATE) &&
197*e14bb325SJeff Bonwick 	    dsl_pool_sync_context(spa_get_dsl(zio->io_spa))));
198*e14bb325SJeff Bonwick 	ASSERT(flags & ZIO_FLAG_CONFIG_WRITER);
199fa9e4066Sahrens 
200fa9e4066Sahrens 	zio_nowait(zio_write_phys(zio, vd,
201fa9e4066Sahrens 	    vdev_label_offset(vd->vdev_psize, l, offset),
202fa9e4066Sahrens 	    size, buf, ZIO_CHECKSUM_LABEL, done, private,
20317f17c2dSbonwick 	    ZIO_PRIORITY_SYNC_WRITE, flags, B_TRUE));
204fa9e4066Sahrens }
205fa9e4066Sahrens 
206fa9e4066Sahrens /*
207fa9e4066Sahrens  * Generate the nvlist representing this vdev's config.
208fa9e4066Sahrens  */
209fa9e4066Sahrens nvlist_t *
21099653d4eSeschrock vdev_config_generate(spa_t *spa, vdev_t *vd, boolean_t getstats,
211fa94a07fSbrendan     boolean_t isspare, boolean_t isl2cache)
212fa9e4066Sahrens {
213fa9e4066Sahrens 	nvlist_t *nv = NULL;
214fa9e4066Sahrens 
215ea8dc4b6Seschrock 	VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
216fa9e4066Sahrens 
217fa9e4066Sahrens 	VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_TYPE,
218fa9e4066Sahrens 	    vd->vdev_ops->vdev_op_type) == 0);
219fa94a07fSbrendan 	if (!isspare && !isl2cache)
22099653d4eSeschrock 		VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ID, vd->vdev_id)
22199653d4eSeschrock 		    == 0);
222fa9e4066Sahrens 	VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_GUID, vd->vdev_guid) == 0);
223fa9e4066Sahrens 
224fa9e4066Sahrens 	if (vd->vdev_path != NULL)
225fa9e4066Sahrens 		VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PATH,
226fa9e4066Sahrens 		    vd->vdev_path) == 0);
227fa9e4066Sahrens 
228fa9e4066Sahrens 	if (vd->vdev_devid != NULL)
229fa9e4066Sahrens 		VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_DEVID,
230fa9e4066Sahrens 		    vd->vdev_devid) == 0);
231fa9e4066Sahrens 
2323d7072f8Seschrock 	if (vd->vdev_physpath != NULL)
2333d7072f8Seschrock 		VERIFY(nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
2343d7072f8Seschrock 		    vd->vdev_physpath) == 0);
2353d7072f8Seschrock 
23699653d4eSeschrock 	if (vd->vdev_nparity != 0) {
23799653d4eSeschrock 		ASSERT(strcmp(vd->vdev_ops->vdev_op_type,
23899653d4eSeschrock 		    VDEV_TYPE_RAIDZ) == 0);
23999653d4eSeschrock 
24099653d4eSeschrock 		/*
24199653d4eSeschrock 		 * Make sure someone hasn't managed to sneak a fancy new vdev
24299653d4eSeschrock 		 * into a crufty old storage pool.
24399653d4eSeschrock 		 */
24499653d4eSeschrock 		ASSERT(vd->vdev_nparity == 1 ||
24599653d4eSeschrock 		    (vd->vdev_nparity == 2 &&
246e7437265Sahrens 		    spa_version(spa) >= SPA_VERSION_RAID6));
24799653d4eSeschrock 
24899653d4eSeschrock 		/*
24999653d4eSeschrock 		 * Note that we'll add the nparity tag even on storage pools
25099653d4eSeschrock 		 * that only support a single parity device -- older software
25199653d4eSeschrock 		 * will just ignore it.
25299653d4eSeschrock 		 */
25399653d4eSeschrock 		VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NPARITY,
25499653d4eSeschrock 		    vd->vdev_nparity) == 0);
25599653d4eSeschrock 	}
25699653d4eSeschrock 
257afefbcddSeschrock 	if (vd->vdev_wholedisk != -1ULL)
258afefbcddSeschrock 		VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK,
259afefbcddSeschrock 		    vd->vdev_wholedisk) == 0);
260afefbcddSeschrock 
261ea8dc4b6Seschrock 	if (vd->vdev_not_present)
262ea8dc4b6Seschrock 		VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 1) == 0);
263ea8dc4b6Seschrock 
26499653d4eSeschrock 	if (vd->vdev_isspare)
26599653d4eSeschrock 		VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 1) == 0);
26699653d4eSeschrock 
267fa94a07fSbrendan 	if (!isspare && !isl2cache && vd == vd->vdev_top) {
268fa9e4066Sahrens 		VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY,
269fa9e4066Sahrens 		    vd->vdev_ms_array) == 0);
270fa9e4066Sahrens 		VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT,
271fa9e4066Sahrens 		    vd->vdev_ms_shift) == 0);
272fa9e4066Sahrens 		VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASHIFT,
273fa9e4066Sahrens 		    vd->vdev_ashift) == 0);
274fa9e4066Sahrens 		VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_ASIZE,
275fa9e4066Sahrens 		    vd->vdev_asize) == 0);
2768654d025Sperrin 		VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_IS_LOG,
2778654d025Sperrin 		    vd->vdev_islog) == 0);
278fa9e4066Sahrens 	}
279fa9e4066Sahrens 
280fa9e4066Sahrens 	if (vd->vdev_dtl.smo_object != 0)
281fa9e4066Sahrens 		VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DTL,
282fa9e4066Sahrens 		    vd->vdev_dtl.smo_object) == 0);
283fa9e4066Sahrens 
284fa9e4066Sahrens 	if (getstats) {
285fa9e4066Sahrens 		vdev_stat_t vs;
286fa9e4066Sahrens 		vdev_get_stats(vd, &vs);
287fa9e4066Sahrens 		VERIFY(nvlist_add_uint64_array(nv, ZPOOL_CONFIG_STATS,
288fa9e4066Sahrens 		    (uint64_t *)&vs, sizeof (vs) / sizeof (uint64_t)) == 0);
289fa9e4066Sahrens 	}
290fa9e4066Sahrens 
291fa9e4066Sahrens 	if (!vd->vdev_ops->vdev_op_leaf) {
292fa9e4066Sahrens 		nvlist_t **child;
293fa9e4066Sahrens 		int c;
294fa9e4066Sahrens 
295fa9e4066Sahrens 		child = kmem_alloc(vd->vdev_children * sizeof (nvlist_t *),
296fa9e4066Sahrens 		    KM_SLEEP);
297fa9e4066Sahrens 
298fa9e4066Sahrens 		for (c = 0; c < vd->vdev_children; c++)
29999653d4eSeschrock 			child[c] = vdev_config_generate(spa, vd->vdev_child[c],
300fa94a07fSbrendan 			    getstats, isspare, isl2cache);
301fa9e4066Sahrens 
302fa9e4066Sahrens 		VERIFY(nvlist_add_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
303fa9e4066Sahrens 		    child, vd->vdev_children) == 0);
304fa9e4066Sahrens 
305fa9e4066Sahrens 		for (c = 0; c < vd->vdev_children; c++)
306fa9e4066Sahrens 			nvlist_free(child[c]);
307fa9e4066Sahrens 
308fa9e4066Sahrens 		kmem_free(child, vd->vdev_children * sizeof (nvlist_t *));
309441d80aaSlling 
310441d80aaSlling 	} else {
311ecc2d604Sbonwick 		if (vd->vdev_offline && !vd->vdev_tmpoffline)
312441d80aaSlling 			VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_OFFLINE,
313ecc2d604Sbonwick 			    B_TRUE) == 0);
3143d7072f8Seschrock 		if (vd->vdev_faulted)
3153d7072f8Seschrock 			VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_FAULTED,
3163d7072f8Seschrock 			    B_TRUE) == 0);
3173d7072f8Seschrock 		if (vd->vdev_degraded)
3183d7072f8Seschrock 			VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_DEGRADED,
3193d7072f8Seschrock 			    B_TRUE) == 0);
3203d7072f8Seschrock 		if (vd->vdev_removed)
3213d7072f8Seschrock 			VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_REMOVED,
3223d7072f8Seschrock 			    B_TRUE) == 0);
3233d7072f8Seschrock 		if (vd->vdev_unspare)
3243d7072f8Seschrock 			VERIFY(nvlist_add_uint64(nv, ZPOOL_CONFIG_UNSPARE,
3253d7072f8Seschrock 			    B_TRUE) == 0);
326fa9e4066Sahrens 	}
327fa9e4066Sahrens 
328fa9e4066Sahrens 	return (nv);
329fa9e4066Sahrens }
330fa9e4066Sahrens 
331fa9e4066Sahrens nvlist_t *
332fa9e4066Sahrens vdev_label_read_config(vdev_t *vd)
333fa9e4066Sahrens {
3340373e76bSbonwick 	spa_t *spa = vd->vdev_spa;
335fa9e4066Sahrens 	nvlist_t *config = NULL;
336fa9e4066Sahrens 	vdev_phys_t *vp;
337fa9e4066Sahrens 	zio_t *zio;
338*e14bb325SJeff Bonwick 	int flags =
339*e14bb325SJeff Bonwick 	    ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE;
340fa9e4066Sahrens 
341*e14bb325SJeff Bonwick 	ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL);
3420373e76bSbonwick 
3430a4e9518Sgw 	if (!vdev_readable(vd))
344fa9e4066Sahrens 		return (NULL);
345fa9e4066Sahrens 
346fa9e4066Sahrens 	vp = zio_buf_alloc(sizeof (vdev_phys_t));
347fa9e4066Sahrens 
348*e14bb325SJeff Bonwick 	for (int l = 0; l < VDEV_LABELS; l++) {
349fa9e4066Sahrens 
350*e14bb325SJeff Bonwick 		zio = zio_root(spa, NULL, NULL, flags);
351fa9e4066Sahrens 
352fa9e4066Sahrens 		vdev_label_read(zio, vd, l, vp,
353fa9e4066Sahrens 		    offsetof(vdev_label_t, vl_vdev_phys),
354*e14bb325SJeff Bonwick 		    sizeof (vdev_phys_t), NULL, NULL, flags);
355fa9e4066Sahrens 
356fa9e4066Sahrens 		if (zio_wait(zio) == 0 &&
357fa9e4066Sahrens 		    nvlist_unpack(vp->vp_nvlist, sizeof (vp->vp_nvlist),
358ea8dc4b6Seschrock 		    &config, 0) == 0)
359fa9e4066Sahrens 			break;
360fa9e4066Sahrens 
361fa9e4066Sahrens 		if (config != NULL) {
362fa9e4066Sahrens 			nvlist_free(config);
363fa9e4066Sahrens 			config = NULL;
364fa9e4066Sahrens 		}
365fa9e4066Sahrens 	}
366fa9e4066Sahrens 
367fa9e4066Sahrens 	zio_buf_free(vp, sizeof (vdev_phys_t));
368fa9e4066Sahrens 
369fa9e4066Sahrens 	return (config);
370fa9e4066Sahrens }
371fa9e4066Sahrens 
37239c23413Seschrock /*
37339c23413Seschrock  * Determine if a device is in use.  The 'spare_guid' parameter will be filled
37439c23413Seschrock  * in with the device guid if this spare is active elsewhere on the system.
37539c23413Seschrock  */
37639c23413Seschrock static boolean_t
37739c23413Seschrock vdev_inuse(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason,
378fa94a07fSbrendan     uint64_t *spare_guid, uint64_t *l2cache_guid)
37939c23413Seschrock {
38039c23413Seschrock 	spa_t *spa = vd->vdev_spa;
38139c23413Seschrock 	uint64_t state, pool_guid, device_guid, txg, spare_pool;
38239c23413Seschrock 	uint64_t vdtxg = 0;
38339c23413Seschrock 	nvlist_t *label;
38439c23413Seschrock 
38539c23413Seschrock 	if (spare_guid)
38639c23413Seschrock 		*spare_guid = 0ULL;
387fa94a07fSbrendan 	if (l2cache_guid)
388fa94a07fSbrendan 		*l2cache_guid = 0ULL;
38939c23413Seschrock 
39039c23413Seschrock 	/*
39139c23413Seschrock 	 * Read the label, if any, and perform some basic sanity checks.
39239c23413Seschrock 	 */
39339c23413Seschrock 	if ((label = vdev_label_read_config(vd)) == NULL)
39439c23413Seschrock 		return (B_FALSE);
39539c23413Seschrock 
39639c23413Seschrock 	(void) nvlist_lookup_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
39739c23413Seschrock 	    &vdtxg);
39839c23413Seschrock 
39939c23413Seschrock 	if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE,
40039c23413Seschrock 	    &state) != 0 ||
40139c23413Seschrock 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID,
40239c23413Seschrock 	    &device_guid) != 0) {
40339c23413Seschrock 		nvlist_free(label);
40439c23413Seschrock 		return (B_FALSE);
40539c23413Seschrock 	}
40639c23413Seschrock 
407fa94a07fSbrendan 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
40839c23413Seschrock 	    (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID,
40939c23413Seschrock 	    &pool_guid) != 0 ||
41039c23413Seschrock 	    nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_TXG,
41139c23413Seschrock 	    &txg) != 0)) {
41239c23413Seschrock 		nvlist_free(label);
41339c23413Seschrock 		return (B_FALSE);
41439c23413Seschrock 	}
41539c23413Seschrock 
41639c23413Seschrock 	nvlist_free(label);
41739c23413Seschrock 
41839c23413Seschrock 	/*
41939c23413Seschrock 	 * Check to see if this device indeed belongs to the pool it claims to
42039c23413Seschrock 	 * be a part of.  The only way this is allowed is if the device is a hot
42139c23413Seschrock 	 * spare (which we check for later on).
42239c23413Seschrock 	 */
423fa94a07fSbrendan 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
42439c23413Seschrock 	    !spa_guid_exists(pool_guid, device_guid) &&
42589a89ebfSlling 	    !spa_spare_exists(device_guid, NULL, NULL) &&
426fa94a07fSbrendan 	    !spa_l2cache_exists(device_guid, NULL))
42739c23413Seschrock 		return (B_FALSE);
42839c23413Seschrock 
42939c23413Seschrock 	/*
43039c23413Seschrock 	 * If the transaction group is zero, then this an initialized (but
43139c23413Seschrock 	 * unused) label.  This is only an error if the create transaction
43239c23413Seschrock 	 * on-disk is the same as the one we're using now, in which case the
43339c23413Seschrock 	 * user has attempted to add the same vdev multiple times in the same
43439c23413Seschrock 	 * transaction.
43539c23413Seschrock 	 */
436fa94a07fSbrendan 	if (state != POOL_STATE_SPARE && state != POOL_STATE_L2CACHE &&
437fa94a07fSbrendan 	    txg == 0 && vdtxg == crtxg)
43839c23413Seschrock 		return (B_TRUE);
43939c23413Seschrock 
44039c23413Seschrock 	/*
44139c23413Seschrock 	 * Check to see if this is a spare device.  We do an explicit check for
44239c23413Seschrock 	 * spa_has_spare() here because it may be on our pending list of spares
443fa94a07fSbrendan 	 * to add.  We also check if it is an l2cache device.
44439c23413Seschrock 	 */
44589a89ebfSlling 	if (spa_spare_exists(device_guid, &spare_pool, NULL) ||
44639c23413Seschrock 	    spa_has_spare(spa, device_guid)) {
44739c23413Seschrock 		if (spare_guid)
44839c23413Seschrock 			*spare_guid = device_guid;
44939c23413Seschrock 
45039c23413Seschrock 		switch (reason) {
45139c23413Seschrock 		case VDEV_LABEL_CREATE:
452fa94a07fSbrendan 		case VDEV_LABEL_L2CACHE:
45339c23413Seschrock 			return (B_TRUE);
45439c23413Seschrock 
45539c23413Seschrock 		case VDEV_LABEL_REPLACE:
45639c23413Seschrock 			return (!spa_has_spare(spa, device_guid) ||
45739c23413Seschrock 			    spare_pool != 0ULL);
45839c23413Seschrock 
45939c23413Seschrock 		case VDEV_LABEL_SPARE:
46039c23413Seschrock 			return (spa_has_spare(spa, device_guid));
46139c23413Seschrock 		}
46239c23413Seschrock 	}
46339c23413Seschrock 
464fa94a07fSbrendan 	/*
465fa94a07fSbrendan 	 * Check to see if this is an l2cache device.
466fa94a07fSbrendan 	 */
467fa94a07fSbrendan 	if (spa_l2cache_exists(device_guid, NULL))
468fa94a07fSbrendan 		return (B_TRUE);
469fa94a07fSbrendan 
47039c23413Seschrock 	/*
47139c23413Seschrock 	 * If the device is marked ACTIVE, then this device is in use by another
47239c23413Seschrock 	 * pool on the system.
47339c23413Seschrock 	 */
47439c23413Seschrock 	return (state == POOL_STATE_ACTIVE);
47539c23413Seschrock }
47639c23413Seschrock 
47739c23413Seschrock /*
47839c23413Seschrock  * Initialize a vdev label.  We check to make sure each leaf device is not in
47939c23413Seschrock  * use, and writable.  We put down an initial label which we will later
48039c23413Seschrock  * overwrite with a complete label.  Note that it's important to do this
48139c23413Seschrock  * sequentially, not in parallel, so that we catch cases of multiple use of the
48239c23413Seschrock  * same leaf vdev in the vdev we're creating -- e.g. mirroring a disk with
48339c23413Seschrock  * itself.
48439c23413Seschrock  */
48539c23413Seschrock int
48639c23413Seschrock vdev_label_init(vdev_t *vd, uint64_t crtxg, vdev_labeltype_t reason)
487fa9e4066Sahrens {
488fa9e4066Sahrens 	spa_t *spa = vd->vdev_spa;
489fa9e4066Sahrens 	nvlist_t *label;
490fa9e4066Sahrens 	vdev_phys_t *vp;
491fa9e4066Sahrens 	vdev_boot_header_t *vb;
492ecc2d604Sbonwick 	uberblock_t *ub;
493fa9e4066Sahrens 	zio_t *zio;
494fa9e4066Sahrens 	char *buf;
495fa9e4066Sahrens 	size_t buflen;
496fa9e4066Sahrens 	int error;
497fa94a07fSbrendan 	uint64_t spare_guid, l2cache_guid;
498*e14bb325SJeff Bonwick 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
499fa9e4066Sahrens 
500*e14bb325SJeff Bonwick 	ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL);
5010373e76bSbonwick 
502*e14bb325SJeff Bonwick 	for (int c = 0; c < vd->vdev_children; c++)
50339c23413Seschrock 		if ((error = vdev_label_init(vd->vdev_child[c],
50439c23413Seschrock 		    crtxg, reason)) != 0)
505fa9e4066Sahrens 			return (error);
506fa9e4066Sahrens 
507fa9e4066Sahrens 	if (!vd->vdev_ops->vdev_op_leaf)
508fa9e4066Sahrens 		return (0);
509fa9e4066Sahrens 
510fa9e4066Sahrens 	/*
51139c23413Seschrock 	 * Dead vdevs cannot be initialized.
512fa9e4066Sahrens 	 */
513fa9e4066Sahrens 	if (vdev_is_dead(vd))
514fa9e4066Sahrens 		return (EIO);
515fa9e4066Sahrens 
516fa9e4066Sahrens 	/*
51739c23413Seschrock 	 * Determine if the vdev is in use.
518fa9e4066Sahrens 	 */
51939c23413Seschrock 	if (reason != VDEV_LABEL_REMOVE &&
520fa94a07fSbrendan 	    vdev_inuse(vd, crtxg, reason, &spare_guid, &l2cache_guid))
52139c23413Seschrock 		return (EBUSY);
52239c23413Seschrock 
52339c23413Seschrock 	ASSERT(reason != VDEV_LABEL_REMOVE ||
524fa94a07fSbrendan 	    vdev_inuse(vd, crtxg, reason, NULL, NULL));
52539c23413Seschrock 
52639c23413Seschrock 	/*
527fa94a07fSbrendan 	 * If this is a request to add or replace a spare or l2cache device
528fa94a07fSbrendan 	 * that is in use elsewhere on the system, then we must update the
529fa94a07fSbrendan 	 * guid (which was initialized to a random value) to reflect the
530fa94a07fSbrendan 	 * actual GUID (which is shared between multiple pools).
53139c23413Seschrock 	 */
532fa94a07fSbrendan 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_L2CACHE &&
533fa94a07fSbrendan 	    spare_guid != 0ULL) {
534*e14bb325SJeff Bonwick 		ASSERT(vd->vdev_parent == NULL);
53599653d4eSeschrock 
53639c23413Seschrock 		vd->vdev_guid = vd->vdev_guid_sum = spare_guid;
53739c23413Seschrock 
53899653d4eSeschrock 		/*
53939c23413Seschrock 		 * If this is a replacement, then we want to fallthrough to the
54039c23413Seschrock 		 * rest of the code.  If we're adding a spare, then it's already
5413d7072f8Seschrock 		 * labeled appropriately and we can just return.
54299653d4eSeschrock 		 */
54339c23413Seschrock 		if (reason == VDEV_LABEL_SPARE)
54439c23413Seschrock 			return (0);
54539c23413Seschrock 		ASSERT(reason == VDEV_LABEL_REPLACE);
546fa9e4066Sahrens 	}
547fa9e4066Sahrens 
548fa94a07fSbrendan 	if (reason != VDEV_LABEL_REMOVE && reason != VDEV_LABEL_SPARE &&
549fa94a07fSbrendan 	    l2cache_guid != 0ULL) {
550*e14bb325SJeff Bonwick 		ASSERT(vd->vdev_parent == NULL);
551fa94a07fSbrendan 
552fa94a07fSbrendan 		vd->vdev_guid = vd->vdev_guid_sum = l2cache_guid;
553fa94a07fSbrendan 
554fa94a07fSbrendan 		/*
555fa94a07fSbrendan 		 * If this is a replacement, then we want to fallthrough to the
556fa94a07fSbrendan 		 * rest of the code.  If we're adding an l2cache, then it's
557fa94a07fSbrendan 		 * already labeled appropriately and we can just return.
558fa94a07fSbrendan 		 */
559fa94a07fSbrendan 		if (reason == VDEV_LABEL_L2CACHE)
560fa94a07fSbrendan 			return (0);
561fa94a07fSbrendan 		ASSERT(reason == VDEV_LABEL_REPLACE);
562fa94a07fSbrendan 	}
563fa94a07fSbrendan 
564fa9e4066Sahrens 	/*
56539c23413Seschrock 	 * Initialize its label.
566fa9e4066Sahrens 	 */
567fa9e4066Sahrens 	vp = zio_buf_alloc(sizeof (vdev_phys_t));
568fa9e4066Sahrens 	bzero(vp, sizeof (vdev_phys_t));
569fa9e4066Sahrens 
570fa9e4066Sahrens 	/*
571fa9e4066Sahrens 	 * Generate a label describing the pool and our top-level vdev.
572fa9e4066Sahrens 	 * We mark it as being from txg 0 to indicate that it's not
573fa9e4066Sahrens 	 * really part of an active pool just yet.  The labels will
574fa9e4066Sahrens 	 * be written again with a meaningful txg by spa_sync().
575fa9e4066Sahrens 	 */
57639c23413Seschrock 	if (reason == VDEV_LABEL_SPARE ||
57739c23413Seschrock 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isspare)) {
57839c23413Seschrock 		/*
57939c23413Seschrock 		 * For inactive hot spares, we generate a special label that
58039c23413Seschrock 		 * identifies as a mutually shared hot spare.  We write the
58139c23413Seschrock 		 * label if we are adding a hot spare, or if we are removing an
58239c23413Seschrock 		 * active hot spare (in which case we want to revert the
58339c23413Seschrock 		 * labels).
58439c23413Seschrock 		 */
58599653d4eSeschrock 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
58699653d4eSeschrock 
58799653d4eSeschrock 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
58899653d4eSeschrock 		    spa_version(spa)) == 0);
58999653d4eSeschrock 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
59099653d4eSeschrock 		    POOL_STATE_SPARE) == 0);
59199653d4eSeschrock 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
59299653d4eSeschrock 		    vd->vdev_guid) == 0);
593fa94a07fSbrendan 	} else if (reason == VDEV_LABEL_L2CACHE ||
594fa94a07fSbrendan 	    (reason == VDEV_LABEL_REMOVE && vd->vdev_isl2cache)) {
595fa94a07fSbrendan 		/*
596fa94a07fSbrendan 		 * For level 2 ARC devices, add a special label.
597fa94a07fSbrendan 		 */
598fa94a07fSbrendan 		VERIFY(nvlist_alloc(&label, NV_UNIQUE_NAME, KM_SLEEP) == 0);
599fa94a07fSbrendan 
600fa94a07fSbrendan 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_VERSION,
601fa94a07fSbrendan 		    spa_version(spa)) == 0);
602fa94a07fSbrendan 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_POOL_STATE,
603fa94a07fSbrendan 		    POOL_STATE_L2CACHE) == 0);
604fa94a07fSbrendan 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_GUID,
605fa94a07fSbrendan 		    vd->vdev_guid) == 0);
60699653d4eSeschrock 	} else {
60799653d4eSeschrock 		label = spa_config_generate(spa, vd, 0ULL, B_FALSE);
60899653d4eSeschrock 
60999653d4eSeschrock 		/*
61099653d4eSeschrock 		 * Add our creation time.  This allows us to detect multiple
61199653d4eSeschrock 		 * vdev uses as described above, and automatically expires if we
61299653d4eSeschrock 		 * fail.
61399653d4eSeschrock 		 */
61499653d4eSeschrock 		VERIFY(nvlist_add_uint64(label, ZPOOL_CONFIG_CREATE_TXG,
61599653d4eSeschrock 		    crtxg) == 0);
61699653d4eSeschrock 	}
617fa9e4066Sahrens 
618fa9e4066Sahrens 	buf = vp->vp_nvlist;
619fa9e4066Sahrens 	buflen = sizeof (vp->vp_nvlist);
620fa9e4066Sahrens 
621a75573b6Smmusante 	error = nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP);
622a75573b6Smmusante 	if (error != 0) {
623fa9e4066Sahrens 		nvlist_free(label);
624fa9e4066Sahrens 		zio_buf_free(vp, sizeof (vdev_phys_t));
625a75573b6Smmusante 		/* EFAULT means nvlist_pack ran out of room */
626a75573b6Smmusante 		return (error == EFAULT ? ENAMETOOLONG : EINVAL);
627fa9e4066Sahrens 	}
628fa9e4066Sahrens 
629fa9e4066Sahrens 	/*
630fa9e4066Sahrens 	 * Initialize boot block header.
631fa9e4066Sahrens 	 */
632fa9e4066Sahrens 	vb = zio_buf_alloc(sizeof (vdev_boot_header_t));
633fa9e4066Sahrens 	bzero(vb, sizeof (vdev_boot_header_t));
634fa9e4066Sahrens 	vb->vb_magic = VDEV_BOOT_MAGIC;
635fa9e4066Sahrens 	vb->vb_version = VDEV_BOOT_VERSION;
636fa9e4066Sahrens 	vb->vb_offset = VDEV_BOOT_OFFSET;
637fa9e4066Sahrens 	vb->vb_size = VDEV_BOOT_SIZE;
638fa9e4066Sahrens 
639fa9e4066Sahrens 	/*
640fa9e4066Sahrens 	 * Initialize uberblock template.
641fa9e4066Sahrens 	 */
642ecc2d604Sbonwick 	ub = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
643ecc2d604Sbonwick 	bzero(ub, VDEV_UBERBLOCK_SIZE(vd));
644ecc2d604Sbonwick 	*ub = spa->spa_uberblock;
645ecc2d604Sbonwick 	ub->ub_txg = 0;
646fa9e4066Sahrens 
647fa9e4066Sahrens 	/*
648fa9e4066Sahrens 	 * Write everything in parallel.
649fa9e4066Sahrens 	 */
65017f17c2dSbonwick 	zio = zio_root(spa, NULL, NULL, flags);
651fa9e4066Sahrens 
652*e14bb325SJeff Bonwick 	for (int l = 0; l < VDEV_LABELS; l++) {
653fa9e4066Sahrens 
654fa9e4066Sahrens 		vdev_label_write(zio, vd, l, vp,
655fa9e4066Sahrens 		    offsetof(vdev_label_t, vl_vdev_phys),
65617f17c2dSbonwick 		    sizeof (vdev_phys_t), NULL, NULL, flags);
657fa9e4066Sahrens 
658fa9e4066Sahrens 		vdev_label_write(zio, vd, l, vb,
659fa9e4066Sahrens 		    offsetof(vdev_label_t, vl_boot_header),
66017f17c2dSbonwick 		    sizeof (vdev_boot_header_t), NULL, NULL, flags);
661fa9e4066Sahrens 
662*e14bb325SJeff Bonwick 		for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
663ecc2d604Sbonwick 			vdev_label_write(zio, vd, l, ub,
664ecc2d604Sbonwick 			    VDEV_UBERBLOCK_OFFSET(vd, n),
66517f17c2dSbonwick 			    VDEV_UBERBLOCK_SIZE(vd), NULL, NULL, flags);
666fa9e4066Sahrens 		}
667fa9e4066Sahrens 	}
668fa9e4066Sahrens 
669fa9e4066Sahrens 	error = zio_wait(zio);
670fa9e4066Sahrens 
671fa9e4066Sahrens 	nvlist_free(label);
672ecc2d604Sbonwick 	zio_buf_free(ub, VDEV_UBERBLOCK_SIZE(vd));
673fa9e4066Sahrens 	zio_buf_free(vb, sizeof (vdev_boot_header_t));
674fa9e4066Sahrens 	zio_buf_free(vp, sizeof (vdev_phys_t));
675fa9e4066Sahrens 
67639c23413Seschrock 	/*
67739c23413Seschrock 	 * If this vdev hasn't been previously identified as a spare, then we
6783d7072f8Seschrock 	 * mark it as such only if a) we are labeling it as a spare, or b) it
679fa94a07fSbrendan 	 * exists as a spare elsewhere in the system.  Do the same for
680fa94a07fSbrendan 	 * level 2 ARC devices.
68139c23413Seschrock 	 */
68239c23413Seschrock 	if (error == 0 && !vd->vdev_isspare &&
68339c23413Seschrock 	    (reason == VDEV_LABEL_SPARE ||
68489a89ebfSlling 	    spa_spare_exists(vd->vdev_guid, NULL, NULL)))
68539c23413Seschrock 		spa_spare_add(vd);
68699653d4eSeschrock 
687fa94a07fSbrendan 	if (error == 0 && !vd->vdev_isl2cache &&
688fa94a07fSbrendan 	    (reason == VDEV_LABEL_L2CACHE ||
689fa94a07fSbrendan 	    spa_l2cache_exists(vd->vdev_guid, NULL)))
690fa94a07fSbrendan 		spa_l2cache_add(vd);
691fa94a07fSbrendan 
69239c23413Seschrock 	return (error);
69399653d4eSeschrock }
69499653d4eSeschrock 
695fa9e4066Sahrens /*
696fa9e4066Sahrens  * ==========================================================================
697fa9e4066Sahrens  * uberblock load/sync
698fa9e4066Sahrens  * ==========================================================================
699fa9e4066Sahrens  */
700fa9e4066Sahrens 
701fa9e4066Sahrens /*
702fa9e4066Sahrens  * Consider the following situation: txg is safely synced to disk.  We've
703fa9e4066Sahrens  * written the first uberblock for txg + 1, and then we lose power.  When we
704fa9e4066Sahrens  * come back up, we fail to see the uberblock for txg + 1 because, say,
705fa9e4066Sahrens  * it was on a mirrored device and the replica to which we wrote txg + 1
706fa9e4066Sahrens  * is now offline.  If we then make some changes and sync txg + 1, and then
707fa9e4066Sahrens  * the missing replica comes back, then for a new seconds we'll have two
708fa9e4066Sahrens  * conflicting uberblocks on disk with the same txg.  The solution is simple:
709fa9e4066Sahrens  * among uberblocks with equal txg, choose the one with the latest timestamp.
710fa9e4066Sahrens  */
711fa9e4066Sahrens static int
712fa9e4066Sahrens vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
713fa9e4066Sahrens {
714fa9e4066Sahrens 	if (ub1->ub_txg < ub2->ub_txg)
715fa9e4066Sahrens 		return (-1);
716fa9e4066Sahrens 	if (ub1->ub_txg > ub2->ub_txg)
717fa9e4066Sahrens 		return (1);
718fa9e4066Sahrens 
719fa9e4066Sahrens 	if (ub1->ub_timestamp < ub2->ub_timestamp)
720fa9e4066Sahrens 		return (-1);
721fa9e4066Sahrens 	if (ub1->ub_timestamp > ub2->ub_timestamp)
722fa9e4066Sahrens 		return (1);
723fa9e4066Sahrens 
724fa9e4066Sahrens 	return (0);
725fa9e4066Sahrens }
726fa9e4066Sahrens 
727fa9e4066Sahrens static void
728fa9e4066Sahrens vdev_uberblock_load_done(zio_t *zio)
729fa9e4066Sahrens {
730*e14bb325SJeff Bonwick 	zio_t *rio = zio->io_private;
731ecc2d604Sbonwick 	uberblock_t *ub = zio->io_data;
732*e14bb325SJeff Bonwick 	uberblock_t *ubbest = rio->io_private;
733fa9e4066Sahrens 
734ecc2d604Sbonwick 	ASSERT3U(zio->io_size, ==, VDEV_UBERBLOCK_SIZE(zio->io_vd));
735fa9e4066Sahrens 
736ea8dc4b6Seschrock 	if (zio->io_error == 0 && uberblock_verify(ub) == 0) {
737*e14bb325SJeff Bonwick 		mutex_enter(&rio->io_lock);
738fa9e4066Sahrens 		if (vdev_uberblock_compare(ub, ubbest) > 0)
739fa9e4066Sahrens 			*ubbest = *ub;
740*e14bb325SJeff Bonwick 		mutex_exit(&rio->io_lock);
741fa9e4066Sahrens 	}
742fa9e4066Sahrens 
743fa9e4066Sahrens 	zio_buf_free(zio->io_data, zio->io_size);
744fa9e4066Sahrens }
745fa9e4066Sahrens 
746fa9e4066Sahrens void
747fa9e4066Sahrens vdev_uberblock_load(zio_t *zio, vdev_t *vd, uberblock_t *ubbest)
748fa9e4066Sahrens {
749*e14bb325SJeff Bonwick 	spa_t *spa = vd->vdev_spa;
750*e14bb325SJeff Bonwick 	vdev_t *rvd = spa->spa_root_vdev;
751*e14bb325SJeff Bonwick 	int flags =
752*e14bb325SJeff Bonwick 	    ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE;
753*e14bb325SJeff Bonwick 
754*e14bb325SJeff Bonwick 	if (vd == rvd) {
755*e14bb325SJeff Bonwick 		ASSERT(zio == NULL);
756*e14bb325SJeff Bonwick 		spa_config_enter(spa, SCL_ALL, FTAG, RW_WRITER);
757*e14bb325SJeff Bonwick 		zio = zio_root(spa, NULL, ubbest, flags);
758*e14bb325SJeff Bonwick 		bzero(ubbest, sizeof (uberblock_t));
759*e14bb325SJeff Bonwick 	}
760fa9e4066Sahrens 
761*e14bb325SJeff Bonwick 	ASSERT(zio != NULL);
762fa9e4066Sahrens 
763*e14bb325SJeff Bonwick 	for (int c = 0; c < vd->vdev_children; c++)
764*e14bb325SJeff Bonwick 		vdev_uberblock_load(zio, vd->vdev_child[c], ubbest);
765fa9e4066Sahrens 
766*e14bb325SJeff Bonwick 	if (vd->vdev_ops->vdev_op_leaf && vdev_readable(vd)) {
767*e14bb325SJeff Bonwick 		for (int l = 0; l < VDEV_LABELS; l++) {
768*e14bb325SJeff Bonwick 			for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
769*e14bb325SJeff Bonwick 				vdev_label_read(zio, vd, l,
770*e14bb325SJeff Bonwick 				    zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd)),
771*e14bb325SJeff Bonwick 				    VDEV_UBERBLOCK_OFFSET(vd, n),
772*e14bb325SJeff Bonwick 				    VDEV_UBERBLOCK_SIZE(vd),
773*e14bb325SJeff Bonwick 				    vdev_uberblock_load_done, zio, flags);
774*e14bb325SJeff Bonwick 			}
775fa9e4066Sahrens 		}
776fa9e4066Sahrens 	}
777*e14bb325SJeff Bonwick 
778*e14bb325SJeff Bonwick 	if (vd == rvd) {
779*e14bb325SJeff Bonwick 		(void) zio_wait(zio);
780*e14bb325SJeff Bonwick 		spa_config_exit(spa, SCL_ALL, FTAG);
781*e14bb325SJeff Bonwick 	}
782fa9e4066Sahrens }
783fa9e4066Sahrens 
784fa9e4066Sahrens /*
78517f17c2dSbonwick  * On success, increment root zio's count of good writes.
7860373e76bSbonwick  * We only get credit for writes to known-visible vdevs; see spa_vdev_add().
787fa9e4066Sahrens  */
788fa9e4066Sahrens static void
789fa9e4066Sahrens vdev_uberblock_sync_done(zio_t *zio)
790fa9e4066Sahrens {
79117f17c2dSbonwick 	uint64_t *good_writes = zio->io_private;
792fa9e4066Sahrens 
7930373e76bSbonwick 	if (zio->io_error == 0 && zio->io_vd->vdev_top->vdev_ms_array != 0)
794fa9e4066Sahrens 		atomic_add_64(good_writes, 1);
795fa9e4066Sahrens }
796fa9e4066Sahrens 
79717f17c2dSbonwick /*
79817f17c2dSbonwick  * Write the uberblock to all labels of all leaves of the specified vdev.
79917f17c2dSbonwick  */
800fa9e4066Sahrens static void
801*e14bb325SJeff Bonwick vdev_uberblock_sync(zio_t *zio, uberblock_t *ub, vdev_t *vd, int flags)
802fa9e4066Sahrens {
80317f17c2dSbonwick 	uberblock_t *ubbuf;
804*e14bb325SJeff Bonwick 	int n;
805fa9e4066Sahrens 
806*e14bb325SJeff Bonwick 	for (int c = 0; c < vd->vdev_children; c++)
807*e14bb325SJeff Bonwick 		vdev_uberblock_sync(zio, ub, vd->vdev_child[c], flags);
808fa9e4066Sahrens 
809fa9e4066Sahrens 	if (!vd->vdev_ops->vdev_op_leaf)
810fa9e4066Sahrens 		return;
811fa9e4066Sahrens 
812*e14bb325SJeff Bonwick 	if (!vdev_writeable(vd))
813fa9e4066Sahrens 		return;
814fa9e4066Sahrens 
81517f17c2dSbonwick 	n = ub->ub_txg & (VDEV_UBERBLOCK_COUNT(vd) - 1);
816fa9e4066Sahrens 
81717f17c2dSbonwick 	ubbuf = zio_buf_alloc(VDEV_UBERBLOCK_SIZE(vd));
81817f17c2dSbonwick 	bzero(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
81917f17c2dSbonwick 	*ubbuf = *ub;
820fa9e4066Sahrens 
821*e14bb325SJeff Bonwick 	for (int l = 0; l < VDEV_LABELS; l++)
82217f17c2dSbonwick 		vdev_label_write(zio, vd, l, ubbuf,
823*e14bb325SJeff Bonwick 		    VDEV_UBERBLOCK_OFFSET(vd, n), VDEV_UBERBLOCK_SIZE(vd),
82417f17c2dSbonwick 		    vdev_uberblock_sync_done, zio->io_private,
825*e14bb325SJeff Bonwick 		    flags | ZIO_FLAG_DONT_PROPAGATE);
826fa9e4066Sahrens 
82717f17c2dSbonwick 	zio_buf_free(ubbuf, VDEV_UBERBLOCK_SIZE(vd));
828fa9e4066Sahrens }
829fa9e4066Sahrens 
83017f17c2dSbonwick int
83117f17c2dSbonwick vdev_uberblock_sync_list(vdev_t **svd, int svdcount, uberblock_t *ub, int flags)
832fa9e4066Sahrens {
83317f17c2dSbonwick 	spa_t *spa = svd[0]->vdev_spa;
834*e14bb325SJeff Bonwick 	zio_t *zio;
83517f17c2dSbonwick 	uint64_t good_writes = 0;
836fa9e4066Sahrens 
837*e14bb325SJeff Bonwick 	zio = zio_root(spa, NULL, &good_writes, flags);
838*e14bb325SJeff Bonwick 
839*e14bb325SJeff Bonwick 	for (int v = 0; v < svdcount; v++)
840*e14bb325SJeff Bonwick 		vdev_uberblock_sync(zio, ub, svd[v], flags);
841fa9e4066Sahrens 
84217f17c2dSbonwick 	(void) zio_wait(zio);
843fa9e4066Sahrens 
844fa9e4066Sahrens 	/*
84517f17c2dSbonwick 	 * Flush the uberblocks to disk.  This ensures that the odd labels
84617f17c2dSbonwick 	 * are no longer needed (because the new uberblocks and the even
84717f17c2dSbonwick 	 * labels are safely on disk), so it is safe to overwrite them.
848fa9e4066Sahrens 	 */
84917f17c2dSbonwick 	zio = zio_root(spa, NULL, NULL, flags);
850fa9e4066Sahrens 
851*e14bb325SJeff Bonwick 	for (int v = 0; v < svdcount; v++)
85217f17c2dSbonwick 		zio_flush(zio, svd[v]);
853fa9e4066Sahrens 
85417f17c2dSbonwick 	(void) zio_wait(zio);
85517f17c2dSbonwick 
85617f17c2dSbonwick 	return (good_writes >= 1 ? 0 : EIO);
857fa9e4066Sahrens }
858fa9e4066Sahrens 
859fa9e4066Sahrens /*
86017f17c2dSbonwick  * On success, increment the count of good writes for our top-level vdev.
861fa9e4066Sahrens  */
862fa9e4066Sahrens static void
86317f17c2dSbonwick vdev_label_sync_done(zio_t *zio)
864fa9e4066Sahrens {
86517f17c2dSbonwick 	uint64_t *good_writes = zio->io_private;
866fa9e4066Sahrens 
867fa9e4066Sahrens 	if (zio->io_error == 0)
868fa9e4066Sahrens 		atomic_add_64(good_writes, 1);
869fa9e4066Sahrens }
870fa9e4066Sahrens 
87117f17c2dSbonwick /*
87217f17c2dSbonwick  * If there weren't enough good writes, indicate failure to the parent.
87317f17c2dSbonwick  */
874fa9e4066Sahrens static void
87517f17c2dSbonwick vdev_label_sync_top_done(zio_t *zio)
87617f17c2dSbonwick {
87717f17c2dSbonwick 	uint64_t *good_writes = zio->io_private;
87817f17c2dSbonwick 
87917f17c2dSbonwick 	if (*good_writes == 0)
88017f17c2dSbonwick 		zio->io_error = EIO;
88117f17c2dSbonwick 
88217f17c2dSbonwick 	kmem_free(good_writes, sizeof (uint64_t));
88317f17c2dSbonwick }
88417f17c2dSbonwick 
88551ece835Seschrock /*
8860430f8daSeschrock  * We ignore errors for log and cache devices, simply free the private data.
88751ece835Seschrock  */
88851ece835Seschrock static void
8890430f8daSeschrock vdev_label_sync_ignore_done(zio_t *zio)
89051ece835Seschrock {
89151ece835Seschrock 	kmem_free(zio->io_private, sizeof (uint64_t));
89251ece835Seschrock }
89351ece835Seschrock 
89417f17c2dSbonwick /*
89517f17c2dSbonwick  * Write all even or odd labels to all leaves of the specified vdev.
89617f17c2dSbonwick  */
89717f17c2dSbonwick static void
898*e14bb325SJeff Bonwick vdev_label_sync(zio_t *zio, vdev_t *vd, int l, uint64_t txg, int flags)
899fa9e4066Sahrens {
900fa9e4066Sahrens 	nvlist_t *label;
901fa9e4066Sahrens 	vdev_phys_t *vp;
902fa9e4066Sahrens 	char *buf;
903fa9e4066Sahrens 	size_t buflen;
904fa9e4066Sahrens 
905*e14bb325SJeff Bonwick 	for (int c = 0; c < vd->vdev_children; c++)
906*e14bb325SJeff Bonwick 		vdev_label_sync(zio, vd->vdev_child[c], l, txg, flags);
907fa9e4066Sahrens 
908fa9e4066Sahrens 	if (!vd->vdev_ops->vdev_op_leaf)
909fa9e4066Sahrens 		return;
910fa9e4066Sahrens 
911*e14bb325SJeff Bonwick 	if (!vdev_writeable(vd))
912fa9e4066Sahrens 		return;
913fa9e4066Sahrens 
914fa9e4066Sahrens 	/*
915fa9e4066Sahrens 	 * Generate a label describing the top-level config to which we belong.
916fa9e4066Sahrens 	 */
9170373e76bSbonwick 	label = spa_config_generate(vd->vdev_spa, vd, txg, B_FALSE);
918fa9e4066Sahrens 
919fa9e4066Sahrens 	vp = zio_buf_alloc(sizeof (vdev_phys_t));
920fa9e4066Sahrens 	bzero(vp, sizeof (vdev_phys_t));
921fa9e4066Sahrens 
922fa9e4066Sahrens 	buf = vp->vp_nvlist;
923fa9e4066Sahrens 	buflen = sizeof (vp->vp_nvlist);
924fa9e4066Sahrens 
92517f17c2dSbonwick 	if (nvlist_pack(label, &buf, &buflen, NV_ENCODE_XDR, KM_SLEEP) == 0) {
92617f17c2dSbonwick 		for (; l < VDEV_LABELS; l += 2) {
92717f17c2dSbonwick 			vdev_label_write(zio, vd, l, vp,
92817f17c2dSbonwick 			    offsetof(vdev_label_t, vl_vdev_phys),
92917f17c2dSbonwick 			    sizeof (vdev_phys_t),
93017f17c2dSbonwick 			    vdev_label_sync_done, zio->io_private,
931*e14bb325SJeff Bonwick 			    flags | ZIO_FLAG_DONT_PROPAGATE);
93217f17c2dSbonwick 		}
93317f17c2dSbonwick 	}
934fa9e4066Sahrens 
935fa9e4066Sahrens 	zio_buf_free(vp, sizeof (vdev_phys_t));
936fa9e4066Sahrens 	nvlist_free(label);
937fa9e4066Sahrens }
938fa9e4066Sahrens 
93917f17c2dSbonwick int
940*e14bb325SJeff Bonwick vdev_label_sync_list(spa_t *spa, int l, uint64_t txg, int flags)
941fa9e4066Sahrens {
942*e14bb325SJeff Bonwick 	list_t *dl = &spa->spa_config_dirty_list;
94317f17c2dSbonwick 	vdev_t *vd;
944*e14bb325SJeff Bonwick 	zio_t *zio;
945fa9e4066Sahrens 	int error;
946fa9e4066Sahrens 
947fa9e4066Sahrens 	/*
948*e14bb325SJeff Bonwick 	 * Write the new labels to disk.
949fa9e4066Sahrens 	 */
950*e14bb325SJeff Bonwick 	zio = zio_root(spa, NULL, NULL, flags);
951fa9e4066Sahrens 
95217f17c2dSbonwick 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd)) {
95317f17c2dSbonwick 		uint64_t *good_writes = kmem_zalloc(sizeof (uint64_t),
95417f17c2dSbonwick 		    KM_SLEEP);
955*e14bb325SJeff Bonwick 		zio_t *vio = zio_null(zio, spa,
9560430f8daSeschrock 		    (vd->vdev_islog || vd->vdev_aux != NULL) ?
9570430f8daSeschrock 		    vdev_label_sync_ignore_done : vdev_label_sync_top_done,
95817f17c2dSbonwick 		    good_writes, flags);
959*e14bb325SJeff Bonwick 		vdev_label_sync(vio, vd, l, txg, flags);
96017f17c2dSbonwick 		zio_nowait(vio);
961fa9e4066Sahrens 	}
962*e14bb325SJeff Bonwick 
963*e14bb325SJeff Bonwick 	error = zio_wait(zio);
964fa9e4066Sahrens 
9658654d025Sperrin 	/*
96617f17c2dSbonwick 	 * Flush the new labels to disk.
9678654d025Sperrin 	 */
96817f17c2dSbonwick 	zio = zio_root(spa, NULL, NULL, flags);
9698654d025Sperrin 
97017f17c2dSbonwick 	for (vd = list_head(dl); vd != NULL; vd = list_next(dl, vd))
97117f17c2dSbonwick 		zio_flush(zio, vd);
97217f17c2dSbonwick 
97317f17c2dSbonwick 	(void) zio_wait(zio);
974fa9e4066Sahrens 
975fa9e4066Sahrens 	return (error);
976fa9e4066Sahrens }
977fa9e4066Sahrens 
978fa9e4066Sahrens /*
97917f17c2dSbonwick  * Sync the uberblock and any changes to the vdev configuration.
980fa9e4066Sahrens  *
981fa9e4066Sahrens  * The order of operations is carefully crafted to ensure that
982fa9e4066Sahrens  * if the system panics or loses power at any time, the state on disk
983fa9e4066Sahrens  * is still transactionally consistent.  The in-line comments below
984fa9e4066Sahrens  * describe the failure semantics at each stage.
985fa9e4066Sahrens  *
98617f17c2dSbonwick  * Moreover, vdev_config_sync() is designed to be idempotent: if it fails
987fa9e4066Sahrens  * at any time, you can just call it again, and it will resume its work.
988fa9e4066Sahrens  */
989*e14bb325SJeff Bonwick int
99017f17c2dSbonwick vdev_config_sync(vdev_t **svd, int svdcount, uint64_t txg)
991fa9e4066Sahrens {
99217f17c2dSbonwick 	spa_t *spa = svd[0]->vdev_spa;
993fa9e4066Sahrens 	uberblock_t *ub = &spa->spa_uberblock;
9940373e76bSbonwick 	vdev_t *vd;
995fa9e4066Sahrens 	zio_t *zio;
996*e14bb325SJeff Bonwick 	int error;
997*e14bb325SJeff Bonwick 	int flags = ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_CANFAIL;
998fa9e4066Sahrens 
999fa9e4066Sahrens 	ASSERT(ub->ub_txg <= txg);
1000fa9e4066Sahrens 
1001fa9e4066Sahrens 	/*
100217f17c2dSbonwick 	 * If this isn't a resync due to I/O errors,
100317f17c2dSbonwick 	 * and nothing changed in this transaction group,
100417f17c2dSbonwick 	 * and the vdev configuration hasn't changed,
10050373e76bSbonwick 	 * then there's nothing to do.
1006fa9e4066Sahrens 	 */
100717f17c2dSbonwick 	if (ub->ub_txg < txg &&
100817f17c2dSbonwick 	    uberblock_update(ub, spa->spa_root_vdev, txg) == B_FALSE &&
1009*e14bb325SJeff Bonwick 	    list_is_empty(&spa->spa_config_dirty_list))
1010*e14bb325SJeff Bonwick 		return (0);
1011fa9e4066Sahrens 
1012fa9e4066Sahrens 	if (txg > spa_freeze_txg(spa))
1013*e14bb325SJeff Bonwick 		return (0);
1014fa9e4066Sahrens 
10150373e76bSbonwick 	ASSERT(txg <= spa->spa_final_txg);
10160373e76bSbonwick 
1017fa9e4066Sahrens 	/*
1018fa9e4066Sahrens 	 * Flush the write cache of every disk that's been written to
1019fa9e4066Sahrens 	 * in this transaction group.  This ensures that all blocks
1020fa9e4066Sahrens 	 * written in this txg will be committed to stable storage
1021fa9e4066Sahrens 	 * before any uberblock that references them.
1022fa9e4066Sahrens 	 */
102317f17c2dSbonwick 	zio = zio_root(spa, NULL, NULL, flags);
102417f17c2dSbonwick 
1025fa9e4066Sahrens 	for (vd = txg_list_head(&spa->spa_vdev_txg_list, TXG_CLEAN(txg)); vd;
102617f17c2dSbonwick 	    vd = txg_list_next(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)))
102717f17c2dSbonwick 		zio_flush(zio, vd);
102817f17c2dSbonwick 
1029fa9e4066Sahrens 	(void) zio_wait(zio);
1030fa9e4066Sahrens 
1031fa9e4066Sahrens 	/*
1032fa9e4066Sahrens 	 * Sync out the even labels (L0, L2) for every dirty vdev.  If the
1033fa9e4066Sahrens 	 * system dies in the middle of this process, that's OK: all of the
1034fa9e4066Sahrens 	 * even labels that made it to disk will be newer than any uberblock,
1035fa9e4066Sahrens 	 * and will therefore be considered invalid.  The odd labels (L1, L3),
103617f17c2dSbonwick 	 * which have not yet been touched, will still be valid.  We flush
103717f17c2dSbonwick 	 * the new labels to disk to ensure that all even-label updates
103817f17c2dSbonwick 	 * are committed to stable storage before the uberblock update.
1039fa9e4066Sahrens 	 */
1040*e14bb325SJeff Bonwick 	if ((error = vdev_label_sync_list(spa, 0, txg, flags)) != 0)
1041*e14bb325SJeff Bonwick 		return (error);
1042fa9e4066Sahrens 
1043fa9e4066Sahrens 	/*
1044*e14bb325SJeff Bonwick 	 * Sync the uberblocks to all vdevs in svd[].
10450373e76bSbonwick 	 * If the system dies in the middle of this step, there are two cases
10460373e76bSbonwick 	 * to consider, and the on-disk state is consistent either way:
1047fa9e4066Sahrens 	 *
1048fa9e4066Sahrens 	 * (1)	If none of the new uberblocks made it to disk, then the
1049fa9e4066Sahrens 	 *	previous uberblock will be the newest, and the odd labels
1050fa9e4066Sahrens 	 *	(which had not yet been touched) will be valid with respect
1051fa9e4066Sahrens 	 *	to that uberblock.
1052fa9e4066Sahrens 	 *
1053fa9e4066Sahrens 	 * (2)	If one or more new uberblocks made it to disk, then they
1054fa9e4066Sahrens 	 *	will be the newest, and the even labels (which had all
1055fa9e4066Sahrens 	 *	been successfully committed) will be valid with respect
1056fa9e4066Sahrens 	 *	to the new uberblocks.
1057fa9e4066Sahrens 	 */
1058*e14bb325SJeff Bonwick 	if ((error = vdev_uberblock_sync_list(svd, svdcount, ub, flags)) != 0)
1059*e14bb325SJeff Bonwick 		return (error);
1060fa9e4066Sahrens 
1061fa9e4066Sahrens 	/*
1062fa9e4066Sahrens 	 * Sync out odd labels for every dirty vdev.  If the system dies
1063fa9e4066Sahrens 	 * in the middle of this process, the even labels and the new
1064fa9e4066Sahrens 	 * uberblocks will suffice to open the pool.  The next time
1065fa9e4066Sahrens 	 * the pool is opened, the first thing we'll do -- before any
1066fa9e4066Sahrens 	 * user data is modified -- is mark every vdev dirty so that
106717f17c2dSbonwick 	 * all labels will be brought up to date.  We flush the new labels
106817f17c2dSbonwick 	 * to disk to ensure that all odd-label updates are committed to
106917f17c2dSbonwick 	 * stable storage before the next transaction group begins.
1070fa9e4066Sahrens 	 */
1071*e14bb325SJeff Bonwick 	return (vdev_label_sync_list(spa, 1, txg, flags));
1072fa9e4066Sahrens }
1073