xref: /illumos-gate/usr/src/uts/common/fs/zfs/sys/uberblock_impl.h (revision 4348eb901228d2f8fa50bb132a34248e8662074e)
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  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2016, 2017 by Delphix. All rights reserved.
24  */
25 
26 #ifndef _SYS_UBERBLOCK_IMPL_H
27 #define	_SYS_UBERBLOCK_IMPL_H
28 
29 #include <sys/uberblock.h>
30 
31 #ifdef	__cplusplus
32 extern "C" {
33 #endif
34 
35 /*
36  * The uberblock version is incremented whenever an incompatible on-disk
37  * format change is made to the SPA, DMU, or ZAP.
38  *
39  * Note: the first two fields should never be moved.  When a storage pool
40  * is opened, the uberblock must be read off the disk before the version
41  * can be checked.  If the ub_version field is moved, we may not detect
42  * version mismatch.  If the ub_magic field is moved, applications that
43  * expect the magic number in the first word won't work.
44  */
45 #define	UBERBLOCK_MAGIC		0x00bab10c		/* oo-ba-bloc!	*/
46 #define	UBERBLOCK_SHIFT		10			/* up to 1K	*/
47 #define	MMP_MAGIC		0xa11cea11		/* all-see-all	*/
48 
49 #define	MMP_INTERVAL_VALID_BIT	0x01
50 #define	MMP_SEQ_VALID_BIT	0x02
51 #define	MMP_FAIL_INT_VALID_BIT	0x04
52 
53 #define	MMP_VALID(ubp)		(ubp->ub_magic == UBERBLOCK_MAGIC && \
54 				    ubp->ub_mmp_magic == MMP_MAGIC)
55 #define	MMP_INTERVAL_VALID(ubp)	(MMP_VALID(ubp) && (ubp->ub_mmp_config & \
56 				    MMP_INTERVAL_VALID_BIT))
57 #define	MMP_SEQ_VALID(ubp)	(MMP_VALID(ubp) && (ubp->ub_mmp_config & \
58 				    MMP_SEQ_VALID_BIT))
59 #define	MMP_FAIL_INT_VALID(ubp)	(MMP_VALID(ubp) && (ubp->ub_mmp_config & \
60 				    MMP_FAIL_INT_VALID_BIT))
61 
62 #define	MMP_INTERVAL(ubp)	((ubp->ub_mmp_config & 0x00000000FFFFFF00) \
63 				    >> 8)
64 #define	MMP_SEQ(ubp)		((ubp->ub_mmp_config & 0x0000FFFF00000000) \
65 				    >> 32)
66 #define	MMP_FAIL_INT(ubp)	((ubp->ub_mmp_config & 0xFFFF000000000000) \
67 				    >> 48)
68 
69 #define	MMP_INTERVAL_SET(write) \
70 	    (((uint64_t)(write & 0xFFFFFF) << 8) | MMP_INTERVAL_VALID_BIT)
71 
72 #define	MMP_SEQ_SET(seq) \
73 	    (((uint64_t)(seq & 0xFFFF) << 32) | MMP_SEQ_VALID_BIT)
74 
75 #define	MMP_FAIL_INT_SET(fail) \
76 	    (((uint64_t)(fail & 0xFFFF) << 48) | MMP_FAIL_INT_VALID_BIT)
77 
78 struct uberblock {
79 	uint64_t	ub_magic;	/* UBERBLOCK_MAGIC		*/
80 	uint64_t	ub_version;	/* SPA_VERSION			*/
81 	uint64_t	ub_txg;		/* txg of last sync		*/
82 	uint64_t	ub_guid_sum;	/* sum of all vdev guids	*/
83 	uint64_t	ub_timestamp;	/* UTC time of last sync	*/
84 	blkptr_t	ub_rootbp;	/* MOS objset_phys_t		*/
85 
86 	/* highest SPA_VERSION supported by software that wrote this txg */
87 	uint64_t	ub_software_version;
88 
89 	/* Maybe missing in uberblocks we read, but always written */
90 	uint64_t	ub_mmp_magic;
91 	/*
92 	 * If ub_mmp_delay == 0 and ub_mmp_magic is valid, MMP is off.
93 	 * Otherwise, nanosec since last MMP write.
94 	 */
95 	uint64_t	ub_mmp_delay;
96 
97 	/*
98 	 * The ub_mmp_config contains the multihost write interval, multihost
99 	 * fail intervals, sequence number for sub-second granularity, and
100 	 * valid bit mask.  This layout is as follows:
101 	 *
102 	 *   64      56      48      40      32      24      16      8       0
103 	 *   +-------+-------+-------+-------+-------+-------+-------+-------+
104 	 * 0 | Fail Intervals|      Seq      |   Write Interval (ms) | VALID |
105 	 *   +-------+-------+-------+-------+-------+-------+-------+-------+
106 	 *
107 	 * This allows a write_interval of (2^24/1000)s, over 4.5 hours
108 	 *
109 	 * VALID Bits:
110 	 * - 0x01 - Write Interval (ms)
111 	 * - 0x02 - Sequence number exists
112 	 * - 0x04 - Fail Intervals
113 	 * - 0xf8 - Reserved
114 	 */
115 	uint64_t	ub_mmp_config;
116 
117 	/*
118 	 * ub_checkpoint_txg indicates two things about the current uberblock:
119 	 *
120 	 * 1] If it is not zero then this uberblock is a checkpoint. If it is
121 	 *    zero, then this uberblock is not a checkpoint.
122 	 *
123 	 * 2] On checkpointed uberblocks, the value of ub_checkpoint_txg is
124 	 *    the ub_txg that the uberblock had at the time we moved it to
125 	 *    the MOS config.
126 	 *
127 	 * The field is set when we checkpoint the uberblock and continues to
128 	 * hold that value even after we've rewound (unlike the ub_txg that
129 	 * is reset to a higher value).
130 	 *
131 	 * Besides checks used to determine whether we are reopening the
132 	 * pool from a checkpointed uberblock [see spa_ld_select_uberblock()],
133 	 * the value of the field is used to determine which ZIL blocks have
134 	 * been allocated according to the ms_sm when we are rewinding to a
135 	 * checkpoint. Specifically, if blk_birth > ub_checkpoint_txg, then
136 	 * the ZIL block is not allocated [see uses of spa_min_claim_txg()].
137 	 */
138 	uint64_t	ub_checkpoint_txg;
139 };
140 
141 #ifdef	__cplusplus
142 }
143 #endif
144 
145 #endif	/* _SYS_UBERBLOCK_IMPL_H */
146