xref: /freebsd/sys/contrib/openzfs/include/sys/uberblock_impl.h (revision b985c9cafd2aedac5cf92428c0211485ea4ede24)
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 https://opensource.org/licenses/CDDL-1.0.
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 /*
79  * RAIDZ expansion reflow information.
80  *
81  *	64      56      48      40      32      24      16      8       0
82  *	+-------+-------+-------+-------+-------+-------+-------+-------+
83  *	|Scratch |                    Reflow                            |
84  *	| State  |                    Offset                            |
85  *	+-------+-------+-------+-------+-------+-------+-------+-------+
86  */
87 typedef enum raidz_reflow_scratch_state {
88 	RRSS_SCRATCH_NOT_IN_USE = 0,
89 	RRSS_SCRATCH_VALID,
90 	RRSS_SCRATCH_INVALID_SYNCED,
91 	RRSS_SCRATCH_INVALID_SYNCED_ON_IMPORT,
92 	RRSS_SCRATCH_INVALID_SYNCED_REFLOW
93 } raidz_reflow_scratch_state_t;
94 
95 #define	RRSS_GET_OFFSET(ub) \
96 	BF64_GET_SB((ub)->ub_raidz_reflow_info, 0, 55, SPA_MINBLOCKSHIFT, 0)
97 #define	RRSS_SET_OFFSET(ub, x) \
98 	BF64_SET_SB((ub)->ub_raidz_reflow_info, 0, 55, SPA_MINBLOCKSHIFT, 0, x)
99 
100 #define	RRSS_GET_STATE(ub) \
101 	BF64_GET((ub)->ub_raidz_reflow_info, 55, 9)
102 #define	RRSS_SET_STATE(ub, x) \
103 	BF64_SET((ub)->ub_raidz_reflow_info, 55, 9, x)
104 
105 #define	RAIDZ_REFLOW_SET(ub, state, offset) do { \
106 	(ub)->ub_raidz_reflow_info = 0; \
107 	RRSS_SET_OFFSET(ub, offset); \
108 	RRSS_SET_STATE(ub, state); \
109 } while (0)
110 
111 struct uberblock {
112 	uint64_t	ub_magic;	/* UBERBLOCK_MAGIC		*/
113 	uint64_t	ub_version;	/* SPA_VERSION			*/
114 	uint64_t	ub_txg;		/* txg of last sync		*/
115 	uint64_t	ub_guid_sum;	/* sum of all vdev guids	*/
116 	uint64_t	ub_timestamp;	/* UTC time of last sync	*/
117 	blkptr_t	ub_rootbp;	/* MOS objset_phys_t		*/
118 
119 	/* highest SPA_VERSION supported by software that wrote this txg */
120 	uint64_t	ub_software_version;
121 
122 	/* Maybe missing in uberblocks we read, but always written */
123 	uint64_t	ub_mmp_magic;	/* MMP_MAGIC			*/
124 	/*
125 	 * If ub_mmp_delay == 0 and ub_mmp_magic is valid, MMP is off.
126 	 * Otherwise, nanosec since last MMP write.
127 	 */
128 	uint64_t	ub_mmp_delay;
129 
130 	/*
131 	 * The ub_mmp_config contains the multihost write interval, multihost
132 	 * fail intervals, sequence number for sub-second granularity, and
133 	 * valid bit mask.  This layout is as follows:
134 	 *
135 	 *   64      56      48      40      32      24      16      8       0
136 	 *   +-------+-------+-------+-------+-------+-------+-------+-------+
137 	 * 0 | Fail Intervals|      Seq      |   Write Interval (ms) | VALID |
138 	 *   +-------+-------+-------+-------+-------+-------+-------+-------+
139 	 *
140 	 * This allows a write_interval of (2^24/1000)s, over 4.5 hours
141 	 *
142 	 * VALID Bits:
143 	 * - 0x01 - Write Interval (ms)
144 	 * - 0x02 - Sequence number exists
145 	 * - 0x04 - Fail Intervals
146 	 * - 0xf8 - Reserved
147 	 */
148 	uint64_t	ub_mmp_config;
149 
150 	/*
151 	 * ub_checkpoint_txg indicates two things about the current uberblock:
152 	 *
153 	 * 1] If it is not zero then this uberblock is a checkpoint. If it is
154 	 *    zero, then this uberblock is not a checkpoint.
155 	 *
156 	 * 2] On checkpointed uberblocks, the value of ub_checkpoint_txg is
157 	 *    the ub_txg that the uberblock had at the time we moved it to
158 	 *    the MOS config.
159 	 *
160 	 * The field is set when we checkpoint the uberblock and continues to
161 	 * hold that value even after we've rewound (unlike the ub_txg that
162 	 * is reset to a higher value).
163 	 *
164 	 * Besides checks used to determine whether we are reopening the
165 	 * pool from a checkpointed uberblock [see spa_ld_select_uberblock()],
166 	 * the value of the field is used to determine which ZIL blocks have
167 	 * been allocated according to the ms_sm when we are rewinding to a
168 	 * checkpoint. Specifically, if logical birth > ub_checkpoint_txg,then
169 	 * the ZIL block is not allocated [see uses of spa_min_claim_txg()].
170 	 */
171 	uint64_t	ub_checkpoint_txg;
172 
173 	uint64_t	ub_raidz_reflow_info;
174 };
175 
176 #ifdef	__cplusplus
177 }
178 #endif
179 
180 #endif	/* _SYS_UBERBLOCK_IMPL_H */
181