1 // SPDX-License-Identifier: CDDL-1.0 2 /* 3 * This file and its contents are supplied under the terms of the 4 * Common Development and Distribution License ("CDDL"), version 1.0. 5 * You may only use this file in accordance with the terms of version 6 * 1.0 of the CDDL. 7 * 8 * A full copy of the text of the CDDL should have accompanied this 9 * source. A copy of the CDDL is also available via the Internet at 10 * https://opensource.org/license/CDDL-1.0. 11 */ 12 /* 13 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 14 * Copyright (c) 2016, 2017 by Delphix. All rights reserved. 15 */ 16 17 #ifndef _SYS_UBERBLOCK_IMPL_H 18 #define _SYS_UBERBLOCK_IMPL_H 19 20 #include <sys/uberblock.h> 21 22 #ifdef __cplusplus 23 extern "C" { 24 #endif 25 26 /* 27 * The uberblock version is incremented whenever an incompatible on-disk 28 * format change is made to the SPA, DMU, or ZAP. 29 * 30 * Note: the first two fields should never be moved. When a storage pool 31 * is opened, the uberblock must be read off the disk before the version 32 * can be checked. If the ub_version field is moved, we may not detect 33 * version mismatch. If the ub_magic field is moved, applications that 34 * expect the magic number in the first word won't work. 35 */ 36 #define UBERBLOCK_MAGIC 0x00bab10c /* oo-ba-bloc! */ 37 #define UBERBLOCK_SHIFT 10 /* up to 1K */ 38 #define MMP_MAGIC 0xa11cea11 /* all-see-all */ 39 40 #define MMP_INTERVAL_VALID_BIT 0x01 41 #define MMP_SEQ_VALID_BIT 0x02 42 #define MMP_FAIL_INT_VALID_BIT 0x04 43 44 #define MMP_INTERVAL_MASK 0x00000000FFFFFF00 45 #define MMP_SEQ_MASK 0x0000FFFF00000000 46 #define MMP_FAIL_INT_MASK 0xFFFF000000000000 47 48 #define MMP_SEQ_MAX UINT16_MAX 49 50 #define MMP_VALID(ubp) ((ubp)->ub_magic == UBERBLOCK_MAGIC && \ 51 (ubp)->ub_mmp_magic == MMP_MAGIC) 52 #define MMP_INTERVAL_VALID(ubp) (MMP_VALID(ubp) && ((ubp)->ub_mmp_config & \ 53 MMP_INTERVAL_VALID_BIT)) 54 #define MMP_SEQ_VALID(ubp) (MMP_VALID(ubp) && ((ubp)->ub_mmp_config & \ 55 MMP_SEQ_VALID_BIT)) 56 #define MMP_FAIL_INT_VALID(ubp) (MMP_VALID(ubp) && ((ubp)->ub_mmp_config & \ 57 MMP_FAIL_INT_VALID_BIT)) 58 59 #define MMP_INTERVAL(ubp) (((ubp)->ub_mmp_config & MMP_INTERVAL_MASK) \ 60 >> 8) 61 #define MMP_SEQ(ubp) (((ubp)->ub_mmp_config & MMP_SEQ_MASK) \ 62 >> 32) 63 #define MMP_FAIL_INT(ubp) (((ubp)->ub_mmp_config & MMP_FAIL_INT_MASK) \ 64 >> 48) 65 66 #define MMP_INTERVAL_SET(write) \ 67 (((uint64_t)((write) & 0xFFFFFF) << 8) | MMP_INTERVAL_VALID_BIT) 68 69 #define MMP_SEQ_SET(seq) \ 70 (((uint64_t)((seq) & 0xFFFF) << 32) | MMP_SEQ_VALID_BIT) 71 72 #define MMP_FAIL_INT_SET(fail) \ 73 (((uint64_t)((fail) & 0xFFFF) << 48) | MMP_FAIL_INT_VALID_BIT) 74 75 76 #define MMP_SEQ_CLEAR(ubp) \ 77 ((ubp)->ub_mmp_config &= ~(MMP_SEQ_MASK | MMP_SEQ_VALID_BIT)) 78 79 /* 80 * RAIDZ expansion reflow information. 81 * 82 * 64 56 48 40 32 24 16 8 0 83 * +-------+-------+-------+-------+-------+-------+-------+-------+ 84 * |Scratch | Reflow | 85 * | State | Offset | 86 * +-------+-------+-------+-------+-------+-------+-------+-------+ 87 */ 88 typedef enum raidz_reflow_scratch_state { 89 RRSS_SCRATCH_NOT_IN_USE = 0, 90 RRSS_SCRATCH_VALID, 91 RRSS_SCRATCH_INVALID_SYNCED, 92 RRSS_SCRATCH_INVALID_SYNCED_ON_IMPORT, 93 RRSS_SCRATCH_INVALID_SYNCED_REFLOW 94 } raidz_reflow_scratch_state_t; 95 96 #define RRSS_GET_OFFSET(ub) \ 97 BF64_GET_SB((ub)->ub_raidz_reflow_info, 0, 55, SPA_MINBLOCKSHIFT, 0) 98 #define RRSS_SET_OFFSET(ub, x) \ 99 BF64_SET_SB((ub)->ub_raidz_reflow_info, 0, 55, SPA_MINBLOCKSHIFT, 0, x) 100 101 #define RRSS_GET_STATE(ub) \ 102 BF64_GET((ub)->ub_raidz_reflow_info, 55, 9) 103 #define RRSS_SET_STATE(ub, x) \ 104 BF64_SET((ub)->ub_raidz_reflow_info, 55, 9, x) 105 106 #define RAIDZ_REFLOW_SET(ub, state, offset) do { \ 107 (ub)->ub_raidz_reflow_info = 0; \ 108 RRSS_SET_OFFSET(ub, offset); \ 109 RRSS_SET_STATE(ub, state); \ 110 } while (0) 111 112 struct uberblock { 113 uint64_t ub_magic; /* UBERBLOCK_MAGIC */ 114 uint64_t ub_version; /* SPA_VERSION */ 115 uint64_t ub_txg; /* txg of last sync */ 116 uint64_t ub_guid_sum; /* sum of all vdev guids */ 117 uint64_t ub_timestamp; /* UTC time of last sync */ 118 blkptr_t ub_rootbp; /* MOS objset_phys_t */ 119 120 /* highest SPA_VERSION supported by software that wrote this txg */ 121 uint64_t ub_software_version; 122 123 /* Maybe missing in uberblocks we read, but always written */ 124 uint64_t ub_mmp_magic; /* MMP_MAGIC */ 125 /* 126 * If ub_mmp_delay == 0 and ub_mmp_magic is valid, MMP is off. 127 * Otherwise, nanosec since last MMP write. 128 */ 129 uint64_t ub_mmp_delay; 130 131 /* 132 * The ub_mmp_config contains the multihost write interval, multihost 133 * fail intervals, sequence number for sub-second granularity, and 134 * valid bit mask. This layout is as follows: 135 * 136 * 64 56 48 40 32 24 16 8 0 137 * +-------+-------+-------+-------+-------+-------+-------+-------+ 138 * 0 | Fail Intervals| Seq | Write Interval (ms) | VALID | 139 * +-------+-------+-------+-------+-------+-------+-------+-------+ 140 * 141 * This allows a write_interval of (2^24/1000)s, over 4.5 hours 142 * 143 * VALID Bits: 144 * - 0x01 - Write Interval (ms) 145 * - 0x02 - Sequence number exists 146 * - 0x04 - Fail Intervals 147 * - 0xf8 - Reserved 148 */ 149 uint64_t ub_mmp_config; 150 151 /* 152 * ub_checkpoint_txg indicates two things about the current uberblock: 153 * 154 * 1] If it is not zero then this uberblock is a checkpoint. If it is 155 * zero, then this uberblock is not a checkpoint. 156 * 157 * 2] On checkpointed uberblocks, the value of ub_checkpoint_txg is 158 * the ub_txg that the uberblock had at the time we moved it to 159 * the MOS config. 160 * 161 * The field is set when we checkpoint the uberblock and continues to 162 * hold that value even after we've rewound (unlike the ub_txg that 163 * is reset to a higher value). 164 * 165 * Besides checks used to determine whether we are reopening the 166 * pool from a checkpointed uberblock [see spa_ld_select_uberblock()], 167 * the value of the field is used to determine which ZIL blocks have 168 * been allocated according to the ms_sm when we are rewinding to a 169 * checkpoint. Specifically, if logical birth > ub_checkpoint_txg,then 170 * the ZIL block is not allocated [see uses of spa_min_claim_txg()]. 171 */ 172 uint64_t ub_checkpoint_txg; 173 174 uint64_t ub_raidz_reflow_info; 175 }; 176 177 #ifdef __cplusplus 178 } 179 #endif 180 181 #endif /* _SYS_UBERBLOCK_IMPL_H */ 182