1 /* 2 * Block Translation Table library 3 * Copyright (c) 2014-2015, Intel Corporation. 4 * 5 * This program is free software; you can redistribute it and/or modify it 6 * under the terms and conditions of the GNU General Public License, 7 * version 2, as published by the Free Software Foundation. 8 * 9 * This program is distributed in the hope it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 12 * more details. 13 */ 14 15 #ifndef _LINUX_BTT_H 16 #define _LINUX_BTT_H 17 18 #include <linux/badblocks.h> 19 #include <linux/types.h> 20 21 #define BTT_SIG_LEN 16 22 #define BTT_SIG "BTT_ARENA_INFO\0" 23 #define MAP_ENT_SIZE 4 24 #define MAP_TRIM_SHIFT 31 25 #define MAP_TRIM_MASK (1 << MAP_TRIM_SHIFT) 26 #define MAP_ERR_SHIFT 30 27 #define MAP_ERR_MASK (1 << MAP_ERR_SHIFT) 28 #define MAP_LBA_MASK (~((1 << MAP_TRIM_SHIFT) | (1 << MAP_ERR_SHIFT))) 29 #define MAP_ENT_NORMAL 0xC0000000 30 #define LOG_GRP_SIZE sizeof(struct log_group) 31 #define LOG_ENT_SIZE sizeof(struct log_entry) 32 #define ARENA_MIN_SIZE (1UL << 24) /* 16 MB */ 33 #define ARENA_MAX_SIZE (1ULL << 39) /* 512 GB */ 34 #define RTT_VALID (1UL << 31) 35 #define RTT_INVALID 0 36 #define BTT_PG_SIZE 4096 37 #define BTT_DEFAULT_NFREE ND_MAX_LANES 38 #define LOG_SEQ_INIT 1 39 40 #define IB_FLAG_ERROR 0x00000001 41 #define IB_FLAG_ERROR_MASK 0x00000001 42 43 #define ent_lba(ent) (ent & MAP_LBA_MASK) 44 #define ent_e_flag(ent) (!!(ent & MAP_ERR_MASK)) 45 #define ent_z_flag(ent) (!!(ent & MAP_TRIM_MASK)) 46 #define set_e_flag(ent) (ent |= MAP_ERR_MASK) 47 48 enum btt_init_state { 49 INIT_UNCHECKED = 0, 50 INIT_NOTFOUND, 51 INIT_READY 52 }; 53 54 /* 55 * A log group represents one log 'lane', and consists of four log entries. 56 * Two of the four entries are valid entries, and the remaining two are 57 * padding. Due to an old bug in the padding location, we need to perform a 58 * test to determine the padding scheme being used, and use that scheme 59 * thereafter. 60 * 61 * In kernels prior to 4.15, 'log group' would have actual log entries at 62 * indices (0, 2) and padding at indices (1, 3), where as the correct/updated 63 * format has log entries at indices (0, 1) and padding at indices (2, 3). 64 * 65 * Old (pre 4.15) format: 66 * +-----------------+-----------------+ 67 * | ent[0] | ent[1] | 68 * | 16B | 16B | 69 * | lba/old/new/seq | pad | 70 * +-----------------------------------+ 71 * | ent[2] | ent[3] | 72 * | 16B | 16B | 73 * | lba/old/new/seq | pad | 74 * +-----------------+-----------------+ 75 * 76 * New format: 77 * +-----------------+-----------------+ 78 * | ent[0] | ent[1] | 79 * | 16B | 16B | 80 * | lba/old/new/seq | lba/old/new/seq | 81 * +-----------------------------------+ 82 * | ent[2] | ent[3] | 83 * | 16B | 16B | 84 * | pad | pad | 85 * +-----------------+-----------------+ 86 * 87 * We detect during start-up which format is in use, and set 88 * arena->log_index[(0, 1)] with the detected format. 89 */ 90 91 struct log_entry { 92 __le32 lba; 93 __le32 old_map; 94 __le32 new_map; 95 __le32 seq; 96 }; 97 98 struct log_group { 99 struct log_entry ent[4]; 100 }; 101 102 struct btt_sb { 103 u8 signature[BTT_SIG_LEN]; 104 u8 uuid[16]; 105 u8 parent_uuid[16]; 106 __le32 flags; 107 __le16 version_major; 108 __le16 version_minor; 109 __le32 external_lbasize; 110 __le32 external_nlba; 111 __le32 internal_lbasize; 112 __le32 internal_nlba; 113 __le32 nfree; 114 __le32 infosize; 115 __le64 nextoff; 116 __le64 dataoff; 117 __le64 mapoff; 118 __le64 logoff; 119 __le64 info2off; 120 u8 padding[3968]; 121 __le64 checksum; 122 }; 123 124 struct free_entry { 125 u32 block; 126 u8 sub; 127 u8 seq; 128 u8 has_err; 129 }; 130 131 struct aligned_lock { 132 union { 133 spinlock_t lock; 134 u8 cacheline_padding[L1_CACHE_BYTES]; 135 }; 136 }; 137 138 /** 139 * struct arena_info - handle for an arena 140 * @size: Size in bytes this arena occupies on the raw device. 141 * This includes arena metadata. 142 * @external_lba_start: The first external LBA in this arena. 143 * @internal_nlba: Number of internal blocks available in the arena 144 * including nfree reserved blocks 145 * @internal_lbasize: Internal and external lba sizes may be different as 146 * we can round up 'odd' external lbasizes such as 520B 147 * to be aligned. 148 * @external_nlba: Number of blocks contributed by the arena to the number 149 * reported to upper layers. (internal_nlba - nfree) 150 * @external_lbasize: LBA size as exposed to upper layers. 151 * @nfree: A reserve number of 'free' blocks that is used to 152 * handle incoming writes. 153 * @version_major: Metadata layout version major. 154 * @version_minor: Metadata layout version minor. 155 * @sector_size: The Linux sector size - 512 or 4096 156 * @nextoff: Offset in bytes to the start of the next arena. 157 * @infooff: Offset in bytes to the info block of this arena. 158 * @dataoff: Offset in bytes to the data area of this arena. 159 * @mapoff: Offset in bytes to the map area of this arena. 160 * @logoff: Offset in bytes to the log area of this arena. 161 * @info2off: Offset in bytes to the backup info block of this arena. 162 * @freelist: Pointer to in-memory list of free blocks 163 * @rtt: Pointer to in-memory "Read Tracking Table" 164 * @map_locks: Spinlocks protecting concurrent map writes 165 * @nd_btt: Pointer to parent nd_btt structure. 166 * @list: List head for list of arenas 167 * @debugfs_dir: Debugfs dentry 168 * @flags: Arena flags - may signify error states. 169 * @err_lock: Mutex for synchronizing error clearing. 170 * @log_index: Indices of the valid log entries in a log_group 171 * 172 * arena_info is a per-arena handle. Once an arena is narrowed down for an 173 * IO, this struct is passed around for the duration of the IO. 174 */ 175 struct arena_info { 176 u64 size; /* Total bytes for this arena */ 177 u64 external_lba_start; 178 u32 internal_nlba; 179 u32 internal_lbasize; 180 u32 external_nlba; 181 u32 external_lbasize; 182 u32 nfree; 183 u16 version_major; 184 u16 version_minor; 185 u32 sector_size; 186 /* Byte offsets to the different on-media structures */ 187 u64 nextoff; 188 u64 infooff; 189 u64 dataoff; 190 u64 mapoff; 191 u64 logoff; 192 u64 info2off; 193 /* Pointers to other in-memory structures for this arena */ 194 struct free_entry *freelist; 195 u32 *rtt; 196 struct aligned_lock *map_locks; 197 struct nd_btt *nd_btt; 198 struct list_head list; 199 struct dentry *debugfs_dir; 200 /* Arena flags */ 201 u32 flags; 202 struct mutex err_lock; 203 int log_index[2]; 204 }; 205 206 /** 207 * struct btt - handle for a BTT instance 208 * @btt_disk: Pointer to the gendisk for BTT device 209 * @btt_queue: Pointer to the request queue for the BTT device 210 * @arena_list: Head of the list of arenas 211 * @debugfs_dir: Debugfs dentry 212 * @nd_btt: Parent nd_btt struct 213 * @nlba: Number of logical blocks exposed to the upper layers 214 * after removing the amount of space needed by metadata 215 * @rawsize: Total size in bytes of the available backing device 216 * @lbasize: LBA size as requested and presented to upper layers. 217 * This is sector_size + size of any metadata. 218 * @sector_size: The Linux sector size - 512 or 4096 219 * @lanes: Per-lane spinlocks 220 * @init_lock: Mutex used for the BTT initialization 221 * @init_state: Flag describing the initialization state for the BTT 222 * @num_arenas: Number of arenas in the BTT instance 223 * @phys_bb: Pointer to the namespace's badblocks structure 224 */ 225 struct btt { 226 struct gendisk *btt_disk; 227 struct request_queue *btt_queue; 228 struct list_head arena_list; 229 struct dentry *debugfs_dir; 230 struct nd_btt *nd_btt; 231 u64 nlba; 232 unsigned long long rawsize; 233 u32 lbasize; 234 u32 sector_size; 235 struct nd_region *nd_region; 236 struct mutex init_lock; 237 int init_state; 238 int num_arenas; 239 struct badblocks *phys_bb; 240 }; 241 242 bool nd_btt_arena_is_valid(struct nd_btt *nd_btt, struct btt_sb *super); 243 int nd_btt_version(struct nd_btt *nd_btt, struct nd_namespace_common *ndns, 244 struct btt_sb *btt_sb); 245 246 #endif 247