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/* 'normal' is both e and z flags set */
48#define ent_normal(ent) (ent_e_flag(ent) && ent_z_flag(ent))
49
50enum btt_init_state {
51 INIT_UNCHECKED = 0,
52 INIT_NOTFOUND,
53 INIT_READY
54};
55
56/*
57 * A log group represents one log 'lane', and consists of four log entries.
58 * Two of the four entries are valid entries, and the remaining two are
59 * padding. Due to an old bug in the padding location, we need to perform a
60 * test to determine the padding scheme being used, and use that scheme
61 * thereafter.
62 *
63 * In kernels prior to 4.15, 'log group' would have actual log entries at
64 * indices (0, 2) and padding at indices (1, 3), where as the correct/updated
65 * format has log entries at indices (0, 1) and padding at indices (2, 3).
66 *
67 * Old (pre 4.15) format:
68 * +-----------------+-----------------+
69 * | ent[0] | ent[1] |
70 * | 16B | 16B |
71 * | lba/old/new/seq | pad |
72 * +-----------------------------------+
73 * | ent[2] | ent[3] |
74 * | 16B | 16B |
75 * | lba/old/new/seq | pad |
76 * +-----------------+-----------------+
77 *
78 * New format:
79 * +-----------------+-----------------+
80 * | ent[0] | ent[1] |
81 * | 16B | 16B |
82 * | lba/old/new/seq | lba/old/new/seq |
83 * +-----------------------------------+
84 * | ent[2] | ent[3] |
85 * | 16B | 16B |
86 * | pad | pad |
87 * +-----------------+-----------------+
88 *
89 * We detect during start-up which format is in use, and set
90 * arena->log_index[(0, 1)] with the detected format.
91 */
92
93struct log_entry {
94 __le32 lba;
95 __le32 old_map;
96 __le32 new_map;
97 __le32 seq;
98};
99
100struct log_group {
101 struct log_entry ent[4];
102};
103
104struct btt_sb {
105 u8 signature[BTT_SIG_LEN];
106 u8 uuid[16];
107 u8 parent_uuid[16];
108 __le32 flags;
109 __le16 version_major;
110 __le16 version_minor;
111 __le32 external_lbasize;
112 __le32 external_nlba;
113 __le32 internal_lbasize;
114 __le32 internal_nlba;
115 __le32 nfree;
116 __le32 infosize;
117 __le64 nextoff;
118 __le64 dataoff;
119 __le64 mapoff;
120 __le64 logoff;
121 __le64 info2off;
122 u8 padding[3968];
123 __le64 checksum;
124};
125
126struct free_entry {
127 u32 block;
128 u8 sub;
129 u8 seq;
130 u8 has_err;
131};
132
133struct aligned_lock {
134 union {
135 spinlock_t lock;
136 u8 cacheline_padding[L1_CACHE_BYTES];
137 };
138};
139
140/**
141 * struct arena_info - handle for an arena
142 * @size: Size in bytes this arena occupies on the raw device.
143 * This includes arena metadata.
144 * @external_lba_start: The first external LBA in this arena.
145 * @internal_nlba: Number of internal blocks available in the arena
146 * including nfree reserved blocks
147 * @internal_lbasize: Internal and external lba sizes may be different as
148 * we can round up 'odd' external lbasizes such as 520B
149 * to be aligned.
150 * @external_nlba: Number of blocks contributed by the arena to the number
151 * reported to upper layers. (internal_nlba - nfree)
152 * @external_lbasize: LBA size as exposed to upper layers.
153 * @nfree: A reserve number of 'free' blocks that is used to
154 * handle incoming writes.
155 * @version_major: Metadata layout version major.
156 * @version_minor: Metadata layout version minor.
157 * @sector_size: The Linux sector size - 512 or 4096
158 * @nextoff: Offset in bytes to the start of the next arena.
159 * @infooff: Offset in bytes to the info block of this arena.
160 * @dataoff: Offset in bytes to the data area of this arena.
161 * @mapoff: Offset in bytes to the map area of this arena.
162 * @logoff: Offset in bytes to the log area of this arena.
163 * @info2off: Offset in bytes to the backup info block of this arena.
164 * @freelist: Pointer to in-memory list of free blocks
165 * @rtt: Pointer to in-memory "Read Tracking Table"
166 * @map_locks: Spinlocks protecting concurrent map writes
167 * @nd_btt: Pointer to parent nd_btt structure.
168 * @list: List head for list of arenas
169 * @debugfs_dir: Debugfs dentry
170 * @flags: Arena flags - may signify error states.
171 * @err_lock: Mutex for synchronizing error clearing.
172 * @log_index: Indices of the valid log entries in a log_group
173 *
174 * arena_info is a per-arena handle. Once an arena is narrowed down for an
175 * IO, this struct is passed around for the duration of the IO.
176 */
177struct arena_info {
178 u64 size; /* Total bytes for this arena */
179 u64 external_lba_start;
180 u32 internal_nlba;
181 u32 internal_lbasize;
182 u32 external_nlba;
183 u32 external_lbasize;
184 u32 nfree;
185 u16 version_major;
186 u16 version_minor;
187 u32 sector_size;
188 /* Byte offsets to the different on-media structures */
189 u64 nextoff;
190 u64 infooff;
191 u64 dataoff;
192 u64 mapoff;
193 u64 logoff;
194 u64 info2off;
195 /* Pointers to other in-memory structures for this arena */
196 struct free_entry *freelist;
197 u32 *rtt;
198 struct aligned_lock *map_locks;
199 struct nd_btt *nd_btt;
200 struct list_head list;
201 struct dentry *debugfs_dir;
202 /* Arena flags */
203 u32 flags;
204 struct mutex err_lock;
205 int log_index[2];
206};
207
208/**
209 * struct btt - handle for a BTT instance
210 * @btt_disk: Pointer to the gendisk for BTT device
211 * @btt_queue: Pointer to the request queue for the BTT device
212 * @arena_list: Head of the list of arenas
213 * @debugfs_dir: Debugfs dentry
214 * @nd_btt: Parent nd_btt struct
215 * @nlba: Number of logical blocks exposed to the upper layers
216 * after removing the amount of space needed by metadata
217 * @rawsize: Total size in bytes of the available backing device
218 * @lbasize: LBA size as requested and presented to upper layers.
219 * This is sector_size + size of any metadata.
220 * @sector_size: The Linux sector size - 512 or 4096
221 * @lanes: Per-lane spinlocks
222 * @init_lock: Mutex used for the BTT initialization
223 * @init_state: Flag describing the initialization state for the BTT
224 * @num_arenas: Number of arenas in the BTT instance
225 * @phys_bb: Pointer to the namespace's badblocks structure
226 */
227struct btt {
228 struct gendisk *btt_disk;
229 struct request_queue *btt_queue;
230 struct list_head arena_list;
231 struct dentry *debugfs_dir;
232 struct nd_btt *nd_btt;
233 u64 nlba;
234 unsigned long long rawsize;
235 u32 lbasize;
236 u32 sector_size;
237 struct nd_region *nd_region;
238 struct mutex init_lock;
239 int init_state;
240 int num_arenas;
241 struct badblocks *phys_bb;
242};
243
244bool nd_btt_arena_is_valid(struct nd_btt *nd_btt, struct btt_sb *super);
245int nd_btt_version(struct nd_btt *nd_btt, struct nd_namespace_common *ndns,
246 struct btt_sb *btt_sb);
247
248#endif
249