1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _BCACHEFS_JOURNAL_TYPES_H 3 #define _BCACHEFS_JOURNAL_TYPES_H 4 5 #include <linux/cache.h> 6 #include <linux/workqueue.h> 7 8 #include "alloc_types.h" 9 #include "super_types.h" 10 #include "fifo.h" 11 12 /* btree write buffer steals 8 bits for its own purposes: */ 13 #define JOURNAL_SEQ_MAX ((1ULL << 56) - 1) 14 15 #define JOURNAL_BUF_BITS 2 16 #define JOURNAL_BUF_NR (1U << JOURNAL_BUF_BITS) 17 #define JOURNAL_BUF_MASK (JOURNAL_BUF_NR - 1) 18 19 /* 20 * We put JOURNAL_BUF_NR of these in struct journal; we used them for writes to 21 * the journal that are being staged or in flight. 22 */ 23 struct journal_buf { 24 struct closure io; 25 struct jset *data; 26 27 __BKEY_PADDED(key, BCH_REPLICAS_MAX); 28 struct bch_devs_list devs_written; 29 30 struct closure_waitlist wait; 31 u64 last_seq; /* copy of data->last_seq */ 32 long expires; 33 u64 flush_time; 34 35 unsigned buf_size; /* size in bytes of @data */ 36 unsigned sectors; /* maximum size for current entry */ 37 unsigned disk_sectors; /* maximum size entry could have been, if 38 buf_size was bigger */ 39 unsigned u64s_reserved; 40 bool noflush:1; /* write has already been kicked off, and was noflush */ 41 bool must_flush:1; /* something wants a flush */ 42 bool separate_flush:1; 43 bool need_flush_to_write_buffer:1; 44 bool write_started:1; 45 bool write_allocated:1; 46 bool write_done:1; 47 u8 idx; 48 }; 49 50 /* 51 * Something that makes a journal entry dirty - i.e. a btree node that has to be 52 * flushed: 53 */ 54 55 enum journal_pin_type { 56 JOURNAL_PIN_TYPE_btree, 57 JOURNAL_PIN_TYPE_key_cache, 58 JOURNAL_PIN_TYPE_other, 59 JOURNAL_PIN_TYPE_NR, 60 }; 61 62 struct journal_entry_pin_list { 63 struct list_head unflushed[JOURNAL_PIN_TYPE_NR]; 64 struct list_head flushed[JOURNAL_PIN_TYPE_NR]; 65 atomic_t count; 66 struct bch_devs_list devs; 67 }; 68 69 struct journal; 70 struct journal_entry_pin; 71 typedef int (*journal_pin_flush_fn)(struct journal *j, 72 struct journal_entry_pin *, u64); 73 74 struct journal_entry_pin { 75 struct list_head list; 76 journal_pin_flush_fn flush; 77 u64 seq; 78 }; 79 80 struct journal_res { 81 bool ref; 82 u8 idx; 83 u16 u64s; 84 u32 offset; 85 u64 seq; 86 }; 87 88 union journal_res_state { 89 struct { 90 atomic64_t counter; 91 }; 92 93 struct { 94 u64 v; 95 }; 96 97 struct { 98 u64 cur_entry_offset:20, 99 idx:2, 100 unwritten_idx:2, 101 buf0_count:10, 102 buf1_count:10, 103 buf2_count:10, 104 buf3_count:10; 105 }; 106 }; 107 108 /* bytes: */ 109 #define JOURNAL_ENTRY_SIZE_MIN (64U << 10) /* 64k */ 110 #define JOURNAL_ENTRY_SIZE_MAX (4U << 20) /* 4M */ 111 112 /* 113 * We stash some journal state as sentinal values in cur_entry_offset: 114 * note - cur_entry_offset is in units of u64s 115 */ 116 #define JOURNAL_ENTRY_OFFSET_MAX ((1U << 20) - 1) 117 118 #define JOURNAL_ENTRY_BLOCKED_VAL (JOURNAL_ENTRY_OFFSET_MAX - 2) 119 #define JOURNAL_ENTRY_CLOSED_VAL (JOURNAL_ENTRY_OFFSET_MAX - 1) 120 #define JOURNAL_ENTRY_ERROR_VAL (JOURNAL_ENTRY_OFFSET_MAX) 121 122 struct journal_space { 123 /* Units of 512 bytes sectors: */ 124 unsigned next_entry; /* How big the next journal entry can be */ 125 unsigned total; 126 }; 127 128 enum journal_space_from { 129 journal_space_discarded, 130 journal_space_clean_ondisk, 131 journal_space_clean, 132 journal_space_total, 133 journal_space_nr, 134 }; 135 136 #define JOURNAL_FLAGS() \ 137 x(replay_done) \ 138 x(running) \ 139 x(may_skip_flush) \ 140 x(need_flush_write) \ 141 x(space_low) 142 143 enum journal_flags { 144 #define x(n) JOURNAL_##n, 145 JOURNAL_FLAGS() 146 #undef x 147 }; 148 149 /* Reasons we may fail to get a journal reservation: */ 150 #define JOURNAL_ERRORS() \ 151 x(ok) \ 152 x(retry) \ 153 x(blocked) \ 154 x(max_in_flight) \ 155 x(journal_full) \ 156 x(journal_pin_full) \ 157 x(journal_stuck) \ 158 x(insufficient_devices) 159 160 enum journal_errors { 161 #define x(n) JOURNAL_ERR_##n, 162 JOURNAL_ERRORS() 163 #undef x 164 }; 165 166 typedef DARRAY(u64) darray_u64; 167 168 struct journal_bio { 169 struct bch_dev *ca; 170 unsigned buf_idx; 171 172 struct bio bio; 173 }; 174 175 /* Embedded in struct bch_fs */ 176 struct journal { 177 /* Fastpath stuff up front: */ 178 struct { 179 180 union journal_res_state reservations; 181 enum bch_watermark watermark; 182 183 } __aligned(SMP_CACHE_BYTES); 184 185 unsigned long flags; 186 187 /* Max size of current journal entry */ 188 unsigned cur_entry_u64s; 189 unsigned cur_entry_sectors; 190 191 /* Reserved space in journal entry to be used just prior to write */ 192 unsigned entry_u64s_reserved; 193 194 195 /* 196 * 0, or -ENOSPC if waiting on journal reclaim, or -EROFS if 197 * insufficient devices: 198 */ 199 enum journal_errors cur_entry_error; 200 unsigned cur_entry_offset_if_blocked; 201 202 unsigned buf_size_want; 203 /* 204 * We may queue up some things to be journalled (log messages) before 205 * the journal has actually started - stash them here: 206 */ 207 darray_u64 early_journal_entries; 208 209 /* 210 * Protects journal_buf->data, when accessing without a jorunal 211 * reservation: for synchronization between the btree write buffer code 212 * and the journal write path: 213 */ 214 struct mutex buf_lock; 215 /* 216 * Two journal entries -- one is currently open for new entries, the 217 * other is possibly being written out. 218 */ 219 struct journal_buf buf[JOURNAL_BUF_NR]; 220 221 spinlock_t lock; 222 223 /* if nonzero, we may not open a new journal entry: */ 224 unsigned blocked; 225 226 /* Used when waiting because the journal was full */ 227 wait_queue_head_t wait; 228 struct closure_waitlist async_wait; 229 struct closure_waitlist reclaim_flush_wait; 230 231 struct delayed_work write_work; 232 struct workqueue_struct *wq; 233 234 /* Sequence number of most recent journal entry (last entry in @pin) */ 235 atomic64_t seq; 236 237 /* seq, last_seq from the most recent journal entry successfully written */ 238 u64 seq_ondisk; 239 u64 flushed_seq_ondisk; 240 u64 flushing_seq; 241 u64 last_seq_ondisk; 242 u64 err_seq; 243 u64 last_empty_seq; 244 u64 oldest_seq_found_ondisk; 245 246 /* 247 * FIFO of journal entries whose btree updates have not yet been 248 * written out. 249 * 250 * Each entry is a reference count. The position in the FIFO is the 251 * entry's sequence number relative to @seq. 252 * 253 * The journal entry itself holds a reference count, put when the 254 * journal entry is written out. Each btree node modified by the journal 255 * entry also holds a reference count, put when the btree node is 256 * written. 257 * 258 * When a reference count reaches zero, the journal entry is no longer 259 * needed. When all journal entries in the oldest journal bucket are no 260 * longer needed, the bucket can be discarded and reused. 261 */ 262 struct { 263 u64 front, back, size, mask; 264 struct journal_entry_pin_list *data; 265 } pin; 266 267 struct journal_space space[journal_space_nr]; 268 269 u64 replay_journal_seq; 270 u64 replay_journal_seq_end; 271 272 struct write_point wp; 273 spinlock_t err_lock; 274 275 struct mutex reclaim_lock; 276 /* 277 * Used for waiting until journal reclaim has freed up space in the 278 * journal: 279 */ 280 wait_queue_head_t reclaim_wait; 281 struct task_struct *reclaim_thread; 282 bool reclaim_kicked; 283 unsigned long next_reclaim; 284 u64 nr_direct_reclaim; 285 u64 nr_background_reclaim; 286 287 unsigned long last_flushed; 288 struct journal_entry_pin *flush_in_progress; 289 bool flush_in_progress_dropped; 290 wait_queue_head_t pin_flush_wait; 291 292 /* protects advancing ja->discard_idx: */ 293 struct mutex discard_lock; 294 bool can_discard; 295 296 unsigned long last_flush_write; 297 298 u64 write_start_time; 299 300 u64 nr_flush_writes; 301 u64 nr_noflush_writes; 302 u64 entry_bytes_written; 303 304 struct bch2_time_stats *flush_write_time; 305 struct bch2_time_stats *noflush_write_time; 306 struct bch2_time_stats *flush_seq_time; 307 308 #ifdef CONFIG_DEBUG_LOCK_ALLOC 309 struct lockdep_map res_map; 310 #endif 311 } __aligned(SMP_CACHE_BYTES); 312 313 /* 314 * Embedded in struct bch_dev. First three fields refer to the array of journal 315 * buckets, in bch_sb. 316 */ 317 struct journal_device { 318 /* 319 * For each journal bucket, contains the max sequence number of the 320 * journal writes it contains - so we know when a bucket can be reused. 321 */ 322 u64 *bucket_seq; 323 324 unsigned sectors_free; 325 326 /* 327 * discard_idx <= dirty_idx_ondisk <= dirty_idx <= cur_idx: 328 */ 329 unsigned discard_idx; /* Next bucket to discard */ 330 unsigned dirty_idx_ondisk; 331 unsigned dirty_idx; 332 unsigned cur_idx; /* Journal bucket we're currently writing to */ 333 unsigned nr; 334 335 u64 *buckets; 336 337 /* Bio for journal reads/writes to this device */ 338 struct journal_bio *bio[JOURNAL_BUF_NR]; 339 340 /* for bch_journal_read_device */ 341 struct closure read; 342 u64 highest_seq_found; 343 }; 344 345 /* 346 * journal_entry_res - reserve space in every journal entry: 347 */ 348 struct journal_entry_res { 349 unsigned u64s; 350 }; 351 352 #endif /* _BCACHEFS_JOURNAL_TYPES_H */ 353