1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _BCACHEFS_JOURNAL_H 3 #define _BCACHEFS_JOURNAL_H 4 5 /* 6 * THE JOURNAL: 7 * 8 * The primary purpose of the journal is to log updates (insertions) to the 9 * b-tree, to avoid having to do synchronous updates to the b-tree on disk. 10 * 11 * Without the journal, the b-tree is always internally consistent on 12 * disk - and in fact, in the earliest incarnations bcache didn't have a journal 13 * but did handle unclean shutdowns by doing all index updates synchronously 14 * (with coalescing). 15 * 16 * Updates to interior nodes still happen synchronously and without the journal 17 * (for simplicity) - this may change eventually but updates to interior nodes 18 * are rare enough it's not a huge priority. 19 * 20 * This means the journal is relatively separate from the b-tree; it consists of 21 * just a list of keys and journal replay consists of just redoing those 22 * insertions in same order that they appear in the journal. 23 * 24 * PERSISTENCE: 25 * 26 * For synchronous updates (where we're waiting on the index update to hit 27 * disk), the journal entry will be written out immediately (or as soon as 28 * possible, if the write for the previous journal entry was still in flight). 29 * 30 * Synchronous updates are specified by passing a closure (@flush_cl) to 31 * bch2_btree_insert() or bch_btree_insert_node(), which then pass that parameter 32 * down to the journalling code. That closure will wait on the journal write to 33 * complete (via closure_wait()). 34 * 35 * If the index update wasn't synchronous, the journal entry will be 36 * written out after 10 ms have elapsed, by default (the delay_ms field 37 * in struct journal). 38 * 39 * JOURNAL ENTRIES: 40 * 41 * A journal entry is variable size (struct jset), it's got a fixed length 42 * header and then a variable number of struct jset_entry entries. 43 * 44 * Journal entries are identified by monotonically increasing 64 bit sequence 45 * numbers - jset->seq; other places in the code refer to this sequence number. 46 * 47 * A jset_entry entry contains one or more bkeys (which is what gets inserted 48 * into the b-tree). We need a container to indicate which b-tree the key is 49 * for; also, the roots of the various b-trees are stored in jset_entry entries 50 * (one for each b-tree) - this lets us add new b-tree types without changing 51 * the on disk format. 52 * 53 * We also keep some things in the journal header that are logically part of the 54 * superblock - all the things that are frequently updated. This is for future 55 * bcache on raw flash support; the superblock (which will become another 56 * journal) can't be moved or wear leveled, so it contains just enough 57 * information to find the main journal, and the superblock only has to be 58 * rewritten when we want to move/wear level the main journal. 59 * 60 * JOURNAL LAYOUT ON DISK: 61 * 62 * The journal is written to a ringbuffer of buckets (which is kept in the 63 * superblock); the individual buckets are not necessarily contiguous on disk 64 * which means that journal entries are not allowed to span buckets, but also 65 * that we can resize the journal at runtime if desired (unimplemented). 66 * 67 * The journal buckets exist in the same pool as all the other buckets that are 68 * managed by the allocator and garbage collection - garbage collection marks 69 * the journal buckets as metadata buckets. 70 * 71 * OPEN/DIRTY JOURNAL ENTRIES: 72 * 73 * Open/dirty journal entries are journal entries that contain b-tree updates 74 * that have not yet been written out to the b-tree on disk. We have to track 75 * which journal entries are dirty, and we also have to avoid wrapping around 76 * the journal and overwriting old but still dirty journal entries with new 77 * journal entries. 78 * 79 * On disk, this is represented with the "last_seq" field of struct jset; 80 * last_seq is the first sequence number that journal replay has to replay. 81 * 82 * To avoid overwriting dirty journal entries on disk, we keep a mapping (in 83 * journal_device->seq) of for each journal bucket, the highest sequence number 84 * any journal entry it contains. Then, by comparing that against last_seq we 85 * can determine whether that journal bucket contains dirty journal entries or 86 * not. 87 * 88 * To track which journal entries are dirty, we maintain a fifo of refcounts 89 * (where each entry corresponds to a specific sequence number) - when a ref 90 * goes to 0, that journal entry is no longer dirty. 91 * 92 * Journalling of index updates is done at the same time as the b-tree itself is 93 * being modified (see btree_insert_key()); when we add the key to the journal 94 * the pending b-tree write takes a ref on the journal entry the key was added 95 * to. If a pending b-tree write would need to take refs on multiple dirty 96 * journal entries, it only keeps the ref on the oldest one (since a newer 97 * journal entry will still be replayed if an older entry was dirty). 98 * 99 * JOURNAL FILLING UP: 100 * 101 * There are two ways the journal could fill up; either we could run out of 102 * space to write to, or we could have too many open journal entries and run out 103 * of room in the fifo of refcounts. Since those refcounts are decremented 104 * without any locking we can't safely resize that fifo, so we handle it the 105 * same way. 106 * 107 * If the journal fills up, we start flushing dirty btree nodes until we can 108 * allocate space for a journal write again - preferentially flushing btree 109 * nodes that are pinning the oldest journal entries first. 110 */ 111 112 #include <linux/hash.h> 113 114 #include "journal_types.h" 115 116 struct bch_fs; 117 118 static inline void journal_wake(struct journal *j) 119 { 120 wake_up(&j->wait); 121 closure_wake_up(&j->async_wait); 122 } 123 124 static inline struct journal_buf *journal_cur_buf(struct journal *j) 125 { 126 return j->buf + j->reservations.idx; 127 } 128 129 /* Sequence number of oldest dirty journal entry */ 130 131 static inline u64 journal_last_seq(struct journal *j) 132 { 133 return j->pin.front; 134 } 135 136 static inline u64 journal_cur_seq(struct journal *j) 137 { 138 return atomic64_read(&j->seq); 139 } 140 141 static inline u64 journal_last_unwritten_seq(struct journal *j) 142 { 143 return j->seq_ondisk + 1; 144 } 145 146 static inline int journal_state_count(union journal_res_state s, int idx) 147 { 148 switch (idx) { 149 case 0: return s.buf0_count; 150 case 1: return s.buf1_count; 151 case 2: return s.buf2_count; 152 case 3: return s.buf3_count; 153 } 154 BUG(); 155 } 156 157 static inline void journal_state_inc(union journal_res_state *s) 158 { 159 s->buf0_count += s->idx == 0; 160 s->buf1_count += s->idx == 1; 161 s->buf2_count += s->idx == 2; 162 s->buf3_count += s->idx == 3; 163 } 164 165 /* 166 * Amount of space that will be taken up by some keys in the journal (i.e. 167 * including the jset header) 168 */ 169 static inline unsigned jset_u64s(unsigned u64s) 170 { 171 return u64s + sizeof(struct jset_entry) / sizeof(u64); 172 } 173 174 static inline int journal_entry_overhead(struct journal *j) 175 { 176 return sizeof(struct jset) / sizeof(u64) + j->entry_u64s_reserved; 177 } 178 179 static inline struct jset_entry * 180 bch2_journal_add_entry_noreservation(struct journal_buf *buf, size_t u64s) 181 { 182 struct jset *jset = buf->data; 183 struct jset_entry *entry = vstruct_idx(jset, le32_to_cpu(jset->u64s)); 184 185 memset(entry, 0, sizeof(*entry)); 186 entry->u64s = cpu_to_le16(u64s); 187 188 le32_add_cpu(&jset->u64s, jset_u64s(u64s)); 189 190 return entry; 191 } 192 193 static inline struct jset_entry * 194 journal_res_entry(struct journal *j, struct journal_res *res) 195 { 196 return vstruct_idx(j->buf[res->idx].data, res->offset); 197 } 198 199 static inline unsigned journal_entry_init(struct jset_entry *entry, unsigned type, 200 enum btree_id id, unsigned level, 201 unsigned u64s) 202 { 203 entry->u64s = cpu_to_le16(u64s); 204 entry->btree_id = id; 205 entry->level = level; 206 entry->type = type; 207 entry->pad[0] = 0; 208 entry->pad[1] = 0; 209 entry->pad[2] = 0; 210 return jset_u64s(u64s); 211 } 212 213 static inline unsigned journal_entry_set(struct jset_entry *entry, unsigned type, 214 enum btree_id id, unsigned level, 215 const void *data, unsigned u64s) 216 { 217 unsigned ret = journal_entry_init(entry, type, id, level, u64s); 218 219 memcpy_u64s_small(entry->_data, data, u64s); 220 return ret; 221 } 222 223 static inline struct jset_entry * 224 bch2_journal_add_entry(struct journal *j, struct journal_res *res, 225 unsigned type, enum btree_id id, 226 unsigned level, unsigned u64s) 227 { 228 struct jset_entry *entry = journal_res_entry(j, res); 229 unsigned actual = journal_entry_init(entry, type, id, level, u64s); 230 231 EBUG_ON(!res->ref); 232 EBUG_ON(actual > res->u64s); 233 234 res->offset += actual; 235 res->u64s -= actual; 236 return entry; 237 } 238 239 static inline bool journal_entry_empty(struct jset *j) 240 { 241 if (j->seq != j->last_seq) 242 return false; 243 244 vstruct_for_each(j, i) 245 if (i->type == BCH_JSET_ENTRY_btree_keys && i->u64s) 246 return false; 247 return true; 248 } 249 250 /* 251 * Drop reference on a buffer index and return true if the count has hit zero. 252 */ 253 static inline union journal_res_state journal_state_buf_put(struct journal *j, unsigned idx) 254 { 255 union journal_res_state s; 256 257 s.v = atomic64_sub_return(((union journal_res_state) { 258 .buf0_count = idx == 0, 259 .buf1_count = idx == 1, 260 .buf2_count = idx == 2, 261 .buf3_count = idx == 3, 262 }).v, &j->reservations.counter); 263 return s; 264 } 265 266 bool bch2_journal_entry_close(struct journal *); 267 void bch2_journal_buf_put_final(struct journal *, u64, bool); 268 269 static inline void __bch2_journal_buf_put(struct journal *j, unsigned idx, u64 seq) 270 { 271 union journal_res_state s; 272 273 s = journal_state_buf_put(j, idx); 274 if (!journal_state_count(s, idx)) 275 bch2_journal_buf_put_final(j, seq, idx == s.unwritten_idx); 276 } 277 278 static inline void bch2_journal_buf_put(struct journal *j, unsigned idx, u64 seq) 279 { 280 union journal_res_state s; 281 282 s = journal_state_buf_put(j, idx); 283 if (!journal_state_count(s, idx)) { 284 spin_lock(&j->lock); 285 bch2_journal_buf_put_final(j, seq, idx == s.unwritten_idx); 286 spin_unlock(&j->lock); 287 } 288 } 289 290 /* 291 * This function releases the journal write structure so other threads can 292 * then proceed to add their keys as well. 293 */ 294 static inline void bch2_journal_res_put(struct journal *j, 295 struct journal_res *res) 296 { 297 if (!res->ref) 298 return; 299 300 lock_release(&j->res_map, _THIS_IP_); 301 302 while (res->u64s) 303 bch2_journal_add_entry(j, res, 304 BCH_JSET_ENTRY_btree_keys, 305 0, 0, 0); 306 307 bch2_journal_buf_put(j, res->idx, res->seq); 308 309 res->ref = 0; 310 } 311 312 int bch2_journal_res_get_slowpath(struct journal *, struct journal_res *, 313 unsigned); 314 315 /* First bits for BCH_WATERMARK: */ 316 enum journal_res_flags { 317 __JOURNAL_RES_GET_NONBLOCK = BCH_WATERMARK_BITS, 318 __JOURNAL_RES_GET_CHECK, 319 }; 320 321 #define JOURNAL_RES_GET_NONBLOCK (1 << __JOURNAL_RES_GET_NONBLOCK) 322 #define JOURNAL_RES_GET_CHECK (1 << __JOURNAL_RES_GET_CHECK) 323 324 static inline int journal_res_get_fast(struct journal *j, 325 struct journal_res *res, 326 unsigned flags) 327 { 328 union journal_res_state old, new; 329 u64 v = atomic64_read(&j->reservations.counter); 330 331 do { 332 old.v = new.v = v; 333 334 /* 335 * Check if there is still room in the current journal 336 * entry: 337 */ 338 if (new.cur_entry_offset + res->u64s > j->cur_entry_u64s) 339 return 0; 340 341 EBUG_ON(!journal_state_count(new, new.idx)); 342 343 if ((flags & BCH_WATERMARK_MASK) < j->watermark) 344 return 0; 345 346 new.cur_entry_offset += res->u64s; 347 journal_state_inc(&new); 348 349 /* 350 * If the refcount would overflow, we have to wait: 351 * XXX - tracepoint this: 352 */ 353 if (!journal_state_count(new, new.idx)) 354 return 0; 355 356 if (flags & JOURNAL_RES_GET_CHECK) 357 return 1; 358 } while ((v = atomic64_cmpxchg(&j->reservations.counter, 359 old.v, new.v)) != old.v); 360 361 res->ref = true; 362 res->idx = old.idx; 363 res->offset = old.cur_entry_offset; 364 res->seq = le64_to_cpu(j->buf[old.idx].data->seq); 365 return 1; 366 } 367 368 static inline int bch2_journal_res_get(struct journal *j, struct journal_res *res, 369 unsigned u64s, unsigned flags) 370 { 371 int ret; 372 373 EBUG_ON(res->ref); 374 EBUG_ON(!test_bit(JOURNAL_STARTED, &j->flags)); 375 376 res->u64s = u64s; 377 378 if (journal_res_get_fast(j, res, flags)) 379 goto out; 380 381 ret = bch2_journal_res_get_slowpath(j, res, flags); 382 if (ret) 383 return ret; 384 out: 385 if (!(flags & JOURNAL_RES_GET_CHECK)) { 386 lock_acquire_shared(&j->res_map, 0, 387 (flags & JOURNAL_RES_GET_NONBLOCK) != 0, 388 NULL, _THIS_IP_); 389 EBUG_ON(!res->ref); 390 } 391 return 0; 392 } 393 394 /* journal_entry_res: */ 395 396 void bch2_journal_entry_res_resize(struct journal *, 397 struct journal_entry_res *, 398 unsigned); 399 400 int bch2_journal_flush_seq_async(struct journal *, u64, struct closure *); 401 void bch2_journal_flush_async(struct journal *, struct closure *); 402 403 int bch2_journal_flush_seq(struct journal *, u64); 404 int bch2_journal_flush(struct journal *); 405 bool bch2_journal_noflush_seq(struct journal *, u64); 406 int bch2_journal_meta(struct journal *); 407 408 void bch2_journal_halt(struct journal *); 409 410 static inline int bch2_journal_error(struct journal *j) 411 { 412 return j->reservations.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL 413 ? -EIO : 0; 414 } 415 416 struct bch_dev; 417 418 static inline void bch2_journal_set_replay_done(struct journal *j) 419 { 420 BUG_ON(!test_bit(JOURNAL_STARTED, &j->flags)); 421 set_bit(JOURNAL_REPLAY_DONE, &j->flags); 422 } 423 424 void bch2_journal_unblock(struct journal *); 425 void bch2_journal_block(struct journal *); 426 struct journal_buf *bch2_next_write_buffer_flush_journal_buf(struct journal *j, u64 max_seq); 427 428 void __bch2_journal_debug_to_text(struct printbuf *, struct journal *); 429 void bch2_journal_debug_to_text(struct printbuf *, struct journal *); 430 void bch2_journal_pins_to_text(struct printbuf *, struct journal *); 431 bool bch2_journal_seq_pins_to_text(struct printbuf *, struct journal *, u64 *); 432 433 int bch2_set_nr_journal_buckets(struct bch_fs *, struct bch_dev *, 434 unsigned nr); 435 int bch2_dev_journal_alloc(struct bch_dev *); 436 int bch2_fs_journal_alloc(struct bch_fs *); 437 438 void bch2_dev_journal_stop(struct journal *, struct bch_dev *); 439 440 void bch2_fs_journal_stop(struct journal *); 441 int bch2_fs_journal_start(struct journal *, u64); 442 443 void bch2_dev_journal_exit(struct bch_dev *); 444 int bch2_dev_journal_init(struct bch_dev *, struct bch_sb *); 445 void bch2_fs_journal_exit(struct journal *); 446 int bch2_fs_journal_init(struct journal *); 447 448 #endif /* _BCACHEFS_JOURNAL_H */ 449