1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHEFS_JOURNAL_H
3 #define _BCACHEFS_JOURNAL_H
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.
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
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.
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.
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).
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()).
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
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.
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.
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
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.
60 * JOURNAL LAYOUT ON DISK:
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).
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.
71 * OPEN/DIRTY JOURNAL ENTRIES:
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
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.
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
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.
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).
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
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.
112 #include <linux/hash.h>
114 #include "journal_types.h"
118 static inline void journal_wake(struct journal *j)
121 closure_wake_up(&j->async_wait);
122 closure_wake_up(&j->preres_wait);
125 static inline struct journal_buf *journal_cur_buf(struct journal *j)
127 return j->buf + j->reservations.idx;
130 /* Sequence number of oldest dirty journal entry */
132 static inline u64 journal_last_seq(struct journal *j)
137 static inline u64 journal_cur_seq(struct journal *j)
139 EBUG_ON(j->pin.back - 1 != atomic64_read(&j->seq));
141 return j->pin.back - 1;
144 static inline u64 journal_last_unwritten_seq(struct journal *j)
146 return j->seq_ondisk + 1;
149 static inline int journal_state_count(union journal_res_state s, int idx)
152 case 0: return s.buf0_count;
153 case 1: return s.buf1_count;
154 case 2: return s.buf2_count;
155 case 3: return s.buf3_count;
160 static inline void journal_state_inc(union journal_res_state *s)
162 s->buf0_count += s->idx == 0;
163 s->buf1_count += s->idx == 1;
164 s->buf2_count += s->idx == 2;
165 s->buf3_count += s->idx == 3;
169 * Amount of space that will be taken up by some keys in the journal (i.e.
170 * including the jset header)
172 static inline unsigned jset_u64s(unsigned u64s)
174 return u64s + sizeof(struct jset_entry) / sizeof(u64);
177 static inline int journal_entry_overhead(struct journal *j)
179 return sizeof(struct jset) / sizeof(u64) + j->entry_u64s_reserved;
182 static inline struct jset_entry *
183 bch2_journal_add_entry_noreservation(struct journal_buf *buf, size_t u64s)
185 struct jset *jset = buf->data;
186 struct jset_entry *entry = vstruct_idx(jset, le32_to_cpu(jset->u64s));
188 memset(entry, 0, sizeof(*entry));
189 entry->u64s = cpu_to_le16(u64s);
191 le32_add_cpu(&jset->u64s, jset_u64s(u64s));
196 static inline struct jset_entry *
197 journal_res_entry(struct journal *j, struct journal_res *res)
199 return vstruct_idx(j->buf[res->idx].data, res->offset);
202 static inline unsigned journal_entry_init(struct jset_entry *entry, unsigned type,
203 enum btree_id id, unsigned level,
206 entry->u64s = cpu_to_le16(u64s);
207 entry->btree_id = id;
208 entry->level = level;
213 return jset_u64s(u64s);
216 static inline unsigned journal_entry_set(struct jset_entry *entry, unsigned type,
217 enum btree_id id, unsigned level,
218 const void *data, unsigned u64s)
220 unsigned ret = journal_entry_init(entry, type, id, level, u64s);
222 memcpy_u64s_small(entry->_data, data, u64s);
226 static inline struct jset_entry *
227 bch2_journal_add_entry(struct journal *j, struct journal_res *res,
228 unsigned type, enum btree_id id,
229 unsigned level, unsigned u64s)
231 struct jset_entry *entry = journal_res_entry(j, res);
232 unsigned actual = journal_entry_init(entry, type, id, level, u64s);
235 EBUG_ON(actual > res->u64s);
237 res->offset += actual;
242 static inline bool journal_entry_empty(struct jset *j)
244 struct jset_entry *i;
246 if (j->seq != j->last_seq)
249 vstruct_for_each(j, i)
250 if (i->type == BCH_JSET_ENTRY_btree_keys && i->u64s)
255 void __bch2_journal_buf_put(struct journal *);
257 static inline void bch2_journal_buf_put(struct journal *j, unsigned idx)
259 union journal_res_state s;
261 s.v = atomic64_sub_return(((union journal_res_state) {
262 .buf0_count = idx == 0,
263 .buf1_count = idx == 1,
264 .buf2_count = idx == 2,
265 .buf3_count = idx == 3,
266 }).v, &j->reservations.counter);
268 if (!journal_state_count(s, idx) && idx == s.unwritten_idx)
269 __bch2_journal_buf_put(j);
273 * This function releases the journal write structure so other threads can
274 * then proceed to add their keys as well.
276 static inline void bch2_journal_res_put(struct journal *j,
277 struct journal_res *res)
282 lock_release(&j->res_map, _THIS_IP_);
285 bch2_journal_add_entry(j, res,
286 BCH_JSET_ENTRY_btree_keys,
289 bch2_journal_buf_put(j, res->idx);
294 int bch2_journal_res_get_slowpath(struct journal *, struct journal_res *,
297 /* First two bits for JOURNAL_WATERMARK: */
298 #define JOURNAL_RES_GET_NONBLOCK (1 << 2)
299 #define JOURNAL_RES_GET_CHECK (1 << 3)
301 static inline int journal_res_get_fast(struct journal *j,
302 struct journal_res *res,
305 union journal_res_state old, new;
306 u64 v = atomic64_read(&j->reservations.counter);
312 * Check if there is still room in the current journal
315 if (new.cur_entry_offset + res->u64s > j->cur_entry_u64s)
318 EBUG_ON(!journal_state_count(new, new.idx));
320 if ((flags & JOURNAL_WATERMARK_MASK) < j->watermark)
323 new.cur_entry_offset += res->u64s;
324 journal_state_inc(&new);
327 * If the refcount would overflow, we have to wait:
328 * XXX - tracepoint this:
330 if (!journal_state_count(new, new.idx))
333 if (flags & JOURNAL_RES_GET_CHECK)
335 } while ((v = atomic64_cmpxchg(&j->reservations.counter,
336 old.v, new.v)) != old.v);
340 res->offset = old.cur_entry_offset;
341 res->seq = le64_to_cpu(j->buf[old.idx].data->seq);
345 static inline int bch2_journal_res_get(struct journal *j, struct journal_res *res,
346 unsigned u64s, unsigned flags)
351 EBUG_ON(!test_bit(JOURNAL_STARTED, &j->flags));
355 if (journal_res_get_fast(j, res, flags))
358 ret = bch2_journal_res_get_slowpath(j, res, flags);
362 if (!(flags & JOURNAL_RES_GET_CHECK)) {
363 lock_acquire_shared(&j->res_map, 0,
364 (flags & JOURNAL_RES_GET_NONBLOCK) != 0,
371 /* journal_preres: */
373 static inline void journal_set_watermark(struct journal *j)
375 union journal_preres_state s = READ_ONCE(j->prereserved);
376 unsigned watermark = JOURNAL_WATERMARK_any;
378 if (fifo_free(&j->pin) < j->pin.size / 4)
379 watermark = max_t(unsigned, watermark, JOURNAL_WATERMARK_copygc);
380 if (fifo_free(&j->pin) < j->pin.size / 8)
381 watermark = max_t(unsigned, watermark, JOURNAL_WATERMARK_reserved);
383 if (s.reserved > s.remaining)
384 watermark = max_t(unsigned, watermark, JOURNAL_WATERMARK_copygc);
386 watermark = max_t(unsigned, watermark, JOURNAL_WATERMARK_reserved);
388 if (watermark == j->watermark)
391 swap(watermark, j->watermark);
392 if (watermark > j->watermark)
396 static inline void bch2_journal_preres_put(struct journal *j,
397 struct journal_preres *res)
399 union journal_preres_state s = { .reserved = res->u64s };
404 s.v = atomic64_sub_return(s.v, &j->prereserved.counter);
407 if (unlikely(s.waiting)) {
408 clear_bit(ilog2((((union journal_preres_state) { .waiting = 1 }).v)),
409 (unsigned long *) &j->prereserved.v);
410 closure_wake_up(&j->preres_wait);
413 if (s.reserved <= s.remaining && j->watermark)
414 journal_set_watermark(j);
417 int __bch2_journal_preres_get(struct journal *,
418 struct journal_preres *, unsigned, unsigned);
420 static inline int bch2_journal_preres_get_fast(struct journal *j,
421 struct journal_preres *res,
426 int d = new_u64s - res->u64s;
427 union journal_preres_state old, new;
428 u64 v = atomic64_read(&j->prereserved.counter);
435 if ((flags & JOURNAL_WATERMARK_reserved) ||
436 new.reserved + d < new.remaining) {
439 } else if (set_waiting && !new.waiting)
443 } while ((v = atomic64_cmpxchg(&j->prereserved.counter,
444 old.v, new.v)) != old.v);
451 static inline int bch2_journal_preres_get(struct journal *j,
452 struct journal_preres *res,
456 if (new_u64s <= res->u64s)
459 if (bch2_journal_preres_get_fast(j, res, new_u64s, flags, false))
462 if (flags & JOURNAL_RES_GET_NONBLOCK)
463 return -BCH_ERR_journal_preres_get_blocked;
465 return __bch2_journal_preres_get(j, res, new_u64s, flags);
468 /* journal_entry_res: */
470 void bch2_journal_entry_res_resize(struct journal *,
471 struct journal_entry_res *,
474 int bch2_journal_flush_seq_async(struct journal *, u64, struct closure *);
475 void bch2_journal_flush_async(struct journal *, struct closure *);
477 int bch2_journal_flush_seq(struct journal *, u64);
478 int bch2_journal_flush(struct journal *);
479 bool bch2_journal_noflush_seq(struct journal *, u64);
480 int bch2_journal_meta(struct journal *);
482 void bch2_journal_halt(struct journal *);
484 static inline int bch2_journal_error(struct journal *j)
486 return j->reservations.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL
492 static inline void bch2_journal_set_replay_done(struct journal *j)
494 BUG_ON(!test_bit(JOURNAL_STARTED, &j->flags));
495 set_bit(JOURNAL_REPLAY_DONE, &j->flags);
498 void bch2_journal_unblock(struct journal *);
499 void bch2_journal_block(struct journal *);
501 void __bch2_journal_debug_to_text(struct printbuf *, struct journal *);
502 void bch2_journal_debug_to_text(struct printbuf *, struct journal *);
503 void bch2_journal_pins_to_text(struct printbuf *, struct journal *);
504 bool bch2_journal_seq_pins_to_text(struct printbuf *, struct journal *, u64 *);
506 int bch2_set_nr_journal_buckets(struct bch_fs *, struct bch_dev *,
508 int bch2_dev_journal_alloc(struct bch_dev *);
510 void bch2_dev_journal_stop(struct journal *, struct bch_dev *);
512 void bch2_fs_journal_stop(struct journal *);
513 int bch2_fs_journal_start(struct journal *, u64);
515 void bch2_dev_journal_exit(struct bch_dev *);
516 int bch2_dev_journal_init(struct bch_dev *, struct bch_sb *);
517 void bch2_fs_journal_exit(struct journal *);
518 int bch2_fs_journal_init(struct journal *);
520 #endif /* _BCACHEFS_JOURNAL_H */