1 #ifndef _BCACHEFS_JOURNAL_H
2 #define _BCACHEFS_JOURNAL_H
7 * The primary purpose of the journal is to log updates (insertions) to the
8 * b-tree, to avoid having to do synchronous updates to the b-tree on disk.
10 * Without the journal, the b-tree is always internally consistent on
11 * disk - and in fact, in the earliest incarnations bcache didn't have a journal
12 * but did handle unclean shutdowns by doing all index updates synchronously
15 * Updates to interior nodes still happen synchronously and without the journal
16 * (for simplicity) - this may change eventually but updates to interior nodes
17 * are rare enough it's not a huge priority.
19 * This means the journal is relatively separate from the b-tree; it consists of
20 * just a list of keys and journal replay consists of just redoing those
21 * insertions in same order that they appear in the journal.
25 * For synchronous updates (where we're waiting on the index update to hit
26 * disk), the journal entry will be written out immediately (or as soon as
27 * possible, if the write for the previous journal entry was still in flight).
29 * Synchronous updates are specified by passing a closure (@flush_cl) to
30 * bch2_btree_insert() or bch_btree_insert_node(), which then pass that parameter
31 * down to the journalling code. That closure will will wait on the journal
32 * write to complete (via closure_wait()).
34 * If the index update wasn't synchronous, the journal entry will be
35 * written out after 10 ms have elapsed, by default (the delay_ms field
40 * A journal entry is variable size (struct jset), it's got a fixed length
41 * header and then a variable number of struct jset_entry entries.
43 * Journal entries are identified by monotonically increasing 64 bit sequence
44 * numbers - jset->seq; other places in the code refer to this sequence number.
46 * A jset_entry entry contains one or more bkeys (which is what gets inserted
47 * into the b-tree). We need a container to indicate which b-tree the key is
48 * for; also, the roots of the various b-trees are stored in jset_entry entries
49 * (one for each b-tree) - this lets us add new b-tree types without changing
52 * We also keep some things in the journal header that are logically part of the
53 * superblock - all the things that are frequently updated. This is for future
54 * bcache on raw flash support; the superblock (which will become another
55 * journal) can't be moved or wear leveled, so it contains just enough
56 * information to find the main journal, and the superblock only has to be
57 * rewritten when we want to move/wear level the main journal.
59 * JOURNAL LAYOUT ON DISK:
61 * The journal is written to a ringbuffer of buckets (which is kept in the
62 * superblock); the individual buckets are not necessarily contiguous on disk
63 * which means that journal entries are not allowed to span buckets, but also
64 * that we can resize the journal at runtime if desired (unimplemented).
66 * The journal buckets exist in the same pool as all the other buckets that are
67 * managed by the allocator and garbage collection - garbage collection marks
68 * the journal buckets as metadata buckets.
70 * OPEN/DIRTY JOURNAL ENTRIES:
72 * Open/dirty journal entries are journal entries that contain b-tree updates
73 * that have not yet been written out to the b-tree on disk. We have to track
74 * which journal entries are dirty, and we also have to avoid wrapping around
75 * the journal and overwriting old but still dirty journal entries with new
78 * On disk, this is represented with the "last_seq" field of struct jset;
79 * last_seq is the first sequence number that journal replay has to replay.
81 * To avoid overwriting dirty journal entries on disk, we keep a mapping (in
82 * journal_device->seq) of for each journal bucket, the highest sequence number
83 * any journal entry it contains. Then, by comparing that against last_seq we
84 * can determine whether that journal bucket contains dirty journal entries or
87 * To track which journal entries are dirty, we maintain a fifo of refcounts
88 * (where each entry corresponds to a specific sequence number) - when a ref
89 * goes to 0, that journal entry is no longer dirty.
91 * Journalling of index updates is done at the same time as the b-tree itself is
92 * being modified (see btree_insert_key()); when we add the key to the journal
93 * the pending b-tree write takes a ref on the journal entry the key was added
94 * to. If a pending b-tree write would need to take refs on multiple dirty
95 * journal entries, it only keeps the ref on the oldest one (since a newer
96 * journal entry will still be replayed if an older entry was dirty).
100 * There are two ways the journal could fill up; either we could run out of
101 * space to write to, or we could have too many open journal entries and run out
102 * of room in the fifo of refcounts. Since those refcounts are decremented
103 * without any locking we can't safely resize that fifo, so we handle it the
106 * If the journal fills up, we start flushing dirty btree nodes until we can
107 * allocate space for a journal write again - preferentially flushing btree
108 * nodes that are pinning the oldest journal entries first.
111 #include <linux/hash.h>
113 #include "journal_types.h"
116 * Only used for holding the journal entries we read in btree_journal_read()
117 * during cache_registration
119 struct journal_replay {
120 struct list_head list;
121 struct bch_devs_list devs;
126 static inline struct jset_entry *__jset_entry_type_next(struct jset *jset,
127 struct jset_entry *entry, unsigned type)
129 while (entry < vstruct_last(jset)) {
130 if (entry->type == type)
133 entry = vstruct_next(entry);
139 #define for_each_jset_entry_type(entry, jset, type) \
140 for (entry = (jset)->start; \
141 (entry = __jset_entry_type_next(jset, entry, type)); \
142 entry = vstruct_next(entry))
144 #define for_each_jset_key(k, _n, entry, jset) \
145 for_each_jset_entry_type(entry, jset, JOURNAL_ENTRY_BTREE_KEYS) \
146 vstruct_for_each_safe(entry, k, _n)
148 #define JOURNAL_PIN (32 * 1024)
150 static inline bool journal_pin_active(struct journal_entry_pin *pin)
152 return pin->pin_list != NULL;
155 static inline struct journal_entry_pin_list *
156 journal_seq_pin(struct journal *j, u64 seq)
158 return &j->pin.data[seq & j->pin.mask];
161 u64 bch2_journal_pin_seq(struct journal *, struct journal_entry_pin *);
163 void bch2_journal_pin_add(struct journal *, struct journal_res *,
164 struct journal_entry_pin *, journal_pin_flush_fn);
165 void bch2_journal_pin_drop(struct journal *, struct journal_entry_pin *);
166 void bch2_journal_pin_add_if_older(struct journal *,
167 struct journal_entry_pin *,
168 struct journal_entry_pin *,
169 journal_pin_flush_fn);
170 int bch2_journal_flush_pins(struct journal *, u64);
171 int bch2_journal_flush_all_pins(struct journal *);
177 struct bkey_i *bch2_journal_find_btree_root(struct bch_fs *, struct jset *,
178 enum btree_id, unsigned *);
180 int bch2_journal_seq_should_ignore(struct bch_fs *, u64, struct btree *);
182 u64 bch2_inode_journal_seq(struct journal *, u64);
184 static inline int journal_state_count(union journal_res_state s, int idx)
186 return idx == 0 ? s.buf0_count : s.buf1_count;
189 static inline void journal_state_inc(union journal_res_state *s)
191 s->buf0_count += s->idx == 0;
192 s->buf1_count += s->idx == 1;
195 static inline void bch2_journal_set_has_inode(struct journal *j,
196 struct journal_res *res,
199 struct journal_buf *buf = &j->buf[res->idx];
200 unsigned long bit = hash_64(inum, ilog2(sizeof(buf->has_inode) * 8));
202 /* avoid atomic op if possible */
203 if (unlikely(!test_bit(bit, buf->has_inode)))
204 set_bit(bit, buf->has_inode);
208 * Amount of space that will be taken up by some keys in the journal (i.e.
209 * including the jset header)
211 static inline unsigned jset_u64s(unsigned u64s)
213 return u64s + sizeof(struct jset_entry) / sizeof(u64);
216 static inline void bch2_journal_add_entry_at(struct journal_buf *buf,
218 unsigned type, enum btree_id id,
220 const void *data, size_t u64s)
222 struct jset_entry *entry = vstruct_idx(buf->data, offset);
224 memset(entry, 0, sizeof(*entry));
225 entry->u64s = cpu_to_le16(u64s);
226 entry->btree_id = id;
227 entry->level = level;
230 memcpy_u64s(entry->_data, data, u64s);
233 static inline void bch2_journal_add_entry(struct journal *j, struct journal_res *res,
234 unsigned type, enum btree_id id,
236 const void *data, unsigned u64s)
238 struct journal_buf *buf = &j->buf[res->idx];
239 unsigned actual = jset_u64s(u64s);
242 EBUG_ON(actual > res->u64s);
244 bch2_journal_add_entry_at(buf, res->offset, type,
245 id, level, data, u64s);
246 res->offset += actual;
250 static inline void bch2_journal_add_keys(struct journal *j, struct journal_res *res,
251 enum btree_id id, const struct bkey_i *k)
253 bch2_journal_add_entry(j, res, JOURNAL_ENTRY_BTREE_KEYS,
254 id, 0, k, k->k.u64s);
257 void bch2_journal_buf_put_slowpath(struct journal *, bool);
259 static inline void bch2_journal_buf_put(struct journal *j, unsigned idx,
260 bool need_write_just_set)
262 union journal_res_state s;
264 s.v = atomic64_sub_return(((union journal_res_state) {
265 .buf0_count = idx == 0,
266 .buf1_count = idx == 1,
267 }).v, &j->reservations.counter);
269 EBUG_ON(s.idx != idx && !s.prev_buf_unwritten);
272 * Do not initiate a journal write if the journal is in an error state
273 * (previous journal entry write may have failed)
276 !journal_state_count(s, idx) &&
277 s.cur_entry_offset != JOURNAL_ENTRY_ERROR_VAL)
278 bch2_journal_buf_put_slowpath(j, need_write_just_set);
282 * This function releases the journal write structure so other threads can
283 * then proceed to add their keys as well.
285 static inline void bch2_journal_res_put(struct journal *j,
286 struct journal_res *res)
291 lock_release(&j->res_map, 0, _RET_IP_);
294 bch2_journal_add_entry(j, res,
295 JOURNAL_ENTRY_BTREE_KEYS,
298 bch2_journal_buf_put(j, res->idx, false);
303 int bch2_journal_res_get_slowpath(struct journal *, struct journal_res *,
306 static inline int journal_res_get_fast(struct journal *j,
307 struct journal_res *res,
311 union journal_res_state old, new;
312 u64 v = atomic64_read(&j->reservations.counter);
318 * Check if there is still room in the current journal
321 if (old.cur_entry_offset + u64s_min > j->cur_entry_u64s)
324 res->offset = old.cur_entry_offset;
325 res->u64s = min(u64s_max, j->cur_entry_u64s -
326 old.cur_entry_offset);
328 journal_state_inc(&new);
329 new.cur_entry_offset += res->u64s;
330 } while ((v = atomic64_cmpxchg(&j->reservations.counter,
331 old.v, new.v)) != old.v);
335 res->seq = le64_to_cpu(j->buf[res->idx].data->seq);
339 static inline int bch2_journal_res_get(struct journal *j, struct journal_res *res,
340 unsigned u64s_min, unsigned u64s_max)
345 EBUG_ON(u64s_max < u64s_min);
346 EBUG_ON(!test_bit(JOURNAL_STARTED, &j->flags));
348 if (journal_res_get_fast(j, res, u64s_min, u64s_max))
351 ret = bch2_journal_res_get_slowpath(j, res, u64s_min, u64s_max);
355 lock_acquire_shared(&j->res_map, 0, 0, NULL, _THIS_IP_);
360 u64 bch2_journal_last_unwritten_seq(struct journal *);
361 int bch2_journal_open_seq_async(struct journal *, u64, struct closure *);
363 void bch2_journal_wait_on_seq(struct journal *, u64, struct closure *);
364 void bch2_journal_flush_seq_async(struct journal *, u64, struct closure *);
365 void bch2_journal_flush_async(struct journal *, struct closure *);
366 void bch2_journal_meta_async(struct journal *, struct closure *);
368 int bch2_journal_flush_seq(struct journal *, u64);
369 int bch2_journal_flush(struct journal *);
370 int bch2_journal_meta(struct journal *);
371 int bch2_journal_flush_device(struct journal *, unsigned);
373 void bch2_journal_halt(struct journal *);
375 static inline int bch2_journal_error(struct journal *j)
377 return j->reservations.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL
383 static inline bool journal_flushes_device(struct bch_dev *ca)
388 void bch2_journal_start(struct bch_fs *);
389 int bch2_journal_mark(struct bch_fs *, struct list_head *);
390 void bch2_journal_entries_free(struct list_head *);
391 int bch2_journal_read(struct bch_fs *, struct list_head *);
392 int bch2_journal_replay(struct bch_fs *, struct list_head *);
394 static inline void bch2_journal_set_replay_done(struct journal *j)
396 BUG_ON(!test_bit(JOURNAL_STARTED, &j->flags));
397 set_bit(JOURNAL_REPLAY_DONE, &j->flags);
400 ssize_t bch2_journal_print_debug(struct journal *, char *);
401 ssize_t bch2_journal_print_pins(struct journal *, char *);
403 int bch2_dev_journal_alloc(struct bch_fs *, struct bch_dev *);
405 void bch2_dev_journal_stop(struct journal *, struct bch_dev *);
406 void bch2_fs_journal_stop(struct journal *);
407 void bch2_dev_journal_exit(struct bch_dev *);
408 int bch2_dev_journal_init(struct bch_dev *, struct bch_sb *);
409 void bch2_fs_journal_exit(struct journal *);
410 int bch2_fs_journal_init(struct journal *);
412 #endif /* _BCACHEFS_JOURNAL_H */