xref: /linux/fs/bcachefs/journal.h (revision eed4edda910fe34dfae8c6bfbcf57f4593a54295)
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