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