1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BCACHEFS_JOURNAL_TYPES_H
3 #define _BCACHEFS_JOURNAL_TYPES_H
5 #include <linux/cache.h>
6 #include <linux/workqueue.h>
8 #include "alloc_types.h"
9 #include "super_types.h"
12 #define JOURNAL_BUF_BITS 2
13 #define JOURNAL_BUF_NR (1U << JOURNAL_BUF_BITS)
14 #define JOURNAL_BUF_MASK (JOURNAL_BUF_NR - 1)
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.
23 __BKEY_PADDED(key, BCH_REPLICAS_MAX);
24 struct bch_devs_list devs_written;
26 struct closure_waitlist wait;
27 u64 last_seq; /* copy of data->last_seq */
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 */
39 bool need_flush_to_write_buffer;
43 * Something that makes a journal entry dirty - i.e. a btree node that has to be
47 enum journal_pin_type {
49 JOURNAL_PIN_key_cache,
54 struct journal_entry_pin_list {
55 struct list_head list[JOURNAL_PIN_NR];
56 struct list_head flushed;
58 struct bch_devs_list devs;
62 struct journal_entry_pin;
63 typedef int (*journal_pin_flush_fn)(struct journal *j,
64 struct journal_entry_pin *, u64);
66 struct journal_entry_pin {
67 struct list_head list;
68 journal_pin_flush_fn flush;
80 union journal_res_state {
90 u64 cur_entry_offset:20,
101 #define JOURNAL_ENTRY_SIZE_MIN (64U << 10) /* 64k */
102 #define JOURNAL_ENTRY_SIZE_MAX (4U << 20) /* 4M */
105 * We stash some journal state as sentinal values in cur_entry_offset:
106 * note - cur_entry_offset is in units of u64s
108 #define JOURNAL_ENTRY_OFFSET_MAX ((1U << 20) - 1)
110 #define JOURNAL_ENTRY_CLOSED_VAL (JOURNAL_ENTRY_OFFSET_MAX - 1)
111 #define JOURNAL_ENTRY_ERROR_VAL (JOURNAL_ENTRY_OFFSET_MAX)
113 struct journal_space {
114 /* Units of 512 bytes sectors: */
115 unsigned next_entry; /* How big the next journal entry can be */
119 enum journal_space_from {
120 journal_space_discarded,
121 journal_space_clean_ondisk,
130 JOURNAL_MAY_SKIP_FLUSH,
131 JOURNAL_NEED_FLUSH_WRITE,
134 /* Reasons we may fail to get a journal reservation: */
135 #define JOURNAL_ERRORS() \
140 x(journal_pin_full) \
142 x(insufficient_devices)
144 enum journal_errors {
145 #define x(n) JOURNAL_ERR_##n,
150 typedef DARRAY(u64) darray_u64;
152 /* Embedded in struct bch_fs */
154 /* Fastpath stuff up front: */
157 union journal_res_state reservations;
158 enum bch_watermark watermark;
160 } __aligned(SMP_CACHE_BYTES);
164 /* Max size of current journal entry */
165 unsigned cur_entry_u64s;
166 unsigned cur_entry_sectors;
168 /* Reserved space in journal entry to be used just prior to write */
169 unsigned entry_u64s_reserved;
173 * 0, or -ENOSPC if waiting on journal reclaim, or -EROFS if
174 * insufficient devices:
176 enum journal_errors cur_entry_error;
178 unsigned buf_size_want;
180 * We may queue up some things to be journalled (log messages) before
181 * the journal has actually started - stash them here:
183 darray_u64 early_journal_entries;
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:
190 struct mutex buf_lock;
192 * Two journal entries -- one is currently open for new entries, the
193 * other is possibly being written out.
195 struct journal_buf buf[JOURNAL_BUF_NR];
199 /* if nonzero, we may not open a new journal entry: */
202 /* Used when waiting because the journal was full */
203 wait_queue_head_t wait;
204 struct closure_waitlist async_wait;
207 struct delayed_work write_work;
209 /* Sequence number of most recent journal entry (last entry in @pin) */
212 /* seq, last_seq from the most recent journal entry successfully written */
214 u64 flushed_seq_ondisk;
220 * FIFO of journal entries whose btree updates have not yet been
223 * Each entry is a reference count. The position in the FIFO is the
224 * entry's sequence number relative to @seq.
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
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.
236 u64 front, back, size, mask;
237 struct journal_entry_pin_list *data;
240 struct journal_space space[journal_space_nr];
242 u64 replay_journal_seq;
243 u64 replay_journal_seq_end;
245 struct write_point wp;
248 struct mutex reclaim_lock;
250 * Used for waiting until journal reclaim has freed up space in the
253 wait_queue_head_t reclaim_wait;
254 struct task_struct *reclaim_thread;
256 unsigned long next_reclaim;
257 u64 nr_direct_reclaim;
258 u64 nr_background_reclaim;
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;
265 /* protects advancing ja->discard_idx: */
266 struct mutex discard_lock;
269 unsigned long last_flush_write;
271 u64 write_start_time;
274 u64 nr_noflush_writes;
275 u64 entry_bytes_written;
277 u64 low_on_space_start;
278 u64 low_on_pin_start;
279 u64 max_in_flight_start;
280 u64 write_buffer_full_start;
282 struct bch2_time_stats *flush_write_time;
283 struct bch2_time_stats *noflush_write_time;
284 struct bch2_time_stats *flush_seq_time;
286 #ifdef CONFIG_DEBUG_LOCK_ALLOC
287 struct lockdep_map res_map;
289 } __aligned(SMP_CACHE_BYTES);
292 * Embedded in struct bch_dev. First three fields refer to the array of journal
293 * buckets, in bch_sb.
295 struct journal_device {
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.
302 unsigned sectors_free;
305 * discard_idx <= dirty_idx_ondisk <= dirty_idx <= cur_idx:
307 unsigned discard_idx; /* Next bucket to discard */
308 unsigned dirty_idx_ondisk;
310 unsigned cur_idx; /* Journal bucket we're currently writing to */
315 /* Bio for journal reads/writes to this device */
318 /* for bch_journal_read_device */
323 * journal_entry_res - reserve space in every journal entry:
325 struct journal_entry_res {
329 #endif /* _BCACHEFS_JOURNAL_TYPES_H */