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 */
29 unsigned buf_size; /* size in bytes of @data */
30 unsigned sectors; /* maximum size for current entry */
31 unsigned disk_sectors; /* maximum size entry could have been, if
32 buf_size was bigger */
33 unsigned u64s_reserved;
34 bool noflush; /* write has already been kicked off, and was noflush */
35 bool must_flush; /* something wants a flush */
40 * Something that makes a journal entry dirty - i.e. a btree node that has to be
44 struct journal_entry_pin_list {
45 struct list_head list;
46 struct list_head key_cache_list;
47 struct list_head flushed;
49 struct bch_devs_list devs;
53 struct journal_entry_pin;
54 typedef int (*journal_pin_flush_fn)(struct journal *j,
55 struct journal_entry_pin *, u64);
57 struct journal_entry_pin {
58 struct list_head list;
59 journal_pin_flush_fn flush;
72 * For reserving space in the journal prior to getting a reservation on a
73 * particular journal entry:
75 struct journal_preres {
79 union journal_res_state {
89 u64 cur_entry_offset:20,
99 union journal_preres_state {
116 #define JOURNAL_ENTRY_SIZE_MIN (64U << 10) /* 64k */
117 #define JOURNAL_ENTRY_SIZE_MAX (4U << 20) /* 4M */
120 * We stash some journal state as sentinal values in cur_entry_offset:
121 * note - cur_entry_offset is in units of u64s
123 #define JOURNAL_ENTRY_OFFSET_MAX ((1U << 20) - 1)
125 #define JOURNAL_ENTRY_CLOSED_VAL (JOURNAL_ENTRY_OFFSET_MAX - 1)
126 #define JOURNAL_ENTRY_ERROR_VAL (JOURNAL_ENTRY_OFFSET_MAX)
128 struct journal_space {
129 /* Units of 512 bytes sectors: */
130 unsigned next_entry; /* How big the next journal entry can be */
134 enum journal_space_from {
135 journal_space_discarded,
136 journal_space_clean_ondisk,
143 * JOURNAL_NEED_WRITE - current (pending) journal entry should be written ASAP,
144 * either because something's waiting on the write to complete or because it's
145 * been dirty too long and the timer's expired.
151 JOURNAL_RECLAIM_STARTED,
153 JOURNAL_MAY_GET_UNRESERVED,
154 JOURNAL_MAY_SKIP_FLUSH,
158 /* Embedded in struct bch_fs */
160 /* Fastpath stuff up front: */
164 union journal_res_state reservations;
166 /* Max size of current journal entry */
167 unsigned cur_entry_u64s;
168 unsigned cur_entry_sectors;
171 * 0, or -ENOSPC if waiting on journal reclaim, or -EROFS if
172 * insufficient devices:
177 cur_entry_journal_full,
178 cur_entry_journal_pin_full,
179 cur_entry_journal_stuck,
180 cur_entry_insufficient_devices,
183 union journal_preres_state prereserved;
185 /* Reserved space in journal entry to be used just prior to write */
186 unsigned entry_u64s_reserved;
188 unsigned buf_size_want;
191 * Two journal entries -- one is currently open for new entries, the
192 * other is possibly being written out.
194 struct journal_buf buf[JOURNAL_BUF_NR];
198 /* if nonzero, we may not open a new journal entry: */
201 /* Used when waiting because the journal was full */
202 wait_queue_head_t wait;
203 struct closure_waitlist async_wait;
204 struct closure_waitlist preres_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;
249 wait_queue_head_t reclaim_wait;
250 struct task_struct *reclaim_thread;
252 unsigned long next_reclaim;
253 u64 nr_direct_reclaim;
254 u64 nr_background_reclaim;
256 unsigned long last_flushed;
257 struct journal_entry_pin *flush_in_progress;
258 bool flush_in_progress_dropped;
259 wait_queue_head_t pin_flush_wait;
261 /* protects advancing ja->discard_idx: */
262 struct mutex discard_lock;
265 unsigned write_delay_ms;
266 unsigned reclaim_delay_ms;
267 unsigned long last_flush_write;
269 u64 res_get_blocked_start;
271 u64 write_start_time;
274 u64 nr_noflush_writes;
276 struct time_stats *write_time;
277 struct time_stats *delay_time;
278 struct time_stats *blocked_time;
279 struct time_stats *flush_seq_time;
281 #ifdef CONFIG_DEBUG_LOCK_ALLOC
282 struct lockdep_map res_map;
287 * Embedded in struct bch_dev. First three fields refer to the array of journal
288 * buckets, in bch_sb.
290 struct journal_device {
292 * For each journal bucket, contains the max sequence number of the
293 * journal writes it contains - so we know when a bucket can be reused.
297 unsigned sectors_free;
300 * discard_idx <= dirty_idx_ondisk <= dirty_idx <= cur_idx:
302 unsigned discard_idx; /* Next bucket to discard */
303 unsigned dirty_idx_ondisk;
305 unsigned cur_idx; /* Journal bucket we're currently writing to */
310 /* Bio for journal reads/writes to this device */
313 /* for bch_journal_read_device */
318 * journal_entry_res - reserve space in every journal entry:
320 struct journal_entry_res {
324 #endif /* _BCACHEFS_JOURNAL_TYPES_H */