git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/journal_types.h

   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 struct journal_res;
  13
  14 /*
  15  * We put two of these in struct journal; we used them for writes to the
  16  * journal that are being staged or in flight.
  17  */
  18 struct journal_buf {
  19         struct jset             *data;
  20
  21         BKEY_PADDED(key);
  22
  23         struct closure_waitlist wait;
  24
  25         unsigned                buf_size;       /* size in bytes of @data */
  26         unsigned                sectors;        /* maximum size for current entry */
  27         unsigned                disk_sectors;   /* maximum size entry could have been, if
  28                                                    buf_size was bigger */
  29         unsigned                u64s_reserved;
  30         /* bloom filter: */
  31         unsigned long           has_inode[1024 / sizeof(unsigned long)];
  32 };
  33
  34 /*
  35  * Something that makes a journal entry dirty - i.e. a btree node that has to be
  36  * flushed:
  37  */
  38
  39 struct journal_entry_pin_list {
  40         struct list_head                list;
  41         struct list_head                flushed;
  42         atomic_t                        count;
  43         struct bch_devs_list            devs;
  44 };
  45
  46 struct journal;
  47 struct journal_entry_pin;
  48 typedef void (*journal_pin_flush_fn)(struct journal *j,
  49                                 struct journal_entry_pin *, u64);
  50
  51 struct journal_entry_pin {
  52         struct list_head                list;
  53         journal_pin_flush_fn            flush;
  54         u64                             seq;
  55 };
  56
  57 struct journal_res {
  58         bool                    ref;
  59         u8                      idx;
  60         u16                     u64s;
  61         u32                     offset;
  62         u64                     seq;
  63 };
  64
  65 /*
  66  * For reserving space in the journal prior to getting a reservation on a
  67  * particular journal entry:
  68  */
  69 struct journal_preres {
  70         unsigned                u64s;
  71 };
  72
  73 union journal_res_state {
  74         struct {
  75                 atomic64_t      counter;
  76         };
  77
  78         struct {
  79                 u64             v;
  80         };
  81
  82         struct {
  83                 u64             cur_entry_offset:20,
  84                                 idx:1,
  85                                 prev_buf_unwritten:1,
  86                                 buf0_count:21,
  87                                 buf1_count:21;
  88         };
  89 };
  90
  91 union journal_preres_state {
  92         struct {
  93                 atomic64_t      counter;
  94         };
  95
  96         struct {
  97                 u64             v;
  98         };
  99
 100         struct {
 101                 u32             reserved;
 102                 u32             remaining;
 103         };
 104 };
 105
 106 /* bytes: */
 107 #define JOURNAL_ENTRY_SIZE_MIN          (64U << 10) /* 64k */
 108 #define JOURNAL_ENTRY_SIZE_MAX          (4U  << 20) /* 4M */
 109
 110 /*
 111  * We stash some journal state as sentinal values in cur_entry_offset:
 112  * note - cur_entry_offset is in units of u64s
 113  */
 114 #define JOURNAL_ENTRY_OFFSET_MAX        ((1U << 20) - 1)
 115
 116 #define JOURNAL_ENTRY_CLOSED_VAL        (JOURNAL_ENTRY_OFFSET_MAX - 1)
 117 #define JOURNAL_ENTRY_ERROR_VAL         (JOURNAL_ENTRY_OFFSET_MAX)
 118
 119 /*
 120  * JOURNAL_NEED_WRITE - current (pending) journal entry should be written ASAP,
 121  * either because something's waiting on the write to complete or because it's
 122  * been dirty too long and the timer's expired.
 123  */
 124
 125 enum {
 126         JOURNAL_REPLAY_DONE,
 127         JOURNAL_STARTED,
 128         JOURNAL_RECLAIM_STARTED,
 129         JOURNAL_NEED_WRITE,
 130         JOURNAL_MAY_GET_UNRESERVED,
 131 };
 132
 133 /* Embedded in struct bch_fs */
 134 struct journal {
 135         /* Fastpath stuff up front: */
 136
 137         unsigned long           flags;
 138
 139         union journal_res_state reservations;
 140
 141         /* Max size of current journal entry */
 142         unsigned                cur_entry_u64s;
 143         unsigned                cur_entry_sectors;
 144
 145         /*
 146          * 0, or -ENOSPC if waiting on journal reclaim, or -EROFS if
 147          * insufficient devices:
 148          */
 149         int                     cur_entry_error;
 150
 151         union journal_preres_state prereserved;
 152
 153         /* Reserved space in journal entry to be used just prior to write */
 154         unsigned                entry_u64s_reserved;
 155
 156         unsigned                buf_size_want;
 157
 158         /*
 159          * Two journal entries -- one is currently open for new entries, the
 160          * other is possibly being written out.
 161          */
 162         struct journal_buf      buf[2];
 163
 164         spinlock_t              lock;
 165
 166         /* if nonzero, we may not open a new journal entry: */
 167         unsigned                blocked;
 168
 169         /* Used when waiting because the journal was full */
 170         wait_queue_head_t       wait;
 171         struct closure_waitlist async_wait;
 172         struct closure_waitlist preres_wait;
 173
 174         struct closure          io;
 175         struct delayed_work     write_work;
 176
 177         /* Sequence number of most recent journal entry (last entry in @pin) */
 178         atomic64_t              seq;
 179
 180         /* seq, last_seq from the most recent journal entry successfully written */
 181         u64                     seq_ondisk;
 182         u64                     last_seq_ondisk;
 183         u64                     err_seq;
 184         u64                     last_empty_seq;
 185
 186         /*
 187          * FIFO of journal entries whose btree updates have not yet been
 188          * written out.
 189          *
 190          * Each entry is a reference count. The position in the FIFO is the
 191          * entry's sequence number relative to @seq.
 192          *
 193          * The journal entry itself holds a reference count, put when the
 194          * journal entry is written out. Each btree node modified by the journal
 195          * entry also holds a reference count, put when the btree node is
 196          * written.
 197          *
 198          * When a reference count reaches zero, the journal entry is no longer
 199          * needed. When all journal entries in the oldest journal bucket are no
 200          * longer needed, the bucket can be discarded and reused.
 201          */
 202         struct {
 203                 u64 front, back, size, mask;
 204                 struct journal_entry_pin_list *data;
 205         }                       pin;
 206
 207         u64                     replay_journal_seq;
 208         u64                     replay_journal_seq_end;
 209
 210         struct write_point      wp;
 211         spinlock_t              err_lock;
 212
 213         struct delayed_work     reclaim_work;
 214         struct mutex            reclaim_lock;
 215         unsigned long           last_flushed;
 216         struct journal_entry_pin *flush_in_progress;
 217         wait_queue_head_t       pin_flush_wait;
 218
 219         /* protects advancing ja->discard_idx: */
 220         struct mutex            discard_lock;
 221         bool                    can_discard;
 222
 223         unsigned                write_delay_ms;
 224         unsigned                reclaim_delay_ms;
 225
 226         u64                     res_get_blocked_start;
 227         u64                     need_write_time;
 228         u64                     write_start_time;
 229
 230         struct time_stats       *write_time;
 231         struct time_stats       *delay_time;
 232         struct time_stats       *blocked_time;
 233         struct time_stats       *flush_seq_time;
 234
 235 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 236         struct lockdep_map      res_map;
 237 #endif
 238 };
 239
 240 /*
 241  * Embedded in struct bch_dev. First three fields refer to the array of journal
 242  * buckets, in bch_sb.
 243  */
 244 struct journal_device {
 245         /*
 246          * For each journal bucket, contains the max sequence number of the
 247          * journal writes it contains - so we know when a bucket can be reused.
 248          */
 249         u64                     *bucket_seq;
 250
 251         unsigned                sectors_free;
 252
 253         /*
 254          * discard_idx <= dirty_idx_ondisk <= dirty_idx <= cur_idx:
 255          */
 256         unsigned                discard_idx;            /* Next bucket to discard */
 257         unsigned                dirty_idx_ondisk;
 258         unsigned                dirty_idx;
 259         unsigned                cur_idx;                /* Journal bucket we're currently writing to */
 260         unsigned                nr;
 261
 262         u64                     *buckets;
 263
 264         /* Bio for journal reads/writes to this device */
 265         struct bio              *bio;
 266
 267         /* for bch_journal_read_device */
 268         struct closure          read;
 269 };
 270
 271 /*
 272  * journal_entry_res - reserve space in every journal entry:
 273  */
 274 struct journal_entry_res {
 275         unsigned                u64s;
 276 };
 277
 278 #endif /* _BCACHEFS_JOURNAL_TYPES_H */