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

   1 #ifndef _BCACHEFS_JOURNAL_TYPES_H
   2 #define _BCACHEFS_JOURNAL_TYPES_H
   3
   4 #include <linux/cache.h>
   5 #include <linux/workqueue.h>
   6
   7 #include "alloc_types.h"
   8 #include "super_types.h"
   9 #include "fifo.h"
  10
  11 struct journal_res;
  12
  13 /*
  14  * We put two of these in struct journal; we used them for writes to the
  15  * journal that are being staged or in flight.
  16  */
  17 struct journal_buf {
  18         struct jset             *data;
  19
  20         BKEY_PADDED(key);
  21
  22         struct closure_waitlist wait;
  23
  24         unsigned                buf_size;       /* size in bytes of @data */
  25         unsigned                sectors;        /* maximum size for current entry */
  26         unsigned                disk_sectors;   /* maximum size entry could have been, if
  27                                                    buf_size was bigger */
  28         unsigned                u64s_reserved;
  29         /* bloom filter: */
  30         unsigned long           has_inode[1024 / sizeof(unsigned long)];
  31 };
  32
  33 /*
  34  * Something that makes a journal entry dirty - i.e. a btree node that has to be
  35  * flushed:
  36  */
  37
  38 struct journal_entry_pin_list {
  39         struct list_head                list;
  40         struct list_head                flushed;
  41         atomic_t                        count;
  42         struct bch_devs_list            devs;
  43 };
  44
  45 struct journal;
  46 struct journal_entry_pin;
  47 typedef void (*journal_pin_flush_fn)(struct journal *j,
  48                                 struct journal_entry_pin *, u64);
  49
  50 struct journal_entry_pin {
  51         struct list_head                list;
  52         journal_pin_flush_fn            flush;
  53         u64                             seq;
  54 };
  55
  56 struct journal_res {
  57         bool                    ref;
  58         u8                      idx;
  59         u16                     u64s;
  60         u32                     offset;
  61         u64                     seq;
  62 };
  63
  64 /*
  65  * For reserving space in the journal prior to getting a reservation on a
  66  * particular journal entry:
  67  */
  68 struct journal_preres {
  69         unsigned                u64s;
  70 };
  71
  72 union journal_res_state {
  73         struct {
  74                 atomic64_t      counter;
  75         };
  76
  77         struct {
  78                 u64             v;
  79         };
  80
  81         struct {
  82                 u64             cur_entry_offset:20,
  83                                 idx:1,
  84                                 prev_buf_unwritten:1,
  85                                 buf0_count:21,
  86                                 buf1_count:21;
  87         };
  88 };
  89
  90 union journal_preres_state {
  91         struct {
  92                 atomic64_t      counter;
  93         };
  94
  95         struct {
  96                 u64             v;
  97         };
  98
  99         struct {
 100                 u32             reserved;
 101                 u32             remaining;
 102         };
 103 };
 104
 105 /* bytes: */
 106 #define JOURNAL_ENTRY_SIZE_MIN          (64U << 10) /* 64k */
 107 #define JOURNAL_ENTRY_SIZE_MAX          (4U  << 20) /* 4M */
 108
 109 /*
 110  * We stash some journal state as sentinal values in cur_entry_offset:
 111  * note - cur_entry_offset is in units of u64s
 112  */
 113 #define JOURNAL_ENTRY_OFFSET_MAX        ((1U << 20) - 1)
 114
 115 #define JOURNAL_ENTRY_CLOSED_VAL        (JOURNAL_ENTRY_OFFSET_MAX - 1)
 116 #define JOURNAL_ENTRY_ERROR_VAL         (JOURNAL_ENTRY_OFFSET_MAX)
 117
 118 /*
 119  * JOURNAL_NEED_WRITE - current (pending) journal entry should be written ASAP,
 120  * either because something's waiting on the write to complete or because it's
 121  * been dirty too long and the timer's expired.
 122  */
 123
 124 enum {
 125         JOURNAL_REPLAY_DONE,
 126         JOURNAL_STARTED,
 127         JOURNAL_NEED_WRITE,
 128         JOURNAL_NOT_EMPTY,
 129         JOURNAL_MAY_GET_UNRESERVED,
 130 };
 131
 132 /* Embedded in struct bch_fs */
 133 struct journal {
 134         /* Fastpath stuff up front: */
 135
 136         unsigned long           flags;
 137
 138         union journal_res_state reservations;
 139
 140         /* Max size of current journal entry */
 141         unsigned                cur_entry_u64s;
 142         unsigned                cur_entry_sectors;
 143
 144         /*
 145          * 0, or -ENOSPC if waiting on journal reclaim, or -EROFS if
 146          * insufficient devices:
 147          */
 148         int                     cur_entry_error;
 149
 150         union journal_preres_state prereserved;
 151
 152         /* Reserved space in journal entry to be used just prior to write */
 153         unsigned                entry_u64s_reserved;
 154
 155         unsigned                buf_size_want;
 156
 157         /*
 158          * Two journal entries -- one is currently open for new entries, the
 159          * other is possibly being written out.
 160          */
 161         struct journal_buf      buf[2];
 162
 163         spinlock_t              lock;
 164
 165         /* if nonzero, we may not open a new journal entry: */
 166         unsigned                blocked;
 167
 168         /* Used when waiting because the journal was full */
 169         wait_queue_head_t       wait;
 170         struct closure_waitlist async_wait;
 171         struct closure_waitlist preres_wait;
 172
 173         struct closure          io;
 174         struct delayed_work     write_work;
 175
 176         /* Sequence number of most recent journal entry (last entry in @pin) */
 177         atomic64_t              seq;
 178
 179         /* seq, last_seq from the most recent journal entry successfully written */
 180         u64                     seq_ondisk;
 181         u64                     last_seq_ondisk;
 182
 183         /*
 184          * FIFO of journal entries whose btree updates have not yet been
 185          * written out.
 186          *
 187          * Each entry is a reference count. The position in the FIFO is the
 188          * entry's sequence number relative to @seq.
 189          *
 190          * The journal entry itself holds a reference count, put when the
 191          * journal entry is written out. Each btree node modified by the journal
 192          * entry also holds a reference count, put when the btree node is
 193          * written.
 194          *
 195          * When a reference count reaches zero, the journal entry is no longer
 196          * needed. When all journal entries in the oldest journal bucket are no
 197          * longer needed, the bucket can be discarded and reused.
 198          */
 199         struct {
 200                 u64 front, back, size, mask;
 201                 struct journal_entry_pin_list *data;
 202         }                       pin;
 203
 204         u64                     replay_journal_seq;
 205
 206         struct write_point      wp;
 207         spinlock_t              err_lock;
 208
 209         struct delayed_work     reclaim_work;
 210         struct mutex            reclaim_lock;
 211         unsigned long           last_flushed;
 212         struct journal_entry_pin *flush_in_progress;
 213         wait_queue_head_t       pin_flush_wait;
 214
 215         /* protects advancing ja->discard_idx: */
 216         struct mutex            discard_lock;
 217         bool                    can_discard;
 218
 219         unsigned                write_delay_ms;
 220         unsigned                reclaim_delay_ms;
 221
 222         u64                     res_get_blocked_start;
 223         u64                     need_write_time;
 224         u64                     write_start_time;
 225
 226         struct time_stats       *write_time;
 227         struct time_stats       *delay_time;
 228         struct time_stats       *blocked_time;
 229         struct time_stats       *flush_seq_time;
 230
 231 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 232         struct lockdep_map      res_map;
 233 #endif
 234 };
 235
 236 /*
 237  * Embedded in struct bch_dev. First three fields refer to the array of journal
 238  * buckets, in bch_sb.
 239  */
 240 struct journal_device {
 241         /*
 242          * For each journal bucket, contains the max sequence number of the
 243          * journal writes it contains - so we know when a bucket can be reused.
 244          */
 245         u64                     *bucket_seq;
 246
 247         unsigned                sectors_free;
 248
 249         /*
 250          * discard_idx <= dirty_idx_ondisk <= dirty_idx <= cur_idx:
 251          */
 252         unsigned                discard_idx;            /* Next bucket to discard */
 253         unsigned                dirty_idx_ondisk;
 254         unsigned                dirty_idx;
 255         unsigned                cur_idx;                /* Journal bucket we're currently writing to */
 256         unsigned                nr;
 257
 258         u64                     *buckets;
 259
 260         /* Bio for journal reads/writes to this device */
 261         struct bio              *bio;
 262
 263         /* for bch_journal_read_device */
 264         struct closure          read;
 265 };
 266
 267 /*
 268  * journal_entry_res - reserve space in every journal entry:
 269  */
 270 struct journal_entry_res {
 271         unsigned                u64s;
 272 };
 273
 274 #endif /* _BCACHEFS_JOURNAL_TYPES_H */