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                size;
  25         unsigned                disk_sectors;
  26         unsigned                u64s_reserved;
  27         /* bloom filter: */
  28         unsigned long           has_inode[1024 / sizeof(unsigned long)];
  29 };
  30
  31 /*
  32  * Something that makes a journal entry dirty - i.e. a btree node that has to be
  33  * flushed:
  34  */
  35
  36 struct journal_entry_pin_list {
  37         struct list_head                list;
  38         struct list_head                flushed;
  39         atomic_t                        count;
  40         struct bch_devs_list            devs;
  41 };
  42
  43 struct journal;
  44 struct journal_entry_pin;
  45 typedef void (*journal_pin_flush_fn)(struct journal *j,
  46                                 struct journal_entry_pin *, u64);
  47
  48 struct journal_entry_pin {
  49         struct list_head                list;
  50         journal_pin_flush_fn            flush;
  51         u64                             seq;
  52 };
  53
  54 /* corresponds to a btree node with a blacklisted bset: */
  55 struct blacklisted_node {
  56         __le64                  seq;
  57         enum btree_id           btree_id;
  58         struct bpos             pos;
  59 };
  60
  61 struct journal_seq_blacklist {
  62         struct list_head        list;
  63         u64                     start;
  64         u64                     end;
  65
  66         struct journal_entry_pin pin;
  67
  68         struct blacklisted_node *entries;
  69         size_t                  nr_entries;
  70 };
  71
  72 struct journal_res {
  73         bool                    ref;
  74         u8                      idx;
  75         u16                     u64s;
  76         u32                     offset;
  77         u64                     seq;
  78 };
  79
  80 union 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:1,
  92                                 prev_buf_unwritten:1,
  93                                 buf0_count:21,
  94                                 buf1_count:21;
  95         };
  96 };
  97
  98 /* bytes: */
  99 #define JOURNAL_ENTRY_SIZE_MIN          (64U << 10) /* 64k */
 100 #define JOURNAL_ENTRY_SIZE_MAX          (4U  << 20) /* 4M */
 101
 102 /*
 103  * We stash some journal state as sentinal values in cur_entry_offset:
 104  * note - cur_entry_offset is in units of u64s
 105  */
 106 #define JOURNAL_ENTRY_OFFSET_MAX        ((1U << 20) - 1)
 107
 108 #define JOURNAL_ENTRY_CLOSED_VAL        (JOURNAL_ENTRY_OFFSET_MAX - 1)
 109 #define JOURNAL_ENTRY_ERROR_VAL         (JOURNAL_ENTRY_OFFSET_MAX)
 110
 111 /*
 112  * JOURNAL_NEED_WRITE - current (pending) journal entry should be written ASAP,
 113  * either because something's waiting on the write to complete or because it's
 114  * been dirty too long and the timer's expired.
 115  */
 116
 117 enum {
 118         JOURNAL_REPLAY_DONE,
 119         JOURNAL_STARTED,
 120         JOURNAL_NEED_WRITE,
 121         JOURNAL_NOT_EMPTY,
 122 };
 123
 124 /* Embedded in struct bch_fs */
 125 struct journal {
 126         /* Fastpath stuff up front: */
 127
 128         unsigned long           flags;
 129
 130         union journal_res_state reservations;
 131         unsigned                cur_entry_u64s;
 132         unsigned                prev_buf_sectors;
 133         unsigned                cur_buf_sectors;
 134         unsigned                buf_size_want;
 135
 136         /*
 137          * Two journal entries -- one is currently open for new entries, the
 138          * other is possibly being written out.
 139          */
 140         struct journal_buf      buf[2];
 141
 142         spinlock_t              lock;
 143
 144         /* Used when waiting because the journal was full */
 145         wait_queue_head_t       wait;
 146         struct closure_waitlist async_wait;
 147
 148         struct closure          io;
 149         struct delayed_work     write_work;
 150
 151         /* Sequence number of most recent journal entry (last entry in @pin) */
 152         atomic64_t              seq;
 153
 154         /* seq, last_seq from the most recent journal entry successfully written */
 155         u64                     seq_ondisk;
 156         u64                     last_seq_ondisk;
 157
 158         /* Reserved space in journal entry to be used just prior to write */
 159         unsigned                entry_u64s_reserved;
 160
 161         /*
 162          * FIFO of journal entries whose btree updates have not yet been
 163          * written out.
 164          *
 165          * Each entry is a reference count. The position in the FIFO is the
 166          * entry's sequence number relative to @seq.
 167          *
 168          * The journal entry itself holds a reference count, put when the
 169          * journal entry is written out. Each btree node modified by the journal
 170          * entry also holds a reference count, put when the btree node is
 171          * written.
 172          *
 173          * When a reference count reaches zero, the journal entry is no longer
 174          * needed. When all journal entries in the oldest journal bucket are no
 175          * longer needed, the bucket can be discarded and reused.
 176          */
 177         struct {
 178                 u64 front, back, size, mask;
 179                 struct journal_entry_pin_list *data;
 180         }                       pin;
 181
 182         struct journal_entry_pin *flush_in_progress;
 183         wait_queue_head_t       pin_flush_wait;
 184
 185         u64                     replay_journal_seq;
 186
 187         struct mutex            blacklist_lock;
 188         struct list_head        seq_blacklist;
 189         struct journal_seq_blacklist *new_blacklist;
 190
 191         struct write_point      wp;
 192         spinlock_t              err_lock;
 193
 194         struct delayed_work     reclaim_work;
 195         unsigned long           last_flushed;
 196
 197         /* protects advancing ja->last_idx: */
 198         struct mutex            reclaim_lock;
 199         unsigned                write_delay_ms;
 200         unsigned                reclaim_delay_ms;
 201
 202         u64                     res_get_blocked_start;
 203         u64                     need_write_time;
 204         u64                     write_start_time;
 205
 206         struct time_stats       *write_time;
 207         struct time_stats       *delay_time;
 208         struct time_stats       *blocked_time;
 209         struct time_stats       *flush_seq_time;
 210
 211 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 212         struct lockdep_map      res_map;
 213 #endif
 214 };
 215
 216 /*
 217  * Embedded in struct bch_dev. First three fields refer to the array of journal
 218  * buckets, in bch_sb.
 219  */
 220 struct journal_device {
 221         /*
 222          * For each journal bucket, contains the max sequence number of the
 223          * journal writes it contains - so we know when a bucket can be reused.
 224          */
 225         u64                     *bucket_seq;
 226
 227         unsigned                sectors_free;
 228
 229         /* Journal bucket we're currently writing to */
 230         unsigned                cur_idx;
 231
 232         /* Last journal bucket that still contains an open journal entry */
 233
 234         /*
 235          * j->lock and j->reclaim_lock must both be held to modify, j->lock
 236          * sufficient to read:
 237          */
 238         unsigned                last_idx;
 239         unsigned                nr;
 240         u64                     *buckets;
 241
 242         /* Bio for journal reads/writes to this device */
 243         struct bio              *bio;
 244
 245         /* for bch_journal_read_device */
 246         struct closure          read;
 247 };
 248
 249 /*
 250  * journal_entry_res - reserve space in every journal entry:
 251  */
 252 struct journal_entry_res {
 253         unsigned                u64s;
 254 };
 255
 256 #endif /* _BCACHEFS_JOURNAL_TYPES_H */