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 /* corresponds to a btree node with a blacklisted bset: */
  57 struct blacklisted_node {
  58         __le64                  seq;
  59         enum btree_id           btree_id;
  60         struct bpos             pos;
  61 };
  62
  63 struct journal_seq_blacklist {
  64         struct list_head        list;
  65         u64                     start;
  66         u64                     end;
  67
  68         struct journal_entry_pin pin;
  69
  70         struct blacklisted_node *entries;
  71         size_t                  nr_entries;
  72 };
  73
  74 struct journal_res {
  75         bool                    ref;
  76         u8                      idx;
  77         u16                     u64s;
  78         u32                     offset;
  79         u64                     seq;
  80 };
  81
  82 /*
  83  * For reserving space in the journal prior to getting a reservation on a
  84  * particular journal entry:
  85  */
  86 struct journal_preres {
  87         unsigned                u64s;
  88 };
  89
  90 union journal_res_state {
  91         struct {
  92                 atomic64_t      counter;
  93         };
  94
  95         struct {
  96                 u64             v;
  97         };
  98
  99         struct {
 100                 u64             cur_entry_offset:20,
 101                                 idx:1,
 102                                 prev_buf_unwritten:1,
 103                                 buf0_count:21,
 104                                 buf1_count:21;
 105         };
 106 };
 107
 108 union journal_preres_state {
 109         struct {
 110                 atomic64_t      counter;
 111         };
 112
 113         struct {
 114                 u64             v;
 115         };
 116
 117         struct {
 118                 u32             reserved;
 119                 u32             remaining;
 120         };
 121 };
 122
 123 /* bytes: */
 124 #define JOURNAL_ENTRY_SIZE_MIN          (64U << 10) /* 64k */
 125 #define JOURNAL_ENTRY_SIZE_MAX          (4U  << 20) /* 4M */
 126
 127 /*
 128  * We stash some journal state as sentinal values in cur_entry_offset:
 129  * note - cur_entry_offset is in units of u64s
 130  */
 131 #define JOURNAL_ENTRY_OFFSET_MAX        ((1U << 20) - 1)
 132
 133 #define JOURNAL_ENTRY_CLOSED_VAL        (JOURNAL_ENTRY_OFFSET_MAX - 1)
 134 #define JOURNAL_ENTRY_ERROR_VAL         (JOURNAL_ENTRY_OFFSET_MAX)
 135
 136 /*
 137  * JOURNAL_NEED_WRITE - current (pending) journal entry should be written ASAP,
 138  * either because something's waiting on the write to complete or because it's
 139  * been dirty too long and the timer's expired.
 140  */
 141
 142 enum {
 143         JOURNAL_REPLAY_DONE,
 144         JOURNAL_STARTED,
 145         JOURNAL_NEED_WRITE,
 146         JOURNAL_NOT_EMPTY,
 147         JOURNAL_MAY_GET_UNRESERVED,
 148 };
 149
 150 /* Embedded in struct bch_fs */
 151 struct journal {
 152         /* Fastpath stuff up front: */
 153
 154         unsigned long           flags;
 155
 156         union journal_res_state reservations;
 157
 158         /* Max size of current journal entry */
 159         unsigned                cur_entry_u64s;
 160         unsigned                cur_entry_sectors;
 161
 162         /*
 163          * 0, or -ENOSPC if waiting on journal reclaim, or -EROFS if
 164          * insufficient devices:
 165          */
 166         int                     cur_entry_error;
 167
 168         union journal_preres_state prereserved;
 169
 170         /* Reserved space in journal entry to be used just prior to write */
 171         unsigned                entry_u64s_reserved;
 172
 173         unsigned                buf_size_want;
 174
 175         /*
 176          * Two journal entries -- one is currently open for new entries, the
 177          * other is possibly being written out.
 178          */
 179         struct journal_buf      buf[2];
 180
 181         spinlock_t              lock;
 182
 183         /* if nonzero, we may not open a new journal entry: */
 184         unsigned                blocked;
 185
 186         /* Used when waiting because the journal was full */
 187         wait_queue_head_t       wait;
 188         struct closure_waitlist async_wait;
 189         struct closure_waitlist preres_wait;
 190
 191         struct closure          io;
 192         struct delayed_work     write_work;
 193
 194         /* Sequence number of most recent journal entry (last entry in @pin) */
 195         atomic64_t              seq;
 196
 197         /* seq, last_seq from the most recent journal entry successfully written */
 198         u64                     seq_ondisk;
 199         u64                     last_seq_ondisk;
 200
 201         /*
 202          * FIFO of journal entries whose btree updates have not yet been
 203          * written out.
 204          *
 205          * Each entry is a reference count. The position in the FIFO is the
 206          * entry's sequence number relative to @seq.
 207          *
 208          * The journal entry itself holds a reference count, put when the
 209          * journal entry is written out. Each btree node modified by the journal
 210          * entry also holds a reference count, put when the btree node is
 211          * written.
 212          *
 213          * When a reference count reaches zero, the journal entry is no longer
 214          * needed. When all journal entries in the oldest journal bucket are no
 215          * longer needed, the bucket can be discarded and reused.
 216          */
 217         struct {
 218                 u64 front, back, size, mask;
 219                 struct journal_entry_pin_list *data;
 220         }                       pin;
 221
 222         u64                     replay_journal_seq;
 223
 224         struct mutex            blacklist_lock;
 225         struct list_head        seq_blacklist;
 226         struct journal_seq_blacklist *new_blacklist;
 227
 228         struct write_point      wp;
 229         spinlock_t              err_lock;
 230
 231         struct delayed_work     reclaim_work;
 232         struct mutex            reclaim_lock;
 233         unsigned long           last_flushed;
 234         struct journal_entry_pin *flush_in_progress;
 235         wait_queue_head_t       pin_flush_wait;
 236
 237         /* protects advancing ja->discard_idx: */
 238         struct mutex            discard_lock;
 239         bool                    can_discard;
 240
 241         unsigned                write_delay_ms;
 242         unsigned                reclaim_delay_ms;
 243
 244         u64                     res_get_blocked_start;
 245         u64                     need_write_time;
 246         u64                     write_start_time;
 247
 248         struct time_stats       *write_time;
 249         struct time_stats       *delay_time;
 250         struct time_stats       *blocked_time;
 251         struct time_stats       *flush_seq_time;
 252
 253 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 254         struct lockdep_map      res_map;
 255 #endif
 256 };
 257
 258 /*
 259  * Embedded in struct bch_dev. First three fields refer to the array of journal
 260  * buckets, in bch_sb.
 261  */
 262 struct journal_device {
 263         /*
 264          * For each journal bucket, contains the max sequence number of the
 265          * journal writes it contains - so we know when a bucket can be reused.
 266          */
 267         u64                     *bucket_seq;
 268
 269         unsigned                sectors_free;
 270
 271         /*
 272          * discard_idx <= dirty_idx_ondisk <= dirty_idx <= cur_idx:
 273          */
 274         unsigned                discard_idx;            /* Next bucket to discard */
 275         unsigned                dirty_idx_ondisk;
 276         unsigned                dirty_idx;
 277         unsigned                cur_idx;                /* Journal bucket we're currently writing to */
 278         unsigned                nr;
 279
 280         u64                     *buckets;
 281
 282         /* Bio for journal reads/writes to this device */
 283         struct bio              *bio;
 284
 285         /* for bch_journal_read_device */
 286         struct closure          read;
 287 };
 288
 289 /*
 290  * journal_entry_res - reserve space in every journal entry:
 291  */
 292 struct journal_entry_res {
 293         unsigned                u64s;
 294 };
 295
 296 #endif /* _BCACHEFS_JOURNAL_TYPES_H */