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         /* bloom filter: */
  27         unsigned long           has_inode[1024 / sizeof(unsigned long)];
  28 };
  29
  30 /*
  31  * Something that makes a journal entry dirty - i.e. a btree node that has to be
  32  * flushed:
  33  */
  34
  35 struct journal_entry_pin_list {
  36         struct list_head                list;
  37         struct list_head                flushed;
  38         atomic_t                        count;
  39         struct bch_devs_list            devs;
  40 };
  41
  42 struct journal;
  43 struct journal_entry_pin;
  44 typedef void (*journal_pin_flush_fn)(struct journal *j,
  45                                 struct journal_entry_pin *, u64);
  46
  47 struct journal_entry_pin {
  48         struct list_head                list;
  49         journal_pin_flush_fn            flush;
  50         struct journal_entry_pin_list   *pin_list;
  51 };
  52
  53 /* corresponds to a btree node with a blacklisted bset: */
  54 struct blacklisted_node {
  55         __le64                  seq;
  56         enum btree_id           btree_id;
  57         struct bpos             pos;
  58 };
  59
  60 struct journal_seq_blacklist {
  61         struct list_head        list;
  62         u64                     start;
  63         u64                     end;
  64
  65         struct journal_entry_pin pin;
  66
  67         struct blacklisted_node *entries;
  68         size_t                  nr_entries;
  69 };
  70
  71 struct journal_res {
  72         bool                    ref;
  73         u8                      idx;
  74         u16                     u64s;
  75         u32                     offset;
  76         u64                     seq;
  77 };
  78
  79 union journal_res_state {
  80         struct {
  81                 atomic64_t      counter;
  82         };
  83
  84         struct {
  85                 u64             v;
  86         };
  87
  88         struct {
  89                 u64             cur_entry_offset:20,
  90                                 idx:1,
  91                                 prev_buf_unwritten:1,
  92                                 buf0_count:21,
  93                                 buf1_count:21;
  94         };
  95 };
  96
  97 /* bytes: */
  98 #define JOURNAL_ENTRY_SIZE_MIN          (64U << 10) /* 64k */
  99 #define JOURNAL_ENTRY_SIZE_MAX          (4U  << 20) /* 4M */
 100
 101 /*
 102  * We stash some journal state as sentinal values in cur_entry_offset:
 103  * note - cur_entry_offset is in units of u64s
 104  */
 105 #define JOURNAL_ENTRY_OFFSET_MAX        ((1U << 20) - 1)
 106
 107 #define JOURNAL_ENTRY_CLOSED_VAL        (JOURNAL_ENTRY_OFFSET_MAX - 1)
 108 #define JOURNAL_ENTRY_ERROR_VAL         (JOURNAL_ENTRY_OFFSET_MAX)
 109
 110 /*
 111  * JOURNAL_NEED_WRITE - current (pending) journal entry should be written ASAP,
 112  * either because something's waiting on the write to complete or because it's
 113  * been dirty too long and the timer's expired.
 114  */
 115
 116 enum {
 117         JOURNAL_REPLAY_DONE,
 118         JOURNAL_STARTED,
 119         JOURNAL_NEED_WRITE,
 120 };
 121
 122 /* Embedded in struct bch_fs */
 123 struct journal {
 124         /* Fastpath stuff up front: */
 125
 126         unsigned long           flags;
 127
 128         union journal_res_state reservations;
 129         unsigned                cur_entry_u64s;
 130         unsigned                prev_buf_sectors;
 131         unsigned                cur_buf_sectors;
 132         unsigned                buf_size_want;
 133
 134         /*
 135          * Two journal entries -- one is currently open for new entries, the
 136          * other is possibly being written out.
 137          */
 138         struct journal_buf      buf[2];
 139
 140         spinlock_t              lock;
 141
 142         /* Used when waiting because the journal was full */
 143         wait_queue_head_t       wait;
 144         struct closure_waitlist async_wait;
 145
 146         struct closure          io;
 147         struct delayed_work     write_work;
 148
 149         /* Sequence number of most recent journal entry (last entry in @pin) */
 150         atomic64_t              seq;
 151
 152         /* last_seq from the most recent journal entry written */
 153         u64                     last_seq_ondisk;
 154
 155         /*
 156          * FIFO of journal entries whose btree updates have not yet been
 157          * written out.
 158          *
 159          * Each entry is a reference count. The position in the FIFO is the
 160          * entry's sequence number relative to @seq.
 161          *
 162          * The journal entry itself holds a reference count, put when the
 163          * journal entry is written out. Each btree node modified by the journal
 164          * entry also holds a reference count, put when the btree node is
 165          * written.
 166          *
 167          * When a reference count reaches zero, the journal entry is no longer
 168          * needed. When all journal entries in the oldest journal bucket are no
 169          * longer needed, the bucket can be discarded and reused.
 170          */
 171         struct {
 172                 u64 front, back, size, mask;
 173                 struct journal_entry_pin_list *data;
 174         }                       pin;
 175         u64                     replay_journal_seq;
 176
 177         struct mutex            blacklist_lock;
 178         struct list_head        seq_blacklist;
 179         struct journal_seq_blacklist *new_blacklist;
 180
 181         BKEY_PADDED(key);
 182         struct write_point      wp;
 183         spinlock_t              err_lock;
 184
 185         struct delayed_work     reclaim_work;
 186         unsigned long           last_flushed;
 187
 188         /* protects advancing ja->last_idx: */
 189         struct mutex            reclaim_lock;
 190         unsigned                write_delay_ms;
 191         unsigned                reclaim_delay_ms;
 192
 193         u64                     res_get_blocked_start;
 194         u64                     need_write_time;
 195         u64                     write_start_time;
 196
 197         struct time_stats       *write_time;
 198         struct time_stats       *delay_time;
 199         struct time_stats       *blocked_time;
 200         struct time_stats       *flush_seq_time;
 201
 202 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 203         struct lockdep_map      res_map;
 204 #endif
 205 };
 206
 207 /*
 208  * Embedded in struct bch_dev. First three fields refer to the array of journal
 209  * buckets, in bch_sb.
 210  */
 211 struct journal_device {
 212         /*
 213          * For each journal bucket, contains the max sequence number of the
 214          * journal writes it contains - so we know when a bucket can be reused.
 215          */
 216         u64                     *bucket_seq;
 217
 218         unsigned                sectors_free;
 219
 220         /* Journal bucket we're currently writing to */
 221         unsigned                cur_idx;
 222
 223         /* Last journal bucket that still contains an open journal entry */
 224
 225         /*
 226          * j->lock and j->reclaim_lock must both be held to modify, j->lock
 227          * sufficient to read:
 228          */
 229         unsigned                last_idx;
 230         unsigned                nr;
 231         u64                     *buckets;
 232
 233         /* Bio for journal reads/writes to this device */
 234         struct bio              *bio;
 235
 236         /* for bch_journal_read_device */
 237         struct closure          read;
 238 };
 239
 240 #endif /* _BCACHEFS_JOURNAL_TYPES_H */