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

   1 #ifndef _BCACHE_JOURNAL_TYPES_H
   2 #define _BCACHE_JOURNAL_TYPES_H
   3
   4 #include <linux/cache.h>
   5 #include <linux/workqueue.h>
   6
   7 #include "alloc_types.h"
   8 #include "fifo.h"
   9
  10 struct journal_res;
  11
  12 /*
  13  * We put two of these in struct journal; we used them for writes to the
  14  * journal that are being staged or in flight.
  15  */
  16 struct journal_buf {
  17         struct jset             *data;
  18
  19         struct closure_waitlist wait;
  20
  21         unsigned                size;
  22         unsigned                disk_sectors;
  23         /* bloom filter: */
  24         unsigned long           has_inode[1024 / sizeof(unsigned long)];
  25 };
  26
  27 /*
  28  * Something that makes a journal entry dirty - i.e. a btree node that has to be
  29  * flushed:
  30  */
  31
  32 struct journal_entry_pin_list {
  33         struct list_head                list;
  34         struct list_head                flushed;
  35         atomic_t                        count;
  36 };
  37
  38 struct journal;
  39 struct journal_entry_pin;
  40 typedef void (*journal_pin_flush_fn)(struct journal *j,
  41                                 struct journal_entry_pin *, u64);
  42
  43 struct journal_entry_pin {
  44         struct list_head                list;
  45         journal_pin_flush_fn            flush;
  46         struct journal_entry_pin_list   *pin_list;
  47 };
  48
  49 /* corresponds to a btree node with a blacklisted bset: */
  50 struct blacklisted_node {
  51         __le64                  seq;
  52         enum btree_id           btree_id;
  53         struct bpos             pos;
  54 };
  55
  56 struct journal_seq_blacklist {
  57         struct list_head        list;
  58         u64                     seq;
  59         bool                    written;
  60         struct journal_entry_pin pin;
  61
  62         struct blacklisted_node *entries;
  63         size_t                  nr_entries;
  64 };
  65
  66 struct journal_res {
  67         bool                    ref;
  68         u8                      idx;
  69         u16                     u64s;
  70         u32                     offset;
  71         u64                     seq;
  72 };
  73
  74 union journal_res_state {
  75         struct {
  76                 atomic64_t      counter;
  77         };
  78
  79         struct {
  80                 u64             v;
  81         };
  82
  83         struct {
  84                 u64             cur_entry_offset:20,
  85                                 idx:1,
  86                                 prev_buf_unwritten:1,
  87                                 buf0_count:21,
  88                                 buf1_count:21;
  89         };
  90 };
  91
  92 /* bytes: */
  93 #define JOURNAL_ENTRY_SIZE_MIN          (64U << 10) /* 64k */
  94 #define JOURNAL_ENTRY_SIZE_MAX          (4U  << 20) /* 4M */
  95
  96 /*
  97  * We stash some journal state as sentinal values in cur_entry_offset:
  98  * note - cur_entry_offset is in units of u64s
  99  */
 100 #define JOURNAL_ENTRY_OFFSET_MAX        ((1U << 20) - 1)
 101
 102 #define JOURNAL_ENTRY_CLOSED_VAL        (JOURNAL_ENTRY_OFFSET_MAX - 1)
 103 #define JOURNAL_ENTRY_ERROR_VAL         (JOURNAL_ENTRY_OFFSET_MAX)
 104
 105 /*
 106  * JOURNAL_NEED_WRITE - current (pending) journal entry should be written ASAP,
 107  * either because something's waiting on the write to complete or because it's
 108  * been dirty too long and the timer's expired.
 109  */
 110
 111 enum {
 112         JOURNAL_REPLAY_DONE,
 113         JOURNAL_STARTED,
 114         JOURNAL_NEED_WRITE,
 115 };
 116
 117 /* Embedded in struct bch_fs */
 118 struct journal {
 119         /* Fastpath stuff up front: */
 120
 121         unsigned long           flags;
 122
 123         union journal_res_state reservations;
 124         unsigned                cur_entry_u64s;
 125         unsigned                prev_buf_sectors;
 126         unsigned                cur_buf_sectors;
 127         unsigned                buf_size_want;
 128
 129         /*
 130          * Two journal entries -- one is currently open for new entries, the
 131          * other is possibly being written out.
 132          */
 133         struct journal_buf      buf[2];
 134
 135         spinlock_t              lock;
 136
 137         /* Used when waiting because the journal was full */
 138         wait_queue_head_t       wait;
 139
 140         struct closure          io;
 141         struct delayed_work     write_work;
 142         unsigned long           replicas_failed;
 143
 144         /* Sequence number of most recent journal entry (last entry in @pin) */
 145         atomic64_t              seq;
 146
 147         /* last_seq from the most recent journal entry written */
 148         u64                     last_seq_ondisk;
 149
 150         /*
 151          * FIFO of journal entries whose btree updates have not yet been
 152          * written out.
 153          *
 154          * Each entry is a reference count. The position in the FIFO is the
 155          * entry's sequence number relative to @seq.
 156          *
 157          * The journal entry itself holds a reference count, put when the
 158          * journal entry is written out. Each btree node modified by the journal
 159          * entry also holds a reference count, put when the btree node is
 160          * written.
 161          *
 162          * When a reference count reaches zero, the journal entry is no longer
 163          * needed. When all journal entries in the oldest journal bucket are no
 164          * longer needed, the bucket can be discarded and reused.
 165          */
 166         DECLARE_FIFO(struct journal_entry_pin_list, pin);
 167         struct journal_entry_pin_list *replay_pin_list;
 168
 169         /*
 170          * Protects the pin lists - the fifo itself is still protected by
 171          * j->lock though:
 172          */
 173         spinlock_t              pin_lock;
 174
 175         struct mutex            blacklist_lock;
 176         struct list_head        seq_blacklist;
 177
 178         BKEY_PADDED(key);
 179         struct dev_group        devs;
 180
 181         struct delayed_work     reclaim_work;
 182         unsigned long           last_flushed;
 183
 184         /* protects advancing ja->last_idx: */
 185         struct mutex            reclaim_lock;
 186         unsigned                write_delay_ms;
 187         unsigned                reclaim_delay_ms;
 188
 189         u64                     res_get_blocked_start;
 190         u64                     need_write_time;
 191         u64                     write_start_time;
 192
 193         struct time_stats       *write_time;
 194         struct time_stats       *delay_time;
 195         struct time_stats       *blocked_time;
 196         struct time_stats       *flush_seq_time;
 197
 198 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 199         struct lockdep_map      res_map;
 200 #endif
 201 };
 202
 203 /*
 204  * Embedded in struct bch_dev. First three fields refer to the array of journal
 205  * buckets, in bch_sb.
 206  */
 207 struct journal_device {
 208         /*
 209          * For each journal bucket, contains the max sequence number of the
 210          * journal writes it contains - so we know when a bucket can be reused.
 211          */
 212         u64                     *bucket_seq;
 213
 214         unsigned                sectors_free;
 215
 216         /* Journal bucket we're currently writing to */
 217         unsigned                cur_idx;
 218
 219         /* Last journal bucket that still contains an open journal entry */
 220
 221         /*
 222          * j->lock and j->reclaim_lock must both be held to modify, j->lock
 223          * sufficient to read:
 224          */
 225         unsigned                last_idx;
 226         unsigned                nr;
 227         u64                     *buckets;
 228
 229         /* Bio for journal reads/writes to this device */
 230         struct bio              *bio;
 231         u8                      ptr_idx;
 232
 233         /* for bch_journal_read_device */
 234         struct closure          read;
 235 };
 236
 237 #endif /* _BCACHE_JOURNAL_TYPES_H */