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