git.sesse.net Git - bcachefs-tools-debian/blob - libbcache/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         struct closure_waitlist wait;
  19
  20         /*
  21          * ugh, prio_buckets are stupid - need to convert them to new
  22          * transaction machinery when it arrives
  23          */
  24         unsigned                nr_prio_buckets;
  25
  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         atomic_t                        count;
  38 };
  39
  40 struct journal;
  41 struct journal_entry_pin;
  42 typedef void (*journal_pin_flush_fn)(struct journal *j, struct journal_entry_pin *);
  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 /* 4 mb, in bytes: */
  94 #define JOURNAL_ENTRY_SIZE_MAX          (4U << 20)
  95
  96 /*
  97  * We stash some journal state as sentinal values in cur_entry_offset:
  98  */
  99 #define JOURNAL_ENTRY_OFFSET_MAX        ((1U << 20) - 1)
 100
 101 #define JOURNAL_ENTRY_CLOSED_VAL        (JOURNAL_ENTRY_OFFSET_MAX - 1)
 102 #define JOURNAL_ENTRY_ERROR_VAL         (JOURNAL_ENTRY_OFFSET_MAX)
 103
 104 /*
 105  * JOURNAL_NEED_WRITE - current (pending) journal entry should be written ASAP,
 106  * either because something's waiting on the write to complete or because it's
 107  * been dirty too long and the timer's expired.
 108  */
 109
 110 enum {
 111         JOURNAL_REPLAY_DONE,
 112         JOURNAL_STARTED,
 113         JOURNAL_NEED_WRITE,
 114 };
 115
 116 /* Embedded in struct cache_set */
 117 struct journal {
 118         /* Fastpath stuff up front: */
 119
 120         unsigned long           flags;
 121
 122         union journal_res_state reservations;
 123         unsigned                cur_entry_u64s;
 124         unsigned                prev_buf_sectors;
 125         unsigned                cur_buf_sectors;
 126         unsigned                entry_size_max; /* bytes */
 127
 128         /*
 129          * Two journal entries -- one is currently open for new entries, the
 130          * other is possibly being written out.
 131          */
 132         struct journal_buf      buf[2];
 133
 134         spinlock_t              lock;
 135
 136         /* Used when waiting because the journal was full */
 137         wait_queue_head_t       wait;
 138
 139         struct closure          io;
 140         struct delayed_work     write_work;
 141
 142         /* Sequence number of most recent journal entry (last entry in @pin) */
 143         atomic64_t              seq;
 144
 145         /* last_seq from the most recent journal entry written */
 146         u64                     last_seq_ondisk;
 147
 148         /*
 149          * FIFO of journal entries whose btree updates have not yet been
 150          * written out.
 151          *
 152          * Each entry is a reference count. The position in the FIFO is the
 153          * entry's sequence number relative to @seq.
 154          *
 155          * The journal entry itself holds a reference count, put when the
 156          * journal entry is written out. Each btree node modified by the journal
 157          * entry also holds a reference count, put when the btree node is
 158          * written.
 159          *
 160          * When a reference count reaches zero, the journal entry is no longer
 161          * needed. When all journal entries in the oldest journal bucket are no
 162          * longer needed, the bucket can be discarded and reused.
 163          */
 164         DECLARE_FIFO(struct journal_entry_pin_list, pin);
 165         struct journal_entry_pin_list *cur_pin_list;
 166
 167         /*
 168          * Protects the pin lists - the fifo itself is still protected by
 169          * j->lock though:
 170          */
 171         spinlock_t              pin_lock;
 172
 173         struct mutex            blacklist_lock;
 174         struct list_head        seq_blacklist;
 175
 176         BKEY_PADDED(key);
 177         struct cache_group      devs;
 178
 179         struct delayed_work     reclaim_work;
 180         unsigned long           last_flushed;
 181
 182         /* protects advancing ja->last_idx: */
 183         struct mutex            reclaim_lock;
 184
 185         /*
 186          * ugh: need to get prio_buckets converted over to the eventual new
 187          * transaction machinery
 188          */
 189         __le64                  prio_buckets[MAX_CACHES_PER_SET];
 190         unsigned                nr_prio_buckets;
 191
 192         unsigned                write_delay_ms;
 193         unsigned                reclaim_delay_ms;
 194
 195         u64                     res_get_blocked_start;
 196         u64                     need_write_time;
 197         u64                     write_start_time;
 198
 199         struct time_stats       *write_time;
 200         struct time_stats       *delay_time;
 201         struct time_stats       *blocked_time;
 202         struct time_stats       *flush_seq_time;
 203
 204 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 205         struct lockdep_map      res_map;
 206 #endif
 207 };
 208
 209 /*
 210  * Embedded in struct cache. First three fields refer to the array of journal
 211  * buckets, in cache_sb.
 212  */
 213 struct journal_device {
 214         /*
 215          * For each journal bucket, contains the max sequence number of the
 216          * journal writes it contains - so we know when a bucket can be reused.
 217          */
 218         u64                     *bucket_seq;
 219
 220         unsigned                sectors_free;
 221
 222         /* Journal bucket we're currently writing to */
 223         unsigned                cur_idx;
 224
 225         /* Last journal bucket that still contains an open journal entry */
 226
 227         /*
 228          * j->lock and j->reclaim_lock must both be held to modify, j->lock
 229          * sufficient to read:
 230          */
 231         unsigned                last_idx;
 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 /* _BCACHE_JOURNAL_TYPES_H */