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