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

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 #ifndef _BCACHEFS_JOURNAL_TYPES_H
   3 #define _BCACHEFS_JOURNAL_TYPES_H
   4
   5 #include <linux/cache.h>
   6 #include <linux/workqueue.h>
   7
   8 #include "alloc_types.h"
   9 #include "super_types.h"
  10 #include "fifo.h"
  11
  12 #define JOURNAL_BUF_BITS        2
  13 #define JOURNAL_BUF_NR          (1U << JOURNAL_BUF_BITS)
  14 #define JOURNAL_BUF_MASK        (JOURNAL_BUF_NR - 1)
  15
  16 /*
  17  * We put JOURNAL_BUF_NR of these in struct journal; we used them for writes to
  18  * the journal that are being staged or in flight.
  19  */
  20 struct journal_buf {
  21         struct closure          io;
  22         struct jset             *data;
  23
  24         __BKEY_PADDED(key, BCH_REPLICAS_MAX);
  25         struct bch_devs_list    devs_written;
  26
  27         struct closure_waitlist wait;
  28         u64                     last_seq;       /* copy of data->last_seq */
  29         long                    expires;
  30         u64                     flush_time;
  31
  32         unsigned                buf_size;       /* size in bytes of @data */
  33         unsigned                sectors;        /* maximum size for current entry */
  34         unsigned                disk_sectors;   /* maximum size entry could have been, if
  35                                                    buf_size was bigger */
  36         unsigned                u64s_reserved;
  37         bool                    noflush:1;      /* write has already been kicked off, and was noflush */
  38         bool                    must_flush:1;   /* something wants a flush */
  39         bool                    separate_flush:1;
  40         bool                    need_flush_to_write_buffer:1;
  41         bool                    write_started:1;
  42         bool                    write_allocated:1;
  43         bool                    write_done:1;
  44         u8                      idx;
  45 };
  46
  47 /*
  48  * Something that makes a journal entry dirty - i.e. a btree node that has to be
  49  * flushed:
  50  */
  51
  52 enum journal_pin_type {
  53         JOURNAL_PIN_btree,
  54         JOURNAL_PIN_key_cache,
  55         JOURNAL_PIN_other,
  56         JOURNAL_PIN_NR,
  57 };
  58
  59 struct journal_entry_pin_list {
  60         struct list_head                list[JOURNAL_PIN_NR];
  61         struct list_head                flushed;
  62         atomic_t                        count;
  63         struct bch_devs_list            devs;
  64 };
  65
  66 struct journal;
  67 struct journal_entry_pin;
  68 typedef int (*journal_pin_flush_fn)(struct journal *j,
  69                                 struct journal_entry_pin *, u64);
  70
  71 struct journal_entry_pin {
  72         struct list_head                list;
  73         journal_pin_flush_fn            flush;
  74         u64                             seq;
  75 };
  76
  77 struct journal_res {
  78         bool                    ref;
  79         u8                      idx;
  80         u16                     u64s;
  81         u32                     offset;
  82         u64                     seq;
  83 };
  84
  85 union journal_res_state {
  86         struct {
  87                 atomic64_t      counter;
  88         };
  89
  90         struct {
  91                 u64             v;
  92         };
  93
  94         struct {
  95                 u64             cur_entry_offset:20,
  96                                 idx:2,
  97                                 unwritten_idx:2,
  98                                 buf0_count:10,
  99                                 buf1_count:10,
 100                                 buf2_count:10,
 101                                 buf3_count:10;
 102         };
 103 };
 104
 105 /* bytes: */
 106 #define JOURNAL_ENTRY_SIZE_MIN          (64U << 10) /* 64k */
 107 #define JOURNAL_ENTRY_SIZE_MAX          (4U  << 20) /* 4M */
 108
 109 /*
 110  * We stash some journal state as sentinal values in cur_entry_offset:
 111  * note - cur_entry_offset is in units of u64s
 112  */
 113 #define JOURNAL_ENTRY_OFFSET_MAX        ((1U << 20) - 1)
 114
 115 #define JOURNAL_ENTRY_CLOSED_VAL        (JOURNAL_ENTRY_OFFSET_MAX - 1)
 116 #define JOURNAL_ENTRY_ERROR_VAL         (JOURNAL_ENTRY_OFFSET_MAX)
 117
 118 struct journal_space {
 119         /* Units of 512 bytes sectors: */
 120         unsigned        next_entry; /* How big the next journal entry can be */
 121         unsigned        total;
 122 };
 123
 124 enum journal_space_from {
 125         journal_space_discarded,
 126         journal_space_clean_ondisk,
 127         journal_space_clean,
 128         journal_space_total,
 129         journal_space_nr,
 130 };
 131
 132 enum journal_flags {
 133         JOURNAL_REPLAY_DONE,
 134         JOURNAL_STARTED,
 135         JOURNAL_MAY_SKIP_FLUSH,
 136         JOURNAL_NEED_FLUSH_WRITE,
 137 };
 138
 139 /* Reasons we may fail to get a journal reservation: */
 140 #define JOURNAL_ERRORS()                \
 141         x(ok)                           \
 142         x(retry)                        \
 143         x(blocked)                      \
 144         x(max_in_flight)                \
 145         x(journal_full)                 \
 146         x(journal_pin_full)             \
 147         x(journal_stuck)                \
 148         x(insufficient_devices)
 149
 150 enum journal_errors {
 151 #define x(n)    JOURNAL_ERR_##n,
 152         JOURNAL_ERRORS()
 153 #undef x
 154 };
 155
 156 typedef DARRAY(u64)             darray_u64;
 157
 158 struct journal_bio {
 159         struct bch_dev          *ca;
 160         unsigned                buf_idx;
 161
 162         struct bio              bio;
 163 };
 164
 165 /* Embedded in struct bch_fs */
 166 struct journal {
 167         /* Fastpath stuff up front: */
 168         struct {
 169
 170         union journal_res_state reservations;
 171         enum bch_watermark      watermark;
 172
 173         } __aligned(SMP_CACHE_BYTES);
 174
 175         unsigned long           flags;
 176
 177         /* Max size of current journal entry */
 178         unsigned                cur_entry_u64s;
 179         unsigned                cur_entry_sectors;
 180
 181         /* Reserved space in journal entry to be used just prior to write */
 182         unsigned                entry_u64s_reserved;
 183
 184
 185         /*
 186          * 0, or -ENOSPC if waiting on journal reclaim, or -EROFS if
 187          * insufficient devices:
 188          */
 189         enum journal_errors     cur_entry_error;
 190
 191         unsigned                buf_size_want;
 192         /*
 193          * We may queue up some things to be journalled (log messages) before
 194          * the journal has actually started - stash them here:
 195          */
 196         darray_u64              early_journal_entries;
 197
 198         /*
 199          * Protects journal_buf->data, when accessing without a jorunal
 200          * reservation: for synchronization between the btree write buffer code
 201          * and the journal write path:
 202          */
 203         struct mutex            buf_lock;
 204         /*
 205          * Two journal entries -- one is currently open for new entries, the
 206          * other is possibly being written out.
 207          */
 208         struct journal_buf      buf[JOURNAL_BUF_NR];
 209
 210         spinlock_t              lock;
 211
 212         /* if nonzero, we may not open a new journal entry: */
 213         unsigned                blocked;
 214
 215         /* Used when waiting because the journal was full */
 216         wait_queue_head_t       wait;
 217         struct closure_waitlist async_wait;
 218
 219         struct delayed_work     write_work;
 220         struct workqueue_struct *wq;
 221
 222         /* Sequence number of most recent journal entry (last entry in @pin) */
 223         atomic64_t              seq;
 224
 225         /* seq, last_seq from the most recent journal entry successfully written */
 226         u64                     seq_ondisk;
 227         u64                     flushed_seq_ondisk;
 228         u64                     last_seq_ondisk;
 229         u64                     err_seq;
 230         u64                     last_empty_seq;
 231
 232         /*
 233          * FIFO of journal entries whose btree updates have not yet been
 234          * written out.
 235          *
 236          * Each entry is a reference count. The position in the FIFO is the
 237          * entry's sequence number relative to @seq.
 238          *
 239          * The journal entry itself holds a reference count, put when the
 240          * journal entry is written out. Each btree node modified by the journal
 241          * entry also holds a reference count, put when the btree node is
 242          * written.
 243          *
 244          * When a reference count reaches zero, the journal entry is no longer
 245          * needed. When all journal entries in the oldest journal bucket are no
 246          * longer needed, the bucket can be discarded and reused.
 247          */
 248         struct {
 249                 u64 front, back, size, mask;
 250                 struct journal_entry_pin_list *data;
 251         }                       pin;
 252
 253         struct journal_space    space[journal_space_nr];
 254
 255         u64                     replay_journal_seq;
 256         u64                     replay_journal_seq_end;
 257
 258         struct write_point      wp;
 259         spinlock_t              err_lock;
 260
 261         struct mutex            reclaim_lock;
 262         /*
 263          * Used for waiting until journal reclaim has freed up space in the
 264          * journal:
 265          */
 266         wait_queue_head_t       reclaim_wait;
 267         struct task_struct      *reclaim_thread;
 268         bool                    reclaim_kicked;
 269         unsigned long           next_reclaim;
 270         u64                     nr_direct_reclaim;
 271         u64                     nr_background_reclaim;
 272
 273         unsigned long           last_flushed;
 274         struct journal_entry_pin *flush_in_progress;
 275         bool                    flush_in_progress_dropped;
 276         wait_queue_head_t       pin_flush_wait;
 277
 278         /* protects advancing ja->discard_idx: */
 279         struct mutex            discard_lock;
 280         bool                    can_discard;
 281
 282         unsigned long           last_flush_write;
 283
 284         u64                     write_start_time;
 285
 286         u64                     nr_flush_writes;
 287         u64                     nr_noflush_writes;
 288         u64                     entry_bytes_written;
 289
 290         struct time_stats       *flush_write_time;
 291         struct time_stats       *noflush_write_time;
 292         struct time_stats       *flush_seq_time;
 293
 294 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 295         struct lockdep_map      res_map;
 296 #endif
 297 } __aligned(SMP_CACHE_BYTES);
 298
 299 /*
 300  * Embedded in struct bch_dev. First three fields refer to the array of journal
 301  * buckets, in bch_sb.
 302  */
 303 struct journal_device {
 304         /*
 305          * For each journal bucket, contains the max sequence number of the
 306          * journal writes it contains - so we know when a bucket can be reused.
 307          */
 308         u64                     *bucket_seq;
 309
 310         unsigned                sectors_free;
 311
 312         /*
 313          * discard_idx <= dirty_idx_ondisk <= dirty_idx <= cur_idx:
 314          */
 315         unsigned                discard_idx;            /* Next bucket to discard */
 316         unsigned                dirty_idx_ondisk;
 317         unsigned                dirty_idx;
 318         unsigned                cur_idx;                /* Journal bucket we're currently writing to */
 319         unsigned                nr;
 320
 321         u64                     *buckets;
 322
 323         /* Bio for journal reads/writes to this device */
 324         struct journal_bio      *bio[JOURNAL_BUF_NR];
 325
 326         /* for bch_journal_read_device */
 327         struct closure          read;
 328 };
 329
 330 /*
 331  * journal_entry_res - reserve space in every journal entry:
 332  */
 333 struct journal_entry_res {
 334         unsigned                u64s;
 335 };
 336
 337 #endif /* _BCACHEFS_JOURNAL_TYPES_H */