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 jset             *data;
  22
  23         __BKEY_PADDED(key, BCH_REPLICAS_MAX);
  24         struct bch_devs_list    devs_written;
  25
  26         struct closure_waitlist wait;
  27         u64                     last_seq;       /* copy of data->last_seq */
  28         long                    expires;
  29         u64                     flush_time;
  30
  31         unsigned                buf_size;       /* size in bytes of @data */
  32         unsigned                sectors;        /* maximum size for current entry */
  33         unsigned                disk_sectors;   /* maximum size entry could have been, if
  34                                                    buf_size was bigger */
  35         unsigned                u64s_reserved;
  36         bool                    noflush;        /* write has already been kicked off, and was noflush */
  37         bool                    must_flush;     /* something wants a flush */
  38         bool                    separate_flush;
  39 };
  40
  41 /*
  42  * Something that makes a journal entry dirty - i.e. a btree node that has to be
  43  * flushed:
  44  */
  45
  46 enum journal_pin_type {
  47         JOURNAL_PIN_btree,
  48         JOURNAL_PIN_key_cache,
  49         JOURNAL_PIN_other,
  50         JOURNAL_PIN_NR,
  51 };
  52
  53 struct journal_entry_pin_list {
  54         struct list_head                list[JOURNAL_PIN_NR];
  55         struct list_head                flushed;
  56         atomic_t                        count;
  57         struct bch_devs_list            devs;
  58 };
  59
  60 struct journal;
  61 struct journal_entry_pin;
  62 typedef int (*journal_pin_flush_fn)(struct journal *j,
  63                                 struct journal_entry_pin *, u64);
  64
  65 struct journal_entry_pin {
  66         struct list_head                list;
  67         journal_pin_flush_fn            flush;
  68         u64                             seq;
  69 };
  70
  71 struct journal_res {
  72         bool                    ref;
  73         u8                      idx;
  74         u16                     u64s;
  75         u32                     offset;
  76         u64                     seq;
  77 };
  78
  79 /*
  80  * For reserving space in the journal prior to getting a reservation on a
  81  * particular journal entry:
  82  */
  83 struct journal_preres {
  84         unsigned                u64s;
  85 };
  86
  87 union journal_res_state {
  88         struct {
  89                 atomic64_t      counter;
  90         };
  91
  92         struct {
  93                 u64             v;
  94         };
  95
  96         struct {
  97                 u64             cur_entry_offset:20,
  98                                 idx:2,
  99                                 unwritten_idx:2,
 100                                 buf0_count:10,
 101                                 buf1_count:10,
 102                                 buf2_count:10,
 103                                 buf3_count:10;
 104         };
 105 };
 106
 107 union journal_preres_state {
 108         struct {
 109                 atomic64_t      counter;
 110         };
 111
 112         struct {
 113                 u64             v;
 114         };
 115
 116         struct {
 117                 u64             waiting:1,
 118                                 reserved:31,
 119                                 remaining:32;
 120         };
 121 };
 122
 123 /* bytes: */
 124 #define JOURNAL_ENTRY_SIZE_MIN          (64U << 10) /* 64k */
 125 #define JOURNAL_ENTRY_SIZE_MAX          (4U  << 20) /* 4M */
 126
 127 /*
 128  * We stash some journal state as sentinal values in cur_entry_offset:
 129  * note - cur_entry_offset is in units of u64s
 130  */
 131 #define JOURNAL_ENTRY_OFFSET_MAX        ((1U << 20) - 1)
 132
 133 #define JOURNAL_ENTRY_CLOSED_VAL        (JOURNAL_ENTRY_OFFSET_MAX - 1)
 134 #define JOURNAL_ENTRY_ERROR_VAL         (JOURNAL_ENTRY_OFFSET_MAX)
 135
 136 struct journal_space {
 137         /* Units of 512 bytes sectors: */
 138         unsigned        next_entry; /* How big the next journal entry can be */
 139         unsigned        total;
 140 };
 141
 142 enum journal_space_from {
 143         journal_space_discarded,
 144         journal_space_clean_ondisk,
 145         journal_space_clean,
 146         journal_space_total,
 147         journal_space_nr,
 148 };
 149
 150 enum journal_flags {
 151         JOURNAL_REPLAY_DONE,
 152         JOURNAL_STARTED,
 153         JOURNAL_MAY_SKIP_FLUSH,
 154         JOURNAL_NEED_FLUSH_WRITE,
 155 };
 156
 157 #define JOURNAL_WATERMARKS()            \
 158         x(any)                          \
 159         x(copygc)                       \
 160         x(reserved)
 161
 162 enum journal_watermark {
 163 #define x(n)    JOURNAL_WATERMARK_##n,
 164         JOURNAL_WATERMARKS()
 165 #undef x
 166 };
 167
 168 #define JOURNAL_WATERMARK_MASK  3
 169
 170 /* Reasons we may fail to get a journal reservation: */
 171 #define JOURNAL_ERRORS()                \
 172         x(ok)                           \
 173         x(blocked)                      \
 174         x(max_in_flight)                \
 175         x(journal_full)                 \
 176         x(journal_pin_full)             \
 177         x(journal_stuck)                \
 178         x(insufficient_devices)
 179
 180 enum journal_errors {
 181 #define x(n)    JOURNAL_ERR_##n,
 182         JOURNAL_ERRORS()
 183 #undef x
 184 };
 185
 186 typedef DARRAY(u64)             darray_u64;
 187
 188 /* Embedded in struct bch_fs */
 189 struct journal {
 190         /* Fastpath stuff up front: */
 191         struct {
 192
 193         union journal_res_state reservations;
 194         enum journal_watermark  watermark;
 195
 196         union journal_preres_state prereserved;
 197
 198         } __aligned(SMP_CACHE_BYTES);
 199
 200         unsigned long           flags;
 201
 202         /* Max size of current journal entry */
 203         unsigned                cur_entry_u64s;
 204         unsigned                cur_entry_sectors;
 205
 206         /* Reserved space in journal entry to be used just prior to write */
 207         unsigned                entry_u64s_reserved;
 208
 209
 210         /*
 211          * 0, or -ENOSPC if waiting on journal reclaim, or -EROFS if
 212          * insufficient devices:
 213          */
 214         enum journal_errors     cur_entry_error;
 215
 216         unsigned                buf_size_want;
 217         /*
 218          * We may queue up some things to be journalled (log messages) before
 219          * the journal has actually started - stash them here:
 220          */
 221         darray_u64              early_journal_entries;
 222
 223         /*
 224          * Two journal entries -- one is currently open for new entries, the
 225          * other is possibly being written out.
 226          */
 227         struct journal_buf      buf[JOURNAL_BUF_NR];
 228
 229         spinlock_t              lock;
 230
 231         /* if nonzero, we may not open a new journal entry: */
 232         unsigned                blocked;
 233
 234         /* Used when waiting because the journal was full */
 235         wait_queue_head_t       wait;
 236         struct closure_waitlist async_wait;
 237         struct closure_waitlist preres_wait;
 238
 239         struct closure          io;
 240         struct delayed_work     write_work;
 241
 242         /* Sequence number of most recent journal entry (last entry in @pin) */
 243         atomic64_t              seq;
 244
 245         /* seq, last_seq from the most recent journal entry successfully written */
 246         u64                     seq_ondisk;
 247         u64                     flushed_seq_ondisk;
 248         u64                     last_seq_ondisk;
 249         u64                     err_seq;
 250         u64                     last_empty_seq;
 251
 252         /*
 253          * FIFO of journal entries whose btree updates have not yet been
 254          * written out.
 255          *
 256          * Each entry is a reference count. The position in the FIFO is the
 257          * entry's sequence number relative to @seq.
 258          *
 259          * The journal entry itself holds a reference count, put when the
 260          * journal entry is written out. Each btree node modified by the journal
 261          * entry also holds a reference count, put when the btree node is
 262          * written.
 263          *
 264          * When a reference count reaches zero, the journal entry is no longer
 265          * needed. When all journal entries in the oldest journal bucket are no
 266          * longer needed, the bucket can be discarded and reused.
 267          */
 268         struct {
 269                 u64 front, back, size, mask;
 270                 struct journal_entry_pin_list *data;
 271         }                       pin;
 272
 273         struct journal_space    space[journal_space_nr];
 274
 275         u64                     replay_journal_seq;
 276         u64                     replay_journal_seq_end;
 277
 278         struct write_point      wp;
 279         spinlock_t              err_lock;
 280
 281         struct mutex            reclaim_lock;
 282         /*
 283          * Used for waiting until journal reclaim has freed up space in the
 284          * journal:
 285          */
 286         wait_queue_head_t       reclaim_wait;
 287         struct task_struct      *reclaim_thread;
 288         bool                    reclaim_kicked;
 289         unsigned long           next_reclaim;
 290         u64                     nr_direct_reclaim;
 291         u64                     nr_background_reclaim;
 292
 293         unsigned long           last_flushed;
 294         struct journal_entry_pin *flush_in_progress;
 295         bool                    flush_in_progress_dropped;
 296         wait_queue_head_t       pin_flush_wait;
 297
 298         /* protects advancing ja->discard_idx: */
 299         struct mutex            discard_lock;
 300         bool                    can_discard;
 301
 302         unsigned long           last_flush_write;
 303
 304         u64                     res_get_blocked_start;
 305         u64                     write_start_time;
 306
 307         u64                     nr_flush_writes;
 308         u64                     nr_noflush_writes;
 309
 310         struct bch2_time_stats  *flush_write_time;
 311         struct bch2_time_stats  *noflush_write_time;
 312         struct bch2_time_stats  *blocked_time;
 313         struct bch2_time_stats  *flush_seq_time;
 314
 315 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 316         struct lockdep_map      res_map;
 317 #endif
 318 } __aligned(SMP_CACHE_BYTES);
 319
 320 /*
 321  * Embedded in struct bch_dev. First three fields refer to the array of journal
 322  * buckets, in bch_sb.
 323  */
 324 struct journal_device {
 325         /*
 326          * For each journal bucket, contains the max sequence number of the
 327          * journal writes it contains - so we know when a bucket can be reused.
 328          */
 329         u64                     *bucket_seq;
 330
 331         unsigned                sectors_free;
 332
 333         /*
 334          * discard_idx <= dirty_idx_ondisk <= dirty_idx <= cur_idx:
 335          */
 336         unsigned                discard_idx;            /* Next bucket to discard */
 337         unsigned                dirty_idx_ondisk;
 338         unsigned                dirty_idx;
 339         unsigned                cur_idx;                /* Journal bucket we're currently writing to */
 340         unsigned                nr;
 341
 342         u64                     *buckets;
 343
 344         /* Bio for journal reads/writes to this device */
 345         struct bio              *bio;
 346
 347         /* for bch_journal_read_device */
 348         struct closure          read;
 349 };
 350
 351 /*
 352  * journal_entry_res - reserve space in every journal entry:
 353  */
 354 struct journal_entry_res {
 355         unsigned                u64s;
 356 };
 357
 358 #endif /* _BCACHEFS_JOURNAL_TYPES_H */