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 /* Reasons we may fail to get a journal reservation: */
 158 #define JOURNAL_ERRORS()                \
 159         x(ok)                           \
 160         x(blocked)                      \
 161         x(max_in_flight)                \
 162         x(journal_full)                 \
 163         x(journal_pin_full)             \
 164         x(journal_stuck)                \
 165         x(insufficient_devices)
 166
 167 enum journal_errors {
 168 #define x(n)    JOURNAL_ERR_##n,
 169         JOURNAL_ERRORS()
 170 #undef x
 171 };
 172
 173 typedef DARRAY(u64)             darray_u64;
 174
 175 /* Embedded in struct bch_fs */
 176 struct journal {
 177         /* Fastpath stuff up front: */
 178         struct {
 179
 180         union journal_res_state reservations;
 181         enum bch_watermark      watermark;
 182
 183         union journal_preres_state prereserved;
 184
 185         } __aligned(SMP_CACHE_BYTES);
 186
 187         unsigned long           flags;
 188
 189         /* Max size of current journal entry */
 190         unsigned                cur_entry_u64s;
 191         unsigned                cur_entry_sectors;
 192
 193         /* Reserved space in journal entry to be used just prior to write */
 194         unsigned                entry_u64s_reserved;
 195
 196
 197         /*
 198          * 0, or -ENOSPC if waiting on journal reclaim, or -EROFS if
 199          * insufficient devices:
 200          */
 201         enum journal_errors     cur_entry_error;
 202
 203         unsigned                buf_size_want;
 204         /*
 205          * We may queue up some things to be journalled (log messages) before
 206          * the journal has actually started - stash them here:
 207          */
 208         darray_u64              early_journal_entries;
 209
 210         /*
 211          * Two journal entries -- one is currently open for new entries, the
 212          * other is possibly being written out.
 213          */
 214         struct journal_buf      buf[JOURNAL_BUF_NR];
 215
 216         spinlock_t              lock;
 217
 218         /* if nonzero, we may not open a new journal entry: */
 219         unsigned                blocked;
 220
 221         /* Used when waiting because the journal was full */
 222         wait_queue_head_t       wait;
 223         struct closure_waitlist async_wait;
 224         struct closure_waitlist preres_wait;
 225
 226         struct closure          io;
 227         struct delayed_work     write_work;
 228
 229         /* Sequence number of most recent journal entry (last entry in @pin) */
 230         atomic64_t              seq;
 231
 232         /* seq, last_seq from the most recent journal entry successfully written */
 233         u64                     seq_ondisk;
 234         u64                     flushed_seq_ondisk;
 235         u64                     last_seq_ondisk;
 236         u64                     err_seq;
 237         u64                     last_empty_seq;
 238
 239         /*
 240          * FIFO of journal entries whose btree updates have not yet been
 241          * written out.
 242          *
 243          * Each entry is a reference count. The position in the FIFO is the
 244          * entry's sequence number relative to @seq.
 245          *
 246          * The journal entry itself holds a reference count, put when the
 247          * journal entry is written out. Each btree node modified by the journal
 248          * entry also holds a reference count, put when the btree node is
 249          * written.
 250          *
 251          * When a reference count reaches zero, the journal entry is no longer
 252          * needed. When all journal entries in the oldest journal bucket are no
 253          * longer needed, the bucket can be discarded and reused.
 254          */
 255         struct {
 256                 u64 front, back, size, mask;
 257                 struct journal_entry_pin_list *data;
 258         }                       pin;
 259
 260         struct journal_space    space[journal_space_nr];
 261
 262         u64                     replay_journal_seq;
 263         u64                     replay_journal_seq_end;
 264
 265         struct write_point      wp;
 266         spinlock_t              err_lock;
 267
 268         struct mutex            reclaim_lock;
 269         /*
 270          * Used for waiting until journal reclaim has freed up space in the
 271          * journal:
 272          */
 273         wait_queue_head_t       reclaim_wait;
 274         struct task_struct      *reclaim_thread;
 275         bool                    reclaim_kicked;
 276         unsigned long           next_reclaim;
 277         u64                     nr_direct_reclaim;
 278         u64                     nr_background_reclaim;
 279
 280         unsigned long           last_flushed;
 281         struct journal_entry_pin *flush_in_progress;
 282         bool                    flush_in_progress_dropped;
 283         wait_queue_head_t       pin_flush_wait;
 284
 285         /* protects advancing ja->discard_idx: */
 286         struct mutex            discard_lock;
 287         bool                    can_discard;
 288
 289         unsigned long           last_flush_write;
 290
 291         u64                     res_get_blocked_start;
 292         u64                     write_start_time;
 293
 294         u64                     nr_flush_writes;
 295         u64                     nr_noflush_writes;
 296
 297         struct bch2_time_stats  *flush_write_time;
 298         struct bch2_time_stats  *noflush_write_time;
 299         struct bch2_time_stats  *blocked_time;
 300         struct bch2_time_stats  *flush_seq_time;
 301
 302 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 303         struct lockdep_map      res_map;
 304 #endif
 305 } __aligned(SMP_CACHE_BYTES);
 306
 307 /*
 308  * Embedded in struct bch_dev. First three fields refer to the array of journal
 309  * buckets, in bch_sb.
 310  */
 311 struct journal_device {
 312         /*
 313          * For each journal bucket, contains the max sequence number of the
 314          * journal writes it contains - so we know when a bucket can be reused.
 315          */
 316         u64                     *bucket_seq;
 317
 318         unsigned                sectors_free;
 319
 320         /*
 321          * discard_idx <= dirty_idx_ondisk <= dirty_idx <= cur_idx:
 322          */
 323         unsigned                discard_idx;            /* Next bucket to discard */
 324         unsigned                dirty_idx_ondisk;
 325         unsigned                dirty_idx;
 326         unsigned                cur_idx;                /* Journal bucket we're currently writing to */
 327         unsigned                nr;
 328
 329         u64                     *buckets;
 330
 331         /* Bio for journal reads/writes to this device */
 332         struct bio              *bio;
 333
 334         /* for bch_journal_read_device */
 335         struct closure          read;
 336 };
 337
 338 /*
 339  * journal_entry_res - reserve space in every journal entry:
 340  */
 341 struct journal_entry_res {
 342         unsigned                u64s;
 343 };
 344
 345 #endif /* _BCACHEFS_JOURNAL_TYPES_H */