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