fs/bcachefs/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 #define JOURNAL_FLAGS()                 \
 133         x(replay_done)                  \
 134         x(running)                      \
 135         x(may_skip_flush)               \
 136         x(need_flush_write)             \
 137         x(space_low)
 138
 139 enum journal_flags {
 140 #define x(n)    JOURNAL_##n,
 141         JOURNAL_FLAGS()
 142 #undef x
 143 };
 144
 145 /* Reasons we may fail to get a journal reservation: */
 146 #define JOURNAL_ERRORS()                \
 147         x(ok)                           \
 148         x(retry)                        \
 149         x(blocked)                      \
 150         x(max_in_flight)                \
 151         x(journal_full)                 \
 152         x(journal_pin_full)             \
 153         x(journal_stuck)                \
 154         x(insufficient_devices)
 155
 156 enum journal_errors {
 157 #define x(n)    JOURNAL_ERR_##n,
 158         JOURNAL_ERRORS()
 159 #undef x
 160 };
 161
 162 typedef DARRAY(u64)             darray_u64;
 163
 164 struct journal_bio {
 165         struct bch_dev          *ca;
 166         unsigned                buf_idx;
 167
 168         struct bio              bio;
 169 };
 170
 171 /* Embedded in struct bch_fs */
 172 struct journal {
 173         /* Fastpath stuff up front: */
 174         struct {
 175
 176         union journal_res_state reservations;
 177         enum bch_watermark      watermark;
 178
 179         } __aligned(SMP_CACHE_BYTES);
 180
 181         unsigned long           flags;
 182
 183         /* Max size of current journal entry */
 184         unsigned                cur_entry_u64s;
 185         unsigned                cur_entry_sectors;
 186
 187         /* Reserved space in journal entry to be used just prior to write */
 188         unsigned                entry_u64s_reserved;
 189
 190
 191         /*
 192          * 0, or -ENOSPC if waiting on journal reclaim, or -EROFS if
 193          * insufficient devices:
 194          */
 195         enum journal_errors     cur_entry_error;
 196
 197         unsigned                buf_size_want;
 198         /*
 199          * We may queue up some things to be journalled (log messages) before
 200          * the journal has actually started - stash them here:
 201          */
 202         darray_u64              early_journal_entries;
 203
 204         /*
 205          * Protects journal_buf->data, when accessing without a jorunal
 206          * reservation: for synchronization between the btree write buffer code
 207          * and the journal write path:
 208          */
 209         struct mutex            buf_lock;
 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
 225         struct delayed_work     write_work;
 226         struct workqueue_struct *wq;
 227
 228         /* Sequence number of most recent journal entry (last entry in @pin) */
 229         atomic64_t              seq;
 230
 231         /* seq, last_seq from the most recent journal entry successfully written */
 232         u64                     seq_ondisk;
 233         u64                     flushed_seq_ondisk;
 234         u64                     last_seq_ondisk;
 235         u64                     err_seq;
 236         u64                     last_empty_seq;
 237         u64                     oldest_seq_found_ondisk;
 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                     write_start_time;
 292
 293         u64                     nr_flush_writes;
 294         u64                     nr_noflush_writes;
 295         u64                     entry_bytes_written;
 296
 297         struct bch2_time_stats  *flush_write_time;
 298         struct bch2_time_stats  *noflush_write_time;
 299         struct bch2_time_stats  *flush_seq_time;
 300
 301 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 302         struct lockdep_map      res_map;
 303 #endif
 304 } __aligned(SMP_CACHE_BYTES);
 305
 306 /*
 307  * Embedded in struct bch_dev. First three fields refer to the array of journal
 308  * buckets, in bch_sb.
 309  */
 310 struct journal_device {
 311         /*
 312          * For each journal bucket, contains the max sequence number of the
 313          * journal writes it contains - so we know when a bucket can be reused.
 314          */
 315         u64                     *bucket_seq;
 316
 317         unsigned                sectors_free;
 318
 319         /*
 320          * discard_idx <= dirty_idx_ondisk <= dirty_idx <= cur_idx:
 321          */
 322         unsigned                discard_idx;            /* Next bucket to discard */
 323         unsigned                dirty_idx_ondisk;
 324         unsigned                dirty_idx;
 325         unsigned                cur_idx;                /* Journal bucket we're currently writing to */
 326         unsigned                nr;
 327
 328         u64                     *buckets;
 329
 330         /* Bio for journal reads/writes to this device */
 331         struct journal_bio      *bio[JOURNAL_BUF_NR];
 332
 333         /* for bch_journal_read_device */
 334         struct closure          read;
 335         u64                     highest_seq_found;
 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 */