#include "super_types.h"
#include "fifo.h"
-struct journal_res;
+#define JOURNAL_BUF_BITS 2
+#define JOURNAL_BUF_NR (1U << JOURNAL_BUF_BITS)
+#define JOURNAL_BUF_MASK (JOURNAL_BUF_NR - 1)
/*
- * We put two of these in struct journal; we used them for writes to the
- * journal that are being staged or in flight.
+ * We put JOURNAL_BUF_NR of these in struct journal; we used them for writes to
+ * the journal that are being staged or in flight.
*/
struct journal_buf {
+ struct closure io;
struct jset *data;
- BKEY_PADDED(key);
+ __BKEY_PADDED(key, BCH_REPLICAS_MAX);
+ struct bch_devs_list devs_written;
struct closure_waitlist wait;
+ u64 last_seq; /* copy of data->last_seq */
+ long expires;
+ u64 flush_time;
unsigned buf_size; /* size in bytes of @data */
unsigned sectors; /* maximum size for current entry */
unsigned disk_sectors; /* maximum size entry could have been, if
buf_size was bigger */
unsigned u64s_reserved;
- /* bloom filter: */
- unsigned long has_inode[1024 / sizeof(unsigned long)];
+ bool noflush:1; /* write has already been kicked off, and was noflush */
+ bool must_flush:1; /* something wants a flush */
+ bool separate_flush:1;
+ bool need_flush_to_write_buffer:1;
+ bool write_started:1;
+ bool write_allocated:1;
+ bool write_done:1;
+ u8 idx;
};
/*
* flushed:
*/
+enum journal_pin_type {
+ JOURNAL_PIN_btree,
+ JOURNAL_PIN_key_cache,
+ JOURNAL_PIN_other,
+ JOURNAL_PIN_NR,
+};
+
struct journal_entry_pin_list {
- struct list_head list;
+ struct list_head list[JOURNAL_PIN_NR];
struct list_head flushed;
atomic_t count;
struct bch_devs_list devs;
struct journal;
struct journal_entry_pin;
-typedef void (*journal_pin_flush_fn)(struct journal *j,
+typedef int (*journal_pin_flush_fn)(struct journal *j,
struct journal_entry_pin *, u64);
struct journal_entry_pin {
u64 seq;
};
-/*
- * For reserving space in the journal prior to getting a reservation on a
- * particular journal entry:
- */
-struct journal_preres {
- unsigned u64s;
-};
-
union journal_res_state {
struct {
atomic64_t counter;
struct {
u64 cur_entry_offset:20,
- idx:1,
- prev_buf_unwritten:1,
- buf0_count:21,
- buf1_count:21;
- };
-};
-
-union journal_preres_state {
- struct {
- atomic64_t counter;
- };
-
- struct {
- u64 v;
- };
-
- struct {
- u32 reserved;
- u32 remaining;
+ idx:2,
+ unwritten_idx:2,
+ buf0_count:10,
+ buf1_count:10,
+ buf2_count:10,
+ buf3_count:10;
};
};
#define JOURNAL_ENTRY_CLOSED_VAL (JOURNAL_ENTRY_OFFSET_MAX - 1)
#define JOURNAL_ENTRY_ERROR_VAL (JOURNAL_ENTRY_OFFSET_MAX)
-/*
- * JOURNAL_NEED_WRITE - current (pending) journal entry should be written ASAP,
- * either because something's waiting on the write to complete or because it's
- * been dirty too long and the timer's expired.
- */
+struct journal_space {
+ /* Units of 512 bytes sectors: */
+ unsigned next_entry; /* How big the next journal entry can be */
+ unsigned total;
+};
-enum {
+enum journal_space_from {
+ journal_space_discarded,
+ journal_space_clean_ondisk,
+ journal_space_clean,
+ journal_space_total,
+ journal_space_nr,
+};
+
+enum journal_flags {
JOURNAL_REPLAY_DONE,
JOURNAL_STARTED,
- JOURNAL_NEED_WRITE,
- JOURNAL_NOT_EMPTY,
- JOURNAL_MAY_GET_UNRESERVED,
+ JOURNAL_MAY_SKIP_FLUSH,
+ JOURNAL_NEED_FLUSH_WRITE,
+};
+
+/* Reasons we may fail to get a journal reservation: */
+#define JOURNAL_ERRORS() \
+ x(ok) \
+ x(retry) \
+ x(blocked) \
+ x(max_in_flight) \
+ x(journal_full) \
+ x(journal_pin_full) \
+ x(journal_stuck) \
+ x(insufficient_devices)
+
+enum journal_errors {
+#define x(n) JOURNAL_ERR_##n,
+ JOURNAL_ERRORS()
+#undef x
+};
+
+typedef DARRAY(u64) darray_u64;
+
+struct journal_bio {
+ struct bch_dev *ca;
+ unsigned buf_idx;
+
+ struct bio bio;
};
/* Embedded in struct bch_fs */
struct journal {
/* Fastpath stuff up front: */
-
- unsigned long flags;
+ struct {
union journal_res_state reservations;
+ enum bch_watermark watermark;
+
+ } __aligned(SMP_CACHE_BYTES);
+
+ unsigned long flags;
/* Max size of current journal entry */
unsigned cur_entry_u64s;
unsigned cur_entry_sectors;
+ /* Reserved space in journal entry to be used just prior to write */
+ unsigned entry_u64s_reserved;
+
+
/*
* 0, or -ENOSPC if waiting on journal reclaim, or -EROFS if
* insufficient devices:
*/
- int cur_entry_error;
-
- union journal_preres_state prereserved;
-
- /* Reserved space in journal entry to be used just prior to write */
- unsigned entry_u64s_reserved;
+ enum journal_errors cur_entry_error;
unsigned buf_size_want;
+ /*
+ * We may queue up some things to be journalled (log messages) before
+ * the journal has actually started - stash them here:
+ */
+ darray_u64 early_journal_entries;
+ /*
+ * Protects journal_buf->data, when accessing without a jorunal
+ * reservation: for synchronization between the btree write buffer code
+ * and the journal write path:
+ */
+ struct mutex buf_lock;
/*
* Two journal entries -- one is currently open for new entries, the
* other is possibly being written out.
*/
- struct journal_buf buf[2];
+ struct journal_buf buf[JOURNAL_BUF_NR];
spinlock_t lock;
/* Used when waiting because the journal was full */
wait_queue_head_t wait;
struct closure_waitlist async_wait;
- struct closure_waitlist preres_wait;
- struct closure io;
struct delayed_work write_work;
+ struct workqueue_struct *wq;
/* Sequence number of most recent journal entry (last entry in @pin) */
atomic64_t seq;
/* seq, last_seq from the most recent journal entry successfully written */
u64 seq_ondisk;
+ u64 flushed_seq_ondisk;
u64 last_seq_ondisk;
+ u64 err_seq;
+ u64 last_empty_seq;
/*
* FIFO of journal entries whose btree updates have not yet been
struct journal_entry_pin_list *data;
} pin;
+ struct journal_space space[journal_space_nr];
+
u64 replay_journal_seq;
u64 replay_journal_seq_end;
struct write_point wp;
spinlock_t err_lock;
- struct delayed_work reclaim_work;
struct mutex reclaim_lock;
+ /*
+ * Used for waiting until journal reclaim has freed up space in the
+ * journal:
+ */
+ wait_queue_head_t reclaim_wait;
+ struct task_struct *reclaim_thread;
+ bool reclaim_kicked;
+ unsigned long next_reclaim;
+ u64 nr_direct_reclaim;
+ u64 nr_background_reclaim;
+
unsigned long last_flushed;
struct journal_entry_pin *flush_in_progress;
+ bool flush_in_progress_dropped;
wait_queue_head_t pin_flush_wait;
/* protects advancing ja->discard_idx: */
struct mutex discard_lock;
bool can_discard;
- unsigned write_delay_ms;
- unsigned reclaim_delay_ms;
+ unsigned long last_flush_write;
- u64 res_get_blocked_start;
- u64 need_write_time;
u64 write_start_time;
- struct time_stats *write_time;
- struct time_stats *delay_time;
- struct time_stats *blocked_time;
+ u64 nr_flush_writes;
+ u64 nr_noflush_writes;
+ u64 entry_bytes_written;
+
+ struct time_stats *flush_write_time;
+ struct time_stats *noflush_write_time;
struct time_stats *flush_seq_time;
#ifdef CONFIG_DEBUG_LOCK_ALLOC
struct lockdep_map res_map;
#endif
-};
+} __aligned(SMP_CACHE_BYTES);
/*
* Embedded in struct bch_dev. First three fields refer to the array of journal
u64 *buckets;
/* Bio for journal reads/writes to this device */
- struct bio *bio;
+ struct journal_bio *bio[JOURNAL_BUF_NR];
/* for bch_journal_read_device */
struct closure read;