X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fjournal.h;h=25c6876765ac4b09b05055898cd36e165c851ea0;hb=1e574cb1aa07ab3a796c7d6c5501b96f3056ef4d;hp=3a083748110058fe775c3336470fd54d3c1e8beb;hpb=a10a41fa2b1a917b0f3b34d20175867f968b2d12;p=bcachefs-tools-debian diff --git a/libbcachefs/journal.h b/libbcachefs/journal.h index 3a08374..25c6876 100644 --- a/libbcachefs/journal.h +++ b/libbcachefs/journal.h @@ -1,3 +1,4 @@ +/* SPDX-License-Identifier: GPL-2.0 */ #ifndef _BCACHEFS_JOURNAL_H #define _BCACHEFS_JOURNAL_H @@ -118,6 +119,7 @@ static inline void journal_wake(struct journal *j) { wake_up(&j->wait); closure_wake_up(&j->async_wait); + closure_wake_up(&j->preres_wait); } static inline struct journal_buf *journal_cur_buf(struct journal *j) @@ -145,6 +147,7 @@ static inline u64 journal_cur_seq(struct journal *j) } u64 bch2_inode_journal_seq(struct journal *, u64); +void bch2_journal_set_has_inum(struct journal *, u64, u64); static inline int journal_state_count(union journal_res_state s, int idx) { @@ -178,6 +181,11 @@ static inline unsigned jset_u64s(unsigned u64s) return u64s + sizeof(struct jset_entry) / sizeof(u64); } +static inline int journal_entry_overhead(struct journal *j) +{ + return sizeof(struct jset) / sizeof(u64) + j->entry_u64s_reserved; +} + static inline struct jset_entry * bch2_journal_add_entry_noreservation(struct journal_buf *buf, size_t u64s) { @@ -192,27 +200,39 @@ bch2_journal_add_entry_noreservation(struct journal_buf *buf, size_t u64s) return entry; } +static inline struct jset_entry * +journal_res_entry(struct journal *j, struct journal_res *res) +{ + return vstruct_idx(j->buf[res->idx].data, res->offset); +} + +static inline unsigned journal_entry_set(struct jset_entry *entry, unsigned type, + enum btree_id id, unsigned level, + const void *data, unsigned u64s) +{ + memset(entry, 0, sizeof(*entry)); + entry->u64s = cpu_to_le16(u64s); + entry->type = type; + entry->btree_id = id; + entry->level = level; + memcpy_u64s_small(entry->_data, data, u64s); + + return jset_u64s(u64s); +} + static inline void bch2_journal_add_entry(struct journal *j, struct journal_res *res, unsigned type, enum btree_id id, unsigned level, const void *data, unsigned u64s) { - struct journal_buf *buf = &j->buf[res->idx]; - struct jset_entry *entry = vstruct_idx(buf->data, res->offset); - unsigned actual = jset_u64s(u64s); + unsigned actual = journal_entry_set(journal_res_entry(j, res), + type, id, level, data, u64s); EBUG_ON(!res->ref); EBUG_ON(actual > res->u64s); res->offset += actual; res->u64s -= actual; - - memset(entry, 0, sizeof(*entry)); - entry->u64s = cpu_to_le16(u64s); - entry->type = type; - entry->btree_id = id; - entry->level = level; - memcpy_u64s(entry->_data, data, u64s); } static inline void bch2_journal_add_keys(struct journal *j, struct journal_res *res, @@ -222,7 +242,20 @@ static inline void bch2_journal_add_keys(struct journal *j, struct journal_res * id, 0, k, k->k.u64s); } -void bch2_journal_buf_put_slowpath(struct journal *, bool); +static inline bool journal_entry_empty(struct jset *j) +{ + struct jset_entry *i; + + if (j->seq != j->last_seq) + return false; + + vstruct_for_each(j, i) + if (i->type == BCH_JSET_ENTRY_btree_keys && i->u64s) + return false; + return true; +} + +void __bch2_journal_buf_put(struct journal *, bool); static inline void bch2_journal_buf_put(struct journal *j, unsigned idx, bool need_write_just_set) @@ -233,17 +266,10 @@ static inline void bch2_journal_buf_put(struct journal *j, unsigned idx, .buf0_count = idx == 0, .buf1_count = idx == 1, }).v, &j->reservations.counter); - - EBUG_ON(s.idx != idx && !s.prev_buf_unwritten); - - /* - * Do not initiate a journal write if the journal is in an error state - * (previous journal entry write may have failed) - */ - if (s.idx != idx && - !journal_state_count(s, idx) && - s.cur_entry_offset != JOURNAL_ENTRY_ERROR_VAL) - bch2_journal_buf_put_slowpath(j, need_write_just_set); + if (!journal_state_count(s, idx)) { + EBUG_ON(s.idx == idx || !s.prev_buf_unwritten); + __bch2_journal_buf_put(j, need_write_just_set); + } } /* @@ -256,7 +282,7 @@ static inline void bch2_journal_res_put(struct journal *j, if (!res->ref) return; - lock_release(&j->res_map, 0, _RET_IP_); + lock_release(&j->res_map, _THIS_IP_); while (res->u64s) bch2_journal_add_entry(j, res, @@ -273,6 +299,8 @@ int bch2_journal_res_get_slowpath(struct journal *, struct journal_res *, #define JOURNAL_RES_GET_NONBLOCK (1 << 0) #define JOURNAL_RES_GET_CHECK (1 << 1) +#define JOURNAL_RES_GET_RESERVED (1 << 2) +#define JOURNAL_RES_GET_RECLAIM (1 << 3) static inline int journal_res_get_fast(struct journal *j, struct journal_res *res, @@ -291,6 +319,12 @@ static inline int journal_res_get_fast(struct journal *j, if (new.cur_entry_offset + res->u64s > j->cur_entry_u64s) return 0; + EBUG_ON(!journal_state_count(new, new.idx)); + + if (!(flags & JOURNAL_RES_GET_RESERVED) && + !test_bit(JOURNAL_MAY_GET_UNRESERVED, &j->flags)) + return 0; + if (flags & JOURNAL_RES_GET_CHECK) return 1; @@ -324,19 +358,114 @@ static inline int bch2_journal_res_get(struct journal *j, struct journal_res *re return ret; out: if (!(flags & JOURNAL_RES_GET_CHECK)) { - lock_acquire_shared(&j->res_map, 0, 0, NULL, _THIS_IP_); + lock_acquire_shared(&j->res_map, 0, + (flags & JOURNAL_RES_GET_NONBLOCK) != 0, + NULL, _THIS_IP_); EBUG_ON(!res->ref); } return 0; } -u64 bch2_journal_last_unwritten_seq(struct journal *); -int bch2_journal_open_seq_async(struct journal *, u64, struct closure *); +/* journal_preres: */ -void bch2_journal_wait_on_seq(struct journal *, u64, struct closure *); -void bch2_journal_flush_seq_async(struct journal *, u64, struct closure *); +static inline bool journal_check_may_get_unreserved(struct journal *j) +{ + union journal_preres_state s = READ_ONCE(j->prereserved); + bool ret = s.reserved <= s.remaining && + fifo_free(&j->pin) > 8; + + lockdep_assert_held(&j->lock); + + if (ret != test_bit(JOURNAL_MAY_GET_UNRESERVED, &j->flags)) { + if (ret) { + set_bit(JOURNAL_MAY_GET_UNRESERVED, &j->flags); + journal_wake(j); + } else { + clear_bit(JOURNAL_MAY_GET_UNRESERVED, &j->flags); + } + } + return ret; +} + +static inline void bch2_journal_preres_put(struct journal *j, + struct journal_preres *res) +{ + union journal_preres_state s = { .reserved = res->u64s }; + + if (!res->u64s) + return; + + s.v = atomic64_sub_return(s.v, &j->prereserved.counter); + res->u64s = 0; + closure_wake_up(&j->preres_wait); + + if (s.reserved <= s.remaining && + !test_bit(JOURNAL_MAY_GET_UNRESERVED, &j->flags)) { + spin_lock(&j->lock); + journal_check_may_get_unreserved(j); + spin_unlock(&j->lock); + } +} + +int __bch2_journal_preres_get(struct journal *, + struct journal_preres *, unsigned, unsigned); + +static inline int bch2_journal_preres_get_fast(struct journal *j, + struct journal_preres *res, + unsigned new_u64s, + unsigned flags) +{ + int d = new_u64s - res->u64s; + union journal_preres_state old, new; + u64 v = atomic64_read(&j->prereserved.counter); + + do { + old.v = new.v = v; + + new.reserved += d; + + /* + * If we're being called from the journal reclaim path, we have + * to unconditionally give out the pre-reservation, there's + * nothing else sensible we can do - otherwise we'd recurse back + * into the reclaim path and deadlock: + */ + + if (!(flags & JOURNAL_RES_GET_RECLAIM) && + new.reserved > new.remaining) + return 0; + } while ((v = atomic64_cmpxchg(&j->prereserved.counter, + old.v, new.v)) != old.v); + + res->u64s += d; + return 1; +} + +static inline int bch2_journal_preres_get(struct journal *j, + struct journal_preres *res, + unsigned new_u64s, + unsigned flags) +{ + if (new_u64s <= res->u64s) + return 0; + + if (bch2_journal_preres_get_fast(j, res, new_u64s, flags)) + return 0; + + if (flags & JOURNAL_RES_GET_NONBLOCK) + return -EAGAIN; + + return __bch2_journal_preres_get(j, res, new_u64s, flags); +} + +/* journal_entry_res: */ + +void bch2_journal_entry_res_resize(struct journal *, + struct journal_entry_res *, + unsigned); + +int bch2_journal_flush_seq_async(struct journal *, u64, struct closure *); void bch2_journal_flush_async(struct journal *, struct closure *); -void bch2_journal_meta_async(struct journal *, struct closure *); int bch2_journal_flush_seq(struct journal *, u64); int bch2_journal_flush(struct journal *); @@ -363,16 +492,21 @@ static inline void bch2_journal_set_replay_done(struct journal *j) set_bit(JOURNAL_REPLAY_DONE, &j->flags); } -ssize_t bch2_journal_print_debug(struct journal *, char *); -ssize_t bch2_journal_print_pins(struct journal *, char *); +void bch2_journal_unblock(struct journal *); +void bch2_journal_block(struct journal *); + +void bch2_journal_debug_to_text(struct printbuf *, struct journal *); +void bch2_journal_pins_to_text(struct printbuf *, struct journal *); int bch2_set_nr_journal_buckets(struct bch_fs *, struct bch_dev *, unsigned nr); int bch2_dev_journal_alloc(struct bch_dev *); void bch2_dev_journal_stop(struct journal *, struct bch_dev *); + void bch2_fs_journal_stop(struct journal *); -void bch2_fs_journal_start(struct journal *); +int bch2_fs_journal_start(struct journal *, u64, struct list_head *); + void bch2_dev_journal_exit(struct bch_dev *); int bch2_dev_journal_init(struct bch_dev *, struct bch_sb *); void bch2_fs_journal_exit(struct journal *);