+/* SPDX-License-Identifier: GPL-2.0 */
#ifndef _BCACHEFS_BTREE_IO_H
#define _BCACHEFS_BTREE_IO_H
+#include "bkey_methods.h"
+#include "bset.h"
+#include "btree_locking.h"
+#include "checksum.h"
#include "extents.h"
-#include "io_types.h"
+#include "io_write_types.h"
struct bch_fs;
struct btree_write;
struct btree;
struct btree_iter;
+struct btree_node_read_all;
+
+static inline void set_btree_node_dirty_acct(struct bch_fs *c, struct btree *b)
+{
+ if (!test_and_set_bit(BTREE_NODE_dirty, &b->flags))
+ atomic_inc(&c->btree_cache.dirty);
+}
+
+static inline void clear_btree_node_dirty_acct(struct bch_fs *c, struct btree *b)
+{
+ if (test_and_clear_bit(BTREE_NODE_dirty, &b->flags))
+ atomic_dec(&c->btree_cache.dirty);
+}
+
+static inline unsigned btree_ptr_sectors_written(struct bkey_i *k)
+{
+ return k->k.type == KEY_TYPE_btree_ptr_v2
+ ? le16_to_cpu(bkey_i_to_btree_ptr_v2(k)->v.sectors_written)
+ : 0;
+}
struct btree_read_bio {
struct bch_fs *c;
- unsigned submit_time_us;
+ struct btree *b;
+ struct btree_node_read_all *ra;
u64 start_time;
- struct extent_pick_ptr pick;
+ unsigned have_ioref:1;
+ unsigned idx:7;
+ struct extent_ptr_decoded pick;
struct work_struct work;
struct bio bio;
};
struct btree_write_bio {
- void *data;
struct work_struct work;
+ __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
+ void *data;
+ unsigned data_bytes;
+ unsigned sector_offset;
struct bch_write_bio wbio;
};
-static inline void btree_node_io_unlock(struct btree *b)
-{
- EBUG_ON(!btree_node_write_in_flight(b));
- clear_btree_node_write_in_flight(b);
- wake_up_bit(&b->flags, BTREE_NODE_write_in_flight);
-}
+void bch2_btree_node_io_unlock(struct btree *);
+void bch2_btree_node_io_lock(struct btree *);
+void __bch2_btree_node_wait_on_read(struct btree *);
+void __bch2_btree_node_wait_on_write(struct btree *);
+void bch2_btree_node_wait_on_read(struct btree *);
+void bch2_btree_node_wait_on_write(struct btree *);
-static inline void btree_node_io_lock(struct btree *b)
+enum compact_mode {
+ COMPACT_LAZY,
+ COMPACT_ALL,
+};
+
+bool bch2_compact_whiteouts(struct bch_fs *, struct btree *,
+ enum compact_mode);
+
+static inline bool should_compact_bset_lazy(struct btree *b,
+ struct bset_tree *t)
{
- wait_on_bit_lock_io(&b->flags, BTREE_NODE_write_in_flight,
- TASK_UNINTERRUPTIBLE);
+ unsigned total_u64s = bset_u64s(t);
+ unsigned dead_u64s = bset_dead_u64s(b, t);
+
+ return dead_u64s > 64 && dead_u64s * 3 > total_u64s;
}
-static inline void btree_node_wait_on_io(struct btree *b)
+static inline bool bch2_maybe_compact_whiteouts(struct bch_fs *c, struct btree *b)
{
- wait_on_bit_io(&b->flags, BTREE_NODE_write_in_flight,
- TASK_UNINTERRUPTIBLE);
+ struct bset_tree *t;
+
+ for_each_bset(b, t)
+ if (should_compact_bset_lazy(b, t))
+ return bch2_compact_whiteouts(c, b, COMPACT_LAZY);
+
+ return false;
}
-static inline bool btree_node_may_write(struct btree *b)
+static inline struct nonce btree_nonce(struct bset *i, unsigned offset)
{
- return list_empty_careful(&b->write_blocked) &&
- !b->will_make_reachable;
+ return (struct nonce) {{
+ [0] = cpu_to_le32(offset),
+ [1] = ((__le32 *) &i->seq)[0],
+ [2] = ((__le32 *) &i->seq)[1],
+ [3] = ((__le32 *) &i->journal_seq)[0]^BCH_NONCE_BTREE,
+ }};
}
-enum compact_mode {
- COMPACT_LAZY,
- COMPACT_WRITTEN,
- COMPACT_WRITTEN_NO_WRITE_LOCK,
-};
-
-bool __bch2_compact_whiteouts(struct bch_fs *, struct btree *, enum compact_mode);
-
-static inline bool bch2_maybe_compact_whiteouts(struct bch_fs *c, struct btree *b)
+static inline int bset_encrypt(struct bch_fs *c, struct bset *i, unsigned offset)
{
- struct bset_tree *t;
+ struct nonce nonce = btree_nonce(i, offset);
+ int ret;
+
+ if (!offset) {
+ struct btree_node *bn = container_of(i, struct btree_node, keys);
+ unsigned bytes = (void *) &bn->keys - (void *) &bn->flags;
- for_each_bset(b, t) {
- unsigned live_u64s = b->nr.bset_u64s[t - b->set];
- unsigned bset_u64s = le16_to_cpu(bset(b, t)->u64s);
+ ret = bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce,
+ &bn->flags, bytes);
+ if (ret)
+ return ret;
- if (live_u64s * 4 < bset_u64s * 3)
- goto compact;
+ nonce = nonce_add(nonce, round_up(bytes, CHACHA_BLOCK_SIZE));
}
- return false;
-compact:
- return __bch2_compact_whiteouts(c, b, COMPACT_LAZY);
+ return bch2_encrypt(c, BSET_CSUM_TYPE(i), nonce, i->_data,
+ vstruct_end(i) - (void *) i->_data);
}
void bch2_btree_sort_into(struct bch_fs *, struct btree *, struct btree *);
+void bch2_btree_node_drop_keys_outside_node(struct btree *);
+
void bch2_btree_build_aux_trees(struct btree *);
-void bch2_btree_init_next(struct bch_fs *, struct btree *,
- struct btree_iter *);
+void bch2_btree_init_next(struct btree_trans *, struct btree *);
-int bch2_btree_node_read_done(struct bch_fs *, struct btree *, bool);
-void bch2_btree_node_read(struct bch_fs *, struct btree *, bool);
+int bch2_btree_node_read_done(struct bch_fs *, struct bch_dev *,
+ struct btree *, bool, bool *);
+void bch2_btree_node_read(struct btree_trans *, struct btree *, bool);
int bch2_btree_root_read(struct bch_fs *, enum btree_id,
const struct bkey_i *, unsigned);
-void bch2_btree_complete_write(struct bch_fs *, struct btree *,
- struct btree_write *);
-void bch2_btree_write_error_work(struct work_struct *);
-
-void __bch2_btree_node_write(struct bch_fs *, struct btree *,
- enum six_lock_type);
bool bch2_btree_post_write_cleanup(struct bch_fs *, struct btree *);
+enum btree_write_flags {
+ __BTREE_WRITE_ONLY_IF_NEED = BTREE_WRITE_TYPE_BITS,
+ __BTREE_WRITE_ALREADY_STARTED,
+};
+#define BTREE_WRITE_ONLY_IF_NEED BIT(__BTREE_WRITE_ONLY_IF_NEED)
+#define BTREE_WRITE_ALREADY_STARTED BIT(__BTREE_WRITE_ALREADY_STARTED)
+
+void __bch2_btree_node_write(struct bch_fs *, struct btree *, unsigned);
void bch2_btree_node_write(struct bch_fs *, struct btree *,
- enum six_lock_type);
-
-/*
- * btree_node_dirty() can be cleared with only a read lock,
- * and for bch2_btree_node_write_cond() we want to set need_write iff it's
- * still dirty:
- */
-static inline void set_btree_node_need_write_if_dirty(struct btree *b)
+ enum six_lock_type, unsigned);
+
+static inline void btree_node_write_if_need(struct bch_fs *c, struct btree *b,
+ enum six_lock_type lock_held)
{
- unsigned long old, new, v = READ_ONCE(b->flags);
+ bch2_btree_node_write(c, b, lock_held, BTREE_WRITE_ONLY_IF_NEED);
+}
+
+bool bch2_btree_flush_all_reads(struct bch_fs *);
+bool bch2_btree_flush_all_writes(struct bch_fs *);
+
+static inline void compat_bformat(unsigned level, enum btree_id btree_id,
+ unsigned version, unsigned big_endian,
+ int write, struct bkey_format *f)
+{
+ if (version < bcachefs_metadata_version_inode_btree_change &&
+ btree_id == BTREE_ID_inodes) {
+ swap(f->bits_per_field[BKEY_FIELD_INODE],
+ f->bits_per_field[BKEY_FIELD_OFFSET]);
+ swap(f->field_offset[BKEY_FIELD_INODE],
+ f->field_offset[BKEY_FIELD_OFFSET]);
+ }
+
+ if (version < bcachefs_metadata_version_snapshot &&
+ (level || btree_type_has_snapshots(btree_id))) {
+ u64 max_packed =
+ ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
- do {
- old = new = v;
+ f->field_offset[BKEY_FIELD_SNAPSHOT] = write
+ ? 0
+ : cpu_to_le64(U32_MAX - max_packed);
+ }
+}
+
+static inline void compat_bpos(unsigned level, enum btree_id btree_id,
+ unsigned version, unsigned big_endian,
+ int write, struct bpos *p)
+{
+ if (big_endian != CPU_BIG_ENDIAN)
+ bch2_bpos_swab(p);
- if (!(old & (1 << BTREE_NODE_dirty)))
- return;
+ if (version < bcachefs_metadata_version_inode_btree_change &&
+ btree_id == BTREE_ID_inodes)
+ swap(p->inode, p->offset);
+}
- new |= (1 << BTREE_NODE_need_write);
- } while ((v = cmpxchg(&b->flags, old, new)) != old);
+static inline void compat_btree_node(unsigned level, enum btree_id btree_id,
+ unsigned version, unsigned big_endian,
+ int write,
+ struct btree_node *bn)
+{
+ if (version < bcachefs_metadata_version_inode_btree_change &&
+ btree_id_is_extents(btree_id) &&
+ !bpos_eq(bn->min_key, POS_MIN) &&
+ write)
+ bn->min_key = bpos_nosnap_predecessor(bn->min_key);
+
+ if (version < bcachefs_metadata_version_snapshot &&
+ write)
+ bn->max_key.snapshot = 0;
+
+ compat_bpos(level, btree_id, version, big_endian, write, &bn->min_key);
+ compat_bpos(level, btree_id, version, big_endian, write, &bn->max_key);
+
+ if (version < bcachefs_metadata_version_snapshot &&
+ !write)
+ bn->max_key.snapshot = U32_MAX;
+
+ if (version < bcachefs_metadata_version_inode_btree_change &&
+ btree_id_is_extents(btree_id) &&
+ !bpos_eq(bn->min_key, POS_MIN) &&
+ !write)
+ bn->min_key = bpos_nosnap_successor(bn->min_key);
}
-#define bch2_btree_node_write_cond(_c, _b, cond) \
-do { \
- while ((_b)->written && btree_node_dirty(_b) && (cond)) { \
- if (!btree_node_may_write(_b)) { \
- set_btree_node_need_write_if_dirty(_b); \
- break; \
- } \
- \
- if (!btree_node_write_in_flight(_b)) { \
- bch2_btree_node_write(_c, _b, SIX_LOCK_read); \
- break; \
- } \
- \
- six_unlock_read(&(_b)->lock); \
- btree_node_wait_on_io(_b); \
- six_lock_read(&(_b)->lock); \
- } \
-} while (0)
-
-void bch2_btree_verify_flushed(struct bch_fs *);
+void bch2_btree_write_stats_to_text(struct printbuf *, struct bch_fs *);
#endif /* _BCACHEFS_BTREE_IO_H */