struct btree_write {
struct journal_entry_pin journal;
- struct closure_waitlist wait;
};
struct btree_alloc {
BKEY_PADDED(k);
};
-struct btree {
- /* Hottest entries first */
- struct rhash_head hash;
+struct btree_bkey_cached_common {
+ struct six_lock lock;
+ u8 level;
+ u8 btree_id;
+};
- /* Key/pointer for this btree node */
- __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
+struct btree {
+ struct btree_bkey_cached_common c;
- struct six_lock lock;
+ struct rhash_head hash;
+ u64 hash_val;
unsigned long flags;
u16 written;
- u8 level;
- u8 btree_id;
u8 nsets;
u8 nr_key_bits;
struct btree_nr_keys nr;
u16 sib_u64s[2];
u16 whiteout_u64s;
- u16 uncompacted_whiteout_u64s;
u8 page_order;
u8 unpack_fn_len;
#ifdef CONFIG_BCACHEFS_DEBUG
bool *expensive_debug_checks;
#endif
+
+ /* Key/pointer for this btree node */
+ __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
};
struct btree_cache {
enum btree_iter_type {
BTREE_ITER_KEYS,
- BTREE_ITER_SLOTS,
BTREE_ITER_NODES,
+ BTREE_ITER_CACHED,
};
#define BTREE_ITER_TYPE ((1 << 2) - 1)
-#define BTREE_ITER_INTENT (1 << 2)
-#define BTREE_ITER_PREFETCH (1 << 3)
+/*
+ * Iterate over all possible positions, synthesizing deleted keys for holes:
+ */
+#define BTREE_ITER_SLOTS (1 << 2)
+/*
+ * Indicates that intent locks should be taken on leaf nodes, because we expect
+ * to be doing updates:
+ */
+#define BTREE_ITER_INTENT (1 << 3)
+/*
+ * Causes the btree iterator code to prefetch additional btree nodes from disk:
+ */
+#define BTREE_ITER_PREFETCH (1 << 4)
+/*
+ * Indicates that this iterator should not be reused until transaction commit,
+ * either because a pending update references it or because the update depends
+ * on that particular key being locked (e.g. by the str_hash code, for hash
+ * table consistency)
+ */
+#define BTREE_ITER_KEEP_UNTIL_COMMIT (1 << 5)
/*
* Used in bch2_btree_iter_traverse(), to indicate whether we're searching for
* @pos or the first key strictly greater than @pos
*/
-#define BTREE_ITER_IS_EXTENTS (1 << 4)
-#define BTREE_ITER_ERROR (1 << 5)
+#define BTREE_ITER_IS_EXTENTS (1 << 6)
+#define BTREE_ITER_ERROR (1 << 7)
+#define BTREE_ITER_SET_POS_AFTER_COMMIT (1 << 8)
+#define BTREE_ITER_CACHED_NOFILL (1 << 9)
+#define BTREE_ITER_CACHED_NOCREATE (1 << 10)
+
+#define BTREE_ITER_USER_FLAGS \
+ (BTREE_ITER_SLOTS \
+ |BTREE_ITER_INTENT \
+ |BTREE_ITER_PREFETCH \
+ |BTREE_ITER_CACHED_NOFILL \
+ |BTREE_ITER_CACHED_NOCREATE)
enum btree_iter_uptodate {
BTREE_ITER_UPTODATE = 0,
BTREE_ITER_NEED_TRAVERSE = 3,
};
+#define BTREE_ITER_NO_NODE_GET_LOCKS ((struct btree *) 1)
+#define BTREE_ITER_NO_NODE_DROP ((struct btree *) 2)
+#define BTREE_ITER_NO_NODE_LOCK_ROOT ((struct btree *) 3)
+#define BTREE_ITER_NO_NODE_UP ((struct btree *) 4)
+#define BTREE_ITER_NO_NODE_DOWN ((struct btree *) 5)
+#define BTREE_ITER_NO_NODE_INIT ((struct btree *) 6)
+#define BTREE_ITER_NO_NODE_ERROR ((struct btree *) 7)
+
/*
* @pos - iterator's current position
* @level - current btree depth
* @nodes_intent_locked - bitmask indicating which locks are intent locks
*/
struct btree_iter {
- u8 idx;
-
struct btree_trans *trans;
struct bpos pos;
+ struct bpos pos_after_commit;
+
+ u16 flags;
+ u8 idx;
- u8 flags;
- enum btree_iter_uptodate uptodate:4;
enum btree_id btree_id:4;
+ enum btree_iter_uptodate uptodate:4;
unsigned level:4,
+ min_depth:4,
locks_want:4,
nodes_locked:4,
nodes_intent_locked:4;
* bch2_btree_iter_next_slot() can correctly advance pos.
*/
struct bkey k;
+ unsigned long ip_allocated;
+};
+
+static inline enum btree_iter_type
+btree_iter_type(const struct btree_iter *iter)
+{
+ return iter->flags & BTREE_ITER_TYPE;
+}
+
+static inline struct btree_iter_level *iter_l(struct btree_iter *iter)
+{
+ return iter->l + iter->level;
+}
- u64 id;
+struct btree_key_cache {
+ struct mutex lock;
+ struct rhashtable table;
+ struct list_head freed;
+ struct list_head clean;
};
-struct deferred_update {
- struct journal_preres res;
- struct journal_entry_pin journal;
+struct bkey_cached_key {
+ u32 btree_id;
+ struct bpos pos;
+} __packed;
- spinlock_t lock;
- unsigned dirty:1;
+#define BKEY_CACHED_DIRTY 0
- u8 allocated_u64s;
- enum btree_id btree_id;
+struct bkey_cached {
+ struct btree_bkey_cached_common c;
- /* must be last: */
- struct bkey_i k;
+ unsigned long flags;
+ u8 u64s;
+ bool valid;
+ struct bkey_cached_key key;
+
+ struct rhash_head hash;
+ struct list_head list;
+
+ struct journal_preres res;
+ struct journal_entry_pin journal;
+
+ struct bkey_i *k;
};
struct btree_insert_entry {
+ unsigned trigger_flags;
+ unsigned trans_triggers_run:1;
struct bkey_i *k;
-
- union {
struct btree_iter *iter;
- struct deferred_update *d;
- };
-
- bool deferred;
- bool triggered;
- bool marked;
};
+#ifndef CONFIG_LOCKDEP
#define BTREE_ITER_MAX 64
+#else
+#define BTREE_ITER_MAX 32
+#endif
struct btree_trans {
struct bch_fs *c;
+#ifdef CONFIG_BCACHEFS_DEBUG
+ struct list_head list;
+ struct btree *locking;
+ unsigned locking_iter_idx;
+ struct bpos locking_pos;
+ u8 locking_btree_id;
+ u8 locking_level;
+ pid_t pid;
+#endif
unsigned long ip;
- u64 commit_start;
u64 iters_linked;
u64 iters_live;
u64 iters_touched;
- u64 iters_unlink_on_restart;
- u64 iters_unlink_on_commit;
u8 nr_iters;
u8 nr_updates;
+ u8 nr_updates2;
u8 size;
unsigned used_mempool:1;
unsigned error:1;
unsigned nounlock:1;
+ unsigned need_reset:1;
+ unsigned in_traverse_all:1;
unsigned mem_top;
unsigned mem_bytes;
struct btree_iter *iters;
struct btree_insert_entry *updates;
+ struct btree_insert_entry *updates2;
/* update path: */
+ struct jset_entry *extra_journal_entries;
+ unsigned extra_journal_entry_u64s;
+ struct journal_entry_pin *journal_pin;
+
struct journal_res journal_res;
struct journal_preres journal_preres;
u64 *journal_seq;
struct disk_reservation *disk_res;
unsigned flags;
unsigned journal_u64s;
+ unsigned journal_preres_u64s;
+ struct replicas_delta_list *fs_usage_deltas;
struct btree_iter iters_onstack[2];
- struct btree_insert_entry updates_onstack[6];
-
- struct replicas_delta_list *fs_usage_deltas;
+ struct btree_insert_entry updates_onstack[2];
+ struct btree_insert_entry updates2_onstack[2];
};
#define BTREE_FLAG(flag) \
BTREE_NODE_just_written,
BTREE_NODE_dying,
BTREE_NODE_fake,
+ BTREE_NODE_old_extent_overwrite,
};
BTREE_FLAG(read_in_flight);
BTREE_FLAG(just_written);
BTREE_FLAG(dying);
BTREE_FLAG(fake);
+BTREE_FLAG(old_extent_overwrite);
static inline struct btree_write *btree_current_write(struct btree *b)
{
__btree_node_offset_to_key(_b, (_t)->end_offset); \
})
+static inline unsigned bset_u64s(struct bset_tree *t)
+{
+ return t->end_offset - t->data_offset -
+ sizeof(struct bset) / sizeof(u64);
+}
+
+static inline unsigned bset_dead_u64s(struct btree *b, struct bset_tree *t)
+{
+ return bset_u64s(t) - b->nr.bset_u64s[t - b->set];
+}
+
static inline unsigned bset_byte_offset(struct btree *b, void *i)
{
return i - (void *) b->data;
/* Type of keys @b contains: */
static inline enum btree_node_type btree_node_type(struct btree *b)
{
- return __btree_node_type(b->level, b->btree_id);
+ return __btree_node_type(b->c.level, b->c.btree_id);
}
static inline bool btree_node_type_is_extents(enum btree_node_type type)
{
- return type == BKEY_TYPE_EXTENTS;
+ switch (type) {
+ case BKEY_TYPE_EXTENTS:
+ case BKEY_TYPE_REFLINK:
+ return true;
+ default:
+ return false;
+ }
}
static inline bool btree_node_is_extents(struct btree *b)
return btree_node_type_is_extents(btree_node_type(b));
}
+#define BTREE_NODE_TYPE_HAS_TRIGGERS \
+ ((1U << BKEY_TYPE_EXTENTS)| \
+ (1U << BKEY_TYPE_ALLOC)| \
+ (1U << BKEY_TYPE_INODES)| \
+ (1U << BKEY_TYPE_REFLINK)| \
+ (1U << BKEY_TYPE_EC)| \
+ (1U << BKEY_TYPE_BTREE))
+
+#define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS \
+ ((1U << BKEY_TYPE_EXTENTS)| \
+ (1U << BKEY_TYPE_INODES)| \
+ (1U << BKEY_TYPE_REFLINK))
+
+enum btree_trigger_flags {
+ __BTREE_TRIGGER_NORUN, /* Don't run triggers at all */
+ __BTREE_TRIGGER_NOOVERWRITES, /* Don't run triggers on overwrites */
+
+ __BTREE_TRIGGER_INSERT,
+ __BTREE_TRIGGER_OVERWRITE,
+ __BTREE_TRIGGER_OVERWRITE_SPLIT,
+
+ __BTREE_TRIGGER_GC,
+ __BTREE_TRIGGER_BUCKET_INVALIDATE,
+ __BTREE_TRIGGER_ALLOC_READ,
+ __BTREE_TRIGGER_NOATOMIC,
+};
+
+#define BTREE_TRIGGER_NORUN (1U << __BTREE_TRIGGER_NORUN)
+#define BTREE_TRIGGER_NOOVERWRITES (1U << __BTREE_TRIGGER_NOOVERWRITES)
+
+#define BTREE_TRIGGER_INSERT (1U << __BTREE_TRIGGER_INSERT)
+#define BTREE_TRIGGER_OVERWRITE (1U << __BTREE_TRIGGER_OVERWRITE)
+#define BTREE_TRIGGER_OVERWRITE_SPLIT (1U << __BTREE_TRIGGER_OVERWRITE_SPLIT)
+
+#define BTREE_TRIGGER_GC (1U << __BTREE_TRIGGER_GC)
+#define BTREE_TRIGGER_BUCKET_INVALIDATE (1U << __BTREE_TRIGGER_BUCKET_INVALIDATE)
+#define BTREE_TRIGGER_ALLOC_READ (1U << __BTREE_TRIGGER_ALLOC_READ)
+#define BTREE_TRIGGER_NOATOMIC (1U << __BTREE_TRIGGER_NOATOMIC)
+
static inline bool btree_node_type_needs_gc(enum btree_node_type type)
{
- switch (type) {
- case BKEY_TYPE_ALLOC:
- case BKEY_TYPE_BTREE:
- case BKEY_TYPE_EXTENTS:
- case BKEY_TYPE_INODES:
- case BKEY_TYPE_EC:
- return true;
- default:
- return false;
- }
+ return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);
}
struct btree_root {
struct btree *b;
- struct btree_update *as;
-
/* On disk root - see async splits: */
__BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
u8 level;
projects / bcachefs-tools-debian / blobdiff