u16 data_offset;
u16 aux_data_offset;
u16 end_offset;
-
- struct bpos max_key;
};
struct btree_write {
u16 written;
u8 nsets;
u8 nr_key_bits;
+ u16 version_ondisk;
struct bkey_format format;
u8 byte_order;
u8 unpack_fn_len;
+ struct btree_write writes[2];
+
+ /* Key/pointer for this btree node */
+ __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
+
/*
* XXX: add a delete sequence number, so when bch2_btree_node_relock()
* fails because the lock sequence number has changed - i.e. the
/* lru list */
struct list_head list;
-
- struct btree_write writes[2];
-
- /* Key/pointer for this btree node */
- __BKEY_PADDED(key, BKEY_BTREE_PTR_VAL_U64s_MAX);
};
struct btree_cache {
* @pos or the first key strictly greater than @pos
*/
#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_NOT_EXTENTS (1 << 11)
+#define BTREE_ITER_NOT_EXTENTS (1 << 7)
+#define BTREE_ITER_ERROR (1 << 8)
+#define BTREE_ITER_SET_POS_AFTER_COMMIT (1 << 9)
+#define BTREE_ITER_CACHED_NOFILL (1 << 10)
+#define BTREE_ITER_CACHED_NOCREATE (1 << 11)
+#define BTREE_ITER_WITH_UPDATES (1 << 12)
+#define BTREE_ITER_ALL_SNAPSHOTS (1 << 13)
enum btree_iter_uptodate {
BTREE_ITER_UPTODATE = 0,
#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)
+#define BTREE_ITER_NO_NODE_CACHED ((struct btree *) 8)
/*
* @pos - iterator's current position
*/
struct btree_iter {
struct btree_trans *trans;
+ unsigned long ip_allocated;
+
+ u8 idx;
+ u8 child_idx;
+
+ /* btree_iter_copy starts here: */
+ u16 flags;
+
+ /* When we're filtering by snapshot, the snapshot ID we're looking for: */
+ unsigned snapshot;
+
struct bpos pos;
- /* what we're searching for/what the iterator actually points to: */
struct bpos real_pos;
struct bpos pos_after_commit;
- u16 flags;
- u8 idx;
-
enum btree_id btree_id:4;
- enum btree_iter_uptodate uptodate:4;
+ enum btree_iter_uptodate uptodate:3;
+ /*
+ * True if we've returned a key (and thus are expected to keep it
+ * locked), false after set_pos - for avoiding spurious transaction
+ * restarts in bch2_trans_relock():
+ */
+ bool should_be_locked:1;
unsigned level:4,
min_depth:4,
locks_want:4,
* bch2_btree_iter_next_slot() can correctly advance pos.
*/
struct bkey k;
- unsigned long ip_allocated;
};
static inline enum btree_iter_type
struct rhashtable table;
bool table_init_done;
struct list_head freed;
- struct list_head clean;
- struct list_head dirty;
struct shrinker shrink;
+ unsigned shrink_iter;
size_t nr_freed;
- size_t nr_keys;
- size_t nr_dirty;
+ atomic_long_t nr_keys;
+ atomic_long_t nr_dirty;
};
struct bkey_cached_key {
struct btree_insert_entry {
unsigned trigger_flags;
u8 bkey_type;
- u8 btree_id;
+ enum btree_id btree_id:8;
u8 level;
unsigned trans_triggers_run:1;
- unsigned is_extent:1;
struct bkey_i *k;
struct btree_iter *iter;
};
#define BTREE_ITER_MAX 32
#endif
+struct btree_trans_commit_hook;
+typedef int (btree_trans_commit_hook_fn)(struct btree_trans *, struct btree_trans_commit_hook *);
+
+struct btree_trans_commit_hook {
+ btree_trans_commit_hook_fn *fn;
+ struct btree_trans_commit_hook *next;
+};
+
+#define BTREE_TRANS_MEM_MAX (1U << 14)
+
struct btree_trans {
struct bch_fs *c;
#ifdef CONFIG_BCACHEFS_DEBUG
int srcu_idx;
u8 nr_updates;
- u8 nr_updates2;
unsigned used_mempool:1;
unsigned error:1;
- unsigned nounlock:1;
unsigned in_traverse_all:1;
+ /*
+ * For when bch2_trans_update notices we'll be splitting a compressed
+ * extent:
+ */
+ unsigned extra_journal_res;
u64 iters_linked;
u64 iters_live;
struct btree_iter *iters;
struct btree_insert_entry *updates;
- struct btree_insert_entry *updates2;
/* update path: */
+ struct btree_trans_commit_hook *hooks;
struct jset_entry *extra_journal_entries;
unsigned extra_journal_entry_u64s;
struct journal_entry_pin *journal_pin;
(BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS| \
BTREE_NODE_TYPE_HAS_MEM_TRIGGERS)
+#define BTREE_ID_HAS_SNAPSHOTS \
+ ((1U << BTREE_ID_extents)| \
+ (1U << BTREE_ID_inodes)| \
+ (1U << BTREE_ID_dirents)| \
+ (1U << BTREE_ID_xattrs))
+
+#define BTREE_ID_HAS_PTRS \
+ ((1U << BTREE_ID_extents)| \
+ (1U << BTREE_ID_reflink))
+
+static inline bool btree_type_has_snapshots(enum btree_id id)
+{
+ return (1 << id) & BTREE_ID_HAS_SNAPSHOTS;
+}
+
enum btree_trigger_flags {
__BTREE_TRIGGER_NORUN, /* Don't run triggers at all */
__BTREE_TRIGGER_INSERT,
__BTREE_TRIGGER_OVERWRITE,
- __BTREE_TRIGGER_OVERWRITE_SPLIT,
__BTREE_TRIGGER_GC,
__BTREE_TRIGGER_BUCKET_INVALIDATE,
#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_NOATOMIC (1U << __BTREE_TRIGGER_NOATOMIC)
+#define BTREE_TRIGGER_WANTS_OLD_AND_NEW \
+ ((1U << KEY_TYPE_stripe)| \
+ (1U << KEY_TYPE_inode))
+
static inline bool btree_node_type_needs_gc(enum btree_node_type type)
{
return BTREE_NODE_TYPE_HAS_TRIGGERS & (1U << type);
BTREE_INSERT_OK,
/* leaf node needs to be split */
BTREE_INSERT_BTREE_NODE_FULL,
- BTREE_INSERT_ENOSPC,
BTREE_INSERT_NEED_MARK_REPLICAS,
BTREE_INSERT_NEED_JOURNAL_RES,
BTREE_INSERT_NEED_JOURNAL_RECLAIM,