u16 data_offset;
u16 aux_data_offset;
u16 end_offset;
-
- struct bpos max_key;
};
struct btree_write {
struct btree_alloc {
struct open_buckets ob;
- BKEY_PADDED(k);
+ __BKEY_PADDED(k, BKEY_BTREE_PTR_VAL_U64s_MAX);
};
struct btree_bkey_cached_common {
u16 written;
u8 nsets;
u8 nr_key_bits;
+ u16 version_ondisk;
struct bkey_format format;
struct btree_nr_keys nr;
u16 sib_u64s[2];
u16 whiteout_u64s;
- u8 page_order;
+ 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];
-
-#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 {
/* Number of elements in live + freeable lists */
unsigned used;
unsigned reserve;
+ atomic_t dirty;
struct shrinker shrink;
/*
* @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_USER_FLAGS \
- (BTREE_ITER_SLOTS \
- |BTREE_ITER_INTENT \
- |BTREE_ITER_PREFETCH \
- |BTREE_ITER_CACHED_NOFILL \
- |BTREE_ITER_CACHED_NOCREATE)
+#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;
- struct bpos pos;
- struct bpos pos_after_commit;
+ unsigned long ip_allocated;
- u16 flags;
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;
+ struct bpos real_pos;
+ struct bpos pos_after_commit;
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
return iter->flags & BTREE_ITER_TYPE;
}
+static inline bool btree_iter_is_cached(const struct btree_iter *iter)
+{
+ return btree_iter_type(iter) == BTREE_ITER_CACHED;
+}
+
static inline struct btree_iter_level *iter_l(struct btree_iter *iter)
{
return iter->l + iter->level;
struct btree_key_cache {
struct mutex lock;
struct rhashtable table;
+ bool table_init_done;
struct list_head freed;
- struct list_head clean;
+ struct shrinker shrink;
+ unsigned shrink_iter;
+
+ size_t nr_freed;
+ atomic_long_t nr_keys;
+ atomic_long_t nr_dirty;
};
struct bkey_cached_key {
struct bpos pos;
} __attribute__((packed, aligned(4)));
-#define BKEY_CACHED_DIRTY 0
+#define BKEY_CACHED_ACCESSED 0
+#define BKEY_CACHED_DIRTY 1
struct bkey_cached {
struct btree_bkey_cached_common c;
unsigned long flags;
u8 u64s;
bool valid;
+ u32 btree_trans_barrier_seq;
struct bkey_cached_key key;
struct rhash_head hash;
struct btree_insert_entry {
unsigned trigger_flags;
+ u8 bkey_type;
+ enum btree_id btree_id:8;
+ u8 level;
unsigned trans_triggers_run: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
pid_t pid;
#endif
unsigned long ip;
+ int srcu_idx;
- u64 iters_linked;
- u64 iters_live;
- u64 iters_touched;
-
- 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;
+ /*
+ * For when bch2_trans_update notices we'll be splitting a compressed
+ * extent:
+ */
+ unsigned extra_journal_res;
+
+ u64 iters_linked;
+ u64 iters_live;
+ u64 iters_touched;
unsigned mem_top;
unsigned mem_bytes;
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;
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[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_NODE_need_rewrite,
+ BTREE_NODE_never_write,
};
BTREE_FLAG(read_in_flight);
BTREE_FLAG(read_error);
-BTREE_FLAG(dirty);
BTREE_FLAG(need_write);
BTREE_FLAG(noevict);
BTREE_FLAG(write_idx);
BTREE_FLAG(just_written);
BTREE_FLAG(dying);
BTREE_FLAG(fake);
-BTREE_FLAG(old_extent_overwrite);
+BTREE_FLAG(need_rewrite);
+BTREE_FLAG(never_write);
static inline struct btree_write *btree_current_write(struct btree *b)
{
}
enum btree_node_type {
-#define x(kwd, val, name) BKEY_TYPE_##kwd = val,
+#define x(kwd, val) BKEY_TYPE_##kwd = val,
BCH_BTREE_IDS()
#undef x
- BKEY_TYPE_BTREE,
+ BKEY_TYPE_btree,
};
/* Type of a key in btree @id at level @level: */
static inline enum btree_node_type __btree_node_type(unsigned level, enum btree_id id)
{
- return level ? BKEY_TYPE_BTREE : (enum btree_node_type) id;
+ return level ? BKEY_TYPE_btree : (enum btree_node_type) id;
}
/* Type of keys @b contains: */
static inline bool btree_node_type_is_extents(enum btree_node_type type)
{
switch (type) {
- case BKEY_TYPE_EXTENTS:
- case BKEY_TYPE_REFLINK:
+ case BKEY_TYPE_extents:
+ case BKEY_TYPE_reflink:
return true;
default:
return false;
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))
+static inline enum btree_node_type btree_iter_key_type(struct btree_iter *iter)
+{
+ return __btree_node_type(iter->level, iter->btree_id);
+}
+
+static inline bool btree_iter_is_extents(struct btree_iter *iter)
+{
+ return btree_node_type_is_extents(btree_iter_key_type(iter));
+}
#define BTREE_NODE_TYPE_HAS_TRANS_TRIGGERS \
- ((1U << BKEY_TYPE_EXTENTS)| \
- (1U << BKEY_TYPE_INODES)| \
- (1U << BKEY_TYPE_REFLINK))
+ ((1U << BKEY_TYPE_extents)| \
+ (1U << BKEY_TYPE_inodes)| \
+ (1U << BKEY_TYPE_stripes)| \
+ (1U << BKEY_TYPE_reflink)| \
+ (1U << BKEY_TYPE_btree))
+
+#define BTREE_NODE_TYPE_HAS_MEM_TRIGGERS \
+ ((1U << BKEY_TYPE_alloc)| \
+ (1U << BKEY_TYPE_stripes))
+
+#define BTREE_NODE_TYPE_HAS_TRIGGERS \
+ (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_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)
+#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,
};
enum btree_gc_coalesce_fail_reason {
projects / bcachefs-tools-debian / blobdiff