bcachefs: bucket_gens btree

To improve mount times, add a btree for just bucket gens, 256 of them
per key: this means we'll have to scan drastically less metadata at
startup.

This adds
 - trigger for keeping it in sync with the all btree
 - initialization code, for filesystems from previous versions
 - new path for reading bucket gens
 - new fsck code

And a new on disk format version.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/alloc_background.h

@@ -148,6 +148,16 @@ void bch2_alloc_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 	.atomic_trigger	= bch2_mark_alloc,		\
 })
 
+int bch2_bucket_gens_invalid(const struct bch_fs *, struct bkey_s_c, int, struct printbuf *);
+void bch2_bucket_gens_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
+
+#define bch2_bkey_ops_bucket_gens ((struct bkey_ops) {	\
+	.key_invalid	= bch2_bucket_gens_invalid,	\
+	.val_to_text	= bch2_bucket_gens_to_text,	\
+})
+
+int bch2_bucket_gens_init(struct bch_fs *);
+
 static inline bool bkey_is_alloc(const struct bkey *k)
 {
 	return  k->type == KEY_TYPE_alloc ||
@@ -156,6 +166,7 @@ static inline bool bkey_is_alloc(const struct bkey *k)
 }
 
 int bch2_alloc_read(struct bch_fs *);
+int bch2_bucket_gens_read(struct bch_fs *);
 
 int bch2_trans_mark_alloc(struct btree_trans *, enum btree_id, unsigned,
 			  struct bkey_s_c, struct bkey_i *, unsigned);

fs/bcachefs/bcachefs.h

@@ -441,6 +441,7 @@ enum gc_phase {
 	GC_PHASE_BTREE_freespace,
 	GC_PHASE_BTREE_need_discard,
 	GC_PHASE_BTREE_backpointers,
+	GC_PHASE_BTREE_bucket_gens,
 
 	GC_PHASE_PENDING_DELETE,
 };

fs/bcachefs/bcachefs_format.h

@@ -371,7 +371,8 @@ static inline void bkey_init(struct bkey *k)
 	x(lru,			26)			\
 	x(alloc_v4,		27)			\
 	x(backpointer,		28)			\
-	x(inode_v3,		29)
+	x(inode_v3,		29)			\
+	x(bucket_gens,		30)
 
 enum bch_bkey_type {
 #define x(name, nr) KEY_TYPE_##name	= nr,
@@ -1013,6 +1014,15 @@ struct bch_backpointer {
 	struct bpos		pos;
 } __packed __aligned(8);
 
+#define KEY_TYPE_BUCKET_GENS_BITS	8
+#define KEY_TYPE_BUCKET_GENS_NR		(1U << KEY_TYPE_BUCKET_GENS_BITS)
+#define KEY_TYPE_BUCKET_GENS_MASK	(KEY_TYPE_BUCKET_GENS_NR - 1)
+
+struct bch_bucket_gens {
+	struct bch_val		v;
+	u8			gens[KEY_TYPE_BUCKET_GENS_NR];
+} __packed __aligned(8);
+
 /* Quotas: */
 
 enum quota_types {
@@ -1551,7 +1561,8 @@ struct bch_sb_field_journal_seq_blacklist {
 	x(new_data_types,		21)		\
 	x(backpointers,			22)		\
 	x(inode_v3,			23)		\
-	x(unwritten_extents,		24)
+	x(unwritten_extents,		24)		\
+	x(bucket_gens,			25)
 
 enum bcachefs_metadata_version {
 	bcachefs_metadata_version_min = 9,
@@ -2086,7 +2097,8 @@ LE32_BITMASK(JSET_NO_FLUSH,	struct jset, flags, 5, 6);
 	x(lru,			10)		\
 	x(freespace,		11)		\
 	x(need_discard,		12)		\
-	x(backpointers,		13)
+	x(backpointers,		13)		\
+	x(bucket_gens,		14)
 
 enum btree_id {
 #define x(kwd, val) BTREE_ID_##kwd = val,

fs/bcachefs/bkey_methods.c

@@ -196,6 +196,9 @@ static unsigned bch2_key_types_allowed[] = {
 	[BKEY_TYPE_backpointers] =
 		(1U << KEY_TYPE_deleted)|
 		(1U << KEY_TYPE_backpointer),
+	[BKEY_TYPE_bucket_gens] =
+		(1U << KEY_TYPE_deleted)|
+		(1U << KEY_TYPE_bucket_gens),
 	[BKEY_TYPE_btree] =
 		(1U << KEY_TYPE_deleted)|
 		(1U << KEY_TYPE_btree_ptr)|

fs/bcachefs/recovery.c

@@ -929,6 +929,7 @@ static bool btree_id_is_alloc(enum btree_id id)
 	case BTREE_ID_backpointers:
 	case BTREE_ID_need_discard:
 	case BTREE_ID_freespace:
+	case BTREE_ID_bucket_gens:
 		return true;
 	default:
 		return false;
@@ -1237,7 +1238,9 @@ int bch2_fs_recovery(struct bch_fs *c)
 	err = "error reading allocation information";
 
 	down_read(&c->gc_lock);
-	ret = bch2_alloc_read(c);
+	ret = c->sb.version < bcachefs_metadata_version_bucket_gens
+		? bch2_alloc_read(c)
+		: bch2_bucket_gens_read(c);
 	up_read(&c->gc_lock);
 
 	if (ret)
@@ -1362,6 +1365,16 @@ int bch2_fs_recovery(struct bch_fs *c)
 	if (ret)
 		goto err;
 
+	if (c->sb.version < bcachefs_metadata_version_bucket_gens &&
+	    c->opts.version_upgrade) {
+		bch_info(c, "initializing bucket_gens");
+		err = "error initializing bucket gens";
+		ret = bch2_bucket_gens_init(c);
+		if (ret)
+			goto err;
+		bch_verbose(c, "bucket_gens init done");
+	}
+
 	if (c->sb.version < bcachefs_metadata_version_snapshot_2) {
 		/* set bi_subvol on root inode */
 		err = "error upgrade root inode for subvolumes";