bcachefs: KEY_TYPE_inode_v3, metadata_version_inode_v3

Move bi_size and bi_sectors into the non-varint portion of the inode, so
that the write path can update them without going through the relatively
expensive unpack/pack operations.

Other changes:
 - Add a field for the offset of the varint section, so we can add new
   non-varint fields without needing a new inode type, like alloc_v3
 - Move bi_mode into the flags field, so that the varint section can be
   u64 aligned
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/bcachefs_format.h

@@ -370,7 +370,8 @@ static inline void bkey_init(struct bkey *k)
 	x(set,			25)			\
 	x(lru,			26)			\
 	x(alloc_v4,		27)			\
-	x(backpointer,		28)
+	x(backpointer,		28)			\
+	x(inode_v3,		29)
 
 enum bch_bkey_type {
 #define x(name, nr) KEY_TYPE_##name	= nr,
@@ -721,6 +722,21 @@ struct bch_inode_v2 {
 	__u8			fields[0];
 } __packed __aligned(8);
 
+struct bch_inode_v3 {
+	struct bch_val		v;
+
+	__le64			bi_journal_seq;
+	__le64			bi_hash_seed;
+	__le64			bi_flags;
+	__le64			bi_sectors;
+	__le64			bi_size;
+	__le64			bi_version;
+	__u8			fields[0];
+} __packed __aligned(8);
+
+#define INODEv3_FIELDS_START_INITIAL	6
+#define INODEv3_FIELDS_START_CUR	(offsetof(struct bch_inode_v3, fields) / sizeof(u64))
+
 struct bch_inode_generation {
 	struct bch_val		v;
 
@@ -732,7 +748,7 @@ struct bch_inode_generation {
  * bi_subvol and bi_parent_subvol are only set for subvolume roots:
  */
 
-#define BCH_INODE_FIELDS()			\
+#define BCH_INODE_FIELDS_v2()			\
 	x(bi_atime,			96)	\
 	x(bi_ctime,			96)	\
 	x(bi_mtime,			96)	\
@@ -759,6 +775,31 @@ struct bch_inode_generation {
 	x(bi_subvol,			32)	\
 	x(bi_parent_subvol,		32)
 
+#define BCH_INODE_FIELDS_v3()			\
+	x(bi_atime,			96)	\
+	x(bi_ctime,			96)	\
+	x(bi_mtime,			96)	\
+	x(bi_otime,			96)	\
+	x(bi_uid,			32)	\
+	x(bi_gid,			32)	\
+	x(bi_nlink,			32)	\
+	x(bi_generation,		32)	\
+	x(bi_dev,			32)	\
+	x(bi_data_checksum,		8)	\
+	x(bi_compression,		8)	\
+	x(bi_project,			32)	\
+	x(bi_background_compression,	8)	\
+	x(bi_data_replicas,		8)	\
+	x(bi_promote_target,		16)	\
+	x(bi_foreground_target,		16)	\
+	x(bi_background_target,		16)	\
+	x(bi_erasure_code,		16)	\
+	x(bi_fields_set,		16)	\
+	x(bi_dir,			64)	\
+	x(bi_dir_offset,		64)	\
+	x(bi_subvol,			32)	\
+	x(bi_parent_subvol,		32)
+
 /* subset of BCH_INODE_FIELDS */
 #define BCH_INODE_OPTS()			\
 	x(data_checksum,		8)	\
@@ -815,6 +856,13 @@ LE32_BITMASK(INODE_NEW_VARINT,	struct bch_inode, bi_flags, 31, 32);
 LE64_BITMASK(INODEv2_STR_HASH,	struct bch_inode_v2, bi_flags, 20, 24);
 LE64_BITMASK(INODEv2_NR_FIELDS,	struct bch_inode_v2, bi_flags, 24, 31);
 
+LE64_BITMASK(INODEv3_STR_HASH,	struct bch_inode_v3, bi_flags, 20, 24);
+LE64_BITMASK(INODEv3_NR_FIELDS,	struct bch_inode_v3, bi_flags, 24, 31);
+
+LE64_BITMASK(INODEv3_FIELDS_START,
+				struct bch_inode_v3, bi_flags, 31, 36);
+LE64_BITMASK(INODEv3_MODE,	struct bch_inode_v3, bi_flags, 36, 52);
+
 /* Dirents */
 
 /*
@@ -1499,7 +1547,8 @@ struct bch_sb_field_journal_seq_blacklist {
 	x(freespace,			19)		\
 	x(alloc_v4,			20)		\
 	x(new_data_types,		21)		\
-	x(backpointers,			22)
+	x(backpointers,			22)		\
+	x(inode_v3,			23)
 
 enum bcachefs_metadata_version {
 	bcachefs_metadata_version_min = 9,

fs/bcachefs/bkey_methods.c

@@ -149,6 +149,7 @@ static unsigned bch2_key_types_allowed[] = {
 		(1U << KEY_TYPE_whiteout)|
 		(1U << KEY_TYPE_inode)|
 		(1U << KEY_TYPE_inode_v2)|
+		(1U << KEY_TYPE_inode_v3)|
 		(1U << KEY_TYPE_inode_generation),
 	[BKEY_TYPE_dirents] =
 		(1U << KEY_TYPE_deleted)|

fs/bcachefs/buckets.c

@@ -1123,10 +1123,10 @@ int bch2_mark_inode(struct btree_trans *trans,
 	u64 journal_seq = trans->journal_res.seq;
 
 	if (flags & BTREE_TRIGGER_INSERT) {
-		struct bch_inode_v2 *v = (struct bch_inode_v2 *) new.v;
+		struct bch_inode_v3 *v = (struct bch_inode_v3 *) new.v;
 
 		BUG_ON(!journal_seq);
-		BUG_ON(new.k->type != KEY_TYPE_inode_v2);
+		BUG_ON(new.k->type != KEY_TYPE_inode_v3);
 
 		v->bi_journal_seq = cpu_to_le64(journal_seq);
 	}

fs/bcachefs/inode.c

@@ -60,11 +60,10 @@ static int inode_decode_field(const u8 *in, const u8 *end,
 	return bytes;
 }
 
-static inline void bch2_inode_pack_inlined(struct bch_fs *c,
-					   struct bkey_inode_buf *packed,
+static inline void bch2_inode_pack_inlined(struct bkey_inode_buf *packed,
 					   const struct bch_inode_unpacked *inode)
 {
-	struct bkey_i_inode_v2 *k = &packed->inode;
+	struct bkey_i_inode_v3 *k = &packed->inode;
 	u8 *out = k->v.fields;
 	u8 *end = (void *) &packed[1];
 	u8 *last_nonzero_field = out;
@@ -72,13 +71,17 @@ static inline void bch2_inode_pack_inlined(struct bch_fs *c,
 	unsigned bytes;
 	int ret;
 
-	bkey_inode_v2_init(&packed->inode.k_i);
+	bkey_inode_v3_init(&packed->inode.k_i);
 	packed->inode.k.p.offset	= inode->bi_inum;
 	packed->inode.v.bi_journal_seq	= cpu_to_le64(inode->bi_journal_seq);
 	packed->inode.v.bi_hash_seed	= inode->bi_hash_seed;
 	packed->inode.v.bi_flags	= cpu_to_le64(inode->bi_flags);
-	packed->inode.v.bi_flags	= cpu_to_le64(inode->bi_flags);
-	packed->inode.v.bi_mode		= cpu_to_le16(inode->bi_mode);
+	packed->inode.v.bi_sectors	= cpu_to_le64(inode->bi_sectors);
+	packed->inode.v.bi_size		= cpu_to_le64(inode->bi_size);
+	packed->inode.v.bi_version	= cpu_to_le64(inode->bi_version);
+	SET_INODEv3_MODE(&packed->inode.v, inode->bi_mode);
+	SET_INODEv3_FIELDS_START(&packed->inode.v, INODEv3_FIELDS_START_CUR);
+
 
 #define x(_name, _bits)							\
 	nr_fields++;							\
@@ -99,7 +102,7 @@ static inline void bch2_inode_pack_inlined(struct bch_fs *c,
 			*out++ = 0;					\
 	}
 
-	BCH_INODE_FIELDS()
+	BCH_INODE_FIELDS_v3()
 #undef  x
 	BUG_ON(out > end);
 
@@ -110,7 +113,7 @@ static inline void bch2_inode_pack_inlined(struct bch_fs *c,
 	set_bkey_val_bytes(&packed->inode.k, bytes);
 	memset_u64s_tail(&packed->inode.v, 0, bytes);
 
-	SET_INODEv2_NR_FIELDS(&k->v, nr_fields);
+	SET_INODEv3_NR_FIELDS(&k->v, nr_fields);
 
 	if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG)) {
 		struct bch_inode_unpacked unpacked;
@@ -120,21 +123,23 @@ static inline void bch2_inode_pack_inlined(struct bch_fs *c,
 		BUG_ON(ret);
 		BUG_ON(unpacked.bi_inum		!= inode->bi_inum);
 		BUG_ON(unpacked.bi_hash_seed	!= inode->bi_hash_seed);
+		BUG_ON(unpacked.bi_sectors	!= inode->bi_sectors);
+		BUG_ON(unpacked.bi_size		!= inode->bi_size);
+		BUG_ON(unpacked.bi_version	!= inode->bi_version);
 		BUG_ON(unpacked.bi_mode		!= inode->bi_mode);
 
 #define x(_name, _bits)	if (unpacked._name != inode->_name)		\
 			panic("unpacked %llu should be %llu",		\
 			      (u64) unpacked._name, (u64) inode->_name);
-		BCH_INODE_FIELDS()
+		BCH_INODE_FIELDS_v3()
 #undef  x
 	}
 }
 
-void bch2_inode_pack(struct bch_fs *c,
-		     struct bkey_inode_buf *packed,
+void bch2_inode_pack(struct bkey_inode_buf *packed,
 		     const struct bch_inode_unpacked *inode)
 {
-	bch2_inode_pack_inlined(c, packed, inode);
+	bch2_inode_pack_inlined(packed, inode);
 }
 
 static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
@@ -164,7 +169,7 @@ static noinline int bch2_inode_unpack_v1(struct bkey_s_c_inode inode,
 	unpacked->_name = field[1];					\
 	in += ret;
 
-	BCH_INODE_FIELDS()
+	BCH_INODE_FIELDS_v2()
 #undef  x
 
 	/* XXX: signal if there were more fields than expected? */
@@ -203,15 +208,66 @@ static int bch2_inode_unpack_v2(struct bch_inode_unpacked *unpacked,
 		return -1;						\
 	fieldnr++;
 
-	BCH_INODE_FIELDS()
+	BCH_INODE_FIELDS_v2()
 #undef  x
 
 	/* XXX: signal if there were more fields than expected? */
 	return 0;
 }
 
-int bch2_inode_unpack(struct bkey_s_c k,
-		      struct bch_inode_unpacked *unpacked)
+static int bch2_inode_unpack_v3(struct bkey_s_c k,
+				struct bch_inode_unpacked *unpacked)
+{
+	struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
+	const u8 *in = inode.v->fields;
+	const u8 *end = bkey_val_end(inode);
+	unsigned nr_fields = INODEv3_NR_FIELDS(inode.v);
+	unsigned fieldnr = 0;
+	int ret;
+	u64 v[2];
+
+	unpacked->bi_inum	= inode.k->p.offset;
+	unpacked->bi_journal_seq= le64_to_cpu(inode.v->bi_journal_seq);
+	unpacked->bi_hash_seed	= inode.v->bi_hash_seed;
+	unpacked->bi_flags	= le64_to_cpu(inode.v->bi_flags);
+	unpacked->bi_sectors	= le64_to_cpu(inode.v->bi_sectors);
+	unpacked->bi_size	= le64_to_cpu(inode.v->bi_size);
+	unpacked->bi_version	= le64_to_cpu(inode.v->bi_version);
+	unpacked->bi_mode	= INODEv3_MODE(inode.v);
+
+#define x(_name, _bits)							\
+	if (fieldnr < nr_fields) {					\
+		ret = bch2_varint_decode_fast(in, end, &v[0]);		\
+		if (ret < 0)						\
+			return ret;					\
+		in += ret;						\
+									\
+		if (_bits > 64) {					\
+			ret = bch2_varint_decode_fast(in, end, &v[1]);	\
+			if (ret < 0)					\
+				return ret;				\
+			in += ret;					\
+		} else {						\
+			v[1] = 0;					\
+		}							\
+	} else {							\
+		v[0] = v[1] = 0;					\
+	}								\
+									\
+	unpacked->_name = v[0];						\
+	if (v[1] || v[0] != unpacked->_name)				\
+		return -1;						\
+	fieldnr++;
+
+	BCH_INODE_FIELDS_v3()
+#undef  x
+
+	/* XXX: signal if there were more fields than expected? */
+	return 0;
+}
+
+static noinline int bch2_inode_unpack_slowpath(struct bkey_s_c k,
+					       struct bch_inode_unpacked *unpacked)
 {
 	memset(unpacked, 0, sizeof(*unpacked));
 
@@ -252,6 +308,14 @@ int bch2_inode_unpack(struct bkey_s_c k,
 	}
 }
 
+int bch2_inode_unpack(struct bkey_s_c k,
+		      struct bch_inode_unpacked *unpacked)
+{
+	if (likely(k.k->type == KEY_TYPE_inode_v3))
+		return bch2_inode_unpack_v3(k, unpacked);
+	return bch2_inode_unpack_slowpath(k, unpacked);
+}
+
 int bch2_inode_peek(struct btree_trans *trans,
 		    struct btree_iter *iter,
 		    struct bch_inode_unpacked *inode,
@@ -297,11 +361,32 @@ int bch2_inode_write(struct btree_trans *trans,
 	if (IS_ERR(inode_p))
 		return PTR_ERR(inode_p);
 
-	bch2_inode_pack_inlined(trans->c, inode_p, inode);
+	bch2_inode_pack_inlined(inode_p, inode);
 	inode_p->inode.k.p.snapshot = iter->snapshot;
 	return bch2_trans_update(trans, iter, &inode_p->inode.k_i, 0);
 }
 
+struct bkey_i *bch2_inode_to_v3(struct btree_trans *trans, struct bkey_i *k)
+{
+	struct bch_inode_unpacked u;
+	struct bkey_inode_buf *inode_p;
+	int ret;
+
+	if (!bkey_is_inode(&k->k))
+		return ERR_PTR(-ENOENT);
+
+	inode_p = bch2_trans_kmalloc(trans, sizeof(*inode_p));
+	if (IS_ERR(inode_p))
+		return ERR_CAST(inode_p);
+
+	ret = bch2_inode_unpack(bkey_i_to_s_c(k), &u);
+	if (ret)
+		return ERR_PTR(ret);
+
+	bch2_inode_pack(inode_p, &u);
+	return &inode_p->inode.k_i;
+}
+
 static int __bch2_inode_invalid(struct bkey_s_c k, struct printbuf *err)
 {
 	struct bch_inode_unpacked unpacked;
@@ -387,15 +472,48 @@ int bch2_inode_v2_invalid(const struct bch_fs *c, struct bkey_s_c k,
 	return __bch2_inode_invalid(k, err);
 }
 
-static void __bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked *inode)
+int bch2_inode_v3_invalid(const struct bch_fs *c, struct bkey_s_c k,
+			  int rw, struct printbuf *err)
+{
+	struct bkey_s_c_inode_v3 inode = bkey_s_c_to_inode_v3(k);
+
+	if (bkey_val_bytes(k.k) < sizeof(*inode.v)) {
+		prt_printf(err, "incorrect value size (%zu < %zu)",
+		       bkey_val_bytes(k.k), sizeof(*inode.v));
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (INODEv3_FIELDS_START(inode.v) < INODEv3_FIELDS_START_INITIAL ||
+	    INODEv3_FIELDS_START(inode.v) > bkey_val_u64s(inode.k)) {
+		prt_printf(err, "invalid fields_start (got %llu, min %u max %zu)",
+		       INODEv3_FIELDS_START(inode.v),
+		       INODEv3_FIELDS_START_INITIAL,
+		       bkey_val_u64s(inode.k));
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	if (INODEv3_STR_HASH(inode.v) >= BCH_STR_HASH_NR) {
+		prt_printf(err, "invalid str hash type (%llu >= %u)",
+		       INODEv3_STR_HASH(inode.v), BCH_STR_HASH_NR);
+		return -BCH_ERR_invalid_bkey;
+	}
+
+	return __bch2_inode_invalid(k, err);
+}
+
+static void __bch2_inode_unpacked_to_text(struct printbuf *out,
+					  struct bch_inode_unpacked *inode)
 {
-	prt_printf(out, "mode %o flags %x journal_seq %llu",
+	prt_printf(out, "mode %o flags %x journal_seq %llu bi_size %llu bi_sectors %llu bi_version %llu",
 	       inode->bi_mode, inode->bi_flags,
-	       inode->bi_journal_seq);
+	       inode->bi_journal_seq,
+	       inode->bi_size,
+	       inode->bi_sectors,
+	       inode->bi_version);
 
 #define x(_name, _bits)						\
 	prt_printf(out, " "#_name " %llu", (u64) inode->_name);
-	BCH_INODE_FIELDS()
+	BCH_INODE_FIELDS_v3()
 #undef  x
 }
 
@@ -405,8 +523,7 @@ void bch2_inode_unpacked_to_text(struct printbuf *out, struct bch_inode_unpacked
 	__bch2_inode_unpacked_to_text(out, inode);
 }
 
-void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c,
-		       struct bkey_s_c k)
+void bch2_inode_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
 {
 	struct bch_inode_unpacked inode;
 

fs/bcachefs/inode.h

@@ -9,6 +9,7 @@ extern const char * const bch2_inode_opts[];
 
 int bch2_inode_invalid(const struct bch_fs *, struct bkey_s_c, int, struct printbuf *);
 int bch2_inode_v2_invalid(const struct bch_fs *, struct bkey_s_c, int, struct printbuf *);
+int bch2_inode_v3_invalid(const struct bch_fs *, struct bkey_s_c, int, struct printbuf *);
 void bch2_inode_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 
 #define bch2_bkey_ops_inode ((struct bkey_ops) {	\
@@ -25,10 +26,18 @@ void bch2_inode_to_text(struct printbuf *, struct bch_fs *, struct bkey_s_c);
 	.atomic_trigger	= bch2_mark_inode,		\
 })
 
+#define bch2_bkey_ops_inode_v3 ((struct bkey_ops) {	\
+	.key_invalid	= bch2_inode_v3_invalid,	\
+	.val_to_text	= bch2_inode_to_text,		\
+	.trans_trigger	= bch2_trans_mark_inode,	\
+	.atomic_trigger	= bch2_mark_inode,		\
+})
+
 static inline bool bkey_is_inode(const struct bkey *k)
 {
 	return  k->type == KEY_TYPE_inode ||
-		k->type == KEY_TYPE_inode_v2;
+		k->type == KEY_TYPE_inode_v2 ||
+		k->type == KEY_TYPE_inode_v3;
 }
 
 int bch2_inode_generation_invalid(const struct bch_fs *, struct bkey_s_c,
@@ -52,25 +61,28 @@ struct bch_inode_unpacked {
 	u64			bi_inum;
 	u64			bi_journal_seq;
 	__le64			bi_hash_seed;
+	u64			bi_size;
+	u64			bi_sectors;
+	u64			bi_version;
 	u32			bi_flags;
 	u16			bi_mode;
 
 #define x(_name, _bits)	u##_bits _name;
-	BCH_INODE_FIELDS()
+	BCH_INODE_FIELDS_v3()
 #undef  x
 };
 
 struct bkey_inode_buf {
-	struct bkey_i_inode_v2	inode;
+	struct bkey_i_inode_v3	inode;
 
 #define x(_name, _bits)		+ 8 + _bits / 8
-	u8		_pad[0 + BCH_INODE_FIELDS()];
+	u8		_pad[0 + BCH_INODE_FIELDS_v3()];
 #undef  x
 } __packed __aligned(8);
 
-void bch2_inode_pack(struct bch_fs *, struct bkey_inode_buf *,
-		     const struct bch_inode_unpacked *);
+void bch2_inode_pack(struct bkey_inode_buf *, const struct bch_inode_unpacked *);
 int bch2_inode_unpack(struct bkey_s_c, struct bch_inode_unpacked *);
+struct bkey_i *bch2_inode_to_v3(struct btree_trans *, struct bkey_i *);
 
 void bch2_inode_unpacked_to_text(struct printbuf *, struct bch_inode_unpacked *);
 

fs/bcachefs/io.c

@@ -356,7 +356,7 @@ int bch2_extent_update(struct btree_trans *trans,
 	}
 
 	if (i_sectors_delta || new_i_size) {
-		bch2_inode_pack(trans->c, &inode_p, &inode_u);
+		bch2_inode_pack(&inode_p, &inode_u);
 
 		inode_p.inode.k.p.snapshot = iter->snapshot;
 

fs/bcachefs/recovery.c

@@ -1098,6 +1098,9 @@ int bch2_fs_recovery(struct bch_fs *c)
 			c->opts.version_upgrade	= true;
 			c->opts.fsck		= true;
 			c->opts.fix_errors	= FSCK_OPT_YES;
+		} else if (c->sb.version < bcachefs_metadata_version_inode_v3) {
+			bch_info(c, "version prior to inode_v3, upgrade required");
+			c->opts.version_upgrade	= true;
 		}
 	}
 
@@ -1482,7 +1485,7 @@ int bch2_fs_initialize(struct bch_fs *c)
 	c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
 	c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
 
-	if (c->sb.version < bcachefs_metadata_version_backpointers)
+	if (c->sb.version < bcachefs_metadata_version_inode_v3)
 		c->opts.version_upgrade	= true;
 
 	if (c->opts.version_upgrade) {
@@ -1563,7 +1566,7 @@ int bch2_fs_initialize(struct bch_fs *c)
 	bch2_inode_init(c, &root_inode, 0, 0, S_IFDIR|0755, 0, NULL);
 	root_inode.bi_inum	= BCACHEFS_ROOT_INO;
 	root_inode.bi_subvol	= BCACHEFS_ROOT_SUBVOL;
-	bch2_inode_pack(c, &packed_inode, &root_inode);
+	bch2_inode_pack(&packed_inode, &root_inode);
 	packed_inode.inode.k.p.snapshot = U32_MAX;
 
 	err = "error creating root directory";