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>

1 files changed, 140 insertions, 23 deletions

diff --git a/fs/bcachefs/inode.c b/fs/bcachefs/inode.c
index 9eeabe70aec1..f338cf6fd8b7 100644
--- a/fs/bcachefs/inode.c
+++ b/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;