bcachefs: Indirect inline data extents

When inline data extents were added, reflink was forgotten about - we
need indirect inline data extents for reflink + inline data to work
correctly.

This patch adds them, and a new feature bit that's flipped when they're
used.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/bcachefs_format.h

@@ -344,7 +344,8 @@ static inline void bkey_init(struct bkey *k)
 	x(reflink_p,		15)			\
 	x(reflink_v,		16)			\
 	x(inline_data,		17)			\
-	x(btree_ptr_v2,		18)
+	x(btree_ptr_v2,		18)			\
+	x(indirect_inline_data,	19)
 
 enum bch_bkey_type {
 #define x(name, nr) KEY_TYPE_##name	= nr,
@@ -890,6 +891,12 @@ struct bch_reflink_v {
 	__u64			_data[0];
 };
 
+struct bch_indirect_inline_data {
+	struct bch_val		v;
+	__le64			refcount;
+	u8			data[0];
+};
+
 /* Inline data */
 
 struct bch_inline_data {
@@ -1326,7 +1333,8 @@ LE64_BITMASK(BCH_SB_ERASURE_CODE,	struct bch_sb, flags[3],  0, 16);
 	x(incompressible,		10)	\
 	x(btree_ptr_v2,			11)	\
 	x(extents_above_btree_updates,	12)	\
-	x(btree_updates_journalled,	13)
+	x(btree_updates_journalled,	13)	\
+	x(reflink_inline_data,		14)
 
 #define BCH_SB_FEATURES_ALL				\
 	((1ULL << BCH_FEATURE_new_siphash)|		\

fs/bcachefs/bkey.h

@@ -573,6 +573,7 @@ BKEY_VAL_ACCESSORS(reflink_p);
 BKEY_VAL_ACCESSORS(reflink_v);
 BKEY_VAL_ACCESSORS(inline_data);
 BKEY_VAL_ACCESSORS(btree_ptr_v2);
+BKEY_VAL_ACCESSORS(indirect_inline_data);
 
 /* byte order helpers */
 

fs/bcachefs/bkey_methods.c

@@ -72,7 +72,11 @@ static const char *key_type_inline_data_invalid(const struct bch_fs *c,
 static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c,
 					 struct bkey_s_c k)
 {
-	pr_buf(out, "(%zu bytes)", bkey_val_bytes(k.k));
+	struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k);
+	unsigned datalen = bkey_inline_data_bytes(k.k);
+
+	pr_buf(out, "datalen %u: %*phN",
+	       datalen, min(datalen, 32U), d.v->data);
 }
 
 #define bch2_bkey_ops_inline_data (struct bkey_ops) {	\

fs/bcachefs/buckets.c

@@ -1811,6 +1811,18 @@ static int bch2_trans_mark_stripe(struct btree_trans *trans,
 	return ret;
 }
 
+static __le64 *bkey_refcount(struct bkey_i *k)
+{
+	switch (k->k.type) {
+	case KEY_TYPE_reflink_v:
+		return &bkey_i_to_reflink_v(k)->v.refcount;
+	case KEY_TYPE_indirect_inline_data:
+		return &bkey_i_to_indirect_inline_data(k)->v.refcount;
+	default:
+		return NULL;
+	}
+}
+
 static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
 			struct bkey_s_c_reflink_p p,
 			u64 idx, unsigned sectors,
@@ -1819,7 +1831,8 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
 	struct bch_fs *c = trans->c;
 	struct btree_iter *iter;
 	struct bkey_s_c k;
-	struct bkey_i_reflink_v *r_v;
+	struct bkey_i *n;
+	__le64 *refcount;
 	s64 ret;
 
 	ret = trans_get_key(trans, BTREE_ID_REFLINK,
@@ -1827,14 +1840,6 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
 	if (ret < 0)
 		return ret;
 
-	if (k.k->type != KEY_TYPE_reflink_v) {
-		bch2_fs_inconsistent(c,
-			"%llu:%llu len %u points to nonexistent indirect extent %llu",
-			p.k->p.inode, p.k->p.offset, p.k->size, idx);
-		ret = -EIO;
-		goto err;
-	}
-
 	if ((flags & BTREE_TRIGGER_OVERWRITE) &&
 	    (bkey_start_offset(k.k) < idx ||
 	     k.k->p.offset > idx + sectors))
@@ -1842,25 +1847,33 @@ static int __bch2_trans_mark_reflink_p(struct btree_trans *trans,
 
 	sectors = k.k->p.offset - idx;
 
-	r_v = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
-	ret = PTR_ERR_OR_ZERO(r_v);
+	n = bch2_trans_kmalloc(trans, bkey_bytes(k.k));
+	ret = PTR_ERR_OR_ZERO(n);
 	if (ret)
 		goto err;
 
-	bkey_reassemble(&r_v->k_i, k);
+	bkey_reassemble(n, k);
+
+	refcount = bkey_refcount(n);
+	if (!refcount) {
+		bch2_fs_inconsistent(c,
+			"%llu:%llu len %u points to nonexistent indirect extent %llu",
+			p.k->p.inode, p.k->p.offset, p.k->size, idx);
+		ret = -EIO;
+		goto err;
+	}
 
-	le64_add_cpu(&r_v->v.refcount,
-		     !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1);
+	le64_add_cpu(refcount, !(flags & BTREE_TRIGGER_OVERWRITE) ? 1 : -1);
 
-	if (!r_v->v.refcount) {
-		r_v->k.type = KEY_TYPE_deleted;
-		set_bkey_val_u64s(&r_v->k, 0);
+	if (!*refcount) {
+		n->k.type = KEY_TYPE_deleted;
+		set_bkey_val_u64s(&n->k, 0);
 	}
 
 	bch2_btree_iter_set_pos(iter, bkey_start_pos(k.k));
 	BUG_ON(iter->uptodate > BTREE_ITER_NEED_PEEK);
 
-	bch2_trans_update(trans, iter, &r_v->k_i, 0);
+	bch2_trans_update(trans, iter, n, 0);
 out:
 	ret = sectors;
 err:

fs/bcachefs/extents.c

@@ -1199,14 +1199,14 @@ int bch2_cut_front_s(struct bpos where, struct bkey_s k)
 		le64_add_cpu(&p.v->idx, sub);
 		break;
 	}
-	case KEY_TYPE_inline_data: {
-		struct bkey_s_inline_data d = bkey_s_to_inline_data(k);
+	case KEY_TYPE_inline_data:
+	case KEY_TYPE_indirect_inline_data: {
+		void *p = bkey_inline_data_p(k);
+		unsigned bytes = bkey_inline_data_bytes(k.k);
 
-		sub = min_t(u64, sub << 9, bkey_val_bytes(d.k));
+		sub = min_t(u64, sub << 9, bytes);
 
-		memmove(d.v->data,
-			d.v->data + sub,
-			bkey_val_bytes(d.k) - sub);
+		memmove(p, p + sub, bytes - sub);
 
 		new_val_u64s -= sub >> 3;
 		break;
@@ -1244,7 +1244,9 @@ int bch2_cut_back_s(struct bpos where, struct bkey_s k)
 
 	switch (k.k->type) {
 	case KEY_TYPE_inline_data:
-		new_val_u64s = min(new_val_u64s, k.k->size << 6);
+	case KEY_TYPE_indirect_inline_data:
+		new_val_u64s = (bkey_inline_data_offset(k.k) +
+				min(bkey_inline_data_bytes(k.k), k.k->size << 9)) >> 3;
 		break;
 	}
 

fs/bcachefs/extents.h

@@ -445,10 +445,35 @@ static inline bool bkey_extent_is_direct_data(const struct bkey *k)
 	}
 }
 
+static inline bool bkey_extent_is_inline_data(const struct bkey *k)
+{
+	return  k->type == KEY_TYPE_inline_data ||
+		k->type == KEY_TYPE_indirect_inline_data;
+}
+
+static inline unsigned bkey_inline_data_offset(const struct bkey *k)
+{
+	switch (k->type) {
+	case KEY_TYPE_inline_data:
+		return sizeof(struct bch_inline_data);
+	case KEY_TYPE_indirect_inline_data:
+		return sizeof(struct bch_indirect_inline_data);
+	default:
+		BUG();
+	}
+}
+
+static inline unsigned bkey_inline_data_bytes(const struct bkey *k)
+{
+	return bkey_val_bytes(k) - bkey_inline_data_offset(k);
+}
+
+#define bkey_inline_data_p(_k)	(((void *) (_k).v) + bkey_inline_data_offset((_k).k))
+
 static inline bool bkey_extent_is_data(const struct bkey *k)
 {
-	return bkey_extent_is_direct_data(k) ||
-		k->type == KEY_TYPE_inline_data ||
+	return  bkey_extent_is_direct_data(k) ||
+		bkey_extent_is_inline_data(k) ||
 		k->type == KEY_TYPE_reflink_p;
 }
 
@@ -463,6 +488,7 @@ static inline bool bkey_extent_is_allocation(const struct bkey *k)
 	case KEY_TYPE_reflink_p:
 	case KEY_TYPE_reflink_v:
 	case KEY_TYPE_inline_data:
+	case KEY_TYPE_indirect_inline_data:
 		return true;
 	default:
 		return false;

fs/bcachefs/io.c

@@ -2000,7 +2000,8 @@ int __bch2_read_indirect_extent(struct btree_trans *trans,
 	if (ret)
 		goto err;
 
-	if (k.k->type != KEY_TYPE_reflink_v) {
+	if (k.k->type != KEY_TYPE_reflink_v &&
+	    k.k->type != KEY_TYPE_indirect_inline_data) {
 		__bcache_io_error(trans->c,
 				"pointer to nonexistent indirect extent");
 		ret = -EIO;
@@ -2027,13 +2028,12 @@ int __bch2_read_extent(struct bch_fs *c, struct bch_read_bio *orig,
 	struct bpos pos = bkey_start_pos(k.k);
 	int pick_ret;
 
-	if (k.k->type == KEY_TYPE_inline_data) {
-		struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k);
+	if (bkey_extent_is_inline_data(k.k)) {
 		unsigned bytes = min_t(unsigned, iter.bi_size,
-				       bkey_val_bytes(d.k));
+				       bkey_inline_data_bytes(k.k));
 
 		swap(iter.bi_size, bytes);
-		memcpy_to_bio(&orig->bio, iter, d.v->data);
+		memcpy_to_bio(&orig->bio, iter, bkey_inline_data_p(k));
 		swap(iter.bi_size, bytes);
 		bio_advance_iter(&orig->bio, &iter, bytes);
 		zero_fill_bio_iter(&orig->bio, iter);

fs/bcachefs/opts.h

@@ -185,7 +185,7 @@ enum opt_type {
 	x(inline_data,			u8,				\
 	  OPT_MOUNT|OPT_RUNTIME,					\
 	  OPT_BOOL(),							\
-	  NO_SB_OPT,			false,				\
+	  NO_SB_OPT,			true,				\
 	  NULL,		"Enable inline data extents")			\
 	x(acl,				u8,				\
 	  OPT_FORMAT|OPT_MOUNT,						\

fs/bcachefs/reflink.c

@@ -9,6 +9,18 @@
 
 #include <linux/sched/signal.h>
 
+static inline unsigned bkey_type_to_indirect(const struct bkey *k)
+{
+	switch (k->type) {
+	case KEY_TYPE_extent:
+		return KEY_TYPE_reflink_v;
+	case KEY_TYPE_inline_data:
+		return KEY_TYPE_indirect_inline_data;
+	default:
+		return 0;
+	}
+}
+
 /* reflink pointers */
 
 const char *bch2_reflink_p_invalid(const struct bch_fs *c, struct bkey_s_c k)
@@ -71,17 +83,42 @@ void bch2_reflink_v_to_text(struct printbuf *out, struct bch_fs *c,
 	bch2_bkey_ptrs_to_text(out, c, k);
 }
 
+/* indirect inline data */
+
+const char *bch2_indirect_inline_data_invalid(const struct bch_fs *c,
+					      struct bkey_s_c k)
+{
+	if (bkey_val_bytes(k.k) < sizeof(struct bch_indirect_inline_data))
+		return "incorrect value size";
+	return NULL;
+}
+
+void bch2_indirect_inline_data_to_text(struct printbuf *out,
+					struct bch_fs *c, struct bkey_s_c k)
+{
+	struct bkey_s_c_indirect_inline_data d = bkey_s_c_to_indirect_inline_data(k);
+	unsigned datalen = bkey_inline_data_bytes(k.k);
+
+	pr_buf(out, "refcount %llu datalen %u: %*phN",
+	       le64_to_cpu(d.v->refcount), datalen,
+	       min(datalen, 32U), d.v->data);
+}
+
 static int bch2_make_extent_indirect(struct btree_trans *trans,
 				     struct btree_iter *extent_iter,
-				     struct bkey_i_extent *e)
+				     struct bkey_i *orig)
 {
 	struct bch_fs *c = trans->c;
 	struct btree_iter *reflink_iter;
 	struct bkey_s_c k;
-	struct bkey_i_reflink_v *r_v;
+	struct bkey_i *r_v;
 	struct bkey_i_reflink_p *r_p;
+	__le64 *refcount;
 	int ret;
 
+	if (orig->k.type == KEY_TYPE_inline_data)
+		bch2_check_set_feature(c, BCH_FEATURE_reflink_inline_data);
+
 	for_each_btree_key(trans, reflink_iter, BTREE_ID_REFLINK,
 			   POS(0, c->reflink_hint),
 			   BTREE_ITER_INTENT|BTREE_ITER_SLOTS, k, ret) {
@@ -90,7 +127,7 @@ static int bch2_make_extent_indirect(struct btree_trans *trans,
 			continue;
 		}
 
-		if (bkey_deleted(k.k) && e->k.size <= k.k->size)
+		if (bkey_deleted(k.k) && orig->k.size <= k.k->size)
 			break;
 	}
 
@@ -100,29 +137,31 @@ static int bch2_make_extent_indirect(struct btree_trans *trans,
 	/* rewind iter to start of hole, if necessary: */
 	bch2_btree_iter_set_pos(reflink_iter, bkey_start_pos(k.k));
 
-	r_v = bch2_trans_kmalloc(trans, sizeof(*r_v) + bkey_val_bytes(&e->k));
+	r_v = bch2_trans_kmalloc(trans, sizeof(__le64) + bkey_val_bytes(&orig->k));
 	ret = PTR_ERR_OR_ZERO(r_v);
 	if (ret)
 		goto err;
 
-	bkey_reflink_v_init(&r_v->k_i);
+	bkey_init(&r_v->k);
+	r_v->k.type	= bkey_type_to_indirect(&orig->k);
 	r_v->k.p	= reflink_iter->pos;
-	bch2_key_resize(&r_v->k, e->k.size);
-	r_v->k.version	= e->k.version;
+	bch2_key_resize(&r_v->k, orig->k.size);
+	r_v->k.version	= orig->k.version;
+
+	set_bkey_val_bytes(&r_v->k, sizeof(__le64) + bkey_val_bytes(&orig->k));
 
-	set_bkey_val_u64s(&r_v->k, bkey_val_u64s(&r_v->k) +
-			  bkey_val_u64s(&e->k));
-	r_v->v.refcount	= 0;
-	memcpy(r_v->v.start, e->v.start, bkey_val_bytes(&e->k));
+	refcount	= (void *) &r_v->v;
+	*refcount	= 0;
+	memcpy(refcount + 1, &orig->v, bkey_val_bytes(&orig->k));
 
-	bch2_trans_update(trans, reflink_iter, &r_v->k_i, 0);
+	bch2_trans_update(trans, reflink_iter, r_v, 0);
 
 	r_p = bch2_trans_kmalloc(trans, sizeof(*r_p));
 	if (IS_ERR(r_p))
 		return PTR_ERR(r_p);
 
-	e->k.type = KEY_TYPE_reflink_p;
-	r_p = bkey_i_to_reflink_p(&e->k_i);
+	orig->k.type = KEY_TYPE_reflink_p;
+	r_p = bkey_i_to_reflink_p(orig);
 	set_bkey_val_bytes(&r_p->k, sizeof(r_p->v));
 	r_p->v.idx = cpu_to_le64(bkey_start_offset(&r_v->k));
 
@@ -144,8 +183,7 @@ static struct bkey_s_c get_next_src(struct btree_iter *iter, struct bpos end)
 		if (bkey_cmp(iter->pos, end) >= 0)
 			return bkey_s_c_null;
 
-		if (k.k->type == KEY_TYPE_extent ||
-		    k.k->type == KEY_TYPE_reflink_p)
+		if (bkey_extent_is_data(k.k))
 			break;
 	}
 
@@ -218,7 +256,7 @@ s64 bch2_remap_range(struct bch_fs *c,
 		if (!bkey_cmp(dst_iter->pos, dst_end))
 			break;
 
-		if (src_k.k->type == KEY_TYPE_extent) {
+		if (src_k.k->type != KEY_TYPE_reflink_p) {
 			bkey_on_stack_reassemble(&new_src, c, src_k);
 			src_k = bkey_i_to_s_c(new_src.k);
 
@@ -226,7 +264,7 @@ s64 bch2_remap_range(struct bch_fs *c,
 			bch2_cut_back(src_end,		new_src.k);
 
 			ret = bch2_make_extent_indirect(&trans, src_iter,
-						bkey_i_to_extent(new_src.k));
+						new_src.k);
 			if (ret)
 				goto btree_err;
 

fs/bcachefs/reflink.h

@@ -18,13 +18,22 @@ const char *bch2_reflink_v_invalid(const struct bch_fs *, struct bkey_s_c);
 void bch2_reflink_v_to_text(struct printbuf *, struct bch_fs *,
 			    struct bkey_s_c);
 
-
 #define bch2_bkey_ops_reflink_v (struct bkey_ops) {		\
 	.key_invalid	= bch2_reflink_v_invalid,		\
 	.val_to_text	= bch2_reflink_v_to_text,		\
 	.swab		= bch2_ptr_swab,			\
 }
 
+const char *bch2_indirect_inline_data_invalid(const struct bch_fs *,
+					      struct bkey_s_c);
+void bch2_indirect_inline_data_to_text(struct printbuf *,
+				struct bch_fs *, struct bkey_s_c);
+
+#define bch2_bkey_ops_indirect_inline_data (struct bkey_ops) {	\
+	.key_invalid	= bch2_indirect_inline_data_invalid,	\
+	.val_to_text	= bch2_indirect_inline_data_to_text,	\
+}
+
 s64 bch2_remap_range(struct bch_fs *, struct bpos, struct bpos,
 		     u64, u64 *, u64, s64 *);