bcachefs: don't expose extent whiteouts outside of btree code

6 files changed, 59 insertions, 64 deletions

diff --git a/fs/bcachefs/bset.h b/fs/bcachefs/bset.h
index 5895362e0af3..296c05b4f07a 100644
--- a/fs/bcachefs/bset.h
+++ b/fs/bcachefs/bset.h
@@ -406,7 +406,7 @@ bch2_bkey_prev_all(struct btree *b, struct bset_tree *t, struct bkey_packed *k)
 static inline struct bkey_packed *
 bch2_bkey_prev(struct btree *b, struct bset_tree *t, struct bkey_packed *k)
 {
-	return bch2_bkey_prev_filter(b, t, k, KEY_TYPE_DELETED + 1);
+	return bch2_bkey_prev_filter(b, t, k, KEY_TYPE_DISCARD + 1);
 }
 
 enum bch_extent_overlap {
@@ -552,7 +552,7 @@ bch2_btree_node_iter_peek_all(struct btree_node_iter *iter,
 static inline struct bkey_packed *
 bch2_btree_node_iter_peek(struct btree_node_iter *iter, struct btree *b)
 {
-	return bch2_btree_node_iter_peek_filter(iter, b, KEY_TYPE_DELETED + 1);
+	return bch2_btree_node_iter_peek_filter(iter, b, KEY_TYPE_DISCARD + 1);
 }
 
 static inline struct bkey_packed *
@@ -578,7 +578,7 @@ bch2_btree_node_iter_prev_all(struct btree_node_iter *iter, struct btree *b)
 static inline struct bkey_packed *
 bch2_btree_node_iter_prev(struct btree_node_iter *iter, struct btree *b)
 {
-	return bch2_btree_node_iter_prev_filter(iter, b, KEY_TYPE_DELETED + 1);
+	return bch2_btree_node_iter_prev_filter(iter, b, KEY_TYPE_DISCARD + 1);
 }
 
 /*
diff --git a/fs/bcachefs/btree_iter.c b/fs/bcachefs/btree_iter.c
index c44cadb8e784..a776869f3c78 100644
--- a/fs/bcachefs/btree_iter.c
+++ b/fs/bcachefs/btree_iter.c
@@ -1338,7 +1338,7 @@ struct bkey_s_c bch2_btree_iter_next(struct btree_iter *iter)
 		p = bch2_btree_node_iter_peek_all(&l->iter, l->b);
 		if (unlikely(!p))
 			return bch2_btree_iter_peek_next_leaf(iter);
-	} while (bkey_deleted(p));
+	} while (bkey_whiteout(p));
 
 	k = __btree_iter_unpack(iter, l, &iter->k, p);
 
@@ -1406,13 +1406,6 @@ recheck:
 	       bkey_cmp(bkey_start_pos(k.k), iter->pos) == 0)
 		__btree_iter_advance(l);
 
-	if (k.k && bkey_cmp(bkey_start_pos(k.k), iter->pos) <= 0) {
-		EBUG_ON(bkey_cmp(k.k->p, iter->pos) < 0);
-		EBUG_ON(bkey_deleted(k.k));
-		iter->uptodate = BTREE_ITER_UPTODATE;
-		return k;
-	}
-
 	/*
 	 * If we got to the end of the node, check if we need to traverse to the
 	 * next node:
@@ -1426,6 +1419,15 @@ recheck:
 		goto recheck;
 	}
 
+	if (k.k &&
+	    !bkey_whiteout(k.k) &&
+	    bkey_cmp(bkey_start_pos(k.k), iter->pos) <= 0) {
+		EBUG_ON(bkey_cmp(k.k->p, iter->pos) < 0);
+		EBUG_ON(bkey_deleted(k.k));
+		iter->uptodate = BTREE_ITER_UPTODATE;
+		return k;
+	}
+
 	/* hole */
 	bkey_init(&n);
 	n.p = iter->pos;
@@ -1439,6 +1441,13 @@ recheck:
 			goto recheck;
 		}
 
+		if (k.k && bkey_whiteout(k.k)) {
+			struct btree_node_iter node_iter = l->iter;
+
+			k = __btree_iter_unpack(iter, l, &iter->k,
+				bch2_btree_node_iter_peek(&node_iter, l->b));
+		}
+
 		if (!k.k)
 			k.k = &l->b->key.k;
 
diff --git a/fs/bcachefs/btree_update_leaf.c b/fs/bcachefs/btree_update_leaf.c
index 5149bcd6ca8a..ad3242efb42e 100644
--- a/fs/bcachefs/btree_update_leaf.c
+++ b/fs/bcachefs/btree_update_leaf.c
@@ -428,6 +428,7 @@ int __bch2_btree_insert_at(struct btree_insert *trans)
 		BUG_ON(i->iter->level);
 		BUG_ON(bkey_cmp(bkey_start_pos(&i->k->k), i->iter->pos));
 		BUG_ON(debug_check_bkeys(c) &&
+		       !bkey_deleted(&i->k->k) &&
 		       bch2_bkey_invalid(c, i->iter->btree_id,
 					 bkey_i_to_s_c(i->k)));
 	}
@@ -644,11 +645,6 @@ int bch2_btree_delete_range(struct bch_fs *c, enum btree_id id,
 		if (bkey_cmp(iter.pos, end) >= 0)
 			break;
 
-		if (k.k->type == KEY_TYPE_DISCARD) {
-			bch2_btree_iter_next(&iter);
-			continue;
-		}
-
 		bkey_init(&delete.k);
 
 		/*
@@ -665,15 +661,6 @@ int bch2_btree_delete_range(struct bch_fs *c, enum btree_id id,
 		delete.k.version = version;
 
 		if (iter.flags & BTREE_ITER_IS_EXTENTS) {
-			/*
-			 * The extents btree is special - KEY_TYPE_DISCARD is
-			 * used for deletions, not KEY_TYPE_DELETED. This is an
-			 * internal implementation detail that probably
-			 * shouldn't be exposed (internally, KEY_TYPE_DELETED is
-			 * used as a proxy for k->size == 0):
-			 */
-			delete.k.type = KEY_TYPE_DISCARD;
-
 			/* create the biggest key we can */
 			bch2_key_resize(&delete.k, max_sectors);
 			bch2_cut_back(end, &delete.k);
diff --git a/fs/bcachefs/extents.c b/fs/bcachefs/extents.c
index e65ddb461b0b..fe4bb52717cf 100644
--- a/fs/bcachefs/extents.c
+++ b/fs/bcachefs/extents.c
@@ -1143,10 +1143,13 @@ static void extent_bset_insert(struct bch_fs *c, struct btree_iter *iter,
 	struct bset_tree *t = bset_tree_last(l->b);
 	struct bkey_packed *where =
 		bch2_btree_node_iter_bset_pos(&l->iter, l->b, t);
-	struct bkey_packed *prev = bch2_bkey_prev(l->b, t, where);
+	struct bkey_packed *prev = bch2_bkey_prev_filter(l->b, t, where,
+							 KEY_TYPE_DISCARD);
 	struct bkey_packed *next_live_key = where;
 	unsigned clobber_u64s;
 
+	EBUG_ON(bkey_deleted(&insert->k) || !insert->k.size);
+
 	if (prev)
 		where = bkey_next(prev);
 
@@ -1188,6 +1191,7 @@ static void extent_insert_committed(struct extent_insert_state *s)
 		: &s->whiteout;
 	BKEY_PADDED(k) split;
 
+	EBUG_ON(bkey_deleted(&insert->k) || !insert->k.size);
 	EBUG_ON(bkey_cmp(insert->k.p, s->committed) < 0);
 	EBUG_ON(bkey_cmp(s->committed, bkey_start_pos(&insert->k)) < 0);
 
@@ -1246,8 +1250,6 @@ __extent_insert_advance_pos(struct extent_insert_state *s,
 	else
 		ret = BTREE_INSERT_OK;
 
-	EBUG_ON(bkey_deleted(&s->insert->k->k) || !s->insert->k->k.size);
-
 	if (ret == BTREE_INSERT_OK)
 		s->committed = next_pos;
 
@@ -1446,6 +1448,7 @@ __bch2_delete_fixup_extent(struct extent_insert_state *s)
 	EBUG_ON(bkey_cmp(iter->pos, bkey_start_pos(&insert->k)));
 
 	s->whiteout = *insert;
+	s->whiteout.k.type = KEY_TYPE_DISCARD;
 
 	while (bkey_cmp(s->committed, insert->k.p) < 0 &&
 	       (ret = extent_insert_should_stop(s)) == BTREE_INSERT_OK &&
@@ -1488,6 +1491,8 @@ __bch2_delete_fixup_extent(struct extent_insert_state *s)
 			   bset_written(b, bset(b, t))) {
 			struct bkey_i discard = *insert;
 
+			discard.k.type = KEY_TYPE_DISCARD;
+
 			switch (overlap) {
 			case BCH_EXTENT_OVERLAP_FRONT:
 				bch2_cut_front(bkey_start_pos(k.k), &discard);
@@ -1634,7 +1639,7 @@ bch2_insert_fixup_extent(struct btree_insert *trans,
 	};
 
 	EBUG_ON(iter->level);
-	EBUG_ON(bkey_deleted(&insert->k->k) || !insert->k->k.size);
+	EBUG_ON(!insert->k->k.size);
 
 	/*
 	 * As we process overlapping extents, we advance @iter->pos both to
@@ -2051,11 +2056,6 @@ int bch2_extent_pick_ptr(struct bch_fs *c, struct bkey_s_c k,
 	int ret;
 
 	switch (k.k->type) {
-	case KEY_TYPE_DELETED:
-	case KEY_TYPE_DISCARD:
-	case KEY_TYPE_COOKIE:
-		return 0;
-
 	case KEY_TYPE_ERROR:
 		return -EIO;
 
@@ -2069,11 +2069,8 @@ int bch2_extent_pick_ptr(struct bch_fs *c, struct bkey_s_c k,
 
 		return ret;
 
-	case BCH_RESERVATION:
-		return 0;
-
 	default:
-		BUG();
+		return 0;
 	}
 }
 
@@ -2099,7 +2096,6 @@ enum merge_result bch2_extent_merge(struct bch_fs *c, struct btree *b,
 		return BCH_MERGE_NOMERGE;
 
 	switch (l->k.type) {
-	case KEY_TYPE_DELETED:
 	case KEY_TYPE_DISCARD:
 	case KEY_TYPE_ERROR:
 		/* These types are mergeable, and no val to check */
diff --git a/fs/bcachefs/fs-io.c b/fs/bcachefs/fs-io.c
index 737b9be33e0a..9e78798a4d08 100644
--- a/fs/bcachefs/fs-io.c
+++ b/fs/bcachefs/fs-io.c
@@ -2078,6 +2078,29 @@ out:
 
 /* truncate: */
 
+static inline int range_has_data(struct bch_fs *c,
+				  struct bpos start,
+				  struct bpos end)
+{
+
+	struct btree_iter iter;
+	struct bkey_s_c k;
+	int ret = 0;
+
+	for_each_btree_key(&iter, c, BTREE_ID_EXTENTS,
+			   start, 0, k) {
+		if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
+			break;
+
+		if (bkey_extent_is_data(k.k)) {
+			ret = 1;
+			break;
+		}
+	}
+
+	return bch2_btree_iter_unlock(&iter) ?: ret;
+}
+
 static int __bch2_truncate_page(struct bch_inode_info *inode,
 				pgoff_t index, loff_t start, loff_t end)
 {
@@ -2099,30 +2122,16 @@ static int __bch2_truncate_page(struct bch_inode_info *inode,
 
 	page = find_lock_page(mapping, index);
 	if (!page) {
-		struct btree_iter iter;
-		struct bkey_s_c k = bkey_s_c_null;
-
 		/*
 		 * XXX: we're doing two index lookups when we end up reading the
 		 * page
 		 */
-		for_each_btree_key(&iter, c, BTREE_ID_EXTENTS,
-				   POS(inode->v.i_ino,
-				       index << PAGE_SECTOR_SHIFT), 0, k) {
-			if (bkey_cmp(bkey_start_pos(k.k),
-				     POS(inode->v.i_ino,
-					 (index + 1) << PAGE_SECTOR_SHIFT)) >= 0)
-				break;
+		ret = range_has_data(c,
+				POS(inode->v.i_ino, index << PAGE_SECTOR_SHIFT),
+				POS(inode->v.i_ino, (index + 1) << PAGE_SECTOR_SHIFT));
+		if (ret <= 0)
+			return ret;
 
-			if (k.k->type != KEY_TYPE_DISCARD &&
-			    k.k->type != BCH_RESERVATION) {
-				bch2_btree_iter_unlock(&iter);
-				goto create;
-			}
-		}
-		bch2_btree_iter_unlock(&iter);
-		return 0;
-create:
 		page = find_or_create_page(mapping, index, GFP_KERNEL);
 		if (unlikely(!page)) {
 			ret = -ENOMEM;
@@ -2389,9 +2398,6 @@ static long bch2_fcollapse(struct bch_inode_info *inode,
 
 		bkey_reassemble(&copy.k, k);
 
-		if (bkey_deleted(&copy.k.k))
-			copy.k.k.type = KEY_TYPE_DISCARD;
-
 		bch2_cut_front(src.pos, &copy.k);
 		copy.k.k.p.offset -= len >> 9;
 
diff --git a/fs/bcachefs/fsck.c b/fs/bcachefs/fsck.c
index 048b5c103eb0..edf714f7b98d 100644
--- a/fs/bcachefs/fsck.c
+++ b/fs/bcachefs/fsck.c
@@ -252,9 +252,6 @@ static int check_extents(struct bch_fs *c)
 
 	for_each_btree_key(&iter, c, BTREE_ID_EXTENTS,
 			   POS(BCACHEFS_ROOT_INO, 0), 0, k) {
-		if (k.k->type == KEY_TYPE_DISCARD)
-			continue;
-
 		ret = walk_inode(c, &w, k.k->p.inode);
 		if (ret)
 			break;