bcachefs: Fix repair for extent past end of inode

When we find an extent past an inode's i_size, we need to do the
deletion in the inode's snapshot (which will emit a whiteout if
necessary); and we also need to note that we now have an a key at that
position and snapshot, so that we don't go into an infinite loop.

Also, switch to walking inodes in reverse older, oldest snapshot to
newest, so that we emit the fewest whiteouts possible.

Signed-off-by: Kent Overstreet <[email protected]>

1 files changed, 54 insertions, 14 deletions

diff --git a/fs/bcachefs/fsck.c b/fs/bcachefs/fsck.c
index b401c0913bdc..5cec55edb483 100644
--- a/fs/bcachefs/fsck.c
+++ b/fs/bcachefs/fsck.c
@@ -489,6 +489,28 @@ static inline void snapshots_seen_init(struct snapshots_seen *s)
 	memset(s, 0, sizeof(*s));
 }
 
+static int snapshots_seen_add(struct bch_fs *c, struct snapshots_seen *s, u32 id)
+{
+	struct snapshots_seen_entry *i, n = { id, id };
+	int ret;
+
+	darray_for_each(s->ids, i) {
+		if (n.equiv < i->equiv)
+			break;
+
+		if (i->equiv == n.equiv) {
+			bch_err(c, "adding duplicate snapshot in snapshots_seen_add()");
+			return -EINVAL;
+		}
+	}
+
+	ret = darray_insert_item(&s->ids, i - s->ids.data, n);
+	if (ret)
+		bch_err(c, "error reallocating snapshots_seen table (size %zu)",
+			s->ids.size);
+	return ret;
+}
+
 static int snapshots_seen_update(struct bch_fs *c, struct snapshots_seen *s,
 				 enum btree_id btree_id, struct bpos pos)
 {
@@ -1238,20 +1260,38 @@ peek:
 		goto out;
 	}
 
-	if (!bch2_snapshot_internal_node(c, equiv.snapshot)) {
-		for_each_visible_inode(c, s, inode, equiv.snapshot, i) {
-			if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
-					k.k->type != KEY_TYPE_reservation &&
-					k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9, c,
-					"extent type %u offset %llu past end of inode %llu, i_size %llu",
-					k.k->type, k.k->p.offset, k.k->p.inode, i->inode.bi_size)) {
-				bch2_fs_lazy_rw(c);
-				ret = bch2_btree_delete_range_trans(trans, BTREE_ID_extents,
-						SPOS(k.k->p.inode, round_up(i->inode.bi_size, block_bytes(c)) >> 9,
-						     equiv.snapshot),
-						POS(k.k->p.inode, U64_MAX),
-						0, NULL) ?: -EINTR;
-				goto out;
+	/*
+	 * Check inodes in reverse order, from oldest snapshots to newest, so
+	 * that we emit the fewest number of whiteouts necessary:
+	 */
+	for (i = inode->inodes.data + inode->inodes.nr - 1;
+	     i >= inode->inodes.data;
+	     --i) {
+		if (i->snapshot > equiv.snapshot ||
+		    !key_visible_in_snapshot(c, s, i->snapshot, equiv.snapshot))
+			continue;
+
+		if (fsck_err_on(!(i->inode.bi_flags & BCH_INODE_I_SIZE_DIRTY) &&
+				k.k->type != KEY_TYPE_reservation &&
+				k.k->p.offset > round_up(i->inode.bi_size, block_bytes(c)) >> 9, c,
+				"extent type past end of inode %llu:%u, i_size %llu\n  %s",
+				i->inode.bi_inum, i->snapshot, i->inode.bi_size,
+				(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
+			struct btree_iter iter2;
+
+			bch2_trans_copy_iter(&iter2, iter);
+			bch2_btree_iter_set_snapshot(&iter2, i->snapshot);
+			ret =   bch2_btree_iter_traverse(&iter2) ?:
+				bch2_btree_delete_at(trans, &iter2,
+					BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
+			bch2_trans_iter_exit(trans, &iter2);
+			if (ret)
+				goto err;
+
+			if (i->snapshot != equiv.snapshot) {
+				ret = snapshots_seen_add(c, s, i->snapshot);
+				if (ret)
+					goto err;
 			}
 		}
 	}