In preparation for writing logs, in nilfs_btree_propagate(), which
makes parent and ancestor node blocks dirty starting from a modified
data block or b-tree node block, if the starting block does not belong
to the b-tree, i.e. is isolated, nilfs_btree_do_lookup() called within
the function fails with -ENOENT.
In this case, even though -ENOENT is an internal code, it is
propagated to the log writer via nilfs_bmap_propagate() and may be
erroneously returned to system calls such as fsync().
Fix this issue by changing the error code to -EINVAL in this case, and
having the bmap layer detect metadata corruption and convert the error
code appropriately.
Fixes: 1f5abe7e7dbc ("nilfs2: replace BUG_ON and BUG calls triggerable from ioctl")
Signed-off-by: Ryusuke Konishi <konishi.ryusuke@xxxxxxxxx>
---
fs/nilfs2/btree.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/fs/nilfs2/btree.c b/fs/nilfs2/btree.c
index 0d8f7fb15c2e..dd0c8e560ef6 100644
--- a/fs/nilfs2/btree.c
+++ b/fs/nilfs2/btree.c
@@ -2102,11 +2102,13 @@ static int nilfs_btree_propagate(struct nilfs_bmap *btree,
ret = nilfs_btree_do_lookup(btree, path, key, NULL, level + 1, 0);
if (ret < 0) {
- if (unlikely(ret == -ENOENT))
+ if (unlikely(ret == -ENOENT)) {
nilfs_crit(btree->b_inode->i_sb,
"writing node/leaf block does not appear in b-tree (ino=%lu) at key=%llu, level=%d",
btree->b_inode->i_ino,
(unsigned long long)key, level);
+ ret = -EINVAL;
+ }
goto out;
}
--
2.43.0
