nilfs2: do not propagate ENOENT error from nilfs_sufile_mark_dirty()

nilfs_sufile_mark_dirty(), which marks a block in the sufile metadata file
as dirty in preparation for log writing, returns -ENOENT to the caller if
the block containing the segment usage of the specified segment is
missing.

This internal code can propagate through the log writer to system calls
such as fsync.  To prevent this, treat this case as a filesystem error and
return -EIO instead.

Link: https://lkml.kernel.org/r/20240821154627.11848-6-konishi.ryusuke@gmail.comSigned-off-by: Ryusuke Konishi <konishi.ryusuke@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

fs/nilfs2/sufile.c

@@ -513,8 +513,15 @@ int nilfs_sufile_mark_dirty(struct inode *sufile, __u64 segnum)
 
 	down_write(&NILFS_MDT(sufile)->mi_sem);
 	ret = nilfs_sufile_get_segment_usage_block(sufile, segnum, 0, &bh);
-	if (ret)
+	if (unlikely(ret)) {
+		if (ret == -ENOENT) {
+			nilfs_error(sufile->i_sb,
+				    "segment usage for segment %llu is unreadable due to a hole block",
+				    (unsigned long long)segnum);
+			ret = -EIO;
+		}
 		goto out_sem;
+	}
 
 	kaddr = kmap_local_page(bh->b_page);
 	su = nilfs_sufile_block_get_segment_usage(sufile, segnum, bh, kaddr);