fix: flush SquashFS meta block if *exactely* filled

Otherwise we can get really funky corner cases where an inode offset is
indicated as just 1 byte after the end of a block. The unsquashfs program
doesn't complain but the kernel implementation really doesn't like that.

Signed-off-by: David Oberhollenzer <goliath@infraroot.at>

lib/sqfs/meta_writer.c

@@ -85,5 +85,8 @@ int meta_writer_append(meta_writer_t *m, const void *data, size_t size)
 		data = (const char *)data + diff;
 	}
 
+	if (m->offset == (sizeof(m->data) - 2))
+		return meta_writer_flush(m);
+
 	return 0;
 }