binlog cache I/O error might be reported twice

By code review:

bool MYSQL_BIN_LOG::write_cache(THD *thd, binlog_cache_data *cache_data)
{
...
  IO_CACHE *cache= &cache_data->cache_log;
...
      if ((write_error= do_write_cache(thd, cache)))
        goto err;
...
      if (cache->error) // Error on read
      {
        char errbuf[MYSYS_STRERROR_SIZE];
        sql_print_error(ER(ER_ERROR_ON_READ), cache->file_name,
                        errno, my_strerror(errbuf, sizeof(errbuf), errno));
        write_error=1; // Don't give more errors
        goto err;
      }
...
err:
  if (!write_error)
    write_error= 1;
  {
    char errbuf[MYSYS_STRERROR_SIZE];
    sql_print_error(ER(ER_ERROR_ON_WRITE), name,
                    errno, my_strerror(errbuf, sizeof(errbuf), errno));
  }
  thd->commit_error= THD::CE_FLUSH_ERROR;

  DBUG_RETURN(1);
}

In the case of cache->error will result in two sql_print_errors. This is a regression from upstream 5.6, introduced in the initial 5.6 merge. The upstream has

err:
  if (!write_error)
  {
    char errbuf[MYSYS_STRERROR_SIZE];
    write_error= 1;
    sql_print_error(ER(ER_ERROR_ON_WRITE), name,
                    errno, my_strerror(errbuf, sizeof(errbuf), errno));
  }
  thd->commit_error= THD::CE_FLUSH_ERROR;

  DBUG_RETURN(1);
}

which will result in a single sql_print_error as intended. I am unable to tell right now whether cache->error (presumably by my_b_fill in do_write_cache) might ever possibly happen in the first place. This bug is about restoring the upstream behavior regardless of this.