General open discussion about eCos (embedded configurable operating system)

Graves, Daniel (GE Healthcare | 22 Jun 2012 18:10

Serial Driver imrpovement

Hello,

Note that this might have been fixed already but just in case:

I found that the following code in Serial.c will cause a buffer overflow if there is garbage data on the wire
before a thread has started reading from the serial port:

------------------------------------------------
     static rcv_req_reply_t
     serial_data_rcv_req(serial_channel *chan, int avail, 
                    int* space_avail, unsigned char** space)
    {   
        cbuf_t *cbuf = &chan->in_cbuf;
        int gap;

        #ifdef CYGOPT_IO_SERIAL_FLOW_CONTROL_SOFTWARE
       // When there is software flow-control, force the serial device
      // driver to use the single-char xmt/rcv functions, since these
      // have to make policy decision based on the data. Rcv function
     // may also have to transmit data to throttle the xmitter.
       If (chan->config.flags & (CYGNUM_SERIAL_FLOW_XONXOFF_TX|CYGNUM_SERIAL_FLOW_XONXOFF_RX))
            return CYG_RCV_DISABLED;
        #endif

        CYG_ASSERT(false == cbuf->block_mode_xfer_running,
                   "Attempting new block transfer while another is running");
       // Check for space
       gap = cbuf->nb;
       if (gap == cbuf->len)
           return CYG_RCV_FULL;

(Continue reading)

headers

Jonathan Larmour | 1 Jul 2012 04:37

Re: Serial Driver imrpovement

On 22/06/12 17:10, Graves, Daniel (GE Healthcare) wrote:
> I found that the following code in Serial.c will cause a buffer overflow
> if there is garbage data on the wire before a thread has started reading
> from the serial port:
> 
> ------------------------------------------------
>      static rcv_req_reply_t
>      serial_data_rcv_req(serial_channel *chan, int avail, 
>                     int* space_avail, unsigned char** space)
>     {   
[snip]
>        // Check for space
>        gap = cbuf->nb;
>        if (gap == cbuf->len)
>            return CYG_RCV_FULL;
> 
> ----------------------
> 
> The line gap == cbuf->len will not always evaluate to true if garbage data
> is received.  This will cause the put variable to shoot way past len.

I don't yet follow the situation you are concerned about. This generic layer
doesn't know about garbage data. Garbage data is just data, and whether before
or after a thread has started reading from the serial port, the underlying
hardware serial driver needs to call serial_data_rcv_req() when data is
received and act accordingly based on the return code.

Based on what you're saying, you are implying that cbuf->nb > cbuf->len ? If
so, then something has gone wrong... I imagine that the hardware driver is not
respecting space_avail on return from serial_data_rcv_req().

(Continue reading)