Re: Maybe a bug in libata-core

["Stefan /*St0fF*/ Hübner" <stefan.huebner@stud.tu-ilmenau.de>]

Am 23.08.2010 11:16, schrieb Tejun Heo:
> On 08/23/2010 02:52 AM, Stefan /*St0fF*/ Hübner wrote:
>> Hi Jeff and list,
>>
>> maybe this mail was wrong on the linux-scsi list.  So here we go again:
>>
>> -------- Original-Nachricht --------
>> Betreff: Maybe a bug in libata-core
>> Datum: Fri, 20 Aug 2010 23:48:09 +0200
>> Von: Stefan Hübner  <stefan.huebner@stud.tu-ilmenau.de>
>> Antwort an: stefan.huebner@stud.tu-ilmenau.de
>> Organisation: TU-Ilmenau
>> An: linux-scsi@vger.kernel.org
>>
>>  Hi List!
>>
>> After sending a WRITE_DMA_FUA_EXT ATA-command via SG_IO Passthru to a
>> harddisk (here: /dev/sdd), my kernel panics.  The is what I find in syslog:
> 
> You're sending down data command w/o data.  Panicking probably isn't
> the best response here.  Maybe the BUG_ON() should be changed to
> WARN_ON_ONCE() + goto sg_err.
> 
> Thanks.
> 

As far as I know my code I thought I did send enough data.  Maybe I'm
misunderstanding something: I valloc'ated block_count*logical_block_size
bytes, filled them with data and presented this buffer to the
sg_io_hdr_t structure by setting dxfer_len to the length of the buffer,
setting dxferp to a pointer to the buffer and setting the
dxfer_direction to SG_DXFER_TO_DEV.

The only thing coming to my mind would be overflow of dxfer_len, but as
this is a unsigned int - wouldn't it be 32 bits wide? (and by that
accepting f.e. 64k*512 bytes = 33 554 432 bytes (32M, needing 25 bits))

Any other suggestions, or do I have to present the code?

Thanks,
Stefan