[linux-dvb] [PATCH] 1/3: BUG FIX in dvb_ringbuffer_flush

Oliver Endriss o.endriss at gmx.de
Sat Mar 22 04:56:43 CET 2008


Andrea wrote:
> I've been playing with DMX_SET_BUFFER_SIZE and I've had a problem when I tried to resize the buffer
> to a smaller size.
> 
> dvb_dmxdev_set_buffer_size flushes the ringbuffer and then replaces it with a new one.
> What happens if the current pointer is on a position that would be invalid in the new buffer? An
> access violation.
> 
> This because dvb_ringbuffer_flush resets the 2 pointers to the vaule of pwrite (which could be after
> the end of the new buffer).

Ack, this _is_ a bug. :-(

> I think it is safer to reset them to 0.

Nak. At the first glance one might think that this patch is correct.
Unfortunately, it introduces a subtle bug.

Citing the comments in dvb_ringbuffer.h:
| (2) If there is exactly one reader and one writer, there is no need
|     to lock read or write operations.
|     Two or more readers must be locked against each other.
|     Flushing the buffer counts as a read operation.
      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|     Two or more writers must be locked against each other.

With your patch all dvb_ringbuffer_flush() calls must also be considered
as write operations. As a consequence, you would have to add spinlock
protection at many places in the code...

So I suggest to leave dvb_ringbuffer_flush() as is and zero the read and
write pointers only where it is really required...

CU
Oliver

-- 
----------------------------------------------------------------
VDR Remote Plugin 0.4.0: http://www.escape-edv.de/endriss/vdr/
----------------------------------------------------------------



More information about the linux-dvb mailing list