- Fix allocation problem introduced in 113 as suggested by Jens.
   (Thanks go to  Jens for providing info how to use properly
    blk_get_request in  this case. Nevermind I just got confused by
    __blk_get_request in TCQ  code, which made me worry about queue depth
    strain.)
   (Note: SCSI should propably be using the same mechanism instead of
    sr_request. At least we should check the code in question there.)