On 22:13, Boaz Harrosh wrote: > All the scsi calls do not need any locks. The scsi LLDS never > see these threads since commands are queued through the block > layer. That's what everybody believes, but nobody seems to know for sure. Therefore I did what Andi suggested: Make a zero-semantics change that moves the lock_kernel() to sg_ioctl() to make people aware of the fact that this function runs under the BKL. At least the latter has already succeeded. > What's left is what you see, here in sg.c. you must have the best > knowledge about the possible races between ioctl and open/release > and probe/remove. And all these put_user() copy_user() etc... > Why don't you have a hard look and fix them properly. Because that requires much more knowledge. Al is looking into this which indicates that it is non-trivial issue. I am clearly not the right person to decide this question. > please don't *lock_kernel();* for scsi's sake. The BKL was there all the time. My patch just made it more visable to the scsi people by moving it into sg.c. Andre -- The only person who always got his work done by Friday was Robinson Crusoe
Attachment:
signature.asc
Description: Digital signature
