Process blocking behaviour

Process blocking behaviour

[selvakumar nagendran]

Hello linux-experts,
   
  I am intercepting system calls in Linux kernel
2.4.28.
  
  Pseudo-code:
  ------------
    saved_old_syscall =
sys_call_table[sycallno(read)];
    sys_call_table[read] = my_sys_call;
    
  my_sys_call(file descriptor)
  -------------
     Call saved_old_syscall(file descriptor).
    
     Now at this point, I want to determine whether
the system call blocks waiting for the file descriptor
resource. How can I do that? Should I modify the
kernel code only for this?

   Can I check its state after the call as
     if (task_current_state == INTERRUPTIBLE
            || UNINTERRUPTIBLE) to do this?
             
Thanks,
selva

  

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

Re: Process blocking behaviour

[Daniel Gryniewicz]

On Thu, 2005-01-06 at 22:58 -0800, selvakumar nagendran wrote:
> Hello linux-experts,
>    
>   I am intercepting system calls in Linux kernel
> 2.4.28.
>   
>   Pseudo-code:
>   ------------
>     saved_old_syscall =
> sys_call_table[sycallno(read)];
>     sys_call_table[read] = my_sys_call;
>     
>   my_sys_call(file descriptor)
>   -------------
>      Call saved_old_syscall(file descriptor).
>     
>      Now at this point, I want to determine whether
> the system call blocks waiting for the file descriptor
> resource. How can I do that? Should I modify the
> kernel code only for this?
> 
>    Can I check its state after the call as
>      if (task_current_state == INTERRUPTIBLE
>             || UNINTERRUPTIBLE) to do this?

No, you can't, because by the time the syscall returns to you, it's
blocked, scheduled, unblocked, and been rescheduled.  It's too late.  I
don't think you can do what you want, without modifying all the syscalls
directly.

Besides, most blocks are for data, not on something like a semaphore.
What you really want is priority-inheritance semaphores, not
modification of syscalls.  I seem to remember someone working on such a
beast, and the RTOS versions of Linux presumably have something like
this, so you could check around.

Daniel