[Bug libc/17204] race condition in nice

["carlos at redhat dot com" <sourceware-bugzilla at sourceware dot org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=17204

Carlos O'Donell <carlos at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |SUSPENDED
                 CC|                            |carlos at redhat dot com
           Severity|normal                      |enhancement

--- Comment #2 from Carlos O'Donell <carlos at redhat dot com> ---
(In reply to Andreas Schwab from comment #1)
> It is impossible in general to know the correct return value without calling
> getpriority.

Agreed. At best this bug is an RFC "reduce number of syscalls made in nice()
implementation."

(In reply to Steven Stewart-Gallus from comment #0)
> nice is implemented as calling getpriority and than setting the priority as
> higher and than calling getpriority again but this is racy as other tasks
> can call setpriority concurrently. This bug probably doesn't actually cause
> any problems but it could still be fixed. Fixing this bug would also reduce
> the amount of system calls by one which would be nice. Glibc should not call
> getpriority to get the new value of the priority after.

Agreed. There is a race. To fix the race would require a new kernel syscall or
some kind of userspace lock to make the priority accesses atomic. I expect the
only real solution is a `nice` syscall that does everything in one access
exclusively for the current process. Given that this is really a kernel issue
you could file two bugs, one for the kernel to implement `nice` and another for
glibc to use `nice`, but you'll have to lobby both groups to get the feature
you want and that will be difficult.

Moving to SUSPENDED until the kernel has a `nice` syscall which solves both the
race condition and reduces the implementation from 3 syscalls down to 1.

-- 
You are receiving this mail because:
You are on the CC list for the bug.