[PATCH] Add a choosepid() syscall as a simpler alternative to clone_with_pids()

Oren Laadan orenl at librato.com
Mon Nov 2 13:02:13 PST 2009



Dan Smith wrote:
> OL> 2. What do you expect to gain by splitting the work into two
> OL> separate system calls ?
> 
> Simplicity and avoidance of the arch-specific issues, which seems to
> be what hung us up on $NEW_CLONE_NAME_O'_THE_DAY.

Adding a syscall, a field on current, and a set of rules how those
are managed, is not simpler IMHO.

Some objections in the past argued that this may cause unexpected
behavior to the user or tracing/debugging tools. I'm not sure to
what extend this still is, but those will come again.

Finally, I disagree with the prognosis: the arch-specific issue that
hung the today's-clone was introducing *a* new clone such that its
args are passed incorrectly from a _technical_ point of view.

The thing is, that we need a new clone anyway (if not only for the
flags), and that new clone will have to get its args correctly.

On the other hand, Suka fixed his arch-dependent part and we're left
with agreeing on a new plausible name ...

That said, if you manage to get that to mainline, I'll be fine with
it (and then someone later will need to extend clone somehow). I only
think that chances are slim.

Oren.


More information about the Containers mailing list