Re: [patch -mm] update mq_notify to use a struct pid

[ebiederm@xmission.com (Eric W. Biederman)]

Cedric Le Goater <clg@fr.ibm.com> writes:

> Eric W. Biederman wrote:
>
>> Cedric you mentioned a couple of other patches that are in flight.
>> In the future could you please Cc: the containers list so independent
>> efforts are less likely to duplicate work, and we are more likely
>> to review each others patches instead?
>
> yes sure, i was relying on the openvz wiki to avoid duplicated efforts on
> this topic but i guess email is just the one and only tool for this kind of
> development :)

Sure.  Especially when it comes to helping review each others code :)
Not duplicating work is not really my goal, not submitting a patch
after a patch has been reviewed and accepted is.

Plus we need patch review.

Several people working on a patch in parallel if it is difficult
can frequently find a solution that a single person would miss.

>>>> Filling in a struct pid through sysctl is extremely ugly at the
>>>> moment, plus cad_pid needs some locking.
>>> Which distros use /proc/sys/kernel/cad_pid and why ? I can image the need
>>> but i didn't find much on the topic.
>> 
>> I'm not at all certain, and I'm not even certain I care.  The concept
>> is there in the code so it needs to be dealt with.  
>
> OK. It would be nice to make sure this is still in use before trying to
> deal with /proc/sys/kernel/cad_pid.
>
>> Although if I we extend the cad_pid concept it may make a difference.
>
> what do you mean by extending cad_pid ? kill_init() ?

My meaning was every time we are sending a signal to init.  It is quite
possible we should be using cad_pid instead.

>>> is that about updating the siginfos in collect_signal() to hold the right
>>> pid value depending on the pid namespace they are being received ?
>> 
>> Yes in send_signal, and in collect signal.  To make it work easily I needed
>> to add a struct pid to struct sigqueue.  So in send_signal I generate
>> the struct pid from the pid_t value and in collect signal I regenerate
>> the numeric value.
>
> OK. That's what i imagined also but we need a bit more of the pid namespace
> to regenerate the numerical value. So, how will you convert this 'struct
> pid*' in a pid value using the current pid namespace ?

By calling pid_nr :)  The question I guess is how will pid_nr be implemented.

> thinking aloud :
>
> * if the pid namespace of the sending struct pid and current match,
> 	use nr.
> * if they don't,
> 	if the sending pid namespace is the ancestor of the current pid
> 	namespace
> 		use 0
> 	else
> 		it's a bug.
>
> struct pid* needs a pid namespace attribute and pid namespace needs to know
> its parent.

Yes, that sounds correct.

There is also the case that should not come up with signals where
we have a pid from a child namespace, that we should also be able to
compute the pid for.

In essence I intend to have a list of pid_namespace, pid_t pairs connected
to a struct pid that we can look through to find the appropriate pid.

Eric


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/