[Devel] Re: [PATCH 5/15] Introduce struct upid

[Pavel Emelyanov]

Oleg Nesterov wrote:
> On 07/26, Pavel Emelyanov wrote:
>> --- linux-2.6.23-rc1-mm1.orig/include/linux/pid.h	2007-07-26 16:34:45.000000000 +0400
>> +++ linux-2.6.23-rc1-mm1-7/include/linux/pid.h	2007-07-26 16:36:37.000000000 +0400
>> @@ -40,15 +40,21 @@ enum pid_type
>>   * processes.
>>   */
>>  
>> -struct pid
>> -{
>> -	atomic_t count;
>> +struct upid {
>>  	/* Try to keep pid_chain in the same cacheline as nr for find_pid */
>>  	int nr;
>> +	struct pid_namespace *ns;
>>  	struct hlist_node pid_chain;
>> +};
>> +
>> +struct pid
>> +{
>> +	atomic_t count;
>>  	/* lists of tasks that use this pid */
>>  	struct hlist_head tasks[PIDTYPE_MAX];
>>  	struct rcu_head rcu;
>> +	int level;
>> +	struct upid numbers[1];
>>  };
> 
> Well. Definitely, the kernel can't be compiled with this patch applied,
> this seems to be against the rules...

Yes. U forgot to mention, that this patchset is git-bisect-not-safe :)
I sent the safe split earlier, but it was harder to make and understand,
so I decided not to waste the time and sent a badly-split set just to get
comments about the approach. The ways a big patch is split wouldn't affect
the comments about the ideas, bugs, etc.

> So. The task has a single (PIDTYPE_MAX) pid no matter how many namespaces
> can see it, and "struct pid" has an array of numbers for each namespace.
> 
> Still I can't understand why do we need upid->ns, can't we kill it?
> Suppose we add "struct pid_namespace *parent_ns" to "struct pid_namespace",
> init_pid_ns.parent_ns == NULL.

We already have it :)

> Now,
> 
> 	struct upid {
> 		int nr;
> 		struct hlist_node pid_chain;
> 	};
> 
> 	struct pid
> 	{
> 		atomic_t count;
> 		struct hlist_head tasks[PIDTYPE_MAX];
> 		struct rcu_head rcu;
> 		struct pid_namespace *active_ns;
> 		struct upid numbers[0];
> 	};
> 
> We populate pid->numbers in "reverse" order, so that pid->numbers[0] lives
> in pid->active_ns.
> 
> Now, for example,
> 
> 	void free_pid(struct pid *pid)
> 	{
> 		struct pid_namespace *ns;
> 		unsigned long flags;
> 		int i;
> 
> 		spin_lock_irqsave(&pidmap_lock, flags);
> 		for (i = 0, ns = pid->active_ns; ns; i++, ns = ns->parent_ns)
> 			hlist_del_rcu(&pid->numbers[i].pid_chain);
> 		spin_unlock_irqrestore(&pidmap_lock, flags);
> 
> 		for (i = 0, ns = pid->active_ns; ns; i++, ns = ns->parent_ns)
> 			free_pidmap(ns, pid->numbers[i].nr);
> 
> 		call_rcu(&pid->rcu, delayed_put_pid);
> 	}
> 
> Possible?

Possible, but how will 
struct pid *find_pid_nr_ns(int nr, struct pid_namespace *ns);
look then? The only way (I see) is to make

hlist_for_each_entry (upid, ...)
    if (upid->nr == nr && upid->ns == ns)
             return container_of(upid, struct pid, ...)

> Oleg.

Thanks,
Pavel