Re: [patch 5/5] elf: Add support for loading ET_CKPT files

[Pavel Emelyanov <xemul@parallels.com>]

On 10/20/2011 07:56 PM, Tejun Heo wrote:
> Hello,
> 
> On Thu, Oct 20, 2011 at 12:33:51PM +0400, Pavel Emelyanov wrote:
>>> To shared kernel data structures.
>>
>> Yet again - keeping shared stuctres self-consistent from which point of view? From
>> the kernel's one - already handled with in-kernel locks. From the userspace one -
>> every single kernel API provides a *way* to do things right, but doesn't provide
>> the fool protection.
> 
> Sigh, no, in exec and related paths, the fact that all other threads
> have been zapped are depended upon and the exec'ing task assumes
> exclusive access to resources which in usual cases would require other
> forms of exclusion.  Maybe there are not too many of them and things
> can definitely be audited and updated but it is pretty intricate down
> there and I just don't see good enough justification here.

The amount of new kernel API is good justification. I've sent the list of what
should we allow to change from user-space for the sake of restore only.

Anyway - let's wait for Cyrill's work done in this area.

>>> IMHO, this is a fundamentally broken approach which isn't even
>>> necessary.  So, FWIW, NACK.
>>
>> It's a pity :( Anyway, as I stated above, we'll try to compare two
>> real implementations, not abstract assumptions.
> 
> Given the state of this thread, I'm pretty skeptical this one can
> survive but, hey, who knows?

That's good to hear that you're still open for discussion.

> Thank you.

Thanks,
Pavel