Re: [PATCH 6/7] proc: Introduce the /proc/<pid>/dump file

[Pavel Emelyanov <xemul-bzQdu9zFT3WakBO8gow8eQ@public.gmane.org>]

On 07/23/2011 12:37 PM, Tejun Heo wrote:
> Hello,
> 
> On Sat, Jul 23, 2011 at 12:11:37PM +0400, Pavel Emelyanov wrote:
>> On 07/21/2011 10:44 AM, Tejun Heo wrote:
>>> Most of information dumped here is already available
>>> through /proc and ptrace and we can add the missing pieces like the
>>> suggested proc vma fds.
>>
>> Let's start with the simplest things. Can you suggest the best (from you pov)
>> way for dumping all the registers, tls and the anonymous pages through the 
>> existing interfaces?
> 
> Just use ptrace.  Seizing all threads in a process, gather file and
> memory map info from /proc, inject a parasite to dump memory pages and
> do whatever else.  There's nothing special about TLS, fs/gs base is
> already included in ptrace register dump in x86_64.  Dunno how it's
> handled in 32bit but if it's not available exporting it isn't a big
> deal.  Rebuilding the process image from the captured information
> shouldn't be too hard.

You're talking about your recent set for ptrace? Can you propose a quick scratch
of how you propose to dump and restore registers and memory with this?

> There is simply is no need to put CR into kernel at all.  

I don't want to! I propose to use small set of APIs for it and the execve handler
is just the way to replace VM+regs of a task with another set.

> Just add the
> missing pieces to export the necessary information and missing APIs
> which are required to restore it (e.g. setting TID like you did in
> this patchset).  Approaching it that way would make things useful for
> other use cases && is highly more likely to get merged.  There doesn't
> even need to be one big merge day.  You can just improve things
> piecewise until it works.

Totally agree.

> Thanks.
>