Message-ID: <52283590.8070003@linux.vnet.ibm.com>
Date:	Thu, 05 Sep 2013 13:11:04 +0530
From:	Janani Venkataraman <jananive@...ux.vnet.ibm.com>
To:	Pavel Emelyanov <xemul@...allels.com>
CC:	linux-kernel@...r.kernel.org,
	Jeremy Fitzhardinge <jeremy.fitzhardinge@...rix.com>,
	Daisuke HATAYAMA <d.hatayama@...fujitsu.com>,
	Andi Kleen <andi@...stfloor.org>,
	Roland McGrath <roland@...k.frob.com>,
	Amerigo Wang <amwang@...hat.com>,
	Christoph Hellwig <hch@....de>,
	Linus Torvalds <torvalds@...ux-foundation.org>,
	KOSAKI Motohiro <kosaki.motohiro@...fujitsu.com>,
	Masami Hiramatsu <mhiramat@...hat.com>,
	Andrew Morton <akpm@...ux-foundation.org>,
	Alexey Dobriyan <adobriyan@...il.com>,
	Oleg Nesterov <oleg@...hat.com>, Tejun Heo <tj@...nel.org>,
	avagin@...nvz.org, gorcunov@...nvz.org,
	James Hogan <james.hogan@...tec.com>,
	Mike Frysinger <vapier@...too.org>,
	"Randy.Dunlap" <rdunlap@...otime.net>,
	Eric Paris <eparis@...hat.com>, ananth@...ibm.com,
	suzuki@...ibm.com, aravinda@...ux.vnet.ibm.com,
	tarundeep.singh@...ibm.com
Subject: Re: RFD: Non-Disruptive Core Dump Infrastructure

On 09/04/2013 05:03 PM, Pavel Emelyanov wrote:
> On 09/04/2013 02:53 PM, Janani Venkataraman wrote:
>> On 09/03/2013 04:24 PM, Pavel Emelyanov wrote:
>>> On 09/03/2013 02:47 PM, Janani Venkataraman wrote:
>>>> On 09/03/2013 04:01 PM, Pavel Emelyanov wrote:
>>>>> On 09/03/2013 12:39 PM, Janani Venkataraman wrote:
>>>>>> Hello,
>>>>>>
>>>>>> We are working on an infrastructure to create a system core file of a specific
>>>>>> process at run-time, non-disruptively. It can also be extended to a case where
>>>>>> a process is able to take a self-core dump.
>>>>>
>>>>> This is very close to what we're trying to do in CRIU. And although image files
>>>>> containing info about processes are not ELF files, an ability to generate ELF-cores
>>>>> out of existing CRIU images is one of the features that we were asked for.
>>>>>
>>>>>> 2) CRIU Approach :
>>>>>>
>>>>>> This makes use of the CRIU tool and checkpoints when a dump is called, collects
>>>>>> the required details and continues the running process.
>>>>>> * A self dump cannot be initiated using the command line CRIU which is similar
>>>>>> to the limitation of gcore.
>>>>>
>>>>> This is something we're trying to fix at the moment, as people ask for 'self-dump'
>>>>> ability as well. We plan to have this implemented in v0.8 (the v0.7 is coming out
>>>>> today/tomorrow) in about a month.
>>>>>
>>>>> I can shed more light on this, if required.
>>>>>
>>>>>> * A system call to do the same is being implemented which would help us create
>>>>>> a self dump.The system call is not upstream yet. We could explore that option as
>>>>>> well.
>>>>>
>>>>> Thanks,
>>>>> Pavel
>>>>>
>>>> Hi,
>>>>
>>>> I would like to know more about the "self-dump" ability of CRIU. This is
>>>> the implementation using system calls if I am not wrong.
>>>
>>> Not exactly.
>>>
>>> In CRIU project since it's earliest days, we had to heavily patch the kernel
>>> to make it provide additional APIs for getting more info about running tasks
>>> and kernel objects. You can find all the patches we've created on the page
>>> http://criu.org/Commits
>>>
>>> For almost all the new APIs we proposed the community asked us to restrict them
>>> with CAP_SYS_ADMIN checks, so CRIU even for very basic stuff should be run from
>>> root. The intention was to create the proof-of-concept with maximal and most
>>> strict protection, and then think harder about less strict checks.
>>>
>>> With this the self-dump functionality cannot be implemented as just "CRIU in a
>>> .so file", since this would only be usable by root processes. So, instead of
>>> just wrapping the whole CRIU stuff into a library, we use a trickier approach.
>>> It's described here -- http://criu.org/Self_dump
>>>
>>> Briefly -- we will implement the CRIU service, which is a daemon running from
>>> root and listening on a unix socket. When a task wants to dump himself, it sends
>>> to the service a "dump me" message. The service then goes and dumps the process.
>>>
>>> Thanks,
>>> Pavel
>>>
>>
>> Hi,
>>
>> What we require for our infrastructure is just a register snapshot and a
>> memory dump.Do we require CAP_SYS privileges,if we want to dump the only
>> regset and memory ?
>
> For registers and just the contents of memory you should just have enough
> rights to attach to the "victim" with the debugger. This usually means
> uid-s equivalence or CAP_SYS_PTRACE capability otherwise.
>
>> Is it possible to librarize the dump generation routine so that it is
>> transparent to the user. Also, ideally a single API for dump generation
>> is preferred for generating the dump, irrespective of whether it is a
>> self dump or not.
>
> We're currently developing a protobuf-RPC protocol to talk to criu service.
> Additionally there will be a .so library, that will provide C API above this
> protocol.
>
>> One another aspect we might want to look at is the DoS attacks. Are
>> there any cases where it is prone to such attacks.
>
> Well, checkpoint takes time, memory and disk, so if performed too often, may
> cause starvation on these resources.
>
>> We also looked into the Self-dump page you had mentioned and we would
>> like to know more. Is there any additional information/prototype which
>> you share with us .Also would it be possible for us to test a few
>> patches for the self dump case ?
>
> Currently this is work-in-progress, you can check criu mailing list
> archives at http://lists.openvz.org/pipermail/criu/, the patches from
> Ruslan Kuprieiev <kupruser@> are mostly about it.
>
>> If converting the dump,to ELF-core format from the existing CRIU Image
>> format has not yet been done,we would be happy to contribute towards it.
>
> Oh, that's great! The criu images format is described at http://criu.org/Images,
> feel free to ask questions if you find this information not enough.

    I will look into this and get back to you at the earliest.
>
> Thanks,
> Pavel
>
Thanks,
Janani

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

Powered by blists - more mailing lists