Re: [PATCH 6/7] x86, kaslr: report kernel offset on panic

[HATAYAMA Daisuke <d.hatayama@jp.fujitsu.com>]

(2013/10/08 22:38), Dave Anderson wrote:
>
>
> ----- Original Message -----
>> (2013/10/07 22:21), Dave Anderson wrote:
>>
>>> ----- Original Message -----
>>>> (2013/10/03 22:47), Dave Anderson wrote:
>>
>>>>> ----- Original Message -----
>>>>>> (2013/10/02 18:13), HATAYAMA Daisuke wrote:
>>>>>>> (2013/10/02 16:48), Kees Cook wrote:
>>
>>>>
>>>> Thanks for detailed explanation. So, there's already a feature in crash
>>>> utility
>>>> to address relocation!, though it's better for me to try them to check if
>>>> it's
>>>> really applicable to this feature. My concern is whether --reloc works
>>>> well
>>>> on x86_64 too, because relocation has never done on x86_64 ever, right?
>>>
>>> Correct.
>>>
>>>> Another concern is that in case of relocation, users need to additional information
>>>> regarding runtime symbol information to crash utility. I want to avoid additional
>>>> process, automation is preferable if possible.
>>>
>>> Right.  As I mentioned in the case of 32-bit x86 dumpfiles, there is no automation
>>> available when CONFIG_PHYSICAL_START is larger than CONFIG_PHYSICAL_ALIGN.  The user
>>> either has to be aware of their values in order to calculate the --reloc argument,
>>> or has to capture a copy of the /proc/kallsyms file on the crashed system.  Typically
>>> users/distros using kdump changed their x86 configurations to avoid having to deal
>>> with that.
>>>
>>
>> Sorry, I don't understand why relocation size cannot be calculated when
>> CONFIG_PHYSICALSTART > CONFIG_PHYSICAL_ALIGN. Could you explain that?
>
> I just meant that when CONFIG_PHYSICAL_START > CONFIG_PHYSICAL_ALIGN, the 32-bit x86 kernel
> gets relocated (like the secondary kdump kernel), but that information is not readily available
> from the vmlinux/vmcore pair.
>

My understanding on CONFIG_PHYSICAL_ALIGN was that starting address of kernel text area
is always rounded up to CONFIG_PHYSICAL_ALIGN, only. Your explanation would be part I don't
understand well. I'll reconsider it locally...

>>
>>>> I guess it's enough if there's runtime symbol addresses because we can get relocated
>>>> offset value by comparing it with the compile-time symbol address contained in
>>>> a given debuginfo file. Candidates for such symbols are the ones contained in
>>>> VMCOREINFO note containing some symbol values for makedumpfile to refer to mm-related
>>>> objects in kernel, which is always contained in vmcore generated by current kdump and
>>>> also vmcores converted by makedumpfile from it. How about this idea?
>>>
>>> But how would that differ from using an incorrect (non-matching) vmlinux file?
>>>
>>
>> It seems to me almost similar to what crash currently does even if we do relocation check.
>> The current check crash currently does is trial-and-error since there's no information
>> indicating given vmcore and vmlinuxcertainly match well.
>>
>> For example, the process I imagine is:
>>
>>     1) try to match vmcore and vmlinux with no relocation. If fails, go to 2).
>>     2) try to match vmcore and vmlinux with relocation.
>>
>> The two steps include symbol table initialization so it might actually be difficult to
>> resume back from 2) to 1).
>>
>> Also, if gap due to phys_base and gap due to relocation can happen at the same time,
>> calculating two values automatically might be futher complicated. So, it would be better
>> to add relocation value in VMCOREINFO. Then, what crash utility sholud do becomes very simple.
>
> Yes please...
>
> And while you're at it, the kernel's
>
>    VMCOREINFO_SYMBOL(phys_base);
>
> is pretty much useless, at least w/respect to ELF vmcores, since we need to know its
> value in order to translate the address.  And I don't think that makedumpfile uses
> it when it calculates the phys_base that it stores in compressed kdump headers.  Why
> not put its value instead of its address?
>

Yes, I've also noticed this fact. Anyway, I'll post a patch to improvement this phys_base
and a patch to export relocation information in VMCOREINFO.

-- 
Thanks.
HATAYAMA, Daisuke