Re: [Crash-utility] [PATCH v3 0/2] Fix KASLR problem on sadump
On Wed, Mar 14, 2018 at 03:57:12PM -0400, Dave Anderson wrote:
----- Original Message -----
> Hi,
>
> I know support for guessing the KASLR offset for kdump/diskimages was
> deliberately dropped from this patchset, because upstream was also working on
> the vmcoreinfo device, but I think having that would be *really* useful.
>
> The vmcoreinfo device solution requires explicit support in the virtualization
> Hosts, and it'll take a while until its presence has been generalized among
> products and users. Meanwhile, we already have VMs with KASLR-enabled
> kernels.
>
> So I'd like to ask if a patchset extending the current state of KASLR offset
> guessing for sadump to kdump/diskimages would be considered for merging.
>
> Thanks,
> Sergio.
>
You can always post a patch and it will get consideration.
That being said, the original patch would no longer be applicable given
the recently-introduced support for x86_64 KPTI and 5-level page-tables.
I know, it'll be based on current upstream's implementation.
And BTW, I thought the SADUMP solution was accurate, i.e., not
"offset guessing"?
Yeah, "guessing" may not be the proper term here. Probably "inferring"
would be
more appropriate, in the sense that it's calculated instead of directly read
from vmcoreinfo.
I just ask that the patch be reasonably segregated, so that its code
only runs in
the case of virsh-generated dumpfiles that do not contain the VMCOREINFO note.
Ack, thanks.
Sergio.