Re: migrate_pages() of process with same UID in 4.15-rcX

[Otto Ebeling <otto.ebeling@iki.fi>]

Hi,

[sorry for the even later reply]

I don't have a strong preference either way (between fs creds or real 
creds), having the same behavior as proc_mem_open sounds like a sensible 
option too. Whether moving pages between NUMA nodes is a read-only 
(PTRACE_MODE_READ) activity is debatable, but I'm no NUMA expert.

My concern here was mainly about a) preventing layout discovery and b) 
consistency between move_pages and migrate_pages.

Otto



On Mon, 29 Jan 2018, Michal Hocko wrote:

> [Fixup Christoph email - the thread starts here
> http://lkml.kernel.org/r/1394749328.5225281.1515598510696.JavaMail.zimbra@redhat.com]
>
> On Mon 29-01-18 14:31:51, Michal Hocko wrote:
>> [Sorry for a very late reply]
>>
>> On Wed 10-01-18 10:21:31, Eric W. Biederman wrote:
>> [...]
>>> All of that said.  I am wondering if we should have used
>>> PTRACE_MODE_READ_FSCREDS on these permission checks.
>>
>> If this is really about preventing the layout discovery then we should
>> be in sync with proc_mem_open and that uses PTRACE_MODE_FSCREDS|PTRACE_MODE_READ
>> Should we do the same thing here?
>> --
>> Michal Hocko
>> SUSE Labs
>
> -- 
> Michal Hocko
> SUSE Labs
>