lkml.org 
[lkml]   [2017]   [Aug]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 2/2] mm, oom: fix potential data corruption when oom_reaper races with writer
On Fri 11-08-17 16:54:36, Tetsuo Handa wrote:
> Michal Hocko wrote:
> > On Fri 11-08-17 11:28:52, Tetsuo Handa wrote:
> > > Michal Hocko wrote:
> > > > +/*
> > > > + * Checks whether a page fault on the given mm is still reliable.
> > > > + * This is no longer true if the oom reaper started to reap the
> > > > + * address space which is reflected by MMF_UNSTABLE flag set in
> > > > + * the mm. At that moment any !shared mapping would lose the content
> > > > + * and could cause a memory corruption (zero pages instead of the
> > > > + * original content).
> > > > + *
> > > > + * User should call this before establishing a page table entry for
> > > > + * a !shared mapping and under the proper page table lock.
> > > > + *
> > > > + * Return 0 when the PF is safe VM_FAULT_SIGBUS otherwise.
> > > > + */
> > > > +static inline int check_stable_address_space(struct mm_struct *mm)
> > > > +{
> > > > + if (unlikely(test_bit(MMF_UNSTABLE, &mm->flags)))
> > > > + return VM_FAULT_SIGBUS;
> > > > + return 0;
> > > > +}
> > > > +
> > >
> > > Will you explain the mechanism why random values are written instead of zeros
> > > so that this patch can actually fix the race problem?
> >
> > I am not sure what you mean here. Were you able to see a write with an
> > unexpected content?
>
> Yes. See http://lkml.kernel.org/r/201708072228.FAJ09347.tOOVOFFQJSHMFL@I-love.SAKURA.ne.jp .

Ahh, I've missed that random part of your output. That is really strange
because AFAICS the oom reaper shouldn't really interact here. We are
only unmapping anonymous memory and even if a refault slips through we
should always get zeros.

Your test case doesn't mmap MAP_PRIVATE of a file so we shouldn't even
get any uninitialized data from a file by missing CoWed content. The
only possible explanations would be that a page fault returned a
non-zero data which would be a bug on its own or that a file write
extend the file without actually writing to it which smells like a fs
bug to me.

Anyway I wasn't able to reproduce this and I was running your usecase
in the loop for quite some time (with xfs storage). How reproducible
is this? If you can reproduce easily can you simply comment out
unmap_page_range in __oom_reap_task_mm and see if that makes any change
just to be sure that the oom reaper can be ruled out?
--
Michal Hocko
SUSE Labs

\
 
 \ /
  Last update: 2017-08-11 14:09    [W:0.105 / U:0.660 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site