Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1224002

Re: Multiple potential races on vma->vm_flags

Show all headers | View raw

On Fri, Sep 11, 2015 at 06:27:14PM -0700, Hugh Dickins wrote:
> On Fri, 11 Sep 2015, Kirill A. Shutemov wrote:
> > On Thu, Sep 10, 2015 at 03:27:59PM +0200, Andrey Konovalov wrote:
> > > Can a vma be shared among a few mm's?
> > 
> > Define "shared".
> > 
> > vma can belong only to one process (mm_struct), but it can be accessed
> > from other process like in rmap case below.
> > 
> > rmap uses anon_vma_lock for anon vma and i_mmap_rwsem for file vma to make
> > sure that the vma will not disappear under it.
> > 
> > > If yes, then taking current->mm->mmap_sem to protect vma is not enough.
> > 
> > Depends on what protection you are talking about.
> >  
> > > In the first report below both T378 and T398 take
> > > current->mm->mmap_sem at mm/mlock.c:650, but they turn out to be
> > > different locks (the addresses are different).
> > 
> > See i_mmap_lock_read() in T398. It will guarantee that vma is there.
> > 
> > > In the second report T309 doesn't take any locks at all, since it
> > > assumes that after checking atomic_dec_and_test(&mm->mm_users) the mm
> > > has no other users, but then it does a write to vma.
> > 
> > This one is tricky. I *assume* the mm cannot be generally accessible after
> > mm_users drops to zero, but I'm not entirely sure about it.
> > procfs? ptrace?
> 
> Most of the things (including procfs and ptrace) that need to work on
> a foreign mm do take a hold on mm_users with get_task_mm().  swapoff
> uses atomic_inc_not_zero(&mm->mm_users).  In KSM I managed to get away
> with just a hold on the structure itself, atomic_inc(&mm->mm_count),
> and a check for mm_users 0 wherever it down_reads mmap_sem (but Andrey
> might like to turn KSM on: it wouldn't be entirely shocking if he were
> to discover an anomaly from that).
> 
> > 
> > The VMA is still accessible via rmap at this point. And I think it can be
> > a problem:
> > 
> > 		CPU0					CPU1
> > exit_mmap()
> >   // mmap_sem is *not* taken
> >   munlock_vma_pages_all()
> >     munlock_vma_pages_range()
> >     						try_to_unmap_one()
> > 						  down_read_trylock(&vma->vm_mm->mmap_sem))
> > 						  !!(vma->vm_flags & VM_LOCKED) == true
> >       vma->vm_flags &= ~VM_LOCKED;
> >       <munlock the page>
> >       						  mlock_vma_page(page);
> > 						  // mlocked pages is leaked.
> > 
> > The obvious solution is to take mmap_sem in exit path, but it would cause
> > performance regression.
> > 
> > Any comments?
> 
> I'm inclined to echo Vlastimil's comment from earlier in the thread:
> sounds like an overkill, unless we find something more serious than this.
> 
> I'm not sure whether we'd actually see a regression from taking mmap_sem
> in exit path; but given that it's mmap_sem, yes, history tells us please
> not to take it any more than we have to.
> 
> I do remember wishing, when working out KSM's mm handling, that exit took
> mmap_sem: it would have made it simpler, but that wasn't a change I dared
> to make.
> 
> Maybe an mm_users 0 check after down_read_trylock in try_to_unmap_one() 
> could fix it?

I don't see how. It would shift a picture, but doesn't fix it: exit_mmap()
can happen after down_read_trylock() and mm_users check.
We would only hide the problem.

> But if we were to make a bigger change for this VM_LOCKED issue, and
> something more serious makes it worth all the effort, I'd say that
> what needs to be done is to give mlock/munlock proper locking (haha).
> 
> I have not yet looked at your mlocked THP patch (sorry), but when I
> was doing the same thing for huge tmpfs, what made it so surprisingly
> difficult was all the spongy trylocking, which concealed the rules.
> 
> Maybe I'm completely wrong, but I thought a lot of awkwardness might
> disappear if they were relying on anon_vma->rwsem and i_mmap_rwsem
> throughout instead of mmap_sem.

This can be helpful. But the risk is getting scalability regression on
other front: long anon_vma chain or highly shared files.

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

Back to linux.kernel | Previous | Next — Previous in thread | Next in thread | Find similar | Unroll thread

Thread

Multiple potential races on vma->vm_flags "Kirill A. Shutemov" <kirill@shutemov.name> - 2015-09-07 13:50 +0200
  Re: Multiple potential races on vma->vm_flags Vlastimil Babka <vbabka@suse.cz> - 2015-09-09 17:30 +0200
    Re: Multiple potential races on vma->vm_flags "Kirill A. Shutemov" <kirill@shutemov.name> - 2015-09-09 18:10 +0200
  Re: Multiple potential races on vma->vm_flags Sasha Levin <sasha.levin@oracle.com> - 2015-09-10 03:00 +0200
    Re: Multiple potential races on vma->vm_flags "Kirill A. Shutemov" <kirill@shutemov.name> - 2015-09-10 10:40 +0200
      Re: Multiple potential races on vma->vm_flags Andrey Konovalov <andreyknvl@google.com> - 2015-09-10 15:30 +0200
        Re: Multiple potential races on vma->vm_flags "Kirill A. Shutemov" <kirill@shutemov.name> - 2015-09-11 12:50 +0200
          Re: Multiple potential races on vma->vm_flags Vlastimil Babka <vbabka@suse.cz> - 2015-09-11 17:30 +0200
            Re: Multiple potential races on vma->vm_flags Vlastimil Babka <vbabka@suse.cz> - 2015-09-11 18:10 +0200
          Re: Multiple potential races on vma->vm_flags Hugh Dickins <hughd@google.com> - 2015-09-12 03:30 +0200
            Re: Multiple potential races on vma->vm_flags "Kirill A. Shutemov" <kirill@shutemov.name> - 2015-09-14 12:20 +0200
            Re: Multiple potential races on vma->vm_flags Sasha Levin <sasha.levin@oracle.com> - 2015-09-15 19:40 +0200
              Re: Multiple potential races on vma->vm_flags "Kirill A. Shutemov" <kirill@shutemov.name> - 2015-09-15 21:10 +0200

csiph-web