linux-kernel - Re: [PATCH] mm: fix hang on anon

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Fri, 27 Aug 2010 02:19:26 +0200
From:	Andrea Arcangeli <aarcange@...hat.com>
To:	Hugh Dickins <hughd@...gle.com>
Cc:	David Miller <davem@...emloft.net>, torvalds@...ux-foundation.org,
	akpm@...ux-foundation.org, riel@...hat.com,
	linux-kernel@...r.kernel.org, linux-mm@...ck.org
Subject: Re: [PATCH] mm: fix hang on anon_vma->root->lock

On Thu, Aug 26, 2010 at 03:54:28AM -0700, Hugh Dickins wrote:
> On Wed, Aug 25, 2010 at 11:41 PM, David Miller <davem@...emloft.net> wrote:
> > From: Hugh Dickins <hughd@...gle.com>
> > Date: Wed, 25 Aug 2010 23:12:54 -0700 (PDT)
> >
> >> After several hours, kbuild tests hang with anon_vma_prepare() spinning on
> >> a newly allocated anon_vma's lock - on a box with CONFIG_TREE_PREEMPT_RCU=y
> >> (which makes this very much more likely, but it could happen without).
> >>
> >> The ever-subtle page_lock_anon_vma() now needs a further twist: since
> >> anon_vma_prepare() and anon_vma_fork() are liable to change the ->root
> >> of a reused anon_vma structure at any moment, page_lock_anon_vma()
> >> needs to check page_mapped() again before succeeding, otherwise
> >> page_unlock_anon_vma() might address a different root->lock.
> >>
> >> Signed-off-by: Hugh Dickins <hughd@...gle.com>
> >
> > Interesting, is the condition which allows this to trigger specific
> > to this merge window or was it always possible?
> 
> Just specific to this merge window, which started using
> anon_vma->root->lock in place of anon_vma->lock (anon_vma->root is
> often anon_vma itself, but not always).  I _think_ that change was
> itself a simplification of the locking in 2.6.35, rather than plugging
> a particular hole (it's not been backported to -stable), but I may be
> wrong on that - Rik?

rmap_walk_anon isn't stable without the anon_vma->root->lock. This is
because it only locks the local anon_vma and but the "vma" can be
still modified under it in vma_adjust, so vma_address can fail and
migration will crash. swapping is not issue as it's ok to miss a pte
once in a while if the vma is under vma_adjust by the time the VM
tries to unmap the page.

See the anon_vma_lock added as well to vma_adjust (without it, it'd be
just an useless exercise, but instead it's really plugging a real hole
thanks to the anon_vma_lock addition to vma_adjust).

Again not an issue unless you use migration (incidentally what memory
compaction uses, and you know who the primary user of memory
compaction is :).

Secondly ksm_does_need_to_copy can't be fixed to cow only pages that
are nonlinear as the page->mapping != anon_vma can now generate false
positives. It's now trivial to fix with page->mapping->root !=
anon_vma->root.

About your patch, it's a noop in my view... A single page_mapping
check after rcu_read_lock is enough. And "anon_vma->root" can't change
if page->mapping points to "anon_vma" at any time after rcu_read_lock
returns. rcu_read_lock works for all anon_vma including
anon_vma->root.

If you were running the page_mapped() check on a different "page" then
it could make a difference, but repeating it on the same page in the
same rcu_read_lock protected critical section won't make a difference
as far as the anon_vma freeing is concerned.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/