On Wed 08-04-20 11:24:35, Peter Xu wrote:
> On Wed, Apr 08, 2020 at 04:30:24PM +0200, Michal Hocko wrote:
> > On Wed 08-04-20 10:20:39, Peter Xu wrote:
> > > On Wed, Apr 08, 2020 at 12:21:28PM +0200, Michal Hocko wrote:
> > > > On Tue 07-04-20 21:40:09, Peter Xu wrote:
> > > > > lookup_node() uses gup to pin the page and get node information. It
> > > > > checks against ret>=0 assuming the page will be filled in. However
> > > > > it's also possible that gup will return zero, for example, when the
> > > > > thread is quickly killed with a fatal signal. Teach lookup_node() to
> > > > > gracefully return an error -EFAULT if it happens.
> > > > >
> > > > > Meanwhile, initialize "page" to NULL to avoid potential risk of
> > > > > exploiting the pointer.
> > > > >
> > > > > Reported-by: syzbot+693dc11fcb53120b5559@xxxxxxxxxxxxxxxxxxxxxxxxx
> > > > > Fixes: 4426e945df58 ("mm/gup: allow VM_FAULT_RETRY for multiple times")
> > > >
> > > > I am not familiar with thic commit but shouldn't gup return ERESTARTSYS
> > > > on a fatal signal?
> > >
> > > Hi, Michal,
> > >
> > > I do see quite a few usages on -ERESTARTSYS, but also some others,
> > > majorly -EINTR, or even -EFAULT. I think it could be a more general
> > > question rather than a specific question to this patch only.
> >
> > I am sorry but I was probably not clear enough. I was mostly worried
> > that gup doesn't return ERESTARTSYS or EINTR when it backed off because
> > of fatal signal pending. Your patch is checking for 0 an indicating that
> > this is that condition.
>
> Yeah I just noticed the fact, sorry!
>
> Hillf just posted a fix there for recovering the behavior:
>
> https://lore.kernel.org/lkml/20200408151213.GE66033@xz-x1/
yeah, that is the proper fix.
--
Michal Hocko
SUSE Labs
