Re: get_file() unsafe under epoll (was Re: [syzbot] [fs?] [io-uring?] general protection fault in __ep_remove)

[Daniel Vetter <daniel@ffwll.ch>]

On Fri, May 03, 2024 at 10:53:03PM +0100, Al Viro wrote:
> On Fri, May 03, 2024 at 02:42:22PM -0700, Linus Torvalds wrote:
> > On Fri, 3 May 2024 at 14:36, Al Viro <viro@zeniv.linux.org.uk> wrote:
> > >
> > > ... the last part is no-go - poll_wait() must be able to grab a reference
> > > (well, the callback in it must)
> > 
> > Yeah. I really think that *poll* itself is doing everything right. It
> > knows that it's called with a file pointer with a reference, and it
> > adds its own references as needed.
> 
> Not really.  Note that select's __pollwait() does *NOT* leave a reference
> at the mercy of driver - it's stuck into poll_table_entry->filp and
> the poll_freewait() knows how to take those out.
> 
> 
> dmabuf does something very different - it grabs the damn thing into
> its private data structures and for all we know it could keep it for
> a few hours, until some even materializes.

dma_fence must complete in reasonable amount of time, where "reasonable"
is roughly in line with other i/o (including the option that there's
timeouts if the hw's gone busted).

So definitely not hours (aside from driver bugs when things go really
wrong ofc), but more like a few seconds in a worst case scenario.
-Sima
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch