Re: general protection fault in ucma_connect

[Al Viro]

On Fri, Mar 16, 2018 at 04:59:02PM -0700, syzbot wrote:
> Hello,
> 
> syzbot hit the following crash on upstream commit
> e2c15aff5f353ba80bd3bb49840837f65fa5cc43 (Thu Mar 15 18:07:35 2018 +0000)
> Merge tag 'sound-4.16-rc6' of
> git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound
> 
> So far this crash happened 2 times on upstream.
> C reproducer is attached.
> syzkaller reproducer is attached.
> Raw console output is attached.
> compiler: gcc (GCC) 7.1.1 20170620
> .config is attached.

May I politely inquire why am I getting Cc on the slew of ucma reports?
My last involvement with that thing that wouldn't have been absolutely
trivial had been back in 2012, and even that had been a conversion of
fget() to fdget() - nowhere near the areas implicated by those.  Anything
more recent would have no impact on the object code - replacement of
POLL{IN,RDNORM} with EPOLL{IN,RDNORM} (equal values on x86) and two
replacements of unsigned int with __poll_t, which is typedefed to
unsigned.

It's not that I've objections against helping to debug that thing (other
than general aversion to drivers/infinibarf), but I'm really curious -
just what got me volunteered from the syzbot POV?

Al, digging through tons of unpleasant code in fs/{config,debug}fs
at the moment...