Re: [PATCH] x86/fpu: Simplify fpu__activate_fpstate_read()

From: Ingo Molnar
Date: Sat Sep 23 2017 - 07:29:45 EST



* Ingo Molnar <mingo@xxxxxxxxxx> wrote:

>
> * Eric Biggers <ebiggers3@xxxxxxxxx> wrote:
>
> > On Fri, Sep 22, 2017 at 07:33:14AM +0200, Ingo Molnar wrote:
> > >
> > > * Eric Biggers <ebiggers3@xxxxxxxxx> wrote:
> > >
> > > > From: Eric Biggers <ebiggers@xxxxxxxxxx>
> > > >
> > > > This series fixes the bug found by syzkaller where the ptrace syscall
> > > > can be used to set invalid bits in a task's FPU state. I also found
> > > > that an equivalent bug was reachable using the sigreturn syscall, so the
> > > > first patch fixes the bug in both cases.
> > > >
> > > > The other two patches start validating the other parts of the
> > > > xstate_header and make it so that invalid FPU states can no longer be
> > > > abused to leak the FPU registers of other processes.
> > > >
> > > > Changes since v2:
> > > > - Use an exception handler to handle invalid FPU states
> > > > (suggested by Andy Lutomirski)
> > > > - Check the size of xstate_header.reserved at build time
> > > > (suggested by Dave Hansen)
> > > >
> > > > Eric Biggers (3):
> > > > x86/fpu: don't let userspace set bogus xcomp_bv
> > > > x86/fpu: tighten validation of user-supplied xstate_header
> > > > x86/fpu: reinitialize FPU registers if restoring FPU state fails
> > > >
> > > > arch/x86/include/asm/fpu/internal.h | 51 +++++++++++--------------------------
> > > > arch/x86/include/asm/fpu/xstate.h | 25 ++++++++++++++++++
> > > > arch/x86/kernel/fpu/regset.c | 20 +++++++--------
> > > > arch/x86/kernel/fpu/signal.c | 15 ++++++++---
> > > > arch/x86/kernel/fpu/xstate.c | 27 ++++++++------------
> > > > arch/x86/mm/extable.c | 24 +++++++++++++++++
> > > > 6 files changed, 94 insertions(+), 68 deletions(-)
> > >
> > > Ok - could you please rebase these to to tip:master that is at:
> > >
> > > git git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git master
> > >
> > > In particular this has a WIP.x86/fpu branch with FPU fixes+changes queued up but
> > > not merged upstream (yet), which conflict with these changes. I'd like to merge
> > > them all together.
> > >
> >
> > Working on it, but there is a problem with current tip. PTRACE_GETREGSET is
> > causing the following warning:
>
> Yes, the warning should be harmless, and I fixed it locally earlier today - does
> the patch below solve it for you as well?

Note that this fix is now part of tip:master as well, so if you re-test -tip you
should get all the latest fixes as well (including yours!).

Thanks,

Ingo