Re: [RFC PATCH for 4.18 1/2] rseq: validate rseq_cs fields are < TASK_SIZE

[Andy Lutomirski]

On Mon, Jul 2, 2018 at 12:31 PM, Linus Torvalds
<torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> On Mon, Jul 2, 2018 at 12:02 PM Andy Lutomirski <luto@xxxxxxxxxxxxxx> wrote:
>>
>> Works for me.  Linus, any objection?
>
> I think the 4.19 stage may be overkill, but I don't hate it, so no
> real objections.
>
> If the main reason for this is that we silently clear the upper bits
> when returning to compat mode, I actually think that a better fix
> would be to just fix that. We shouldn't silently ignore bogus data in
> the return path.
>
> But I don't care enough, I think.

Like this:

diff --git a/arch/x86/entry/common.c b/arch/x86/entry/common.c
index 3b2490b81918..ec40223c8856 100644
--- a/arch/x86/entry/common.c
+++ b/arch/x86/entry/common.c
@@ -170,6 +170,26 @@ static void exit_to_usermode_loop(struct pt_regs
*regs, u32 cached_flags)
         if (cached_flags & _TIF_USER_RETURN_NOTIFY)
             fire_user_return_notifiers();

+        if (unlikely(!user_64bit_mode(regs) &&
+                 (regs->ip & 0xffffffff00000000ull))) {
+            siginfo_t info;
+            struct task_struct *tsk = current;
+
+            /* I haven't thought about this *that* hard. */
+            clear_siginfo(&info);
+            tsk->thread.cr2    = regs->ip;
+            tsk->thread.trap_nr = X86_TRAP_PF;
+            tsk->thread.error_code = X86_PF_USER | X86_PF_INSTR;
+            info.si_signo = SIGSEGV;
+            info.si_errno = 0;
+            info.si_code = SEGV_MAPERR;
+            info.si_addr = (void __user *)regs->ip;
+            /* si_addr_lsb? */
+            force_sig_info(SIGSEGV, &info, tsk);
+
+            /* And we'll go through the loop again. */
+        }
+
         /* Disable IRQs and retry */
         local_irq_disable();

It's whitespace damaged and barely tested, but it seems to at least
not be completely busted.  I don't really love doing this, though.