On Fri, 24 Oct 2014, Thomas Gleixner wrote:
> On Sun, 12 Oct 2014, Qiaowei Ren wrote:
> > +int mpx_enable_management(struct task_struct *tsk)
> > +{
> > + struct mm_struct *mm = tsk->mm;
> > + void __user *bd_base = MPX_INVALID_BOUNDS_DIR;
>
> What's the point of initializing bd_base here. I had to look twice to
> figure out that it gets overwritten by task_get_bounds_dir()
>
> > @@ -285,6 +285,7 @@ dotraplinkage void do_bounds(struct pt_regs *regs, long error_code)
> > struct xsave_struct *xsave_buf;
> > struct task_struct *tsk = current;
> > siginfo_t info;
> > + int ret = 0;
> >
> > prev_state = exception_enter();
> > if (notify_die(DIE_TRAP, "bounds", regs, error_code,
> > @@ -312,8 +313,35 @@ dotraplinkage void do_bounds(struct pt_regs *regs, long error_code)
> > */
> > switch (status & MPX_BNDSTA_ERROR_CODE) {
> > case 2: /* Bound directory has invalid entry. */
> > - if (do_mpx_bt_fault(xsave_buf))
> > + down_write(¤t->mm->mmap_sem);
>
> The handling of mm->mmap_sem here is horrible. The only reason why you
> want to hold mmap_sem write locked in the first place is that you want
> to cover the allocation and the mm->bd_addr check.
>
> I think it's wrong to tie this to mmap_sem in the first place. If MPX
> is enabled then you should have mm->bd_addr and an explicit mutex to
> protect it.
>
> So the logic would look like this:
>
> mutex_lock(&mm->bd_mutex);
> if (!kernel_managed(mm))
> do_trap();
> else if (do_mpx_bt_fault())
> force_sig();
> mutex_unlock(&mm->bd_mutex);
>
> No tricks with mmap_sem, no special return value handling. Straight
> forward code instead of a convoluted and error prone mess.
After thinking about the deallocation issue, this would be mm->bd_sem.
Thanks,
tglx
