Re: [RFC PATCH for 4.21 04/16] mm: Introduce vm_map_user_ram, vm_unmap_user_ram

[Steven Rostedt]

On Tue, 16 Oct 2018 15:21:31 -0400 (EDT)
Mathieu Desnoyers <mathieu.desnoyers@xxxxxxxxxxxx> wrote:

> ----- On Oct 16, 2018, at 2:30 PM, rostedt rostedt@xxxxxxxxxxx wrote:
> 
> > On Wed, 10 Oct 2018 15:19:24 -0400
> > Mathieu Desnoyers <mathieu.desnoyers@xxxxxxxxxxxx> wrote:
> >   
> >> + * vm_unmap_user_ram - unmap linear kernel address space set up by
> >> vm_map_user_ram
> >> + * @mem: the pointer returned by vm_map_user_ram
> >> + * @count: the count passed to that vm_map_user_ram call (cannot unmap partial)
> >> + */
> >> +void vm_unmap_user_ram(const void *mem, unsigned int count)
> >> +{
> >> +	unsigned long size = (unsigned long)count << PAGE_SHIFT;
> >> +	unsigned long addr = (unsigned long)mem;
> >> +	struct vmap_area *va;
> >> +
> >> +	might_sleep();
> >> +	BUG_ON(!addr);
> >> +	BUG_ON(addr < VMALLOC_START);
> >> +	BUG_ON(addr > VMALLOC_END);
> >> +	BUG_ON(!PAGE_ALIGNED(addr));
> >> +
> >> +	debug_check_no_locks_freed(mem, size);
> >> +	va = find_vmap_area(addr);
> >> +	BUG_ON(!va);
> >> +	free_unmap_vmap_area(va);
> >> +}
> >> +EXPORT_SYMBOL(vm_unmap_user_ram);
> >> +  
> > 
> > Noticing this from Sergey's question in another patch, why are you
> > using BUG_ON()? That's rather extreme and something we are trying to
> > avoid adding more of (I still need to remove the BUG_ON()s I've added
> > over ten years ago). I don't see why all these BUG_ON's can't be turned
> > into:
> > 
> >	if (WARN_ON(x))
> >		return;  
> 
> I borrowed the code from vm_unmap_ram(), which has the following checks:
> 
>         BUG_ON(!addr);
>         BUG_ON(addr < VMALLOC_START);
>         BUG_ON(addr > VMALLOC_END);
>         BUG_ON(!PAGE_ALIGNED(addr));
> [...]
>         va = find_vmap_area(addr);
>         BUG_ON(!va);
> 
> The expectation here is that inputs to vm_unmap_ram() should always come from
> vm_map_ram(), so an erroneous input is an internal kernel bug. I applied the
> same logic to vm_unmap_user_ram() and vm_map_user_ram().
> 
> Should we turn all those BUG_ON() into if (WARN_ON(x)) return; in vm_{map,unmap}_ram
> as well ?
> 
>

I would argue yes! That code was added in 2008 (which is also the same
year I added BUG_ON() to my code). Back then it wasn't such an issue,
but today we are finding (and Linus has been complaining) that BUG_ON
really shouldn't be necessary. Especially if you can get out of the
function with a simple return.

-- Steve