lkml.org 
[lkml]   [2018]   [Oct]   [16]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC PATCH for 4.21 06/16] cpu_opv: Provide cpu_opv system call (v8)
----- On Oct 16, 2018, at 4:10 AM, Sergey Senozhatsky sergey.senozhatsky.work@gmail.com wrote:

> Hi Mathieu,
>
> On (10/10/18 15:19), Mathieu Desnoyers wrote:
> [..]
>> +SYSCALL_DEFINE4(cpu_opv, struct cpu_op __user *, ucpuopv, int, cpuopcnt,
>> + int, cpu, int, flags)
>> +{
> [..]
>> +again:
>> + ret = cpu_opv_pin_pages(cpuopv, cpuopcnt, &vaddr_ptrs);
>> + if (ret)
>> + goto end;
>> + ret = do_cpu_opv(cpuopv, cpuopcnt, &vaddr_ptrs, cpu);
>> + if (ret == -EAGAIN)
>> + retry = true;
>> +end:
>> + for (i = 0; i < vaddr_ptrs.nr_vaddr; i++) {
>> + struct vaddr *vaddr = &vaddr_ptrs.addr[i];
>> + int j;
>> +
>> + vm_unmap_user_ram((void *)vaddr->mem, vaddr->nr_pages);
>
> A dumb question.
>
> Both vm_unmap_user_ram() and vm_map_user_ram() can BUG_ON().
> So this is
> userspace -> syscall -> cpu_opv() -> vm_unmap_user_ram() -> BUG_ON()
>
> Any chance someone can exploit it?

Hi Sergey,

Let's look at vm_unmap_user_ram() and vm_map_user_ram() separately.

If we look at the input from vm_unmap_user_ram, it's called with the
following parameters by the cpu_opv system call:

for (i = 0; i < vaddr_ptrs.nr_vaddr; i++) {
struct vaddr *vaddr = &vaddr_ptrs.addr[i];
int j;

vm_unmap_user_ram((void *)vaddr->mem, vaddr->nr_pages);
[...]
}

The vaddr_ptrs array content is filled by the call to cpu_opv_pin_pages above:

ret = cpu_opv_pin_pages(cpuopv, cpuopcnt, &vaddr_ptrs);
if (ret)
goto end;

by passing the array to cpu_op_pin_pages(), which appends a virtual address at
the end of the array (on success) and increments nr_vaddr. Those virtual
addresses are returned by vm_map_user_ram(), so they are not user-controlled.

Therefore, only an internal kernel bug between vm_map_user_ram() and
vm_unmap_user_ram() should trigger the BUG_ON(). No user input is passed
to vm_unmap_user_ram().

Now, let's look at vm_map_user_ram(). It calls alloc_vmap_area(), which returns
a vmap_area. Then if vmap_page_range failed, vm_unmap_user_ram is called on the
memory that has just been returned by vm_map_user_ram. Again, only an internal
bug between map/unmap can trigger the BUG_ON() in vm_unmap_user_ram.

Is there another scenario I missed ?

Thanks,

Mathieu


--
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com

\
 
 \ /
  Last update: 2018-10-16 21:18    [W:0.109 / U:8.220 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site