Re: [PATCH] proc: fix pagemap_read() error case (was Re: [PATCH] proc: put check_mem_permission before __get_free_page in mem_read)

[Stephen Wilson <wilsons@start.ca>]

On Tue, Apr 26, 2011 at 02:50:16PM +0900, KOSAKI Motohiro wrote:
> I've finished audit other /proc allocation callsite. If my understand
> is correct, only pagemap_read() has the same issue.

I think there is one additional location that might be worth looking at.
We have a kmalloc(struct numa_maps) happening in show_numa_map() (see
mempolicy.c).  In this case there is an allocation/free cycle happening
for each vma as we generate the seq_file.

Unfortunately a fix might require a little work.  Initial thinking
suggests that we perform a single allocation at numa_maps_open() time.
However, there is an odd layering/dependency issue between mempolicy.c
and task_mmu.c.  A "simple clean fix" does not seem obvious to me.

I am very tempted to suggest we move the proc related stuff out of
mempolicy.c.  However, see 1a75a6c8.

Thoughts?  I can certainly look into this some more if needed.

-- 
steve