Re: [PATCH] avoid null pointer access in vm_struct

[HAYASAKA Mitsuo <mitsuo.hayasaka.hu@hitachi.com>]

Andrew Morton さんは書きました:
> On Wed, 17 Aug 2011 22:28:48 +0900
> Mitsuo Hayasaka <mitsuo.hayasaka.hu@hitachi.com> wrote:
> 
>> The /proc/vmallocinfo shows information about vmalloc allocations in vmlist
>> that is a linklist of vm_struct. It, however, may access pages field of
>> vm_struct where a page was not allocated, which results in a null pointer
>> access and leads to a kernel panic.
>>
>> Why this happen:
>> For example, in __vmalloc_area_node, the nr_pages field of vm_struct are
>> set to the expected number of pages to be allocated, before the actual
>> pages allocations. At the same time, when the /proc/vmallocinfo is read, it
>> accesses the pages field of vm_struct according to the nr_pages field at
>> show_numa_info(). Thus, a null pointer access happens.
>>
>> Patch:
>> This patch avoids accessing the pages field with unallocated page when
>> show_numa_info() is called. So, it can solve this problem.
> 
> Do we have a similar race when running __vunmap() in parallel with
> show_numa_info()?
> 

No. This does not happen when running __vunmap() because the vm_struct
is released after it is removed from vmlist.


>> index 7ef0903..e2ec5b0 100644
>> --- a/mm/vmalloc.c
>> +++ b/mm/vmalloc.c
>> @@ -2472,13 +2472,16 @@ static void show_numa_info(struct seq_file *m, struct vm_struct *v)
>>  	if (NUMA_BUILD) {
>>  		unsigned int nr, *counters = m->private;
>>  
>> -		if (!counters)
>> +		if (!counters || !v->nr_pages || !v->pages)
>>  			return;
>>  
>>  		memset(counters, 0, nr_node_ids * sizeof(unsigned int));
>>  
>> -		for (nr = 0; nr < v->nr_pages; nr++)
>> +		for (nr = 0; nr < v->nr_pages; nr++) {
>> +			if (!v->pages[nr])
>> +				break;
>>  			counters[page_to_nid(v->pages[nr])]++;
>> +		}
>>  
>>  		for_each_node_state(nr, N_HIGH_MEMORY)
>>  			if (counters[nr])
> 
> I think this has memory ordering issues: it requires that this CPU see
> the modification to ->nr_pages and ->pages in the same order as the CPU
> which is writing ->nr_pages, ->pages and ->pages[x].  Perhaps fixable
> by taking vmlist_lock appropriately.
> 
> I suspect that the real bug is that __vmalloc_area_node() and its
> caller made the new vmap_area globally visible before it was fully
> initialised.  If we were to fix that, the /proc/vmallocinfo read would
> not encounter this vm_struct at all.
> 

I agreed.
I'd like to revise __vmalloc_area_node() and submit the patch again.

Thanks.