Re: [PATCH v4 4/4] mm/vmalloc: don't assume vmap_area w/o VM_VM_AREA flag is vm_map_ram allocation

[Wanpeng Li <liwanp@linux.vnet.ibm.com>]

On Tue, Sep 03, 2013 at 05:59:59PM +0900, Joonsoo Kim wrote:
>On Tue, Sep 03, 2013 at 03:51:39PM +0800, Wanpeng Li wrote:
>> On Tue, Sep 03, 2013 at 04:42:21PM +0900, Joonsoo Kim wrote:
>> >On Tue, Sep 03, 2013 at 03:01:46PM +0800, Wanpeng Li wrote:
>> >> There is a race window between vmap_area free and show vmap_area information.
>> >> 
>> >> 	A                                                B
>> >> 
>> >> remove_vm_area
>> >> spin_lock(&vmap_area_lock);
>> >> va->flags &= ~VM_VM_AREA;
>> >> spin_unlock(&vmap_area_lock);
>> >> 						spin_lock(&vmap_area_lock);
>> >> 						if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEZING))
>> >> 							return 0;
>> >> 						if (!(va->flags & VM_VM_AREA)) {
>> >> 							seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n",
>> >> 								(void *)va->va_start, (void *)va->va_end,
>> >> 								va->va_end - va->va_start);
>> >> 							return 0;
>> >> 						}
>> >> free_unmap_vmap_area(va);
>> >> 	flush_cache_vunmap
>> >> 	free_unmap_vmap_area_noflush
>> >> 		unmap_vmap_area
>> >> 		free_vmap_area_noflush
>> >> 			va->flags |= VM_LAZY_FREE 
>> >> 
>> >> The assumption is introduced by commit: d4033afd(mm, vmalloc: iterate vmap_area_list, 
>> >> instead of vmlist, in vmallocinfo()). This patch fix it by drop the assumption and 
>> >> keep not dump vm_map_ram allocation information as the logic before that commit.
>> >> 
>> >> Signed-off-by: Wanpeng Li <liwanp@linux.vnet.ibm.com>
>> >> ---
>> >>  mm/vmalloc.c | 7 -------
>> >>  1 file changed, 7 deletions(-)
>> >> 
>> >> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
>> >> index 5368b17..62b7932 100644
>> >> --- a/mm/vmalloc.c
>> >> +++ b/mm/vmalloc.c
>> >> @@ -2586,13 +2586,6 @@ static int s_show(struct seq_file *m, void *p)
>> >>  	if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
>> >>  		return 0;
>> >>  
>> >> -	if (!(va->flags & VM_VM_AREA)) {
>> >> -		seq_printf(m, "0x%pK-0x%pK %7ld vm_map_ram\n",
>> >> -			(void *)va->va_start, (void *)va->va_end,
>> >> -					va->va_end - va->va_start);
>> >> -		return 0;
>> >> -	}
>> >> -
>> >>  	v = va->vm;
>> >>  
>> >>  	seq_printf(m, "0x%pK-0x%pK %7ld",
>> >
>> >Hello, Wanpeng.
>> >
>> 
>> Hi Joonsoo and Yanfei,
>> 
>> >Did you test this patch?
>> >
>> >I guess that, With this patch, if there are some vm_map areas,
>> >null pointer deference would occurs, since va->vm may be null for it.
>> >
>> >And with this patch, if this race really occur, null pointer deference
>> >would occurs too, since va->vm is set to null in remove_vm_area().
>> >
>> >I think that this is not a right fix for this possible race.
>> >
>> 
>> How about append below to this patch?
>> 
>> if (va->vm)
>> 	v = va->vm;
>> else 
>> 	return 0;
>
>Hello,
>
>I think that appending below code is better to represent it's purpose.
>Maybe some comment is needed.
>
>	/* blablabla */
>	if (!(va->flags & VM_VM_AREA))
>		return 0;
>

Looks reasonable to me. ;-)

>And maybe we can remove below code snippet, since
>either VM_LAZY_FREE or VM_LAZY_FREEING is not possible for !VM_VM_AREA case.
>
>	if (va->flags & (VM_LAZY_FREE | VM_LAZY_FREEING))
>		return 0;
>

Agreed.

I will fold these in my patch and add your suggested-by. Thanks.

Regards,
Wanpeng Li 

>Thanks.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>