Re: [PATCH -v2] rmap: make anon_vma_prepare link in all the anon_vmas of a mergeable VMA

[Rik van Riel <riel@redhat.com>]

On 04/07/2010 11:30 AM, Linus Torvalds wrote:

> I've said this now _three_ times, but let me repeat once more:
>
>   - the locking rules for that anon_vma_chain are very unclear. I _think_
>     you mean for them to be "mmap_sem held for writing, _or_ mmap_sem held
>     for reading and page_table_lock held", but nowhere is that actually
>     documented.

> Why is it so hard for you to just admit that? Especially after you
> yourself got it wrong.

You are right, the idea was to continue use the locking that
the anon_vma code was already using, without introducing any
new locking with the anon_vma patches.

However, it has become clear that this is no longer possible,
due to the need to hold a secondary lock across anon_vma_clone,
when we come from a code path that holds the mmap_sem for read.

>> +		merge_vma = find_mergeable_anon_vma(vma);
>> +		if (merge_vma) {
>> +			int ret;
>> +			spin_lock(&mm->page_table_lock);
>> +			ret = anon_vma_clone(vma, merge_vma);
>> +			if (!ret)
>> +				vma->anon_vma = merge_vma->anon_vma;
>> +			spin_unlock(&mm->page_table_lock);
>> +			return ret;
>> +		}
>
> Rik, the above is obviously total crap.
>
> anon_vma_clone() needs to allocate memory, and it does so with GFP_KERNEL.
> You can't do that with a spinlock held.

Looks like we'll either have to introduce a per-mm semaphore for
the same_vma anon_vma chains, or move the complexity of solving
this bug to anon_vma_merge, where we can ensure that the resulting
VMA has the sum of the anon_vmas of each VMA.