Re: [PATCH - revised] rhashtable: detect when object movement might have invalidated a lookup

[NeilBrown]

On Mon, Jul 16 2018, Herbert Xu wrote:

> On Mon, Jul 16, 2018 at 11:23:43AM +1000, NeilBrown wrote:
>>
>> kmem_cache_free() directly.  For this, I need rhashtable to be safe if
>> an object is deleted and immediately re-inserted into the same hash
>> chain.
>
> This means that
>
> 	rcu_read_lock();
> 	A = rhashtable_lookup();
> 	use(A);
> 	rcu_read_unlock();
>
> A can turn into object B when it is used.  That is just too strange
> for words.  Can we see some actual code on how this works?

Look in Documenation/RCU/rculist_nulls.txt.
The very first example is a typical lookup for a nulls list.
The above sample code would read:

	rcu_read_lock();
    begin:
        A = rhashtable_lookup(table, key);
        if (IS_ERR(A)) {
        	rcu_read_unlock();
        	goto not_found;
        }
        if (!try_get_ref(A))
        	goto again;
        if (A->key != key) {
        	put_ref(A);
        	goto again;
        }
        rcu_read_unlock();
        use(A);

If you particularly need to see real code rather than sample code, I can
have something for you in a day or so.

Thanks,
NeilBrown

>
> For comparison, the existing net code where this happens A doesn't
> actually change and it simply moves from one hashtable to another.
>
> I'm relucant to add semantics that would restrain on how rhashtable
> works unless we have real and valid use-cases for it.
>
> Cheers,
> -- 
> Email: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx>
> Home Page: http://gondor.apana.org.au/~herbert/
> PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Attachment: signature.asc
Description: PGP signature