Pablo Neira Ayuso <pablo@xxxxxxxxxxxxx> wrote:
> On Mon, Mar 11, 2024 at 10:14:54PM +0800, Quan Tian wrote:
> > The NFTA_TABLE_USERDATA attribute was ignored on updates. The patch adds
> > handling for it to support table comment updates.
>
> dump path is lockless:
>
> if (table->udata) {
> if (nla_put(skb, NFTA_TABLE_USERDATA, table->udlen, table->udata))
> goto nla_put_failure;
> }
>
> there are two things to update at the same time here, table->udata and
> table->udlen.
>
> This needs to be reworked fully if updates are required.
See first patch in the series, it makes this a single pointer,
but you are right...
> then, update struct nft_table to have:
>
> struct nft_userdata __rcu *user;
.. this needs an __rcu annotation. I'll respond
to patch 1 too.
> BTW, does swap() ensure rcu semantics?
No, this needs to use rcu_replace_pointer() and manual update
of the old one stored in the transaction update.
