Re: [PATCH mptcp-next v2 5/9] mptcp: infinite mapping sending

[Paolo Abeni <pabeni@redhat.com>]

On Thu, 2021-09-09 at 19:51 +0800, Geliang Tang wrote:
> From: Geliang Tang <geliangtang@xiaomi.com>
> 
> This patch added the infinite mapping sending logic.
> 
> Added a new flag snd_infinite_mapping_enable in mptcp_sock. Set it true
> when a single contiguous subflow is in use in mptcp_pm_mp_fail_received.
> In mptcp_sendmsg_frag, if this flag is true, call the new function
> mptcp_update_infinite_mapping to set the infinite mapping.
> 
> Signed-off-by: Geliang Tang <geliangtang@xiaomi.com>
> ---
>  net/mptcp/pm.c       |  6 ++++++
>  net/mptcp/protocol.c | 18 ++++++++++++++++++
>  net/mptcp/protocol.h |  1 +
>  3 files changed, 25 insertions(+)
> 
> diff --git a/net/mptcp/pm.c b/net/mptcp/pm.c
> index 6ab386ff3294..2830adf64f79 100644
> --- a/net/mptcp/pm.c
> +++ b/net/mptcp/pm.c
> @@ -251,7 +251,13 @@ void mptcp_pm_mp_prio_received(struct sock *sk, u8 bkup)
>  
>  void mptcp_pm_mp_fail_received(struct sock *sk, u64 fail_seq)
>  {
> +	struct mptcp_subflow_context *subflow = mptcp_subflow_ctx(sk);
> +	struct mptcp_sock *msk = mptcp_sk(subflow->conn);
> +
>  	pr_debug("fail_seq=%llu", fail_seq);
> +
> +	if (!mptcp_has_another_subflow(sk) && !READ_ONCE(msk->noncontiguous))
> +		WRITE_ONCE(msk->snd_infinite_mapping_enable, true);
>  }
>  
>  /* path manager helpers */
> diff --git a/net/mptcp/protocol.c b/net/mptcp/protocol.c
> index c7ecd3e3b537..4ebbbc6f1d01 100644
> --- a/net/mptcp/protocol.c
> +++ b/net/mptcp/protocol.c
> @@ -1278,6 +1278,21 @@ static void mptcp_update_data_checksum(struct sk_buff *skb, int added)
>  	mpext->csum = csum_fold(csum_block_add(csum, skb_checksum(skb, offset, added, 0), offset));
>  }
>  
> +static void mptcp_update_infinite_mapping(struct mptcp_sock *msk, struct mptcp_ext *mpext)
> +{
> +	if (!mpext)
> +		return;
> +
> +	mpext->data_seq = READ_ONCE(msk->start_seq);
> +	mpext->subflow_seq = 0;
> +	mpext->data_len = 0;
> +	mpext->csum = 0;
> +
> +	WRITE_ONCE(msk->snd_infinite_mapping_enable, false);
> +	pr_infinite(msk);
> +	__mptcp_do_infinite(msk);
> +}

If you move the 'snd_infinite_mapping_enable' flag at the subflow level
(inside struct mptcp_subflow_context), all the _ONCE annotation will
not be needed. And the flag will be already initializated to 0 at ctx
creation time.

Additionally, I'm unsure the MPTCP_INFINITE_DONE bit is required ?!?
can we simply rely on FALLBACK ???

/P