[net-next,4/4] net/tls: byte swap device req TCP seq no upon setting

To avoid a sparse warning byteswap the be32 sequence number
before it's stored in the atomic value.  While at it drop
unnecessary brackets and use kernel's u64 type.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Reviewed-by: Simon Horman <simon.horman@netronome.com>
---
 include/net/tls.h    | 2 +-
 net/tls/tls_device.c | 2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)

On Thu, 2019-04-25 at 09:56 -0700, Jakub Kicinski wrote:
> To avoid a sparse warning byteswap the be32 sequence number
> before it's stored in the atomic value.  While at it drop
> unnecessary brackets and use kernel's u64 type.
> 
> Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
> Reviewed-by: Reviewed-by: Simon Horman <simon.horman@netronome.com>
> ---
>  include/net/tls.h    | 2 +-
>  net/tls/tls_device.c | 2 +-
>  2 files changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/include/net/tls.h b/include/net/tls.h
> index 41a2ee643fc5..39ea62f0c1f6 100644
> --- a/include/net/tls.h
> +++ b/include/net/tls.h
> @@ -562,7 +562,7 @@ static inline void
> tls_offload_rx_resync_request(struct sock *sk, __be32 seq)
>  	struct tls_context *tls_ctx = tls_get_ctx(sk);
>  	struct tls_offload_context_rx *rx_ctx =
> tls_offload_ctx_rx(tls_ctx);
>  
> -	atomic64_set(&rx_ctx->resync_req, ((((uint64_t)seq) << 32) |
> 1));
> +	atomic64_set(&rx_ctx->resync_req, ((u64)ntohl(seq) << 32) | 1);
>  }
>  
>  
> diff --git a/net/tls/tls_device.c b/net/tls/tls_device.c
> index cb368efe3567..6686013b4e9e 100644
> --- a/net/tls/tls_device.c
> +++ b/net/tls/tls_device.c
> @@ -567,7 +567,7 @@ void (struct sock *sk, u32
> seq, u64 rcd_sn)
>  
>  	rx_ctx = tls_offload_ctx_rx(tls_ctx);
>  	resync_req = atomic64_read(&rx_ctx->resync_req);
> -	req_seq = ntohl(resync_req >> 32) - ((u32)TLS_HEADER_SIZE - 1);
> +	req_seq = (resync_req >> 32) - ((u32)TLS_HEADER_SIZE - 1);

this is not equivalent to what was before, 
resync_req is expected to be in network order, 
(TLS_HEADER_SIZE -1) is still in cpu indianness.


>  	is_req_pending = resync_req;
>  
>  	if (unlikely(is_req_pending) && req_seq == seq &&

On Thu, 25 Apr 2019 19:31:51 +0000, Saeed Mahameed wrote:
> > diff --git a/net/tls/tls_device.c b/net/tls/tls_device.c
> > index cb368efe3567..6686013b4e9e 100644
> > --- a/net/tls/tls_device.c
> > +++ b/net/tls/tls_device.c
> > @@ -567,7 +567,7 @@ void (struct sock *sk, u32
> > seq, u64 rcd_sn)
> >  
> >  	rx_ctx = tls_offload_ctx_rx(tls_ctx);
> >  	resync_req = atomic64_read(&rx_ctx->resync_req);
> > -	req_seq = ntohl(resync_req >> 32) - ((u32)TLS_HEADER_SIZE - 1);
> > +	req_seq = (resync_req >> 32) - ((u32)TLS_HEADER_SIZE - 1);  
> 
> this is not equivalent to what was before, 
> resync_req is expected to be in network order, 
> (TLS_HEADER_SIZE -1) is still in cpu indianness.

Naw, I think they are both in host order.

The driver passes network order.

But the stack has it in host order, this is the call site:

#ifdef CONFIG_TLS_DEVICE
	handle_device_resync(strp->sk, TCP_SKB_CB(skb)->seq + rxm->offset,
			     *(u64*)tls_ctx->rx.rec_seq);
#endif

The value passed by the driver used to be byte swapped when read from
the atomic, but I moved the byte swap to when it's stored to the atomic.
We used to have a weird situation where the atomic would have a be32 on
the top 32bits, and lower 32 bits would store the 1, in host order.

IOW the tls_offload_rx_resync_request() is the only thing setting this
value and I moved the byte swap there.

$ git grep '.->resync_req'
include/net/tls.h:      atomic64_set(&rx_ctx->resync_req, ((u64)ntohl(seq) << 32) | 1);
net/tls/tls_device.c:   resync_req = atomic64_read(&rx_ctx->resync_req);
net/tls/tls_device.c:       atomic64_try_cmpxchg(&rx_ctx->resync_req, &resync_req, 0))

Did I miss something or screw up tls_offload_rx_resync_request()?

On Thu, 2019-04-25 at 12:38 -0700, Jakub Kicinski wrote:
> On Thu, 25 Apr 2019 19:31:51 +0000, Saeed Mahameed wrote:
> > > diff --git a/net/tls/tls_device.c b/net/tls/tls_device.c
> > > index cb368efe3567..6686013b4e9e 100644
> > > --- a/net/tls/tls_device.c
> > > +++ b/net/tls/tls_device.c
> > > @@ -567,7 +567,7 @@ void (struct sock *sk, u32
> > > seq, u64 rcd_sn)
> > >  
> > >  	rx_ctx = tls_offload_ctx_rx(tls_ctx);
> > >  	resync_req = atomic64_read(&rx_ctx->resync_req);
> > > -	req_seq = ntohl(resync_req >> 32) - ((u32)TLS_HEADER_SIZE - 1);
> > > +	req_seq = (resync_req >> 32) - ((u32)TLS_HEADER_SIZE - 1);  
> > 
> > this is not equivalent to what was before, 
> > resync_req is expected to be in network order, 
> > (TLS_HEADER_SIZE -1) is still in cpu indianness.
> 
> Naw, I think they are both in host order.
> 
> The driver passes network order.
> 
> But the stack has it in host order, this is the call site:
> 
> #ifdef CONFIG_TLS_DEVICE
> 	handle_device_resync(strp->sk, TCP_SKB_CB(skb)->seq + rxm-
> >offset,
> 			     *(u64*)tls_ctx->rx.rec_seq);
> #endif
> 
> The value passed by the driver used to be byte swapped when read from
> the atomic, but I moved the byte swap to when it's stored to the
> atomic.
> We used to have a weird situation where the atomic would have a be32
> on
> the top 32bits, and lower 32 bits would store the 1, in host order.
> 
> IOW the tls_offload_rx_resync_request() is the only thing setting
> this
> value and I moved the byte swap there.
> 
> $ git grep '.->resync_req'
> include/net/tls.h:      atomic64_set(&rx_ctx->resync_req,
> ((u64)ntohl(seq) << 32) | 1);
> net/tls/tls_device.c:   resync_req = atomic64_read(&rx_ctx-
> >resync_req);
> net/tls/tls_device.c:       atomic64_try_cmpxchg(&rx_ctx->resync_req, 
> &resync_req, 0))
> 
> Did I miss something or screw up tls_offload_rx_resync_request()?

No, after a second pass, i think all is good, i missed the fact in your
change now rx_ctx->resync_req is storing the value in host order.

Thanks,
Saeed.