Re: [dpdk-dev] [PATCH] net: do fragmented headers check in non-debug build as well

[Olivier Matz <olivier.matz@6wind.com>]

Hi Andrew,

On Wed, May 27, 2020 at 03:32:56PM +0100, Andrew Rybchenko wrote:
> Pseudo-header checksum calculation requires contiguous headers.
> There is no any formal requirements on data location and mbuf
> structure which could be used by the application.
> 
> Make corresponding check to be done in non-debug build as well
> to avoid bad accesses, incorrect checksum caclculation and to

typo: caclculation -> calculation

> return appropriate error from Tx prepare.
> 
> Make no-offloads check more precise and do it in non-debug build
> as well to avoid contiguous headers check and Tx prepare failure
> if it is not actually required.
> 
> Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
> ---
>  lib/librte_net/rte_net.h | 6 +-----
>  1 file changed, 1 insertion(+), 5 deletions(-)
> 
> diff --git a/lib/librte_net/rte_net.h b/lib/librte_net/rte_net.h
> index 1560ecfa46..1edc283a47 100644
> --- a/lib/librte_net/rte_net.h
> +++ b/lib/librte_net/rte_net.h
> @@ -120,20 +120,17 @@ rte_net_intel_cksum_flags_prepare(struct rte_mbuf *m, uint64_t ol_flags)
>  	struct rte_udp_hdr *udp_hdr;
>  	uint64_t inner_l3_offset = m->l2_len;
>  
> -#ifdef RTE_LIBRTE_ETHDEV_DEBUG
>  	/*
>  	 * Does packet set any of available offloads?
>  	 * Mainly it is required to avoid fragmented headers check if
>  	 * no offloads are requested.
>  	 */
> -	if (!(ol_flags & PKT_TX_OFFLOAD_MASK))
> +	if (!(ol_flags & (PKT_TX_IP_CKSUM | PKT_TX_L4_MASK)))
>  		return 0;

Agree, the packet is modified only if one of these flag is set.

> -#endif
>  
>  	if (ol_flags & (PKT_TX_OUTER_IPV4 | PKT_TX_OUTER_IPV6))
>  		inner_l3_offset += m->outer_l2_len + m->outer_l3_len;
>  
> -#ifdef RTE_LIBRTE_ETHDEV_DEBUG
>  	/*
>  	 * Check if headers are fragmented.
>  	 * The check could be less strict depending on which offloads are
> @@ -142,7 +139,6 @@ rte_net_intel_cksum_flags_prepare(struct rte_mbuf *m, uint64_t ol_flags)
>  	if (unlikely(rte_pktmbuf_data_len(m) <
>  		     inner_l3_offset + m->l3_len + m->l4_len))
>  		return -ENOTSUP;
> -#endif

Yes, despite the documentation of thus function says that used headers
should be in the first data segment of the mbuf, when it is used through
the ethdev tx_prepare() API there is no such requirement.

So yes, it looks safer to do these checks even if debug is off. They
will only be done when doing tx offload, so I guess it is ok in terms of
performance.

Maybe it is worth mentioning commit dfc6b2fd8da3 ("mbuf: remove Intel
offload checks from generic API") in the commit log?

>  
>  	if (ol_flags & PKT_TX_IPV4) {
>  		ipv4_hdr = rte_pktmbuf_mtod_offset(m, struct rte_ipv4_hdr *,
> -- 
> 2.17.1
> 

Acked-by: Olivier Matz <olivier.matz@6wind.com>