On Fri, 10 Feb 2023 02:29:44 +0000
"Hu, Jiayu" <jiayu...@intel.com> wrote:
> Hi Stephen,
>
> > -----Original Message-----
> > From: Stephen Hemminger <step...@networkplumber.org>
> > Sent: Friday, February 10, 2023 9:07 AM
> > To: dev@dpdk.org
> > Cc: Stephen Hemminger <step...@networkplumber.org>; Hu, Jiayu
> > <jiayu...@intel.com>; Konstantin Ananyev
> > <konstantin.v.anan...@yandex.ru>; Mark Kavanagh
> > <mark.b.kavan...@intel.com>
> > Subject: [PATCH v3 01/16] gso: remove logtype
> >
> > If a large packet is passed into GSO routines of unknown protocol then
> > library
> > would log a message and pass it through. This is incorrect behaviour on many
> > levels:
> > - it allows oversize packet to get passed on to NIC driver
>
> Applications use SW GSO only if NIC segmentation is not
> supported. For a burst of packets mixed with different packet
> types, GSO doesn't process the packet with unsupported types,
> and the oversize packets go to NIC and get dropped finally.
> I presume this case frequently happen in real cases.
>
> > - no direct return is visible to applications
> > - if it happens once, many more will follow and log will fill.
> > - bonus it is only log message with GSO type.
> >
> > The fix is to just return -EINVAL which is what this library does in many
> > other
> > places when looking at headers.
> >
> > Fixes: 119583797b6a ("gso: support TCP/IPv4 GSO")
> > Cc: jiayu...@intel.com
> > Signed-off-by: Stephen Hemminger <step...@networkplumber.org>
> > ---
> > lib/eal/common/eal_common_log.c | 2 +-
> > lib/eal/include/rte_log.h | 1 -
> > lib/gso/rte_gso.c | 3 +--
> > 3 files changed, 2 insertions(+), 4 deletions(-)
> >
> > diff --git a/lib/eal/common/eal_common_log.c
> > b/lib/eal/common/eal_common_log.c index bd7b188ceb4a..c369154cb1ea
> > 100644
> > --- a/lib/eal/common/eal_common_log.c
> > +++ b/lib/eal/common/eal_common_log.c
> > @@ -368,7 +368,7 @@ static const struct logtype logtype_strings[] = {
> > {RTE_LOGTYPE_CRYPTODEV, "lib.cryptodev"},
> > {RTE_LOGTYPE_EFD, "lib.efd"},
> > {RTE_LOGTYPE_EVENTDEV, "lib.eventdev"},
> > - {RTE_LOGTYPE_GSO, "lib.gso"},
> > +
> > {RTE_LOGTYPE_USER1, "user1"},
> > {RTE_LOGTYPE_USER2, "user2"},
> > {RTE_LOGTYPE_USER3, "user3"},
> > diff --git a/lib/eal/include/rte_log.h b/lib/eal/include/rte_log.h index
> > 6d2b0856a565..97d6b26a9967 100644
> > --- a/lib/eal/include/rte_log.h
> > +++ b/lib/eal/include/rte_log.h
> > @@ -46,7 +46,6 @@ extern "C" {
> > #define RTE_LOGTYPE_CRYPTODEV 17 /**< Log related to cryptodev. */
> > #define RTE_LOGTYPE_EFD 18 /**< Log related to EFD. */
> > #define RTE_LOGTYPE_EVENTDEV 19 /**< Log related to eventdev. */
> > -#define RTE_LOGTYPE_GSO 20 /**< Log related to GSO. */
> >
> > /* these log types can be used in an application */
> > #define RTE_LOGTYPE_USER1 24 /**< User-defined log type 1. */
> > diff --git a/lib/gso/rte_gso.c b/lib/gso/rte_gso.c index
> > 4b59217c16ee..19c351769fcc 100644
> > --- a/lib/gso/rte_gso.c
> > +++ b/lib/gso/rte_gso.c
> > @@ -81,8 +81,7 @@ rte_gso_segment(struct rte_mbuf *pkt,
> > indirect_pool, pkts_out, nb_pkts_out);
> > } else {
> > /* unsupported packet, skip */
> > - RTE_LOG(DEBUG, GSO, "Unsupported packet type\n");
> > - ret = 0;
> > + ret = -EINVAL;
>
> If applications want to know that the packet is failed on SW GSO,
> I think ENOTSUP is better than EINVAL, as it is not invalid input
> from users.
Ok, just wanted to do the same thing as what would happen
earlier in the code path for cases where the arguments are incorrect.
Doing -ENOTSUP makes sense.
