Bug 729

Summary:	Bad mbuf chain handling in the memif driver resulting in truncated packets
Product:	DPDK	Reporter:	Ivan Shvedunov (ivan4th)
Component:	ethdev	Assignee:	dev
Status:	UNCONFIRMED ---
Severity:	normal	CC:	jgrajcia, olivier.matz
Priority:	Normal
Version:	unspecified
Target Milestone:	---
Hardware:	All
OS:	All

Description Ivan Shvedunov 2021-06-10 11:51:11 CEST

In a couple of places in the memif driver [1][2] rte_pktmbuf_is_contiguous() [3] is being used to check whether there are more mbufs in the chain when sending packets. Problem is, rte_pktmbuf_is_contiguous() checks nb_segs in the mbuf, which, as far as I understand, is only set to the number of segments in the first mbuf in the chain. This results in truncated packets as usually, nb_segs is 1 in the mbufs other than the first. Replacing rte_pktmbuf_is_contiguous(mbuf) == 0 with mbuf->next != NULL fixes the issue.

The problem has been observed while trying to use TRex traffic generator with VPP [4].

[1] http://code.dpdk.org/dpdk/v21.05/source/drivers/net/memif/rte_eth_memif.c#L660

[2] http://code.dpdk.org/dpdk/v21.05/source/drivers/net/memif/rte_eth_memif.c#L717

[3] http://code.dpdk.org/dpdk/v21.05/source/lib/mbuf/rte_mbuf.h#L1687

[4] https://github.com/cisco-system-traffic-generator/trex-core/issues/686

Comment 1 Ivan Shvedunov 2021-06-14 17:44:23 CEST

Downstream patch for TRex: https://github.com/cisco-system-traffic-generator/trex-core/pull/723

Comment 2 Olivier Matz 2021-07-27 14:48:50 CEST

Hi Ivan,

This patch makes sense to me, thanks for spotting it.

Can you please submit it on the dev mailing list?

+CC jgrajcia@cisco.com