[v1] net/memif: fix segfault with large burst size

  There will be a segfault when Rx burst size is greater than
MAX_PKT_BURST of memif. Fix the issue by correcting the
wrong mbuf index in eth_memif_rx, which results in accessing
invalid memory address.

Bugzilla ID: 1273
Fixes: aa17df860891 ("net/memif: add a Rx fast path")
Cc: stable@dpdk.org

Signed-off-by: Joyce Kong <joyce.kong@arm.com>
Reviewed-by: Feifei Wang <feifei.wang2@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
---
 drivers/net/memif/rte_eth_memif.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
  

On 9/4/2023 8:10 AM, Joyce Kong wrote:
> There will be a segfault when Rx burst size is greater than
> MAX_PKT_BURST of memif. Fix the issue by correcting the
> wrong mbuf index in eth_memif_rx, which results in accessing
> invalid memory address.
> 
> Bugzilla ID: 1273
> Fixes: aa17df860891 ("net/memif: add a Rx fast path")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Joyce Kong <joyce.kong@arm.com>
> Reviewed-by: Feifei Wang <feifei.wang2@arm.com>
> Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
> ---
>  drivers/net/memif/rte_eth_memif.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/net/memif/rte_eth_memif.c b/drivers/net/memif/rte_eth_memif.c
> index 6a8ff5b4eb..f595656af5 100644
> --- a/drivers/net/memif/rte_eth_memif.c
> +++ b/drivers/net/memif/rte_eth_memif.c
> @@ -356,7 +356,7 @@ eth_memif_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
>  		rx_pkts = 0;
>  		pkts = nb_pkts < MAX_PKT_BURST ? nb_pkts : MAX_PKT_BURST;
>  		while (n_slots && rx_pkts < pkts) {
> -			mbuf_head = mbufs[n_rx_pkts];
> +			mbuf_head = mbufs[rx_pkts];
>  			mbuf = mbuf_head;
>  
>  next_slot1:

Hi Jakub,

Can you please look at this patch?

Thanks,
ferruh
  

On 9/4/2023 8:10 AM, Joyce Kong wrote:
> There will be a segfault when Rx burst size is greater than
> MAX_PKT_BURST of memif. Fix the issue by correcting the
> wrong mbuf index in eth_memif_rx, which results in accessing
> invalid memory address.
> 
> Bugzilla ID: 1273
> Fixes: aa17df860891 ("net/memif: add a Rx fast path")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Joyce Kong <joyce.kong@arm.com>
> Reviewed-by: Feifei Wang <feifei.wang2@arm.com>
> Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
> 

Hi Joyce, good catch.

Reviewed-by: Ferruh Yigit <ferruh.yigit@amd.com>

Applied to dpdk-next-net/main, thanks.



For record, if nb_pkts > MAX_PKT_BURST, memif buffer consumed in chunks
of MAX_PKT_BURST mbufs, next chunk consumption starts with 'goto
next_bulk' call.

For each chunk, MAX_PKT_BURST mbufs allocated and filled, they are
accessed by 'n_rx_pkts' index, but 'n_rx_pkts' is overall received mbuf
number, so it shouldn't be used as index for that chunk, but 'rx_pkts'
should be used which is reset at the beginning of the chunk processing.

For the first chunk using 'n_rx_pkts' or 'rx_pkts' are same, that
explains how issue lived till now, as commit log mentions issue can be
observed when nb_pkts > MAX_PKT_BURST.