[PATCH v4] net/af_xdp: AF_XDP PMD CNI Integration

[Ferruh Yigit]

On 2/10/2023 3:38 PM, Koikkara Reeny, Shibin wrote:
> 
> 
>> -----Original Message-----
>> From: Ferruh Yigit <ferruh.yigit at amd.com>
>> Sent: Friday, February 10, 2023 1:04 PM
>> To: Koikkara Reeny, Shibin <shibin.koikkara.reeny at intel.com>;
>> dev at dpdk.org; Zhang, Qi Z <qi.z.zhang at intel.com>; Burakov, Anatoly
>> <anatoly.burakov at intel.com>; Richardson, Bruce
>> <bruce.richardson at intel.com>; Mcnamara, John
>> <john.mcnamara at intel.com>
>> Cc: Loftus, Ciara <ciara.loftus at intel.com>
>> Subject: Re: [PATCH v4] net/af_xdp: AF_XDP PMD CNI Integration
>>
>> On 2/9/2023 12:05 PM, Shibin Koikkara Reeny wrote:
>>> Integrate support for the AF_XDP CNI and device plugin [1] so that the
>>> DPDK AF_XDP PMD can work in an unprivileged container environment.
>>> Part of the AF_XDP PMD initialization process involves loading an eBPF
>>> program onto the given netdev. This operation requires privileges,
>>> which prevents the PMD from being able to work in an unprivileged
>>> container (without root access). The plugin CNI handles the program
>>> loading. CNI open Unix Domain Socket (UDS) and waits listening for a
>>> client to make requests over that UDS. The client(DPDK) connects and a
>>> "handshake" occurs, then the File Descriptor which points to the
>>> XSKMAP associated with the loaded eBPF program is handed over to the
>>> client. The client can then proceed with creating an AF_XDP socket and
>>> inserting the socket into the XSKMAP pointed to by the FD received on
>>> the UDS.
>>>
>>> A new vdev arg "use_cni" is created to indicate user wishes to run the
>>> PMD in unprivileged mode and to receive the XSKMAP FD from the CNI.
>>> When this flag is set, the XSK_LIBBPF_FLAGS__INHIBIT_PROG_LOAD libbpf
>>> flag should be used when creating the socket, which tells libbpf not
>>> to load the default libbpf program on the netdev. We tell libbpf not
>>> to do this because the loading is handled by the CNI in this scenario.
>>>
>>> Patch include howto doc explain how to configure AF_XDP CNI to working
>>> with DPDK.
>>>
>>> [1]: https://github.com/intel/afxdp-plugins-for-kubernetes
>>>
>>> Signed-off-by: Shibin Koikkara Reeny <shibin.koikkara.reeny at intel.com>
>>
>>
>> Is Anatoly's tested-by tag still valid with this version?
> 
> Yes it is still valid.
> 
>>
>> <...>
>>
>>> @@ -1413,7 +1678,23 @@ xsk_configure(struct pmd_internals *internals,
>> struct pkt_rx_queue *rxq,
>>>  		}
>>>  	}
>>>
>>> -	if (rxq->busy_budget) {
>>> +	if (internals->use_cni) {
>>> +		int err, fd, map_fd;
>>> +
>>> +		/* get socket fd from CNI plugin */
>>> +		map_fd = get_cni_fd(internals->if_name);
>>> +		if (map_fd < 0) {
>>> +			AF_XDP_LOG(ERR, "Failed to receive CNI plugin
>> fd\n");
>>> +			goto out_xsk;
>>> +		}
>>> +		/* get socket fd */
>>> +		fd = xsk_socket__fd(rxq->xsk);
>>> +		err = bpf_map_update_elem(map_fd, &rxq-
>>> xsk_queue_idx, &fd, 0);
>>> +		if (err) {
>>> +			AF_XDP_LOG(ERR, "Failed to insert unprivileged xsk
>> in map.\n");
>>> +			goto out_xsk;
>>> +		}
>>> +	} else if (rxq->busy_budget) {
>>
>>
>> 'use_cni' argument is added as if-else, this result 'use_cni' parameter
>> automatically makes 'busy_budget' argument ineffective, is this intentional?
>> If so can you please describe why?
>> And can you please document this in the driver documentation that 'use_cni'
>> and 'busy_budget' paramters are mutually exclusive.
>> May be this condition can be checked and an error message sent in runtime,
>> not sure.
>>
> 
> When we use "use_cni" option inorder to configure the busy_budget we need to send the request to the CNI plugin
> and CNI plugin will configure the busy_poll. As the dpdk is running inside a container with limited permissions.
> 
>>
>> Similarly, another parameter check above this (not visible in this patch),
>> xdp_prog (custom_prog_configured) is calling same APIs
>> (bpf_map_update_elem()), if both paramters are provided, 'use_cni' will
>> overwrite previous one, is this intentional?
>> Are 'use_cni' & 'xdp_prog' paramters mutually exclusive?
> 
> When we use "use_cni" we don't have the permission to load the xdp_prog. As our privileges are limited inside the container.
> CNI plugin handle the loading of the program.


Yes, but what happens if user provides 'xdp_prog' parameter?

>>
>>
>> Overall is the combination of 'use_cni' paramter with other parameters
>> tested?
> 
> We have tested the communication with CNI plugin which load the program and traffic flow.
>  

I got that, but is the combination of 'use_cni' parameter with other
parameters tested?

Like what happens if user provides both 'xdp_prog' & 'use_cni'?
There is no documentation for this condition or there is no check in the
code that can provide some log message to user.

>>
>>
>>>  		ret = configure_preferred_busy_poll(rxq);
>>>  		if (ret) {
>>>  			AF_XDP_LOG(ERR, "Failed configure busy
>> polling.\n"); @@ -1584,6
>>> +1865,27 @@ static const struct eth_dev_ops ops = {
>>>  	.get_monitor_addr = eth_get_monitor_addr,  };
>>>
>>> +/* CNI option works in unprivileged container environment
>>> + * and ethernet device functionality will be reduced. So
>>> + * additional customiszed eth_dev_ops struct is needed
>>> + * for cni. Promiscuous enable and disable functionality
>>> + * is removed.
>>
>>
>> Why promiscuous enable and disable functionality can't be used with
>> 'use_cni'?
> 
> When we use "use_cni" we are running dpdk_testpmd inside a docker and inside the docker we have only 
> limited permissions only ie the reason I have written it as "unprivileged container environment"
> it the comment.
>>
>> Can you please document the limitation in the driver document, also if
>> possible briefly mention reason of the limitation?
> 
> In the documentation as prerequisites we have added :
> +* The Pod should have enabled the capabilities ``CAP_NET_RAW`` and ``CAP_BPF``
> +  for AF_XDP along with support for hugepages.
> 
> In the Background:
> +The standard `AF_XDP PMD`_ initialization process involves loading an eBPF program
> +onto the kernel netdev to be used by the PMD. This operation requires root or
> +escalated Linux privileges and thus prevents the PMD from working in an
> +unprivileged container. The AF_XDP CNI plugin handles this situation by
> +providing a device plugin that performs the program loading.
> 
> If you think we need to add more please let me know.
> 

Hi Shibin,

Thanks for the update.

I think it would be good to update driver documentation,
'doc/guides/nics/af_xdp.rst', and update where 'use_cni' parameter
documented with following additional information:

- When 'use_cni' parameter is used, 'busy_budget' parameter is not valid
and has no impact
- When 'use_cni' parameter is used, 'xdp_prog' parameter is not valid
and ? (what happens when provided)
- enable and disable promiscuous mode is not supported, and describe
briefly why (I know code has comment for it but less put it in
documentation too).