help with virtio_port

help with virtio_port

[Igor de Paula]

[-- Attachment #1: Type: text/plain, Size: 1338 bytes --]

Hi,
I am running the DPDK version: 21.08.0 and Ubuntu 20.04.3 LTS.
I have an application that uses KNI to interface with the kernel.
I want to replace it with virtio_user ports as KNI will be deprecated in
the future.
Most of the functionality I am able to replace but there is one thing I am
struggling with.
In KNI we can add functions that will be called in case the network stack
makes a request. The following code shows this:
struct rte_kni *kni;
        struct rte_kni_conf conf;
        struct rte_kni_ops ops;
        struct rte_eth_dev_info dev_info;
        int ret;
        /* Clear conf at first */
        memset(&conf, 0, sizeof(conf));
        conf.core_id = 0;
        memset(&ops, 0, sizeof(ops));
        ops.port_id = ppo->id;
        ops.config_promiscusity = ippe_ppo_set_kni_promiscuous_mode;
        ops.change_mtu = ippe_ppo_set_kni_mtu;
        ops.config_network_if = ippe_ppo_set_kni_interface;
        ops.config_mac_address = ippe_ppo_set_kni_mac_address;
        kni = rte_kni_alloc(pktmbuf_pool[0], &conf, &ops);


And there is a handle_request function supplied by KNI that calls these
functions when need be,
I haven't found any documentation on how to replace this functionality. I
am no expert in how to set up and interact with the kernel stack, Some help
on how to achieve this would be appreciated.

[-- Attachment #2: Type: text/html, Size: 1519 bytes --]

Re: help with virtio_port

[Stephen Hemminger]

On Tue, 23 May 2023 16:46:24 +0100
Igor de Paula <igordptx@gmail.com> wrote:

> Hi,
> I am running the DPDK version: 21.08.0 and Ubuntu 20.04.3 LTS.
> I have an application that uses KNI to interface with the kernel.
> I want to replace it with virtio_user ports as KNI will be deprecated in
> the future.
> Most of the functionality I am able to replace but there is one thing I am
> struggling with.
> In KNI we can add functions that will be called in case the network stack
> makes a request. The following code shows this:
> struct rte_kni *kni;
>         struct rte_kni_conf conf;
>         struct rte_kni_ops ops;
>         struct rte_eth_dev_info dev_info;
>         int ret;
>         /* Clear conf at first */
>         memset(&conf, 0, sizeof(conf));
>         conf.core_id = 0;
>         memset(&ops, 0, sizeof(ops));
>         ops.port_id = ppo->id;
>         ops.config_promiscusity = ippe_ppo_set_kni_promiscuous_mode;
>         ops.change_mtu = ippe_ppo_set_kni_mtu;
>         ops.config_network_if = ippe_ppo_set_kni_interface;
>         ops.config_mac_address = ippe_ppo_set_kni_mac_address;
>         kni = rte_kni_alloc(pktmbuf_pool[0], &conf, &ops);
> 
> 
> And there is a handle_request function supplied by KNI that calls these
> functions when need be,
> I haven't found any documentation on how to replace this functionality. I
> am no expert in how to set up and interact with the kernel stack, Some help
> on how to achieve this would be appreciated.

If you want to handle changes to kernel network device, then you
will have to build a netlink listener that monitors these changes.

Re: help with virtio_port

[Igor de Paula]

[-- Attachment #1: Type: text/plain, Size: 3600 bytes --]

Sorry I replied privately. I am replying here again.
WIth the KNI, I have a listener listener as follows:
static void *kni_monitor_system_networking(void *data)
{
    struct mnl_socket *nl;
    char buf[MNL_SOCKET_BUFFER_SIZE];
    int ret;

    nl = mnl_socket_open(NETLINK_ROUTE);
    if (nl == NULL)
    {
        perror("mnl_socket_open");
        exit(EXIT_FAILURE);
    }

    if (mnl_socket_bind(nl, RTMGRP_LINK | RTMGRP_IPV4_IFADDR,
MNL_SOCKET_AUTOPID) < 0)
    {
        perror("mnl_socket_bind");
        exit(EXIT_FAILURE);
    }

    ret = mnl_socket_recvfrom(nl, buf, sizeof(buf));
    while (ret > 0)
    {
        ret = mnl_cb_run(buf, ret, 0, 0, data_cb, NULL);
        if (ret <= 0)
            break;
        ret = mnl_socket_recvfrom(nl, buf, sizeof(buf));
    }
    if (ret == -1)
    {
        perror("error");
        exit(EXIT_FAILURE);
    }

    mnl_socket_close(nl);

    return 0;
}

But also I have a thread looping over rte_kni_handle_request:
while (1)
    {
        for (t = kni_tracker; t; t = t->next)
        {
            struct rte_kni *kni = (  struct rte_kni *  )t->kni;

            if (ppo->kni)
            {
                rte_kni_handle_request(kni);
        }

        usleep(10000);
    }

The rte_kni_handle_request calls the kni_config_promiscusity.
So from what you are saying, I can replace this call with another monitor
on the netlink ports (the work will be on finding the correct flags and to
implement the callback I guess)?
I from what I see,  mnl_socket_bind(nl, RTMGRP_LINK | RTMGRP_IPV4_IFADDR,
MNL_SOCKET_AUTOPID) < 0) this linke should get also requests that are link
related, but it doesn't seem to get any packets when I use tcpdump on the
interface.

On Tue, May 23, 2023 at 5:23 PM Stephen Hemminger <
stephen@networkplumber.org> wrote:

> On Tue, 23 May 2023 16:46:24 +0100
> Igor de Paula <igordptx@gmail.com> wrote:
>
> > Hi,
> > I am running the DPDK version: 21.08.0 and Ubuntu 20.04.3 LTS.
> > I have an application that uses KNI to interface with the kernel.
> > I want to replace it with virtio_user ports as KNI will be deprecated in
> > the future.
> > Most of the functionality I am able to replace but there is one thing I
> am
> > struggling with.
> > In KNI we can add functions that will be called in case the network stack
> > makes a request. The following code shows this:
> > struct rte_kni *kni;
> >         struct rte_kni_conf conf;
> >         struct rte_kni_ops ops;
> >         struct rte_eth_dev_info dev_info;
> >         int ret;
> >         /* Clear conf at first */
> >         memset(&conf, 0, sizeof(conf));
> >         conf.core_id = 0;
> >         memset(&ops, 0, sizeof(ops));
> >         ops.port_id = ppo->id;
> >         ops.config_promiscusity = ippe_ppo_set_kni_promiscuous_mode;
> >         ops.change_mtu = ippe_ppo_set_kni_mtu;
> >         ops.config_network_if = ippe_ppo_set_kni_interface;
> >         ops.config_mac_address = ippe_ppo_set_kni_mac_address;
> >         kni = rte_kni_alloc(pktmbuf_pool[0], &conf, &ops);
> >
> >
> > And there is a handle_request function supplied by KNI that calls these
> > functions when need be,
> > I haven't found any documentation on how to replace this functionality. I
> > am no expert in how to set up and interact with the kernel stack, Some
> help
> > on how to achieve this would be appreciated.
>
> If you want to handle changes to kernel network device, then you
> will have to build a netlink listener that monitors these changes.
>

[-- Attachment #2: Type: text/html, Size: 4492 bytes --]

Re: help with virtio_port

[Igor de Paula]

Thanks!
It's weird because on the KNI version, I am listening on the port and
waiting for a link message. If I run tcpdump there is no packet coming
on that port but the rte_handle_kni_request doe's deal with it.
When I try to listen on the port, without KNI, when I am using
virtio_user I do get a packet when I use tcpdump.
I guess when KNI initialized it takes the traffic associated with the
functions it supports.
Can someone confirm that this is actually the case?
Many thanks for the help so far

On Tue, May 23, 2023 at 5:23 PM Stephen Hemminger
<stephen@networkplumber.org> wrote:
>
> On Tue, 23 May 2023 16:46:24 +0100
> Igor de Paula <igordptx@gmail.com> wrote:
>
> > Hi,
> > I am running the DPDK version: 21.08.0 and Ubuntu 20.04.3 LTS.
> > I have an application that uses KNI to interface with the kernel.
> > I want to replace it with virtio_user ports as KNI will be deprecated in
> > the future.
> > Most of the functionality I am able to replace but there is one thing I am
> > struggling with.
> > In KNI we can add functions that will be called in case the network stack
> > makes a request. The following code shows this:
> > struct rte_kni *kni;
> >         struct rte_kni_conf conf;
> >         struct rte_kni_ops ops;
> >         struct rte_eth_dev_info dev_info;
> >         int ret;
> >         /* Clear conf at first */
> >         memset(&conf, 0, sizeof(conf));
> >         conf.core_id = 0;
> >         memset(&ops, 0, sizeof(ops));
> >         ops.port_id = ppo->id;
> >         ops.config_promiscusity = ippe_ppo_set_kni_promiscuous_mode;
> >         ops.change_mtu = ippe_ppo_set_kni_mtu;
> >         ops.config_network_if = ippe_ppo_set_kni_interface;
> >         ops.config_mac_address = ippe_ppo_set_kni_mac_address;
> >         kni = rte_kni_alloc(pktmbuf_pool[0], &conf, &ops);
> >
> >
> > And there is a handle_request function supplied by KNI that calls these
> > functions when need be,
> > I haven't found any documentation on how to replace this functionality. I
> > am no expert in how to set up and interact with the kernel stack, Some help
> > on how to achieve this would be appreciated.
>
> If you want to handle changes to kernel network device, then you
> will have to build a netlink listener that monitors these changes.