Patchwork [v6] net: L2TPv3 transport

login
register
mail settings
Submitter anton.ivanov@kot-begemot.co.uk
Date March 26, 2014, 11:08 a.m.
Message ID <1395832121-858851-1-git-send-email-anton.ivanov@kot-begemot.co.uk>
Download mbox | patch
Permalink /patch/333892/
State New
Headers show

Comments

anton.ivanov@kot-begemot.co.uk - March 26, 2014, 11:08 a.m.
From: Anton Ivanov <antivano@cisco.com>

This transport allows to connect a QEMU nic to a static Ethernet
over L2TPv3 tunnel. The transport supports all options present
in the Linux kernel implementation. It allows QEMU to connect
to any Linux host running kernel 3.3+, most routers and network
devices as well as other QEMU instances.

Signed-off-by: Anton Ivanov <antivano@cisco.com>
---

Issues Addressed:

1. Correct RX/TX poll handling - address comments from Stephan 
(RX and TX poll changes were mixed up badly in previous version).

2. Separate RX queue (on top of recvmmsg buffer) processing 
from actual recvmmsg RX so it is readable and reusable. 

 net/Makefile.objs |    1 +
 net/clients.h     |    2 +
 net/l2tpv3.c      |  752 +++++++++++++++++++++++++++++++++++++++++++++++++++++
 net/net.c         |    3 +
 qapi-schema.json  |   61 ++++-
 qemu-options.hx   |   82 ++++++
 6 files changed, 900 insertions(+), 1 deletion(-)
 create mode 100644 net/l2tpv3.c
Stefan Hajnoczi - March 27, 2014, 12:30 p.m.
On Wed, Mar 26, 2014 at 11:08:41AM +0000, anton.ivanov@kot-begemot.co.uk wrote:
> +static ssize_t net_l2tpv3_receive_dgram_iov(NetClientState *nc,
> +                    const struct iovec *iov,
> +                    int iovcnt)
> +{
> +    NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
> +
> +    struct msghdr message;
> +    int ret;
> +
> +    if (iovcnt > MAX_L2TPV3_IOVCNT - 1) {
> +        error_report(
> +            "iovec too long %d > %d, change l2tpv3.h",
> +            iovcnt, MAX_L2TPV3_IOVCNT
> +        );
> +        return -1;
> +    }

The alternative is to linearize the buffer using iov_to_buf() and call
net_l2tpv3_receive_dgram().  Something like:

if (iovcnt > MAX_L2TPV3_IOVCNT - 1) {
    size_t size = iov_size(iov, iovcnt);
    uint8_t *buf;

    buf = g_malloc(size);
    iov_to_buf(iov, iovcnt, 0, buf, size);
    ret = net_l2tpv3_receive_dgram(nc, buf, size);
    g_free(buf);
    return ret;
}

> +    l2tpv3_form_header(s);
> +    memcpy(s->vec + 1, iov, iovcnt * sizeof(struct iovec));
> +    s->vec->iov_base = s->header_buf;
> +    s->vec->iov_len = s->offset;
> +    message.msg_name = s->dgram_dst;
> +    message.msg_namelen = s->dst_size;
> +    message.msg_iov = s->vec;
> +    message.msg_iovlen = iovcnt + 1;
> +    message.msg_control = NULL;
> +    message.msg_controllen = 0;
> +    message.msg_flags = 0;
> +    do {
> +        ret = sendmsg(s->fd, &message, 0);
> +    } while ((ret == -1) && (errno == EINTR));
> +    if (ret > 0) {
> +        ret -= s->offset;
> +    } else if (ret == 0) {
> +        /* belt and braces - should not occur on DGRAM
> +        * we should get an error and never a 0 send
> +        */

Space missing before '*'

> +        ret = iov_size(iov, iovcnt);
> +    } else {
> +        /* signal upper layer that socket buffer is full */
> +        ret = -errno;
> +        if (ret == -EAGAIN || ret == -ENOBUFS) {
> +            l2tpv3_write_poll(s, true);
> +            ret = 0;
> +        }
> +    }
> +    return ret;
> +}
> +
> +static ssize_t net_l2tpv3_receive_dgram(NetClientState *nc,
> +                    const uint8_t *buf,
> +                    size_t size)
> +{
> +    NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
> +
> +    struct iovec *vec;
> +    struct msghdr message;
> +    ssize_t ret = 0;
> +
> +    l2tpv3_form_header(s);
> +    vec = s->vec;
> +    vec->iov_base = s->header_buf;
> +    vec->iov_len = s->offset;
> +    vec++;
> +    vec->iov_base = (void *) buf;
> +    vec->iov_len = size;
> +    message.msg_name = s->dgram_dst;
> +    message.msg_namelen = s->dst_size;
> +    message.msg_iov = s->vec;
> +    message.msg_iovlen = 2;
> +    message.msg_control = NULL;
> +    message.msg_controllen = 0;
> +    message.msg_flags = 0;
> +    do {
> +        ret = sendmsg(s->fd, &message, 0);
> +    } while ((ret == -1) && (errno == EINTR));
> +    if (ret > 0) {
> +        ret -= s->offset;
> +    } else if (ret == 0) {
> +        /* belt and braces - should not occur on DGRAM
> +        * we should get an error and never a 0 send
> +        */
> +        ret = size;
> +    } else {
> +        ret = -errno;
> +        if (ret == -EAGAIN || ret == -ENOBUFS) {
> +            /* signal upper layer that socket buffer is full */
> +            l2tpv3_write_poll(s, true);
> +            ret = 0;
> +        }
> +    }
> +    return ret;
> +}

This function duplicates code, how about:

static ssize_t net_l2tpv3_receive_dgram(NetClientState *nc,
                                        const uint8_t *buf,
                                        size_t size)
{
    struct iovec iov = {
        .iov_base = buf,
        .iov_len  = size,
    };

    return net_l2tpv3_receive_dgram_iov(nc, &iov, 1);
}

> +static int l2tpv3_verify_header(NetL2TPV3State *s, uint8_t *buf)
> +{
> +
> +    uint32_t *session;
> +    uint64_t cookie;
> +
> +    if ((!s->udp) && (!s->ipv6)) {
> +        buf += sizeof(struct iphdr) /* fix for ipv4 raw */;
> +    }
> +
> +    /* we do not do a strict check for "data" packets as per
> +    * the RFC spec because the pure IP spec does not have
> +    * that anyway.
> +    */
> +
> +    if (s->cookie) {
> +        if (s->cookie_is_64) {
> +            cookie = ldq_be_p(buf + s->cookie_offset);
> +        } else {
> +            cookie = ldl_be_p(buf + s->cookie_offset);
> +        }
> +        if (cookie != s->rx_cookie) {
> +            if (!s->header_mismatch) {
> +                error_report("unknown cookie id\n");

error_report() messages should not have a newline ('\n') at the end.

> +            }
> +            return -1;
> +        }
> +    }
> +    session = (uint32_t *) (buf + s->session_offset);
> +    if (ldl_be_p(session) != s->rx_session) {
> +        if (!s->header_mismatch) {
> +            error_report("session mismatch");
> +        }
> +        return -1;
> +    }
> +    return 0;
> +}
> +
> +static void net_l2tpv3_process_queue(NetL2TPV3State *s)
> +{
> +    int size = 0;
> +    struct iovec *vec;
> +    bool bad_read;
> +    int data_size;
> +    struct mmsghdr *msgvec;
> +
> +    /* go into ring mode only if there is a "pending" tail */
> +    if (s->queue_depth > 0 && (!s->delivering)) {
> +        do {
> +            msgvec = s->msgvec + s->queue_tail;
> +            if (msgvec->msg_len > 0) {
> +                data_size = msgvec->msg_len - s->header_size;

header_size is never assigned to.

> +                vec = msgvec->msg_hdr.msg_iov;
> +                if ((data_size > 0) &&
> +                    (l2tpv3_verify_header(s, vec->iov_base) == 0)) {
> +                    vec++;
> +                    /* we do not use the "normal" send and send async
> +                     * functions here because we have our own buffer -
> +                     * the message vector. If we use the "usual" ones
> +                     * we will end up double-buffering.
> +                     */
> +                    s->delivering = true;

What is s->delivering guarding against?  This function is only called
from the s->fd read handler function, so I don't see a reentrant code
path.

> +                    /* deliver directly to peer bypassing queueing and
> +                     * buffering
> +                     */
> +                    size = qemu_deliver_packet(
> +                            &s->nc,
> +                            QEMU_NET_PACKET_FLAG_NONE,
> +                            vec->iov_base,
> +                            data_size,
> +                            s->nc.peer
> +                        );

There is no callback when the peer re-enables receive.  In other words,
packets will be stuck in s->msgvec until a new packet arrives and
net_l2tpv3_send() is invoked.

I think you need to modify qemu_flush_queued_packets() to invoke a new
callback.  You could convert the existing code:

if (nc->peer && nc->peer->info->type == NET_CLIENT_OPTIONS_KIND_HUBPORT) {
    if (net_hub_flush(nc->peer)) {
        qemu_notify_event();
    }
}

to:

if (nc->peer && nc->peer->info->peer_flushed) {
    if (nc->peer->info->peer_flushed(nc->peer)) {
        qemu_notify_event();
    }
}

and move the hub flushing code into net/hub.c.

Then you could also implement ->peer_flushed() in l2tpv3.c so that it
calls net_l2tpv3_process_queue() and returns true if packets were
delivered.

Please introduce ->peer_flushed() in a separate patch.

Or you could keep it simple and go back to using qemu_send_packet()
without trying to avoid duplicate buffering.  (You can always send
patches later that eliminate the buffering.)
Eric Blake - March 27, 2014, 1:12 p.m.
On 03/26/2014 05:08 AM, anton.ivanov@kot-begemot.co.uk wrote:
> From: Anton Ivanov <antivano@cisco.com>
> 
> This transport allows to connect a QEMU nic to a static Ethernet
> over L2TPv3 tunnel. The transport supports all options present
> in the Linux kernel implementation. It allows QEMU to connect
> to any Linux host running kernel 3.3+, most routers and network
> devices as well as other QEMU instances.
> 
> Signed-off-by: Anton Ivanov <antivano@cisco.com>
> ---

>  #
>  # Since 1.2
> -##
> +#
> +# Added in 2.1 - l2tpv3
> +#
>  { 'union': 'NetClientOptions',

Minor nit - you lost the ## end-of-comment line.  Also, see how QKeyCode
mentioned added values:

# Since: 1.3.0
#
# 'unmapped' and 'pause' since 2.0
##

We haven't done this sort of addition much to date, but it can't hurt to
use a consistent style when doing so.
Anton Ivanov (antivano) - March 27, 2014, 3:53 p.m.
On 27/03/14 12:30, Stefan Hajnoczi wrote:
> On Wed, Mar 26, 2014 at 11:08:41AM +0000, anton.ivanov@kot-begemot.co.uk wrote:
>> +static ssize_t net_l2tpv3_receive_dgram_iov(NetClientState *nc,
>> +                    const struct iovec *iov,
>> +                    int iovcnt)
>> +{
>> +    NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
>> +
>> +    struct msghdr message;
>> +    int ret;
>> +
>> +    if (iovcnt > MAX_L2TPV3_IOVCNT - 1) {
>> +        error_report(
>> +            "iovec too long %d > %d, change l2tpv3.h",
>> +            iovcnt, MAX_L2TPV3_IOVCNT
>> +        );
>> +        return -1;
>> +    }
> The alternative is to linearize the buffer using iov_to_buf() and call
> net_l2tpv3_receive_dgram().  Something like:
>
> if (iovcnt > MAX_L2TPV3_IOVCNT - 1) {
>     size_t size = iov_size(iov, iovcnt);
>     uint8_t *buf;
>
>     buf = g_malloc(size);
>     iov_to_buf(iov, iovcnt, 0, buf, size);
>     ret = net_l2tpv3_receive_dgram(nc, buf, size);
>     g_free(buf);
>     return ret;
> }

iov_to_buf is a memcpy of the data and a malloc down the call chain.
That is quite a significant cost compared to just shifting the iov a bit
to add an element for header. I tried that early on and it was
introducing a noticeable penalty.

If I am to merge it I would actually merge it the other way around -
express both net_l2tpv3_receive_dgram and net_l2tpv3_receive_dgram_iov
via an iov based backend. That will allow to add sendmmsg one day as the
data structures will be the same in all cases.

>
>> +    l2tpv3_form_header(s);
>> +    memcpy(s->vec + 1, iov, iovcnt * sizeof(struct iovec));
>> +    s->vec->iov_base = s->header_buf;
>> +    s->vec->iov_len = s->offset;
>> +    message.msg_name = s->dgram_dst;
>> +    message.msg_namelen = s->dst_size;
>> +    message.msg_iov = s->vec;
>> +    message.msg_iovlen = iovcnt + 1;
>> +    message.msg_control = NULL;
>> +    message.msg_controllen = 0;
>> +    message.msg_flags = 0;
>> +    do {
>> +        ret = sendmsg(s->fd, &message, 0);
>> +    } while ((ret == -1) && (errno == EINTR));
>> +    if (ret > 0) {
>> +        ret -= s->offset;
>> +    } else if (ret == 0) {
>> +        /* belt and braces - should not occur on DGRAM
>> +        * we should get an error and never a 0 send
>> +        */
> Space missing before '*'

Will fix.

>
>> +        ret = iov_size(iov, iovcnt);
>> +    } else {
>> +        /* signal upper layer that socket buffer is full */
>> +        ret = -errno;
>> +        if (ret == -EAGAIN || ret == -ENOBUFS) {
>> +            l2tpv3_write_poll(s, true);
>> +            ret = 0;
>> +        }
>> +    }
>> +    return ret;
>> +}
>> +
>> +static ssize_t net_l2tpv3_receive_dgram(NetClientState *nc,
>> +                    const uint8_t *buf,
>> +                    size_t size)
>> +{
>> +    NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
>> +
>> +    struct iovec *vec;
>> +    struct msghdr message;
>> +    ssize_t ret = 0;
>> +
>> +    l2tpv3_form_header(s);
>> +    vec = s->vec;
>> +    vec->iov_base = s->header_buf;
>> +    vec->iov_len = s->offset;
>> +    vec++;
>> +    vec->iov_base = (void *) buf;
>> +    vec->iov_len = size;
>> +    message.msg_name = s->dgram_dst;
>> +    message.msg_namelen = s->dst_size;
>> +    message.msg_iov = s->vec;
>> +    message.msg_iovlen = 2;
>> +    message.msg_control = NULL;
>> +    message.msg_controllen = 0;
>> +    message.msg_flags = 0;
>> +    do {
>> +        ret = sendmsg(s->fd, &message, 0);
>> +    } while ((ret == -1) && (errno == EINTR));
>> +    if (ret > 0) {
>> +        ret -= s->offset;
>> +    } else if (ret == 0) {
>> +        /* belt and braces - should not occur on DGRAM
>> +        * we should get an error and never a 0 send
>> +        */
>> +        ret = size;
>> +    } else {
>> +        ret = -errno;
>> +        if (ret == -EAGAIN || ret == -ENOBUFS) {
>> +            /* signal upper layer that socket buffer is full */
>> +            l2tpv3_write_poll(s, true);
>> +            ret = 0;
>> +        }
>> +    }
>> +    return ret;
>> +}
> This function duplicates code, how about:
>
> static ssize_t net_l2tpv3_receive_dgram(NetClientState *nc,
>                                         const uint8_t *buf,
>                                         size_t size)
> {
>     struct iovec iov = {
>         .iov_base = buf,
>         .iov_len  = size,
>     };
>
>     return net_l2tpv3_receive_dgram_iov(nc, &iov, 1);
> }

Agree.

>
>> +static int l2tpv3_verify_header(NetL2TPV3State *s, uint8_t *buf)
>> +{
>> +
>> +    uint32_t *session;
>> +    uint64_t cookie;
>> +
>> +    if ((!s->udp) && (!s->ipv6)) {
>> +        buf += sizeof(struct iphdr) /* fix for ipv4 raw */;
>> +    }
>> +
>> +    /* we do not do a strict check for "data" packets as per
>> +    * the RFC spec because the pure IP spec does not have
>> +    * that anyway.
>> +    */
>> +
>> +    if (s->cookie) {
>> +        if (s->cookie_is_64) {
>> +            cookie = ldq_be_p(buf + s->cookie_offset);
>> +        } else {
>> +            cookie = ldl_be_p(buf + s->cookie_offset);
>> +        }
>> +        if (cookie != s->rx_cookie) {
>> +            if (!s->header_mismatch) {
>> +                error_report("unknown cookie id\n");
> error_report() messages should not have a newline ('\n') at the end.
>
>> +            }
>> +            return -1;
>> +        }
>> +    }
>> +    session = (uint32_t *) (buf + s->session_offset);
>> +    if (ldl_be_p(session) != s->rx_session) {
>> +        if (!s->header_mismatch) {
>> +            error_report("session mismatch");
>> +        }
>> +        return -1;
>> +    }
>> +    return 0;
>> +}
>> +
>> +static void net_l2tpv3_process_queue(NetL2TPV3State *s)
>> +{
>> +    int size = 0;
>> +    struct iovec *vec;
>> +    bool bad_read;
>> +    int data_size;
>> +    struct mmsghdr *msgvec;
>> +
>> +    /* go into ring mode only if there is a "pending" tail */
>> +    if (s->queue_depth > 0 && (!s->delivering)) {
>> +        do {
>> +            msgvec = s->msgvec + s->queue_tail;
>> +            if (msgvec->msg_len > 0) {
>> +                data_size = msgvec->msg_len - s->header_size;
> header_size is never assigned to.

It is - in init. It is used to divorce the "read header size" from
"l2tpv3 header size" as needed by the ipv4 socket API. Using it allows
to remove the check if we are reading from ipv4 raw sockets in most (but
not all) places.

>
>> +                vec = msgvec->msg_hdr.msg_iov;
>> +                if ((data_size > 0) &&
>> +                    (l2tpv3_verify_header(s, vec->iov_base) == 0)) {
>> +                    vec++;
>> +                    /* we do not use the "normal" send and send async
>> +                     * functions here because we have our own buffer -
>> +                     * the message vector. If we use the "usual" ones
>> +                     * we will end up double-buffering.
>> +                     */
>> +                    s->delivering = true;
> What is s->delivering guarding against?  This function is only called
> from the s->fd read handler function, so I don't see a reentrant code
> path.

Cut-n-paste from the queue functions without doing full analysis of what
is it guarding against there.

>
>> +                    /* deliver directly to peer bypassing queueing and
>> +                     * buffering
>> +                     */
>> +                    size = qemu_deliver_packet(
>> +                            &s->nc,
>> +                            QEMU_NET_PACKET_FLAG_NONE,
>> +                            vec->iov_base,
>> +                            data_size,
>> +                            s->nc.peer
>> +                        );
> There is no callback when the peer re-enables receive.  In other words,
> packets will be stuck in s->msgvec until a new packet arrives and
> net_l2tpv3_send() is invoked.
>
> I think you need to modify qemu_flush_queued_packets() to invoke a new
> callback.  You could convert the existing code:

I had the same thoughts, just did not want to touch it just yet.

Something along the lines would have been necessary for sendmmsg too as
there you would want to work off a pre-configured queue in the first place.

>
> if (nc->peer && nc->peer->info->type == NET_CLIENT_OPTIONS_KIND_HUBPORT) {
>     if (net_hub_flush(nc->peer)) {
>         qemu_notify_event();
>     }
> }
>
> to:
>
> if (nc->peer && nc->peer->info->peer_flushed) {
>     if (nc->peer->info->peer_flushed(nc->peer)) {
>         qemu_notify_event();
>     }
> }
>
> and move the hub flushing code into net/hub.c.
>
> Then you could also implement ->peer_flushed() in l2tpv3.c so that it
> calls net_l2tpv3_process_queue() and returns true if packets were
> delivered.
>
> Please introduce ->peer_flushed() in a separate patch.

Understood.

>
> Or you could keep it simple and go back to using qemu_send_packet()
> without trying to avoid duplicate buffering.  (You can always send
> patches later that eliminate the buffering.)

I would probably go back to that for submission so we can do it in stages.

I will branch out the zero copy work and relevant additions to queue.c
for submission at a later date along with zero copy versions of gre,
raw, as well as a revised version of l2tpv3.c

A.
Stefan Hajnoczi - March 27, 2014, 6:25 p.m.
On Thu, Mar 27, 2014 at 4:53 PM, Anton Ivanov (antivano)
<antivano@cisco.com> wrote:
> On 27/03/14 12:30, Stefan Hajnoczi wrote:
>> On Wed, Mar 26, 2014 at 11:08:41AM +0000, anton.ivanov@kot-begemot.co.uk wrote:
>>> +static ssize_t net_l2tpv3_receive_dgram_iov(NetClientState *nc,
>>> +                    const struct iovec *iov,
>>> +                    int iovcnt)
>>> +{
>>> +    NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
>>> +
>>> +    struct msghdr message;
>>> +    int ret;
>>> +
>>> +    if (iovcnt > MAX_L2TPV3_IOVCNT - 1) {
>>> +        error_report(
>>> +            "iovec too long %d > %d, change l2tpv3.h",
>>> +            iovcnt, MAX_L2TPV3_IOVCNT
>>> +        );
>>> +        return -1;
>>> +    }
>> The alternative is to linearize the buffer using iov_to_buf() and call
>> net_l2tpv3_receive_dgram().  Something like:
>>
>> if (iovcnt > MAX_L2TPV3_IOVCNT - 1) {
>>     size_t size = iov_size(iov, iovcnt);
>>     uint8_t *buf;
>>
>>     buf = g_malloc(size);
>>     iov_to_buf(iov, iovcnt, 0, buf, size);
>>     ret = net_l2tpv3_receive_dgram(nc, buf, size);
>>     g_free(buf);
>>     return ret;
>> }
>
> iov_to_buf is a memcpy of the data and a malloc down the call chain.
> That is quite a significant cost compared to just shifting the iov a bit
> to add an element for header. I tried that early on and it was
> introducing a noticeable penalty.
>
> If I am to merge it I would actually merge it the other way around -
> express both net_l2tpv3_receive_dgram and net_l2tpv3_receive_dgram_iov
> via an iov based backend. That will allow to add sendmmsg one day as the
> data structures will be the same in all cases.

I agree with you that the iov code path should be used.  The context
for my comment was "if (iovcnt > MAX_L2TPV3_IOVCNT - 1) {".  I'm just
suggesting that we fall back to linearizing the iovector if it has too
many elements, instead of erroring out.

>>> +static int l2tpv3_verify_header(NetL2TPV3State *s, uint8_t *buf)
>>> +{
>>> +
>>> +    uint32_t *session;
>>> +    uint64_t cookie;
>>> +
>>> +    if ((!s->udp) && (!s->ipv6)) {
>>> +        buf += sizeof(struct iphdr) /* fix for ipv4 raw */;
>>> +    }
>>> +
>>> +    /* we do not do a strict check for "data" packets as per
>>> +    * the RFC spec because the pure IP spec does not have
>>> +    * that anyway.
>>> +    */
>>> +
>>> +    if (s->cookie) {
>>> +        if (s->cookie_is_64) {
>>> +            cookie = ldq_be_p(buf + s->cookie_offset);
>>> +        } else {
>>> +            cookie = ldl_be_p(buf + s->cookie_offset);
>>> +        }
>>> +        if (cookie != s->rx_cookie) {
>>> +            if (!s->header_mismatch) {
>>> +                error_report("unknown cookie id\n");
>> error_report() messages should not have a newline ('\n') at the end.
>>
>>> +            }
>>> +            return -1;
>>> +        }
>>> +    }
>>> +    session = (uint32_t *) (buf + s->session_offset);
>>> +    if (ldl_be_p(session) != s->rx_session) {
>>> +        if (!s->header_mismatch) {
>>> +            error_report("session mismatch");
>>> +        }
>>> +        return -1;
>>> +    }
>>> +    return 0;
>>> +}
>>> +
>>> +static void net_l2tpv3_process_queue(NetL2TPV3State *s)
>>> +{
>>> +    int size = 0;
>>> +    struct iovec *vec;
>>> +    bool bad_read;
>>> +    int data_size;
>>> +    struct mmsghdr *msgvec;
>>> +
>>> +    /* go into ring mode only if there is a "pending" tail */
>>> +    if (s->queue_depth > 0 && (!s->delivering)) {
>>> +        do {
>>> +            msgvec = s->msgvec + s->queue_tail;
>>> +            if (msgvec->msg_len > 0) {
>>> +                data_size = msgvec->msg_len - s->header_size;
>> header_size is never assigned to.
>
> It is - in init. It is used to divorce the "read header size" from
> "l2tpv3 header size" as needed by the ipv4 socket API. Using it allows
> to remove the check if we are reading from ipv4 raw sockets in most (but
> not all) places.

I'm not seeing that in this version of the patch, please check again
and let me know which line number it is on.

>>> +                vec = msgvec->msg_hdr.msg_iov;
>>> +                if ((data_size > 0) &&
>>> +                    (l2tpv3_verify_header(s, vec->iov_base) == 0)) {
>>> +                    vec++;
>>> +                    /* we do not use the "normal" send and send async
>>> +                     * functions here because we have our own buffer -
>>> +                     * the message vector. If we use the "usual" ones
>>> +                     * we will end up double-buffering.
>>> +                     */
>>> +                    s->delivering = true;
>> What is s->delivering guarding against?  This function is only called
>> from the s->fd read handler function, so I don't see a reentrant code
>> path.
>
> Cut-n-paste from the queue functions without doing full analysis of what
> is it guarding against there.

Okay, the problem in the generic net code is that some net clients
(like slirp) will send packets back to their peer from their receive()
function.  For example, imagine a DHCP request packet from the guest
being sent to the slirp net client.  It will send the DHCP reply right
away while we're still in the send and receive functions.  This
creates reentrancy and not all code handles it so we just queue the
packets instead.

But in this case there is no reentrancy since only the fd handler
function can call net_l2tpv3_process_queue().

>>
>>> +                    /* deliver directly to peer bypassing queueing and
>>> +                     * buffering
>>> +                     */
>>> +                    size = qemu_deliver_packet(
>>> +                            &s->nc,
>>> +                            QEMU_NET_PACKET_FLAG_NONE,
>>> +                            vec->iov_base,
>>> +                            data_size,
>>> +                            s->nc.peer
>>> +                        );
>> There is no callback when the peer re-enables receive.  In other words,
>> packets will be stuck in s->msgvec until a new packet arrives and
>> net_l2tpv3_send() is invoked.
>>
>> I think you need to modify qemu_flush_queued_packets() to invoke a new
>> callback.  You could convert the existing code:
>
> I had the same thoughts, just did not want to touch it just yet.
>
> Something along the lines would have been necessary for sendmmsg too as
> there you would want to work off a pre-configured queue in the first place.
>
>>
>> if (nc->peer && nc->peer->info->type == NET_CLIENT_OPTIONS_KIND_HUBPORT) {
>>     if (net_hub_flush(nc->peer)) {
>>         qemu_notify_event();
>>     }
>> }
>>
>> to:
>>
>> if (nc->peer && nc->peer->info->peer_flushed) {
>>     if (nc->peer->info->peer_flushed(nc->peer)) {
>>         qemu_notify_event();
>>     }
>> }
>>
>> and move the hub flushing code into net/hub.c.
>>
>> Then you could also implement ->peer_flushed() in l2tpv3.c so that it
>> calls net_l2tpv3_process_queue() and returns true if packets were
>> delivered.
>>
>> Please introduce ->peer_flushed() in a separate patch.
>
> Understood.
>
>>
>> Or you could keep it simple and go back to using qemu_send_packet()
>> without trying to avoid duplicate buffering.  (You can always send
>> patches later that eliminate the buffering.)
>
> I would probably go back to that for submission so we can do it in stages.
>
> I will branch out the zero copy work and relevant additions to queue.c
> for submission at a later date along with zero copy versions of gre,
> raw, as well as a revised version of l2tpv3.c

Excellent, I was hoping for this, thanks!
anton.ivanov@kot-begemot.co.uk - March 27, 2014, 7:45 p.m.
On 27/03/14 18:25, Stefan Hajnoczi wrote:
> On Thu, Mar 27, 2014 at 4:53 PM, Anton Ivanov (antivano)
> <antivano@cisco.com> wrote:
>> On 27/03/14 12:30, Stefan Hajnoczi wrote:
>>> On Wed, Mar 26, 2014 at 11:08:41AM +0000, anton.ivanov@kot-begemot.co.uk wrote:
>>>> +static ssize_t net_l2tpv3_receive_dgram_iov(NetClientState *nc,
>>>> +                    const struct iovec *iov,
>>>> +                    int iovcnt)
>>>> +{
>>>> +    NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
>>>> +
>>>> +    struct msghdr message;
>>>> +    int ret;
>>>> +
>>>> +    if (iovcnt > MAX_L2TPV3_IOVCNT - 1) {
>>>> +        error_report(
>>>> +            "iovec too long %d > %d, change l2tpv3.h",
>>>> +            iovcnt, MAX_L2TPV3_IOVCNT
>>>> +        );
>>>> +        return -1;
>>>> +    }
>>> The alternative is to linearize the buffer using iov_to_buf() and call
>>> net_l2tpv3_receive_dgram().  Something like:
>>>
>>> if (iovcnt > MAX_L2TPV3_IOVCNT - 1) {
>>>     size_t size = iov_size(iov, iovcnt);
>>>     uint8_t *buf;
>>>
>>>     buf = g_malloc(size);
>>>     iov_to_buf(iov, iovcnt, 0, buf, size);
>>>     ret = net_l2tpv3_receive_dgram(nc, buf, size);
>>>     g_free(buf);
>>>     return ret;
>>> }
>> iov_to_buf is a memcpy of the data and a malloc down the call chain.
>> That is quite a significant cost compared to just shifting the iov a bit
>> to add an element for header. I tried that early on and it was
>> introducing a noticeable penalty.
>>
>> If I am to merge it I would actually merge it the other way around -
>> express both net_l2tpv3_receive_dgram and net_l2tpv3_receive_dgram_iov
>> via an iov based backend. That will allow to add sendmmsg one day as the
>> data structures will be the same in all cases.
> I agree with you that the iov code path should be used.  The context
> for my comment was "if (iovcnt > MAX_L2TPV3_IOVCNT - 1) {".  I'm just
> suggesting that we fall back to linearizing the iovector if it has too
> many elements, instead of erroring out.

Understood. Will do.

>
>>>> +static int l2tpv3_verify_header(NetL2TPV3State *s, uint8_t *buf)
>>>> +{
>>>> +
>>>> +    uint32_t *session;
>>>> +    uint64_t cookie;
>>>> +
>>>> +    if ((!s->udp) && (!s->ipv6)) {
>>>> +        buf += sizeof(struct iphdr) /* fix for ipv4 raw */;
>>>> +    }
>>>> +
>>>> +    /* we do not do a strict check for "data" packets as per
>>>> +    * the RFC spec because the pure IP spec does not have
>>>> +    * that anyway.
>>>> +    */
>>>> +
>>>> +    if (s->cookie) {
>>>> +        if (s->cookie_is_64) {
>>>> +            cookie = ldq_be_p(buf + s->cookie_offset);
>>>> +        } else {
>>>> +            cookie = ldl_be_p(buf + s->cookie_offset);
>>>> +        }
>>>> +        if (cookie != s->rx_cookie) {
>>>> +            if (!s->header_mismatch) {
>>>> +                error_report("unknown cookie id\n");
>>> error_report() messages should not have a newline ('\n') at the end.
>>>
>>>> +            }
>>>> +            return -1;
>>>> +        }
>>>> +    }
>>>> +    session = (uint32_t *) (buf + s->session_offset);
>>>> +    if (ldl_be_p(session) != s->rx_session) {
>>>> +        if (!s->header_mismatch) {
>>>> +            error_report("session mismatch");
>>>> +        }
>>>> +        return -1;
>>>> +    }
>>>> +    return 0;
>>>> +}
>>>> +
>>>> +static void net_l2tpv3_process_queue(NetL2TPV3State *s)
>>>> +{
>>>> +    int size = 0;
>>>> +    struct iovec *vec;
>>>> +    bool bad_read;
>>>> +    int data_size;
>>>> +    struct mmsghdr *msgvec;
>>>> +
>>>> +    /* go into ring mode only if there is a "pending" tail */
>>>> +    if (s->queue_depth > 0 && (!s->delivering)) {
>>>> +        do {
>>>> +            msgvec = s->msgvec + s->queue_tail;
>>>> +            if (msgvec->msg_len > 0) {
>>>> +                data_size = msgvec->msg_len - s->header_size;
>>> header_size is never assigned to.
>> It is - in init. It is used to divorce the "read header size" from
>> "l2tpv3 header size" as needed by the ipv4 socket API. Using it allows
>> to remove the check if we are reading from ipv4 raw sockets in most (but
>> not all) places.
> I'm not seeing that in this version of the patch, please check again
> and let me know which line number it is on.

You are correct.

I failed to cherry-pick that change correctly from my development branch
when creating the patchset for submission.

Mea culpa.

>
>>>> +                vec = msgvec->msg_hdr.msg_iov;
>>>> +                if ((data_size > 0) &&
>>>> +                    (l2tpv3_verify_header(s, vec->iov_base) == 0)) {
>>>> +                    vec++;
>>>> +                    /* we do not use the "normal" send and send async
>>>> +                     * functions here because we have our own buffer -
>>>> +                     * the message vector. If we use the "usual" ones
>>>> +                     * we will end up double-buffering.
>>>> +                     */
>>>> +                    s->delivering = true;
>>> What is s->delivering guarding against?  This function is only called
>>> from the s->fd read handler function, so I don't see a reentrant code
>>> path.
>> Cut-n-paste from the queue functions without doing full analysis of what
>> is it guarding against there.
> Okay, the problem in the generic net code is that some net clients
> (like slirp) will send packets back to their peer from their receive()
> function.  For example, imagine a DHCP request packet from the guest
> being sent to the slirp net client.  It will send the DHCP reply right
> away while we're still in the send and receive functions.  This
> creates reentrancy and not all code handles it so we just queue the
> packets instead.
>
> But in this case there is no reentrancy since only the fd handler
> function can call net_l2tpv3_process_queue().

OK.

>
>>>> +                    /* deliver directly to peer bypassing queueing and
>>>> +                     * buffering
>>>> +                     */
>>>> +                    size = qemu_deliver_packet(
>>>> +                            &s->nc,
>>>> +                            QEMU_NET_PACKET_FLAG_NONE,
>>>> +                            vec->iov_base,
>>>> +                            data_size,
>>>> +                            s->nc.peer
>>>> +                        );
>>> There is no callback when the peer re-enables receive.  In other words,
>>> packets will be stuck in s->msgvec until a new packet arrives and
>>> net_l2tpv3_send() is invoked.
>>>
>>> I think you need to modify qemu_flush_queued_packets() to invoke a new
>>> callback.  You could convert the existing code:
>> I had the same thoughts, just did not want to touch it just yet.
>>
>> Something along the lines would have been necessary for sendmmsg too as
>> there you would want to work off a pre-configured queue in the first place.
>>
>>> if (nc->peer && nc->peer->info->type == NET_CLIENT_OPTIONS_KIND_HUBPORT) {
>>>     if (net_hub_flush(nc->peer)) {
>>>         qemu_notify_event();
>>>     }
>>> }
>>>
>>> to:
>>>
>>> if (nc->peer && nc->peer->info->peer_flushed) {
>>>     if (nc->peer->info->peer_flushed(nc->peer)) {
>>>         qemu_notify_event();
>>>     }
>>> }
>>>
>>> and move the hub flushing code into net/hub.c.
>>>
>>> Then you could also implement ->peer_flushed() in l2tpv3.c so that it
>>> calls net_l2tpv3_process_queue() and returns true if packets were
>>> delivered.
>>>
>>> Please introduce ->peer_flushed() in a separate patch.
>> Understood.
>>
>>> Or you could keep it simple and go back to using qemu_send_packet()
>>> without trying to avoid duplicate buffering.  (You can always send
>>> patches later that eliminate the buffering.)
>> I would probably go back to that for submission so we can do it in stages.
>>
>> I will branch out the zero copy work and relevant additions to queue.c
>> for submission at a later date along with zero copy versions of gre,
>> raw, as well as a revised version of l2tpv3.c
> Excellent, I was hoping for this, thanks!

OK. Will do it this way.

ETA - Monday.


A.
anton.ivanov@kot-begemot.co.uk - March 31, 2014, 3:48 p.m.
>>> if (iovcnt > MAX_L2TPV3_IOVCNT - 1) {
>>>      size_t size = iov_size(iov, iovcnt);
>>>      uint8_t *buf;
>>>
>>>      buf = g_malloc(size);
>>>      iov_to_buf(iov, iovcnt, 0, buf, size);
>>>      ret = net_l2tpv3_receive_dgram(nc, buf, size);
>>>      g_free(buf);
>>>      return ret;
>>> }
>> iov_to_buf is a memcpy of the data and a malloc down the call chain.
>> That is quite a significant cost compared to just shifting the iov a bit
>> to add an element for header. I tried that early on and it was
>> introducing a noticeable penalty.
>>
>> If I am to merge it I would actually merge it the other way around -
>> express both net_l2tpv3_receive_dgram and net_l2tpv3_receive_dgram_iov
>> via an iov based backend. That will allow to add sendmmsg one day as the
>> data structures will be the same in all cases.
> I agree with you that the iov code path should be used.  The context
> for my comment was "if (iovcnt > MAX_L2TPV3_IOVCNT - 1) {".  I'm just
> suggesting that we fall back to linearizing the iovector if it has too
> many elements, instead of erroring out.

I will send this one as incremental in addition to a proposal to improve 
iov buffer lineariziation.

I have some ideas how to avoid copying there for the cases where you get 
a chunk of data in only one of the iov_base pointers.

A.
>>>> +                    /* deliver directly to peer bypassing queueing and
>>>> +                     * buffering
>>>> +                     */
>>>> +                    size = qemu_deliver_packet(
>>>> +                            &s->nc,
>>>> +                            QEMU_NET_PACKET_FLAG_NONE,
>>>> +                            vec->iov_base,
>>>> +                            data_size,
>>>> +                            s->nc.peer
>>>> +                        );
>>> There is no callback when the peer re-enables receive.  In other words,
>>> packets will be stuck in s->msgvec until a new packet arrives and
>>> net_l2tpv3_send() is invoked.
>>>
>>> I think you need to modify qemu_flush_queued_packets() to invoke a new
>>> callback.  You could convert the existing code:
>> I had the same thoughts, just did not want to touch it just yet.
>>
>> Something along the lines would have been necessary for sendmmsg too as
>> there you would want to work off a pre-configured queue in the first place.

I tried it adding a quick and ugly if similar to the one used by HUB, 
this gives ~ 5% improvement on a benchmark, more when the machine is 
under "other" load (so it is easier to trigger the buffering).

However it is better if it is introduced across the board. There are 
other drivers which (if I understand them correctly) should not invoke 
the legacy queue - f.e. netmap. I will submit this separately.

A.

Patch

diff --git a/net/Makefile.objs b/net/Makefile.objs
index 4854a14..160214e 100644
--- a/net/Makefile.objs
+++ b/net/Makefile.objs
@@ -2,6 +2,7 @@  common-obj-y = net.o queue.o checksum.o util.o hub.o
 common-obj-y += socket.o
 common-obj-y += dump.o
 common-obj-y += eth.o
+common-obj-$(CONFIG_LINUX) += l2tpv3.o
 common-obj-$(CONFIG_POSIX) += tap.o
 common-obj-$(CONFIG_LINUX) += tap-linux.o
 common-obj-$(CONFIG_WIN32) += tap-win32.o
diff --git a/net/clients.h b/net/clients.h
index 7793294..bbf177c 100644
--- a/net/clients.h
+++ b/net/clients.h
@@ -47,6 +47,8 @@  int net_init_tap(const NetClientOptions *opts, const char *name,
 int net_init_bridge(const NetClientOptions *opts, const char *name,
                     NetClientState *peer);
 
+int net_init_l2tpv3(const NetClientOptions *opts, const char *name,
+                    NetClientState *peer);
 #ifdef CONFIG_VDE
 int net_init_vde(const NetClientOptions *opts, const char *name,
                  NetClientState *peer);
diff --git a/net/l2tpv3.c b/net/l2tpv3.c
new file mode 100644
index 0000000..467332a
--- /dev/null
+++ b/net/l2tpv3.c
@@ -0,0 +1,752 @@ 
+/*
+ * QEMU System Emulator
+ *
+ * Copyright (c) 2003-2008 Fabrice Bellard
+ * Copyright (c) 2012-2014 Cisco Systems
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+
+#include <linux/ip.h>
+#include <netdb.h>
+#include "config-host.h"
+#include "net/net.h"
+#include "clients.h"
+#include "monitor/monitor.h"
+#include "qemu-common.h"
+#include "qemu/error-report.h"
+#include "qemu/option.h"
+#include "qemu/sockets.h"
+#include "qemu/iov.h"
+#include "qemu/main-loop.h"
+
+
+/* The buffer size needs to be investigated for optimum numbers and
+ * optimum means of paging in on different systems. This size is
+ * chosen to be sufficient to accommodate one packet with some headers
+ */
+
+#define BUFFER_ALIGN sysconf(_SC_PAGESIZE)
+#define BUFFER_SIZE 2048
+#define IOVSIZE 2
+#define MAX_L2TPV3_MSGCNT 64
+#define MAX_L2TPV3_IOVCNT (MAX_L2TPV3_MSGCNT * IOVSIZE)
+
+/* Header set to 0x30000 signifies a data packet */
+
+#define L2TPV3_DATA_PACKET 0x30000
+
+/* IANA-assigned IP protocol ID for L2TPv3 */
+
+#ifndef IPPROTO_L2TP
+#define IPPROTO_L2TP 0x73
+#endif
+
+typedef struct NetL2TPV3State {
+    NetClientState nc;
+    int fd;
+
+    /*
+     * these are used for xmit - that happens packet a time
+     * and for first sign of life packet (easier to parse that once)
+     */
+
+    uint8_t *header_buf;
+    struct iovec *vec;
+
+    /*
+     * these are used for receive - try to "eat" up to 32 packets at a time
+     */
+
+    struct mmsghdr *msgvec;
+
+    /*
+     * peer address
+     */
+
+    struct sockaddr_storage *dgram_dst;
+    uint32_t dst_size;
+
+    /*
+     * L2TPv3 parameters
+     */
+
+    uint64_t rx_cookie;
+    uint64_t tx_cookie;
+    uint32_t rx_session;
+    uint32_t tx_session;
+    uint32_t header_size;
+    uint32_t counter;
+
+    /*
+    * DOS avoidance in error handling
+    */
+
+    bool header_mismatch;
+
+    /*
+     * Ring buffer handling
+     */
+
+    int queue_head;
+    int queue_tail;
+    int queue_depth;
+
+    /*
+     * Precomputed offsets
+     */
+
+    uint32_t offset;
+    uint32_t cookie_offset;
+    uint32_t counter_offset;
+    uint32_t session_offset;
+
+    /* Poll Control */
+
+    bool read_poll;
+    bool write_poll;
+    bool delivering;
+
+    /* Flags */
+
+    bool ipv6;
+    bool udp;
+    bool has_counter;
+    bool pin_counter;
+    bool cookie;
+    bool cookie_is_64;
+
+} NetL2TPV3State;
+
+static int l2tpv3_can_send(void *opaque);
+static void net_l2tpv3_send(void *opaque);
+static void l2tpv3_writable(void *opaque);
+
+static void l2tpv3_update_fd_handler(NetL2TPV3State *s)
+{
+    qemu_set_fd_handler2(s->fd,
+                         s->read_poll ? l2tpv3_can_send : NULL,
+                         s->read_poll ? net_l2tpv3_send     : NULL,
+                         s->write_poll ? l2tpv3_writable : NULL,
+                         s);
+}
+
+static void l2tpv3_read_poll(NetL2TPV3State *s, bool enable)
+{
+    if (s->read_poll != enable) {
+        s->read_poll = enable;
+        l2tpv3_update_fd_handler(s);
+    }
+}
+
+static void l2tpv3_write_poll(NetL2TPV3State *s, bool enable)
+{
+    if (s->write_poll != enable) {
+        s->write_poll = enable;
+        l2tpv3_update_fd_handler(s);
+    }
+}
+
+static void l2tpv3_writable(void *opaque)
+{
+    NetL2TPV3State *s = opaque;
+    l2tpv3_write_poll(s, false);
+    qemu_flush_queued_packets(&s->nc);
+}
+
+static int l2tpv3_can_send(void *opaque)
+{
+    NetL2TPV3State *s = opaque;
+
+    return qemu_can_send_packet(&s->nc);
+}
+
+static void l2tpv3_poll(NetClientState *nc, bool enable)
+{
+    NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
+    l2tpv3_write_poll(s, enable);
+    l2tpv3_read_poll(s, enable);
+}
+
+static void l2tpv3_form_header(NetL2TPV3State *s)
+{
+    uint32_t *counter;
+
+    if (s->udp) {
+        stl_be_p((uint32_t *) s->header_buf, L2TPV3_DATA_PACKET);
+    }
+    stl_be_p(
+            (uint32_t *) (s->header_buf + s->session_offset),
+            s->tx_session
+        );
+    if (s->cookie) {
+        if (s->cookie_is_64) {
+            stq_be_p(
+                (uint64_t *)(s->header_buf + s->cookie_offset),
+                s->tx_cookie
+            );
+        } else {
+            stl_be_p(
+                (uint32_t *) (s->header_buf + s->cookie_offset),
+                s->tx_cookie
+            );
+        }
+    }
+    if (s->has_counter) {
+        counter = (uint32_t *)(s->header_buf + s->counter_offset);
+        if (s->pin_counter) {
+            *counter = 0;
+        } else {
+            stl_be_p(counter, ++s->counter);
+        }
+    }
+}
+
+static ssize_t net_l2tpv3_receive_dgram_iov(NetClientState *nc,
+                    const struct iovec *iov,
+                    int iovcnt)
+{
+    NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
+
+    struct msghdr message;
+    int ret;
+
+    if (iovcnt > MAX_L2TPV3_IOVCNT - 1) {
+        error_report(
+            "iovec too long %d > %d, change l2tpv3.h",
+            iovcnt, MAX_L2TPV3_IOVCNT
+        );
+        return -1;
+    }
+    l2tpv3_form_header(s);
+    memcpy(s->vec + 1, iov, iovcnt * sizeof(struct iovec));
+    s->vec->iov_base = s->header_buf;
+    s->vec->iov_len = s->offset;
+    message.msg_name = s->dgram_dst;
+    message.msg_namelen = s->dst_size;
+    message.msg_iov = s->vec;
+    message.msg_iovlen = iovcnt + 1;
+    message.msg_control = NULL;
+    message.msg_controllen = 0;
+    message.msg_flags = 0;
+    do {
+        ret = sendmsg(s->fd, &message, 0);
+    } while ((ret == -1) && (errno == EINTR));
+    if (ret > 0) {
+        ret -= s->offset;
+    } else if (ret == 0) {
+        /* belt and braces - should not occur on DGRAM
+        * we should get an error and never a 0 send
+        */
+        ret = iov_size(iov, iovcnt);
+    } else {
+        /* signal upper layer that socket buffer is full */
+        ret = -errno;
+        if (ret == -EAGAIN || ret == -ENOBUFS) {
+            l2tpv3_write_poll(s, true);
+            ret = 0;
+        }
+    }
+    return ret;
+}
+
+static ssize_t net_l2tpv3_receive_dgram(NetClientState *nc,
+                    const uint8_t *buf,
+                    size_t size)
+{
+    NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
+
+    struct iovec *vec;
+    struct msghdr message;
+    ssize_t ret = 0;
+
+    l2tpv3_form_header(s);
+    vec = s->vec;
+    vec->iov_base = s->header_buf;
+    vec->iov_len = s->offset;
+    vec++;
+    vec->iov_base = (void *) buf;
+    vec->iov_len = size;
+    message.msg_name = s->dgram_dst;
+    message.msg_namelen = s->dst_size;
+    message.msg_iov = s->vec;
+    message.msg_iovlen = 2;
+    message.msg_control = NULL;
+    message.msg_controllen = 0;
+    message.msg_flags = 0;
+    do {
+        ret = sendmsg(s->fd, &message, 0);
+    } while ((ret == -1) && (errno == EINTR));
+    if (ret > 0) {
+        ret -= s->offset;
+    } else if (ret == 0) {
+        /* belt and braces - should not occur on DGRAM
+        * we should get an error and never a 0 send
+        */
+        ret = size;
+    } else {
+        ret = -errno;
+        if (ret == -EAGAIN || ret == -ENOBUFS) {
+            /* signal upper layer that socket buffer is full */
+            l2tpv3_write_poll(s, true);
+            ret = 0;
+        }
+    }
+    return ret;
+}
+
+static int l2tpv3_verify_header(NetL2TPV3State *s, uint8_t *buf)
+{
+
+    uint32_t *session;
+    uint64_t cookie;
+
+    if ((!s->udp) && (!s->ipv6)) {
+        buf += sizeof(struct iphdr) /* fix for ipv4 raw */;
+    }
+
+    /* we do not do a strict check for "data" packets as per
+    * the RFC spec because the pure IP spec does not have
+    * that anyway.
+    */
+
+    if (s->cookie) {
+        if (s->cookie_is_64) {
+            cookie = ldq_be_p(buf + s->cookie_offset);
+        } else {
+            cookie = ldl_be_p(buf + s->cookie_offset);
+        }
+        if (cookie != s->rx_cookie) {
+            if (!s->header_mismatch) {
+                error_report("unknown cookie id\n");
+            }
+            return -1;
+        }
+    }
+    session = (uint32_t *) (buf + s->session_offset);
+    if (ldl_be_p(session) != s->rx_session) {
+        if (!s->header_mismatch) {
+            error_report("session mismatch");
+        }
+        return -1;
+    }
+    return 0;
+}
+
+static void net_l2tpv3_process_queue(NetL2TPV3State *s)
+{
+    int size = 0;
+    struct iovec *vec;
+    bool bad_read;
+    int data_size;
+    struct mmsghdr *msgvec;
+
+    /* go into ring mode only if there is a "pending" tail */
+    if (s->queue_depth > 0 && (!s->delivering)) {
+        do {
+            msgvec = s->msgvec + s->queue_tail;
+            if (msgvec->msg_len > 0) {
+                data_size = msgvec->msg_len - s->header_size;
+                vec = msgvec->msg_hdr.msg_iov;
+                if ((data_size > 0) &&
+                    (l2tpv3_verify_header(s, vec->iov_base) == 0)) {
+                    vec++;
+                    /* we do not use the "normal" send and send async
+                     * functions here because we have our own buffer -
+                     * the message vector. If we use the "usual" ones
+                     * we will end up double-buffering.
+                     */
+                    s->delivering = true;
+                    /* deliver directly to peer bypassing queueing and
+                     * buffering
+                     */
+                    size = qemu_deliver_packet(
+                            &s->nc,
+                            QEMU_NET_PACKET_FLAG_NONE,
+                            vec->iov_base,
+                            data_size,
+                            s->nc.peer
+                        );
+                    s->delivering = false;
+                    bad_read = false;
+                } else {
+                    bad_read = true;
+                    if (!s->header_mismatch) {
+                        /* report error only once */
+                        error_report("l2tpv3 header verification failed");
+                        s->header_mismatch = true;
+                    }
+                }
+            } else {
+                bad_read = true;
+            }
+            if ((bad_read) || (size > 0)) {
+                s->queue_tail = (s->queue_tail + 1) % MAX_L2TPV3_MSGCNT;
+                s->queue_depth--;
+            }
+        } while (
+                (s->queue_depth > 0) &&
+                 qemu_can_send_packet(&s->nc) &&
+                ((size > 0) || bad_read)
+            );
+    }
+}
+
+static void net_l2tpv3_send(void *opaque)
+{
+    NetL2TPV3State *s = opaque;
+    int target_count, count;
+    struct mmsghdr *msgvec;
+
+    /* go into ring mode only if there is a "pending" tail */
+
+    if (s->queue_depth) {
+
+        /* The ring buffer we use has variable intake
+         * count of how much we can read varies - adjust accordingly
+         */
+
+        target_count = MAX_L2TPV3_MSGCNT - s->queue_depth;
+
+        /* Ensure we do not overrun the ring when we have
+         * a lot of enqueued packets
+         */
+
+        if (s->queue_head + target_count > MAX_L2TPV3_MSGCNT) {
+            target_count = MAX_L2TPV3_MSGCNT - s->queue_head;
+        }
+    } else {
+
+        /* we do not have any pending packets - we can use
+        * the whole message vector linearly instead of using
+        * it as a ring
+        */
+
+        s->queue_head = 0;
+        s->queue_tail = 0;
+        target_count = MAX_L2TPV3_MSGCNT;
+    }
+
+    msgvec = s->msgvec + s->queue_head;
+    if (target_count > 0) {
+        do {
+            count = recvmmsg(
+                s->fd,
+                msgvec,
+                target_count, MSG_DONTWAIT, NULL);
+        } while ((count == -1) && (errno == EINTR));
+        if (count < 0) {
+            /* Recv error - we still need to flush packets here,
+             * (re)set queue head to current position
+             */
+            count = 0;
+        }
+        s->queue_head = (s->queue_head + count) % MAX_L2TPV3_MSGCNT;
+        s->queue_depth += count;
+    }
+    net_l2tpv3_process_queue(s);
+}
+
+static void destroy_vector(struct mmsghdr *msgvec, int count, int iovcount)
+{
+    int i, j;
+    struct iovec *iov;
+    struct mmsghdr *cleanup = msgvec;
+    if (cleanup) {
+        for (i = 0; i < count; i++) {
+            if (cleanup->msg_hdr.msg_iov) {
+                iov = cleanup->msg_hdr.msg_iov;
+                for (j = 0; j < iovcount; j++) {
+                    g_free(iov->iov_base);
+                    iov++;
+                }
+                g_free(cleanup->msg_hdr.msg_iov);
+            }
+            cleanup++;
+        }
+        g_free(msgvec);
+    }
+}
+
+static struct mmsghdr *build_l2tpv3_vector(NetL2TPV3State *s, int count)
+{
+    int i;
+    struct iovec *iov;
+    struct mmsghdr *msgvec, *result;
+
+    msgvec = g_malloc(sizeof(struct mmsghdr) * count);
+    result = msgvec;
+    for (i = 0; i < count ; i++) {
+        msgvec->msg_hdr.msg_name = NULL;
+        msgvec->msg_hdr.msg_namelen = 0;
+        iov =  g_malloc(sizeof(struct iovec) * IOVSIZE);
+        msgvec->msg_hdr.msg_iov = iov;
+        if ((!s->udp) && (!s->ipv6)) {
+            /* fix for ipv4 raw */;
+            iov->iov_base = g_malloc(s->offset + sizeof(struct iphdr));
+            iov->iov_len = s->offset + sizeof(struct iphdr);
+        } else {
+            iov->iov_base = g_malloc(s->offset);
+            iov->iov_len = s->offset;
+        }
+        iov++ ;
+        iov->iov_base = qemu_memalign(BUFFER_ALIGN, BUFFER_SIZE);
+        iov->iov_len = BUFFER_SIZE;
+        msgvec->msg_hdr.msg_iovlen = 2;
+        msgvec->msg_hdr.msg_control = NULL;
+        msgvec->msg_hdr.msg_controllen = 0;
+        msgvec->msg_hdr.msg_flags = 0;
+        msgvec++;
+    }
+    return result;
+}
+
+static void net_l2tpv3_cleanup(NetClientState *nc)
+{
+    NetL2TPV3State *s = DO_UPCAST(NetL2TPV3State, nc, nc);
+    qemu_purge_queued_packets(nc);
+    l2tpv3_read_poll(s, false);
+    l2tpv3_write_poll(s, false);
+    close(s->fd);
+    destroy_vector(s->msgvec, MAX_L2TPV3_MSGCNT, IOVSIZE);
+    g_free(s->header_buf);
+    g_free(s->dgram_dst);
+}
+
+static NetClientInfo net_l2tpv3_info = {
+    .type = NET_CLIENT_OPTIONS_KIND_L2TPV3,
+    .size = sizeof(NetL2TPV3State),
+    .receive = net_l2tpv3_receive_dgram,
+    .receive_iov = net_l2tpv3_receive_dgram_iov,
+    .poll = l2tpv3_poll,
+    .cleanup = net_l2tpv3_cleanup,
+};
+
+int net_init_l2tpv3(const NetClientOptions *opts,
+                    const char *name,
+                    NetClientState *peer)
+{
+
+
+    const NetdevL2TPv3Options *l2tpv3;
+    NetL2TPV3State *s;
+    NetClientState *nc;
+    int fd = -1, gairet;
+    struct addrinfo hints;
+    struct addrinfo *result = NULL;
+    char *srcport, *dstport;
+
+    nc = qemu_new_net_client(&net_l2tpv3_info, peer, "l2tpv3", name);
+
+    s = DO_UPCAST(NetL2TPV3State, nc, nc);
+
+    s->queue_head = 0;
+    s->queue_tail = 0;
+    s->header_mismatch = false;
+    s->delivering = false;
+
+    assert(opts->kind == NET_CLIENT_OPTIONS_KIND_L2TPV3);
+    l2tpv3 = opts->l2tpv3;
+
+    if (l2tpv3->has_ipv6 && l2tpv3->ipv6) {
+        s->ipv6 = l2tpv3->ipv6;
+    } else {
+        s->ipv6 = false;
+    }
+
+    if (l2tpv3->has_rxcookie || l2tpv3->has_txcookie) {
+        if (l2tpv3->has_rxcookie && l2tpv3->has_txcookie) {
+            s->cookie = true;
+        } else {
+            goto outerr;
+        }
+    } else {
+        s->cookie = false;
+    }
+
+    if (l2tpv3->has_cookie64 || l2tpv3->cookie64) {
+        s->cookie_is_64  = true;
+    } else {
+        s->cookie_is_64  = false;
+    }
+
+    if (l2tpv3->has_udp && l2tpv3->udp) {
+        s->udp = true;
+        if (!(l2tpv3->has_srcport && l2tpv3->has_dstport)) {
+            error_report("l2tpv3_open : need both src and dst port for udp");
+            goto outerr;
+        } else {
+            srcport = l2tpv3->srcport;
+            dstport = l2tpv3->dstport;
+        }
+    } else {
+        s->udp = false;
+        srcport = NULL;
+        dstport = NULL;
+    }
+
+
+    s->offset = 4;
+    s->session_offset = 0;
+    s->cookie_offset = 4;
+    s->counter_offset = 4;
+
+    s->tx_session = l2tpv3->txsession;
+    if (l2tpv3->has_rxsession) {
+        s->rx_session = l2tpv3->rxsession;
+    } else {
+        s->rx_session = s->tx_session;
+    }
+
+    if (s->cookie) {
+        s->rx_cookie = l2tpv3->rxcookie;
+        s->tx_cookie = l2tpv3->txcookie;
+        if (s->cookie_is_64 == true) {
+            /* 64 bit cookie */
+            s->offset += 8;
+            s->counter_offset += 8;
+        } else {
+            /* 32 bit cookie */
+            s->offset += 4;
+            s->counter_offset += 4;
+        }
+    }
+
+    memset(&hints, 0, sizeof(hints));
+
+    if (s->ipv6) {
+        hints.ai_family = AF_INET6;
+    } else {
+        hints.ai_family = AF_INET;
+    }
+    if (s->udp) {
+        hints.ai_socktype = SOCK_DGRAM;
+        hints.ai_protocol = 0;
+        s->offset += 4;
+        s->counter_offset += 4;
+        s->session_offset += 4;
+        s->cookie_offset += 4;
+    } else {
+        hints.ai_socktype = SOCK_RAW;
+        hints.ai_protocol = IPPROTO_L2TP;
+    }
+
+    gairet = getaddrinfo(l2tpv3->src, srcport, &hints, &result);
+
+    if ((gairet != 0) || (result == NULL)) {
+        error_report(
+            "l2tpv3_open : could not resolve src, errno = %s",
+            gai_strerror(gairet)
+        );
+        goto outerr;
+    }
+    fd = socket(result->ai_family, result->ai_socktype, result->ai_protocol);
+    if (fd == -1) {
+        fd = -errno;
+        error_report("l2tpv3_open : socket creation failed, errno = %d", -fd);
+        freeaddrinfo(result);
+        goto outerr;
+    }
+    if (bind(fd, (struct sockaddr *) result->ai_addr, result->ai_addrlen)) {
+        error_report("l2tpv3_open :  could not bind socket err=%i", errno);
+        goto outerr;
+    }
+
+    freeaddrinfo(result);
+
+    memset(&hints, 0, sizeof(hints));
+
+    if (s->ipv6) {
+        hints.ai_family = AF_INET6;
+    } else {
+        hints.ai_family = AF_INET;
+    }
+    if (s->udp) {
+        hints.ai_socktype = SOCK_DGRAM;
+        hints.ai_protocol = 0;
+    } else {
+        hints.ai_socktype = SOCK_RAW;
+        hints.ai_protocol = IPPROTO_L2TP;
+    }
+
+    gairet = getaddrinfo(l2tpv3->dst, dstport, &hints, &result);
+    if ((gairet != 0) || (result == NULL)) {
+        error_report(
+            "l2tpv3_open : could not resolve dst, error = %s",
+            gai_strerror(gairet)
+        );
+        goto outerr;
+    }
+
+    s->dgram_dst = g_malloc(sizeof(struct sockaddr_storage));
+    memset(s->dgram_dst, '\0' , sizeof(struct sockaddr_storage));
+    memcpy(s->dgram_dst, result->ai_addr, result->ai_addrlen);
+    s->dst_size = result->ai_addrlen;
+
+    freeaddrinfo(result);
+
+    if (l2tpv3->has_counter && l2tpv3->counter) {
+        s->has_counter = true;
+        s->offset += 4;
+    } else {
+        s->has_counter = false;
+    }
+
+    if (l2tpv3->has_pincounter && l2tpv3->pincounter) {
+        s->has_counter = true;  /* pin counter implies that there is counter */
+        s->pin_counter = true;
+    } else {
+        s->pin_counter = false;
+    }
+
+    if (l2tpv3->has_offset) {
+        /* extra offset */
+        s->offset += l2tpv3->offset;
+    }
+
+    s->msgvec = build_l2tpv3_vector(s, MAX_L2TPV3_MSGCNT);
+    s->vec = g_malloc(sizeof(struct iovec) * MAX_L2TPV3_IOVCNT);
+    if ((!s->udp) && (!s->ipv6)) {
+        s->header_buf = g_malloc(s->offset + sizeof(struct iphdr));
+    } else {
+        s->header_buf = g_malloc(s->offset);
+    }
+
+    qemu_set_nonblock(fd);
+
+    s->fd = fd;
+    s->counter = 0;
+
+    l2tpv3_read_poll(s, true);
+
+    if (!s) {
+        error_report("l2tpv3_open : failed to set fd handler");
+        goto outerr;
+    }
+    snprintf(s->nc.info_str, sizeof(s->nc.info_str),
+             "l2tpv3: connected");
+    return 0;
+outerr:
+    qemu_del_net_client(nc);
+    if (fd > 0) {
+        close(fd);
+    }
+    return -1;
+}
+
diff --git a/net/net.c b/net/net.c
index 0a88e68..749d34c 100644
--- a/net/net.c
+++ b/net/net.c
@@ -731,6 +731,9 @@  static int (* const net_client_init_fun[NET_CLIENT_OPTIONS_KIND_MAX])(
         [NET_CLIENT_OPTIONS_KIND_BRIDGE]    = net_init_bridge,
 #endif
         [NET_CLIENT_OPTIONS_KIND_HUBPORT]   = net_init_hubport,
+#ifdef CONFIG_LINUX
+        [NET_CLIENT_OPTIONS_KIND_L2TPV3]    = net_init_l2tpv3,
+#endif
 };
 
 
diff --git a/qapi-schema.json b/qapi-schema.json
index 83fa485..c033e3b 100644
--- a/qapi-schema.json
+++ b/qapi-schema.json
@@ -2941,6 +2941,62 @@ 
     '*udp':       'str' } }
 
 ##
+# @NetdevL2TPv3Options
+#
+# Connect the VLAN to Ethernet over L2TPv3 Static tunnel
+#
+# @src: source address
+#
+# @dst: destination address
+#
+# @srcport: #optional source port - mandatory for udp, optional for ip
+#
+# @dstport: #optional destination port - mandatory for udp, optional for ip
+#
+# @ipv6: #optional - force the use of ipv6
+#
+# @udp: #optional - use the udp version of l2tpv3 encapsulation
+#
+# @cookie64: #optional - use 64 bit coookies
+#
+# @counter: #optional have sequence counter
+#
+# @pincounter: #optional pin sequence counter to zero -
+#              workaround for buggy implementations or
+#              networks with packet reorder
+#
+# @txcookie: #optional 32 or 64 bit transmit cookie
+#
+# @rxcookie: #optional 32 or 64 bit receive cookie
+#
+# @txsession: 32 bit transmit session
+#
+# @rxsession: #optional 32 bit receive session - if not specified
+#             set to the same value as transmit
+#
+# @offset: #optional additional offset - allows the insertion of
+#          additional application-specific data before the packet payload
+#
+# Since 2.1
+##
+{ 'type': 'NetdevL2TPv3Options',
+  'data': {
+    'src':          'str',
+    'dst':          'str',
+    '*srcport':     'str',
+    '*dstport':     'str',
+    '*ipv6':        'bool',
+    '*udp':         'bool',
+    '*cookie64':    'bool',
+    '*counter':     'bool',
+    '*pincounter':  'bool',
+    '*txcookie':    'uint64',
+    '*rxcookie':    'uint64',
+    'txsession':    'uint32',
+    '*rxsession':   'uint32',
+    '*offset':      'uint32' } }
+
+##
 # @NetdevVdeOptions
 #
 # Connect the VLAN to a vde switch running on the host.
@@ -3014,13 +3070,16 @@ 
 # A discriminated record of network device traits.
 #
 # Since 1.2
-##
+#
+# Added in 2.1 - l2tpv3
+#
 { 'union': 'NetClientOptions',
   'data': {
     'none':     'NetdevNoneOptions',
     'nic':      'NetLegacyNicOptions',
     'user':     'NetdevUserOptions',
     'tap':      'NetdevTapOptions',
+    'l2tpv3':   'NetdevL2TPv3Options',
     'socket':   'NetdevSocketOptions',
     'vde':      'NetdevVdeOptions',
     'dump':     'NetdevDumpOptions',
diff --git a/qemu-options.hx b/qemu-options.hx
index 8b94264..e1caf6f 100644
--- a/qemu-options.hx
+++ b/qemu-options.hx
@@ -1395,6 +1395,29 @@  DEF("net", HAS_ARG, QEMU_OPTION_net,
     "                (default=" DEFAULT_BRIDGE_INTERFACE ") using the program 'helper'\n"
     "                (default=" DEFAULT_BRIDGE_HELPER ")\n"
 #endif
+#ifdef __linux__
+    "-net l2tpv3[,vlan=n][,name=str],src=srcaddr,dst=dstaddr[,srcport=srcport][,dstport=dstport],txsession=txsession[,rxsession=rxsession][,ipv6=on/off][,udp=on/off][,cookie64=on/off][,counter][,pincounter][,txcookie=txcookie][,rxcookie=rxcookie][,offset=offset]\n"
+    "                connect the VLAN to an Ethernet over L2TPv3 pseudowire\n"
+    "                Linux kernel 3.3+ as well as most routers can talk\n"
+    "                L2TPv3. This transport allows to connect a VM to a VM,\n"
+    "                VM to a router and even VM to Host. It is a nearly-universal\n"
+    "                standard (RFC3391). Note - this implementation uses static\n"
+    "                pre-configured tunnels (same as the Linux kernel).\n"
+    "                use 'src=' to specify source address\n"
+    "                use 'dst=' to specify destination address\n"
+    "                use 'udp=on' to specify udp encapsulation\n"
+    "                use 'dstport=' to specify destination udp port\n"
+    "                use 'dstport=' to specify destination udp port\n"
+    "                use 'ipv6=on' to force v6\n"
+    "                L2TPv3 uses cookies to prevent misconfiguration as\n"
+    "                well as a weak security measure\n"
+    "                use 'rxcookie=0x012345678' to specify a rxcookie\n"
+    "                use 'txcookie=0x012345678' to specify a txcookie\n"
+    "                use 'cookie64=on' to set cookie size to 64 bit, otherwise 32\n"
+    "                use 'counter=off' to force a 'cut-down' L2TPv3 with no counter\n"
+    "                use 'pincounter=on' to work around broken counter handling in peer\n"
+    "                use 'offset=X' to add an extra offset between header and data\n"
+#endif
     "-net socket[,vlan=n][,name=str][,fd=h][,listen=[host]:port][,connect=host:port]\n"
     "                connect the vlan 'n' to another VLAN using a socket connection\n"
     "-net socket[,vlan=n][,name=str][,fd=h][,mcast=maddr:port[,localaddr=addr]]\n"
@@ -1730,6 +1753,65 @@  qemu-system-i386 linux.img \
                  -net socket,mcast=239.192.168.1:1102,localaddr=1.2.3.4
 @end example
 
+@item -netdev l2tpv3,id=@var{id},src=@var{srcaddr},dst=@var{dstaddr}[,srcport=@var{srcport}][,dstport=@var{dstport}],txsession=@var{txsession}[,rxsession=@var{rxsession}][,ipv6][,udp][,cookie64][,counter][,pincounter][,txcookie=@var{txcookie}][,rxcookie=@var{rxcookie}][,offset=@var{offset}]
+@item -net l2tpv3[,vlan=@var{n}][,name=@var{name}],src=@var{srcaddr},dst=@var{dstaddr}[,srcport=@var{srcport}][,dstport=@var{dstport}],txsession=@var{txsession}[,rxsession=@var{rxsession}][,ipv6][,udp][,cookie64][,counter][,pincounter][,txcookie=@var{txcookie}][,rxcookie=@var{rxcookie}][,offset=@var{offset}]
+Connect VLAN @var{n} to L2TPv3 pseudowire. L2TPv3 (RFC3391) is a popular
+protocol to transport Ethernet (and other Layer 2) data frames between
+two systems. It is present in routers, firewalls and the Linux kernel
+(from version 3.3 onwards).
+
+This transport allows a VM to communicate to another VM, router or firewall directly.
+
+@item src=@var{srcaddr}
+    source address (mandatory)
+@item dst=@var{dstaddr}
+    destination address (mandatory)
+@item udp
+    select udp encapsulation (default is ip).
+@item srcport=@var{srcport}
+    source udp port.
+@item dstport=@var{dstport}
+    destination udp port.
+@item ipv6
+    force v6, otherwise defaults to v4.
+@item rxcookie=@var{rxcookie}
+@item txcookie=@var{txcookie}
+    Cookies are a weak form of security in the l2tpv3 specification.
+Their function is mostly to prevent misconfiguration. By default they are 32
+bit.
+@item cookie64
+    Set cookie size to 64 bit instead of the default 32
+@item counter=off
+    Force a 'cut-down' L2TPv3 with no counter as in
+draft-mkonstan-l2tpext-keyed-ipv6-tunnel-00
+@item pincounter=on
+    Work around broken counter handling in peer. This may also help on
+networks which have packet reorder.
+@item offset=@var{offset}
+    Add an extra offset between header and data
+
+For example, to attach a VM running on host 4.3.2.1 via L2TPv3 to the bridge br-lan
+on the remote Linux host 1.2.3.4:
+@example
+# Setup tunnel on linux host using raw ip as encapsulation
+# on 1.2.3.4
+ip l2tp add tunnel remote 4.3.2.1 local 1.2.3.4 tunnel_id 1 peer_tunnel_id 1 \
+    encap udp udp_sport 16384 udp_dport 16384
+ip l2tp add session tunnel_id 1 name vmtunnel0 session_id \
+    0xFFFFFFFF peer_session_id 0xFFFFFFFF
+ifconfig vmtunnel0 mtu 1500
+ifconfig vmtunnel0 up
+brctl addif br-lan vmtunnel0
+
+
+# on 4.3.2.1
+# launch QEMU instance - if your network has reorder or is very lossy add ,pincounter
+
+qemu-system-i386 linux.img -net nic -net l2tpv3,src=4.2.3.1,dst=1.2.3.4,udp,srcport=16384,dstport=16384,rxsession=0xffffffff,txsession=0xffffffff,counter
+
+
+@end example
+
 @item -netdev vde,id=@var{id}[,sock=@var{socketpath}][,port=@var{n}][,group=@var{groupname}][,mode=@var{octalmode}]
 @item -net vde[,vlan=@var{n}][,name=@var{name}][,sock=@var{socketpath}] [,port=@var{n}][,group=@var{groupname}][,mode=@var{octalmode}]
 Connect VLAN @var{n} to PORT @var{n} of a vde switch running on host and