[9fans] tun/tap support for 9vx

[9fans] tun/tap support for 9vx

[Tully Gray]

[-- Attachment #1: Type: text/plain, Size: 331 bytes --]

Hi,

I have modified Erik Quanstrom's raw socket ethernet driver
for 9vx so that it uses the Linux kernel's "tap" device.
It seems to work just fine. I create the tap device first
using "tunctl" which comes with the Usermode Linux toolkit
but I don't think this is necessary.

Tully Gray.

ps: I am shadowdaemon on IRC.

[-- Attachment #2: tuntap.diff --]
[-- Type: text/plain, Size: 1076 bytes --]

12a13
> #include <linux/if_tun.h>
38d38
< 	int fd;
40a41
> 	int fd;
42c43
< 	if((fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL))) < 0)
---
> 	if((fd = open("/dev/net/tun", O_RDWR)) < 0){
43a45,46
> 	}
>
45a49
> 		ifr.ifr_flags = IFF_TAP | IFF_NO_PI;
47,55c51
< 		if(ioctl(fd, SIOCGIFINDEX, &ifr) < 0){
< 			close(fd);
< 			return -1;
< 		}
< 		memset(&sa, 0, sizeof sa);
< 		sa.sll_family = AF_PACKET;
< 		sa.sll_protocol = htons(ETH_P_ALL);
< 		sa.sll_ifindex = ifr.ifr_ifindex;
< 		if(bind(fd, (struct sockaddr*)&sa, sizeof sa) < 0){
---
> 		if(ioctl(fd, TUNSETIFF, &ifr) < 0){
61,71d56
< 	memset(&ifr, 0, sizeof ifr);
< 	strncpy(ifr.ifr_name, dev, sizeof ifr.ifr_name);
< 	if(ioctl(fd, SIOCGIFFLAGS, &ifr) < 0){
< 		close(fd);
< 		return -1;
< 	}
< 	ifr.ifr_flags |= IFF_PROMISC;
< 	if(ioctl(fd, SIOCSIFFLAGS, &ifr) < 0){
< 		close(fd);
< 		return -1;
< 	}
149c134
< 	0x00, 0x48, 0x01, 0x23, 0x45, 0x61,
---
> //	0x00, 0x48, 0x01, 0x23, 0x45, 0x61,
158c143
< 	"eth1", "eth0", "eth2", "eth3", "eth4", "eth5",
---
> 	"tap0",

Re: [9fans] tun/tap support for 9vx

[Devon H. O'Dell]

2010/5/7 Tully Gray <tullygray@arc.net.au>:
> Hi,
>
> I have modified Erik Quanstrom's raw socket ethernet driver
> for 9vx so that it uses the Linux kernel's "tap" device.
> It seems to work just fine. I create the tap device first
> using "tunctl" which comes with the Usermode Linux toolkit
> but I don't think this is necessary.

Thanks for this. My summer of code student will be working on wrapping
up some of this so that it's a bit nicer and more portable, and this
is a good start.

--Devon

> Tully Gray.
>
> ps: I am shadowdaemon on IRC.
>

Re: [9fans] tun/tap support for 9vx

[erik quanstrom]

> 2010/5/7 Tully Gray <tullygray@arc.net.au>:
> > Hi,
> >
> > I have modified Erik Quanstrom's raw socket ethernet driver
> > for 9vx so that it uses the Linux kernel's "tap" device.
> > It seems to work just fine. I create the tap device first
> > using "tunctl" which comes with the Usermode Linux toolkit
> > but I don't think this is necessary.
>
> Thanks for this. My summer of code student will be working on wrapping
> up some of this so that it's a bit nicer and more portable, and this
> is a good start.

interesting.  what's the advantage of the tap device?
is the code available online?

- erik

Re: [9fans] tun/tap support for 9vx

[erik quanstrom]

> interesting.  what's the advantage of the tap device?
> is the code available online?

duh.  sorry.  missed the attachment.  having peeked
at the code, do you think that the tap devices is a
complete replacement for the raw socket?

- erik

Re: [9fans] tun/tap support for 9vx

[Devon H. O'Dell]

2010/5/7 erik quanstrom <quanstro@quanstro.net>:
>> interesting.  what's the advantage of the tap device?
>> is the code available online?
>
> duh.  sorry.  missed the attachment.  having peeked
> at the code, do you think that the tap devices is a
> complete replacement for the raw socket?

Yes. The tap(4) interface allows you to clone an existing ethernet
device on the system, and it's what most virtualization platforms use.
I had an ethertap.c in there, but never finished it up. It'll be
something that gets done for GSoC this year anyway.

--dho

> - erik
>
>

Re: [9fans] tun/tap support for 9vx

[Tully Gray]

On Fri, May 07, 2010 at 02:04:47PM -0400, erik quanstrom wrote:
> > 2010/5/7 Tully Gray <tullygray@arc.net.au>:
> > > Hi,
> > >
> > > I have modified Erik Quanstrom's raw socket ethernet driver
> > > for 9vx so that it uses the Linux kernel's "tap" device.
> > > It seems to work just fine. I create the tap device first
> > > using "tunctl" which comes with the Usermode Linux toolkit
> > > but I don't think this is necessary.
> >
> > Thanks for this. My summer of code student will be working on wrapping
> > up some of this so that it's a bit nicer and more portable, and this
> > is a good start.
>
> interesting.  what's the advantage of the tap device?
> is the code available online?
>
> - erik

Hi,

originally I was using your driver and attaching it to either
eth0 or the bridge I have, br0. I was getting problems with
ARP, 9vx could see the traffic but the host could not see the
guest. Also 9vx would eventually crash due to freeing memory
that was in use. Is the raw socket driver meant to be attached
to a unused device? Both eth0 and br0 were being used by the
host system.

Tully Gray.

ps: I got the code to setup the tap device from http://vtun.sourceforge.net/

Re: [9fans] tun/tap support for 9vx

[Bakul Shah]

On Sat, 08 May 2010 03:54:26 +1000 Tully Gray <tullygray@arc.net.au>  wrote:
>
> I have modified Erik Quanstrom's raw socket ethernet driver
> for 9vx so that it uses the Linux kernel's "tap" device.

Neat!

> < 	if((fd = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL))) < 0)
> ---
> > 	if((fd = open("/dev/net/tun", O_RDWR)) < 0){

On *BSD tun is for IP frames, tap is for ethernet frames
so there you'd use /dev/tap.  You may wish to see how Qemu
handles tap differences & multiple tap interfaces on various
platforms.

On Fri, 07 May 2010 19:14:58 BST "Devon H. O'Dell" <devon.odell@gmail.com>  wrot
e:
>
> Yes. The tap(4) interface allows you to clone an existing ethernet
> device on the system, and it's what most virtualization platforms use.

Sorry for nitpicking!  "tap" appears as a completely separate
virtual ethernet interface to the host.  The "other end" of
tap is a device, to be open()ed by a program.  One can bridge
the host side interface to existing physical ethernet
interfaces but you don't have to.