Re: of ethernet names (was [PATCH] Futex Asynchronous

Re: of ethernet names (was [PATCH] Futex Asynchronous

[Paul Menage]

In article <0C01A29FBAE24448A792F5C68F5EA47D29DD32@nasdaq.ms.ensim.com>,
you write:
>Is there any documentation on the netlink API, beyond UTSL(iproute)?
>Reference would be good, but a tutorial would be ideal.

The man pages for netlink(3), netlink(7), rtnetlink(3) and rtnetlink(7)
give a basic reference.

Paul

Re: of ethernet names (was [PATCH] Futex Asynchronous

[Brad Hards]

On Tue, 11 Jun 2002 09:20, Paul Menage wrote:
> In article <0C01A29FBAE24448A792F5C68F5EA47D29DD32@nasdaq.ms.ensim.com>,
>
> you write:
> >Is there any documentation on the netlink API, beyond UTSL(iproute)?
> >Reference would be good, but a tutorial would be ideal.
>
> The man pages for netlink(3), netlink(7), rtnetlink(3) and rtnetlink(7)
> give a basic reference.
Unfortunately my brain is not on the same level and scope as Alexey or davem. 

I simply don't grok those pages. I also note caveats about incompleteness, and 
recommendation to use libnetlink, which is also not documented much.

I sometimes hack on zcip (see ftp://ftp.kernel.org/pub/software/network/zcip 
for an out of date version), and it could really use some netlink-ification. 
But I can't even follow enough of iproute (or zebra, which also uses netlink, 
AFAICT) to figure out how to do basic stuff like a list of configured 
networking devices, or set the default route.

Brad

-- 
http://conf.linux.org.au. 22-25Jan2003. Perth, Australia. Birds in Black.

Re: of ethernet names (was [PATCH] Futex Asynchronous

[Randy.Dunlap]

On Tue, 11 Jun 2002, Brad Hards wrote:

| On Tue, 11 Jun 2002 09:20, Paul Menage wrote:
| > In article <0C01A29FBAE24448A792F5C68F5EA47D29DD32@nasdaq.ms.ensim.com>,
| >
| > you write:
| > >Is there any documentation on the netlink API, beyond UTSL(iproute)?
| > >Reference would be good, but a tutorial would be ideal.
| >
| > The man pages for netlink(3), netlink(7), rtnetlink(3) and rtnetlink(7)
| > give a basic reference.
| Unfortunately my brain is not on the same level and scope as Alexey or davem.
|
| I simply don't grok those pages. I also note caveats about incompleteness, and
| recommendation to use libnetlink, which is also not documented much.

Hi Brad,

There was an announcement earlier today of some networking
documentation.  I think that it's still in its early stages, though.

>From the 'netdev' list:

Hi,

I am working on a doc on "Networking Subsystem" in the Linux kernel.
This
is part of the "Linux Kernel documentation project" (www.lkdp.tk).

Chapter 2 of this doc is almost complete. I was hoping for some feedback
on this doc. Its available at

www.corewars.org/~radical/lkdp

Only Chapter 2 has been started as of now, its nearing completion.

I am looking for feedback on the contents, the general feel of the book,
flow.. etc. Any kind of feedback actually.

Thank you
-Anks

-- 
~Randy

Re: of ethernet names (was [PATCH] Futex Asynchronous

[Brad Hards]

On Tue, 11 Jun 2002 09:36, Randy.Dunlap wrote:
> On Tue, 11 Jun 2002, Brad Hards wrote:
> | On Tue, 11 Jun 2002 09:20, Paul Menage wrote:
> | > In article
> | > <0C01A29FBAE24448A792F5C68F5EA47D29DD32@nasdaq.ms.ensim.com>,
> | >
> | > you write:
> | > >Is there any documentation on the netlink API, beyond UTSL(iproute)?
> | > >Reference would be good, but a tutorial would be ideal.
> | >
> | > The man pages for netlink(3), netlink(7), rtnetlink(3) and rtnetlink(7)
> | > give a basic reference.
> |
> | Unfortunately my brain is not on the same level and scope as Alexey or
> | davem.
> |
> | I simply don't grok those pages. I also note caveats about
> | incompleteness, and recommendation to use libnetlink, which is also not
> | documented much.
>
> Hi Brad,
>
> There was an announcement earlier today of some networking
> documentation.  I think that it's still in its early stages, though.
Very early, from my review :)

> From the 'netdev' list:
>
> Hi,
>
> I am working on a doc on "Networking Subsystem" in the Linux kernel.
> This
> is part of the "Linux Kernel documentation project" (www.lkdp.tk).
And this is the wrong focus (for me). I need documentation from the userspace 
view. The documentation I need shouldn't try to explain how the kernel works, 
except as  needed to understand what to do in userspace. My concept is 
something like the tutorial I am working on for Linux HID programming 
(explaination of the various interfaces and options - see 
http://www.frogmouth.net/hid-doco/linux-hid.html for an early draft).

Not to suggest we shouldn't document the kernel mechanics - this is important 
too.

But kernel programmers need to remember that user space programmers will take 
the path of least resistance when trying to interact with the kernel. Good 
documentation and easy-to-copy examples beats a slightly more powerful API 
(where slightly depends on being able to accomplish task in both APIs, even 
if it gets ugly) every time. If you want people to switch over to your new 
API/ABI, its got to be easy.
-- 
http://conf.linux.org.au. 22-25Jan2003. Perth, Australia. Birds in Black.

Re: netlink documentation (was: of ethernet names)

[Paul Menage]

>But I can't even follow enough of iproute (or zebra, which also uses netlink, 
>AFAICT) to figure out how to do basic stuff like a list of configured 
>networking devices, or set the default route.

E.g. to get the list of devices (untested, lacking error checking, etc), 
use something like:

    struct {
	struct nlmsghdr  hdr;
	struct ifinfomsg info;
    } msg;
    struct sockaddr_nl addr;

    /* Create and bind the netlink socket */
    netlink_sk = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_ROUTE);

    memset(&addr, 0, sizeof(addr));
    
    addr.nl_family = AF_NETLINK;
    addr.nl_pid    = getpid();
    addr.nl_groups = RTMGRP_IPV4_IFADDR;

    bind(netlink_sk, (struct sockaddr *)&addr, sizeof(addr)));

    /* Build the netlink request */
    memset(&msg, 0, sizeof(msg));
    msg.hdr.nlmsg_len   = sizeof(msg);
    msg.hdr.nlmsg_type  = RTM_GETLINK;
    msg.hdr.nlmsg_flags = NLM_F_REQUEST | NLM_F_MATCH;
    msg.hdr.nlmsg_pid   = getpid();
    msg.hdr.nlmsg_seq   = 0;

    msg.info.ifi_family = AF_UNSPEC;
    msg.info.ifi_type   = 0;
    msg.info.ifi_index  = 0;
    msg.info.ifi_change = -1;

    /* Send the message */
    send(netlink_sk, &msg, msg.hdr.nlmsg_len, 0);

    /* Loop, as we might get replies spread over several packets */
    while((bytes = recv(netlink_sk, replybuf, sizeof(replybuf), 0))) {
        struct nlmsghdr *hdr = (struct nlmsghdr *)replybuf;

        if(hdr->nlmsg_type == NLMSG_DONE) {
            break;
        }
	
	/* Loop over the messages in this packet */
        while(bytes) {
            int len = hdr->nlmsg_len;

            struct ifinfomsg *info = NLMSG_DATA(hdr);
            struct rtattr *rta = IFLA_RTA(info);    
            
            
            len -= NLMSG_LENGTH(sizeof(*info));
            
            /* Loop over the attributes in this message */
            while(RTA_OK(rta, len)) {
                switch(rta->rta_type) {
                case IFLA_IFNAME:
                    printf("Found device %s - %u\n", 
                           RTA_DATA(rta), info->ifi_index);

                    break;
                }
                
                rta = RTA_NEXT(rta, len);
            }
            
            bytes -= hdr->nlmsg_len;
            ((void *)hdr) += hdr->nlmsg_len;
        }
        
    }

Paul

Re: netlink documentation (was: of ethernet names)

[Brad Hards]

On Tue, 11 Jun 2002 10:06, Paul Menage wrote:
> >But I can't even follow enough of iproute (or zebra, which also uses
> > netlink, AFAICT) to figure out how to do basic stuff like a list of
> > configured networking devices, or set the default route.
>
> E.g. to get the list of devices (untested, lacking error checking, etc),
> use something like:
Bingo! I think I see what is happening, just by inspection. Beverage of choice 
when we meet up!

Now, how did you get this? Just familarity with the API built up over time, 
pure zen, or copied from some other app?

A couple of days of playing with this, I might even be able to write the 
tutorial myself :)

Brad
-- 
http://conf.linux.org.au. 22-25Jan2003. Perth, Australia. Birds in Black.

Re: netlink documentation (was: of ethernet names)

[Paul Menage]

>Now, how did you get this? Just familarity with the API built up over time, 
>pure zen, or copied from some other app?

A combination of the man pages, the iproute source, and the kernel 
source. It's all pretty straightforward once you get a working skeleton.

Paul

Re: of ethernet names (was [PATCH] Futex Asynchronous

[Andi Kleen]

Brad Hards <bhards@bigpond.net.au> writes:

> I simply don't grok those pages. I also note caveats about incompleteness, and 
> recommendation to use libnetlink, which is also not documented much.

I'm sorry you don't like my manpages.

I also wrote some manpages on libnetlink, but for some reason they were never
merged in the main distribution so you can only find them in the SuSE rpm
or at http://www.firstfloor.org/~andi/man/libnetlink.3.gz (source)
or http://www.firstfloor.org/~andi/man/libnetlink.3.txt (formatted manpage)

Jamal Hadi wrote a ietf draft on netlink.  I don't know if it's publicly
available (if yes probably somewhere on ftp.ietf.org) it describes quite 
a lot of basic concepts in netlink/rtnetlink although in a bit foreign 
to linux terminology.

I also did a presentation on netlink and related topics at previous to
last year's Ottawa Linux symposium which may be also helpful. The slides
are at http://www.firstfloor.org/~andi/ols/OLSpres.htm

-Andi

Re: of ethernet names (was [PATCH] Futex Asynchronous

[Brad Hards]

On Tue, 11 Jun 2002 12:25, Andi Kleen wrote:
> Brad Hards <bhards@bigpond.net.au> writes:
> > I simply don't grok those pages. I also note caveats about
> > incompleteness, and recommendation to use libnetlink, which is also not
> > documented much.
>
> I'm sorry you don't like my manpages.
Make no mistake. I do like them. I am relying on them now. I just need more 
explanation and examples.

> I also wrote some manpages on libnetlink, but for some reason they were
> never merged in the main distribution so you can only find them in the SuSE
> rpm or at http://www.firstfloor.org/~andi/man/libnetlink.3.gz (source) or
> http://www.firstfloor.org/~andi/man/libnetlink.3.txt (formatted manpage)
Thanks.
> Jamal Hadi wrote a ietf draft on netlink.  I don't know if it's publicly
> available (if yes probably somewhere on ftp.ietf.org) it describes quite
> a lot of basic concepts in netlink/rtnetlink although in a bit foreign
> to linux terminology.
Thanks again. Found it at 
ftp://ftp.isi.edu/in-notes/search.ietf.org/internet-drafts/draft-ietf-forces-netlink-03.txt

> I also did a presentation on netlink and related topics at previous to
> last year's Ottawa Linux symposium which may be also helpful. The slides
> are at http://www.firstfloor.org/~andi/ols/OLSpres.htm
Its all coming together. Thanks again.

-- 
http://conf.linux.org.au. 22-25Jan2003. Perth, Australia. Birds in Black.

Re: of ethernet names (was [PATCH] Futex Asynchronous

[Dr. David Alan Gilbert]

Interface)
Reply-To: 
In-Reply-To: <Pine.LNX.4.44.0206091056550.13459-100000@home.transmeta.com>
X-Chocolate: 70 percent or better cocoa solids preferably
X-Operating-System: Linux/2.4.18 (i686)
X-Uptime:  19:19:16 up 1 day,  6:46,  5 users,  load average: 2.00, 2.00, 2.00

* Linus Torvalds (torvalds@transmeta.com) wrote:

> On 9 Jun 2002, Kai Henningsen wrote:
> >
> > However, I don't think that's all that important. What I'd rather see is
> > making the network devices into namespace nodes. The situation of eth0 and
> > friends, from a Unix perspective, is utterly unnatural.
> 
> But what would you _do_ with them? What would be the advantage as compared
> to the current situation?

Personally I would do away with ifconfig and replace it with 
cat in and out of device nodes; ifconfig seems to suffer about having to
know about every protocol on every device type and the kernel has to
provide interfaces for it that only it uses.

Dave
 ---------------- Have a happy GNU millennium! ----------------------   
/ Dr. David Alan Gilbert    | Running GNU/Linux on Alpha,68K| Happy  \ 
\ gro.gilbert @ treblig.org | MIPS,x86,ARM, SPARC and HP-PA | In Hex /
 \ _________________________|_____ http://www.treblig.org   |_______/

Re: of ethernet names (was [PATCH] Futex Asynchronous

[Linus Torvalds]

On Sun, 9 Jun 2002, Dr. David Alan Gilbert wrote:
>
> Personally I would do away with ifconfig and replace it with
> cat in and out of device nodes; ifconfig seems to suffer about having to
> know about every protocol on every device type and the kernel has to
> provide interfaces for it that only it uses.

Well, the kernel would have to provide the same interfaces for "cat" if
you did it that way, and it would probably take up more space and cause
more kernel bloat.

And we'd still have to have the old interfaces for backwards compatibility
for ifconfig.

Is the "magic ioctl" approach ugly? Sure. But it's fairly well contained
to just one program (ifconfig), and everybody else just uses that. I think
it's less horrible than the alternatives right now.

		Linus

Re: of ethernet names (was [PATCH] Futex Asynchronous

[Kai Henningsen]

torvalds@transmeta.com (Linus Torvalds)  wrote on 09.06.02 in <Pine.LNX.4.44.0206091130490.13751-100000@home.transmeta.com>:

> Is the "magic ioctl" approach ugly? Sure. But it's fairly well contained
> to just one program (ifconfig), and everybody else just uses that. I think
> it's less horrible than the alternatives right now.

If it *were* all contained in ifconfig, you'd be right, but that isn't  
even remotely true.

There are a *huge* number of programs that know about network interfaces.  
Apart from ifconfig, we have route, routed, iptables, ip, bootloads of  
admin scripts for device configuration, for firewalling, pppd anmd  
scripts, arp, dhcpd, portsentry, and I haven't even really scratched the  
surface here.

Sure, in the old times, when you could get Unix(tm) with or without the  
"networking option" for extra money, there wasn't much. But that has  
dramatically changed in these Internet times.


MfG Kai

Re: of ethernet names (was [PATCH] Futex Asynchronous

[Jeff Garzik]

Actually, networking is moving in the direction described --
yes, as Linus points out, we will need the magic ioctl stuff for back 
compat.
But the main way to communicate with a net device is netlink, already a 
chardev.  ifconfig actually should be updated to use netlink.

	Jeff

Re: of ethernet names (was [PATCH] Futex Asynchronous

[Brad Hards]

On Tue, 11 Jun 2002 07:44, Jeff Garzik wrote:
> Actually, networking is moving in the direction described --
> yes, as Linus points out, we will need the magic ioctl stuff for back
> compat.
> But the main way to communicate with a net device is netlink, already a
> chardev.  ifconfig actually should be updated to use netlink.
Is there any documentation on the netlink API, beyond UTSL(iproute)?
Reference would be good, but a tutorial would be ideal.

Brad

-- 
http://conf.linux.org.au. 22-25Jan2003. Perth, Australia. Birds in Black.

Re: of ethernet names (was [PATCH] Futex Asynchronous

[Jeff Garzik]

Brad Hards wrote:
> On Tue, 11 Jun 2002 07:44, Jeff Garzik wrote:
> 
>>Actually, networking is moving in the direction described --
>>yes, as Linus points out, we will need the magic ioctl stuff for back
>>compat.
>>But the main way to communicate with a net device is netlink, already a
>>chardev.  ifconfig actually should be updated to use netlink.
> 
> Is there any documentation on the netlink API, beyond UTSL(iproute)?
> Reference would be good, but a tutorial would be ideal.


I don't know of any...  Alexey/DaveM/Jamal are probably the best people 
to ask.

	Jeff

Re: of ethernet names (was [PATCH] Futex Asynchronous

[David Ford]

ifconfig is "so yesterday" as it is, ip is incredibly more capable.  I 
would gently push ifconfig into the dungeons of time.

-d

Jeff Garzik wrote:

> Actually, networking is moving in the direction described --
> yes, as Linus points out, we will need the magic ioctl stuff for back 
> compat.
> But the main way to communicate with a net device is netlink, already 
> a chardev.  ifconfig actually should be updated to use netlink.