Re: Sending sk_buffs without being tied to a struct sock

Patrick McHardy <kaber@trash.net> · Mon, 13 Jan 2003 05:00:14 +0100

I don't think this is of much interest to lkml so i changed back cc to 
linux-net.

this is the basic outline i use to generate tcp packets (unimportant 
parts cut):

       struct rt_key key = { .dst ctuple->dst.ip,

                             .src 0,

                             .oif 0,

                             .tos RT_TOS(IPTOS_LOWDELAY) | RTO_CONN,

#ifdef CONFIG_IP_ROUTE_FWMARK

                             .fwmark 0

#endif

       };

       if ((ip_route_output_key(&rt, &key) != 0))

               return 1;

       dev = rt->u.dst.dev;

       skb = alloc_skb(dev->hard_header_len + sizeof(struct iphdr) +

                                 tcp_hdr_len, GFP_ATOMIC);

       if (!skb) {

               ip_rt_put(rt);

               return -1;

       }

       skb->dev = dev;

       skb->dst = &rt->u.dst;

       skb->protocol  = __constant_htons(ETH_P_IP);

       skb_reserve(skb, dev->hard_header_len);

       skb->nh.raw = skb->data;

       iph  = (struct iphdr*) skb_put(skb, sizeof(struct iphdr));

       tcph = (struct tcphdr*)skb_put(skb, tcp_hdr_len);

... fill in ip & tcp header ...

       tcph->check = iph->check = 0;

       tcph->check = tcp_v4_check(tcph, tcp_hdr_len, iph->saddr, 
iph->daddr,

                                  csum_partial((char *)tcph, 
tcp_hdr_len, 0));

       iph->check = ip_fast_csum((unsigned char *)iph, iph->ihl);

       skb->ip_summed = CHECKSUM_UNNECESSARY;

       return skb->dst->output(skb);

}

this works fine, except when you turn on memory allocation debugging a

BUG_ON somewhere is triggered, i haven't looked into this yet.

While looking over it right now i noticed another thing, i think usually 
dev->hard_header_len

is padded to 16-byte boundaries, you might want to change this.

I hope it helps ..

Patrick

Joshua Stewart wrote:

At this point, if I call skb->dst->output() what should skb->data point

to? A properly crafted IP packet?  Checksummed and all?  I shouldn't

need any layer 2 info in my skb at all to use this solution, should I?

Thanks for the help.  I think this will do the trick.

J

On Sun, 2003-01-12 at 12:28, Patrick McHardy wrote:

Joshua Stewart wrote:

I'm trying to take "hand-built" sk_buffs with little more than some data

and a dev member and push them to the NIC for transmission.  I would

like to simply give them to dev_queue_xmit.  Does anybody know what

state I should have them in before handing them to dev_queue_xmit? 
Should skb->data point to the start of a MAC header or an IP header?

Also, given an IP address in skb->nh.iph->daddr, what's the easiest way

to get the appropriate MAC address?

use ip_route_output_key() and then skb->dst->output(), that way it will work on all media

and you don't have to care about l2 resolving at all.

Patrick

-

: send the line "unsubscribe linux-net" in

the body of a message to majordomo@vger.kernel.org

More majordomo info at  http://vger.kernel.org/majordomo-info.html

-
: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html