Network implementation

From: an7
Date: Mon Oct 20 2003 - 04:24:23 EST



Hi, i am a french engineer who has some work to do on
TCP/IP implementation in the linux kernel.

I have three questions on it :

1) What are the advantages of using a ring buffer for
receiving and queuing packets ?

2) in tcp_minisocks.c in linux i get, in the function
which handles segment arrival in FIN_WAIT_2 state :

if (th->syn && !before(TCP_SKB_CB(skb)->seq,
tw->rcv_nxt))
goto kill_with_rst;

Just to confirm : in the state FIN_WAIT_2 a SYN
segment is automatically killed by a RST, unless its
sequence number is superior to the waited sequence
number. But in this case , what happens ? I think the
RFC tells to send a ACK. Am i right ?

3) the function tcp_recv_skb() (tcp.c)

struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq,
u32 *off)
1383 {
1384 struct sk_buff *skb;
1385 u32 offset;
1386
1387 skb_queue_walk(&sk->receive_queue, skb) {
1388 offset = seq -
TCP_SKB_CB(skb)->seq;
1389 if (skb->h.th->syn)
1390 offset--;
1391 if (offset < skb->len ||
skb->h.th->fin) {
1392 *off = offset;
1393 return skb;
1394 }
1395 }
1396 return NULL;
1397 }


I just can't understand the role of offset (and thus,
i can't undersand why it is decremented with a syn
segment, and why we can return the skb structure when
we have an offset inferior to the skb len, or a FIN
segment)

Could you help please ?

___________________________________________________________
Do You Yahoo!? -- Une adresse @yahoo.fr gratuite et en français !
Yahoo! Mail : http://fr.mail.yahoo.com
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/