tcp_ipv6.c

ipv6地址转换器

	retval = tcp_do_sendmsg(sk, msg);

out:
	return retval;
}

void tcp_v6_err(struct sk_buff *skb, struct ipv6hdr *hdr,
		struct inet6_skb_parm *opt,
		int type, int code, unsigned char *header, __u32 info)
{
	struct in6_addr *saddr = &hdr->saddr;
	struct in6_addr *daddr = &hdr->daddr;
	struct tcphdr *th = (struct tcphdr *)header;
	struct ipv6_pinfo *np;
	struct sock *sk;
	int err;
	struct tcp_opt *tp; 
	__u32 seq; 

	if (header + 8 > skb->tail)
		return;

	sk = tcp_v6_lookup(daddr, th->dest, saddr, th->source, skb->dev->ifindex);

	if (sk == NULL || sk->state == TCP_TIME_WAIT) {
		/* XXX: Update ICMP error count */
		return;
	}

	tp = &sk->tp_pinfo.af_tcp;
	seq = ntohl(th->seq); 
	if (sk->state != TCP_LISTEN && !between(seq, tp->snd_una, tp->snd_nxt)) {
		net_statistics.OutOfWindowIcmps++;
		return; 
	}

	np = &sk->net_pinfo.af_inet6;
	if (type == ICMPV6_PKT_TOOBIG) {
		struct dst_entry *dst = NULL;

		if (atomic_read(&sk->sock_readers))
			return;

		if (sk->state == TCP_LISTEN)
			return;

		/* icmp should have updated the destination cache entry */
		if (sk->dst_cache)
			dst = dst_check(&sk->dst_cache, np->dst_cookie);

		if (dst == NULL) {
			struct flowi fl;

			/* BUGGG_FUTURE: Again, it is not clear how
			   to handle rthdr case. Ignore this complexity
			   for now.
			 */
			fl.proto = IPPROTO_TCP;
			fl.nl_u.ip6_u.daddr = &np->daddr;
			fl.nl_u.ip6_u.saddr = &np->saddr;
			fl.oif = sk->bound_dev_if;
			fl.uli_u.ports.dport = sk->dport;
			fl.uli_u.ports.sport = sk->sport;

			dst = ip6_route_output(sk, &fl);
		} else
			dst = dst_clone(dst);

		if (dst->error) {
			sk->err_soft = -dst->error;
		} else if (tp->pmtu_cookie > dst->pmtu) {
			tcp_sync_mss(sk, dst->pmtu);
			tcp_simple_retransmit(sk);
		} /* else let the usual retransmit timer handle it */
		dst_release(dst);
		return;
	}

	icmpv6_err_convert(type, code, &err);

	/* Might be for an open_request */
	switch (sk->state) {
		struct open_request *req, *prev;
		struct ipv6hdr hd;
	case TCP_LISTEN:
		if (atomic_read(&sk->sock_readers)) {
			net_statistics.LockDroppedIcmps++;
			 /* If too many ICMPs get dropped on busy
			  * servers this needs to be solved differently.
			  */
 			return;
		}

		/* Grrrr - fix this later. */
		ipv6_addr_copy(&hd.saddr, saddr);
		ipv6_addr_copy(&hd.daddr, daddr); 
		req = tcp_v6_search_req(tp, &hd, th, tcp_v6_iif(skb), &prev);
		if (!req)
			return;
		if (seq != req->snt_isn) {
			net_statistics.OutOfWindowIcmps++;
			return;
		}
		if (req->sk) {
			sk = req->sk; /* report error in accept */
		} else {
			tp->syn_backlog--;
			tcp_synq_unlink(tp, req, prev);
			req->class->destructor(req);
			tcp_openreq_free(req);
		}
		/* FALL THROUGH */ 
	case TCP_SYN_SENT:
	case TCP_SYN_RECV:  /* Cannot happen */ 
		tcp_statistics.TcpAttemptFails++;
		sk->err = err;
		sk->zapped = 1;
		mb();
		sk->error_report(sk);
		return;
	}

	if (np->recverr) {
		/* This code isn't serialized with the socket code */
		/* ANK (980927) ... which is harmless now,
		   sk->err's may be safely lost.
		 */
		sk->err = err;
		mb();
		sk->error_report(sk);
	} else {
		sk->err_soft = err;
		mb();
	}
}


static void tcp_v6_send_synack(struct sock *sk, struct open_request *req)
{
	struct sk_buff * skb;
	struct dst_entry *dst;
	struct ipv6_txoptions *opt = NULL;
	struct flowi fl;
	int mss;

	fl.proto = IPPROTO_TCP;
	fl.nl_u.ip6_u.daddr = &req->af.v6_req.rmt_addr;
	fl.nl_u.ip6_u.saddr = &req->af.v6_req.loc_addr;
	fl.fl6_flowlabel = 0;
	fl.oif = req->af.v6_req.iif;
	fl.uli_u.ports.dport = req->rmt_port;
	fl.uli_u.ports.sport = sk->sport;

	opt = sk->net_pinfo.af_inet6.opt;
	if (opt == NULL &&
	    sk->net_pinfo.af_inet6.rxopt.bits.srcrt == 2 &&
	    req->af.v6_req.pktopts) {
		struct sk_buff *pktopts = req->af.v6_req.pktopts;
		struct inet6_skb_parm *rxopt = (struct inet6_skb_parm *)pktopts->cb;
		if (rxopt->srcrt)
			opt = ipv6_invert_rthdr(sk, (struct ipv6_rt_hdr*)(pktopts->nh.raw + rxopt->srcrt));
	}

	if (opt && opt->srcrt) {
		struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
		fl.nl_u.ip6_u.daddr = rt0->addr;
	}

	dst = ip6_route_output(sk, &fl);
	if (dst->error)
		goto done;

	mss = dst->pmtu - sizeof(struct ipv6hdr) - sizeof(struct tcphdr);

	skb = tcp_make_synack(sk, dst, req, mss);
	if (skb) {
		struct tcphdr *th = skb->h.th;

		th->check = tcp_v6_check(th, skb->len,
					 &req->af.v6_req.loc_addr, &req->af.v6_req.rmt_addr,
					 csum_partial((char *)th, skb->len, skb->csum));

		fl.nl_u.ip6_u.daddr = &req->af.v6_req.rmt_addr;
		ip6_xmit(sk, skb, &fl, opt);
	}

done:
	dst_release(dst);
        if (opt && opt != sk->net_pinfo.af_inet6.opt)
		sock_kfree_s(sk, opt, opt->tot_len);
}

static void tcp_v6_or_free(struct open_request *req)
{
	if (req->af.v6_req.pktopts) {
		kfree_skb(req->af.v6_req.pktopts);
		req->af.v6_req.pktopts = NULL;
	}
}

static struct or_calltable or_ipv6 = {
	tcp_v6_send_synack,
	tcp_v6_or_free,
	tcp_v6_send_reset
};

static int ipv6_opt_accepted(struct sock *sk, struct sk_buff *skb)
{
	struct inet6_skb_parm *opt = (struct inet6_skb_parm *)skb->cb;

	if (sk->net_pinfo.af_inet6.rxopt.all) {
		if ((opt->hop && sk->net_pinfo.af_inet6.rxopt.bits.hopopts) ||
		    ((IPV6_FLOWINFO_MASK&*(u32*)skb->nh.raw) &&
		     sk->net_pinfo.af_inet6.rxopt.bits.rxflow) ||
		    (opt->srcrt && sk->net_pinfo.af_inet6.rxopt.bits.srcrt) ||
		    ((opt->dst1 || opt->dst0) && sk->net_pinfo.af_inet6.rxopt.bits.dstopts))
			return 1;
	}
	return 0;
}


#define BACKLOG(sk) ((sk)->tp_pinfo.af_tcp.syn_backlog) /* lvalue! */
#define BACKLOGMAX(sk) sysctl_max_syn_backlog

/* FIXME: this is substantially similar to the ipv4 code.
 * Can some kind of merge be done? -- erics
 */
static int tcp_v6_conn_request(struct sock *sk, struct sk_buff *skb, __u32 isn)
{
	struct tcp_opt tp;
	struct open_request *req;
	
	/* If the socket is dead, don't accept the connection.	*/
	if (sk->dead) {
		SOCK_DEBUG(sk, "Reset on %p: Connect on dead socket.\n", sk);
		tcp_statistics.TcpAttemptFails++;
		return -ENOTCONN;
	}

	if (skb->protocol == __constant_htons(ETH_P_IP))
		return tcp_v4_conn_request(sk, skb, isn);

	/* FIXME: do the same check for anycast */
	if (ipv6_addr_is_multicast(&skb->nh.ipv6h->daddr))
		goto drop; 

	if (isn == 0) 
		isn = tcp_v6_init_sequence(sk,skb);

	/*
	 *	There are no SYN attacks on IPv6, yet...	
	 */
	if (BACKLOG(sk) >= BACKLOGMAX(sk)) {
		(void)(net_ratelimit() && 
		       printk(KERN_INFO "droping syn ack:%d max:%d\n",
			       BACKLOG(sk), BACKLOGMAX(sk)));
		goto drop;		
	}

	req = tcp_openreq_alloc();
	if (req == NULL) {
		goto drop;
	}

	BACKLOG(sk)++;

	req->rcv_wnd = 0;		/* So that tcp_send_synack() knows! */

	req->rcv_isn = TCP_SKB_CB(skb)->seq;
	req->snt_isn = isn;
	tp.tstamp_ok = tp.sack_ok = tp.wscale_ok = tp.snd_wscale = 0;
	tp.mss_clamp = 65535;
	tcp_parse_options(NULL, skb->h.th, &tp, 0);
	if (tp.mss_clamp == 65535)
		tp.mss_clamp = 576 - sizeof(struct ipv6hdr) - sizeof(struct iphdr);
	if (sk->tp_pinfo.af_tcp.user_mss && sk->tp_pinfo.af_tcp.user_mss < tp.mss_clamp)
		tp.mss_clamp = sk->tp_pinfo.af_tcp.user_mss;

        req->mss = tp.mss_clamp;
	if (tp.saw_tstamp)
                req->ts_recent = tp.rcv_tsval;
        req->tstamp_ok = tp.tstamp_ok;
	req->sack_ok = tp.sack_ok;
        req->snd_wscale = tp.snd_wscale;
        req->wscale_ok = tp.wscale_ok;
	req->rmt_port = skb->h.th->source;
	ipv6_addr_copy(&req->af.v6_req.rmt_addr, &skb->nh.ipv6h->saddr);
	ipv6_addr_copy(&req->af.v6_req.loc_addr, &skb->nh.ipv6h->daddr);
	req->af.v6_req.pktopts = NULL;
	if (ipv6_opt_accepted(sk, skb)) {
		atomic_inc(&skb->users);
		req->af.v6_req.pktopts = skb;
	}
	req->af.v6_req.iif = sk->bound_dev_if;

	/* So that link locals have meaning */
	if (!sk->bound_dev_if && ipv6_addr_type(&req->af.v6_req.rmt_addr)&IPV6_ADDR_LINKLOCAL)
		req->af.v6_req.iif = tcp_v6_iif(skb);

	req->class = &or_ipv6;
	req->retrans = 0;
	req->sk = NULL;

	tcp_v6_send_synack(sk, req);

	req->expires = jiffies + TCP_TIMEOUT_INIT;
	tcp_inc_slow_timer(TCP_SLT_SYNACK);
	tcp_synq_queue(&sk->tp_pinfo.af_tcp, req);	

	return 0;

drop:
	tcp_statistics.TcpAttemptFails++;
	return 0; /* don't send reset */
}

static void tcp_v6_send_check(struct sock *sk, struct tcphdr *th, int len, 
			      struct sk_buff *skb)
{
	struct ipv6_pinfo *np = &sk->net_pinfo.af_inet6;
	th->check = 0;
	
	th->check = csum_ipv6_magic(&np->saddr, &np->daddr, len, IPPROTO_TCP, 
				    csum_partial((char *)th, th->doff<<2, 
						 skb->csum));
}

static struct sock * tcp_v6_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
					  struct open_request *req,
					  struct dst_entry *dst)
{
	struct ipv6_pinfo *np;
	struct flowi fl;
	struct tcp_opt *newtp;
	struct sock *newsk;
	struct ipv6_txoptions *opt;

	if (skb->protocol == __constant_htons(ETH_P_IP)) {
		/*
		 *	v6 mapped
		 */

		newsk = tcp_v4_syn_recv_sock(sk, skb, req, dst);

		if (newsk == NULL) 
			return NULL;
	
		np = &newsk->net_pinfo.af_inet6;

		ipv6_addr_set(&np->daddr, 0, 0, __constant_htonl(0x0000FFFF),
			      newsk->daddr);

		ipv6_addr_set(&np->saddr, 0, 0, __constant_htonl(0x0000FFFF),
			      newsk->saddr);

		ipv6_addr_copy(&np->rcv_saddr, &np->saddr);

		newsk->tp_pinfo.af_tcp.af_specific = &ipv6_mapped;
		newsk->backlog_rcv = tcp_v4_do_rcv;
		newsk->net_pinfo.af_inet6.pktoptions = NULL;
		newsk->net_pinfo.af_inet6.opt = NULL;

		/* It is tricky place. Until this moment IPv4 tcp
		   worked with IPv6 af_tcp.af_specific.
		   Sync it now.
		 */
		tcp_sync_mss(newsk, newsk->tp_pinfo.af_tcp.pmtu_cookie);

		return newsk;
	}

	opt = sk->net_pinfo.af_inet6.opt;

	if (sk->ack_backlog > sk->max_ack_backlog)
		goto out;

	if (sk->net_pinfo.af_inet6.rxopt.bits.srcrt == 2 &&
	    opt == NULL && req->af.v6_req.pktopts) {
		struct inet6_skb_parm *rxopt = (struct inet6_skb_parm *)req->af.v6_req.pktopts->cb;
		if (rxopt->srcrt)
			opt = ipv6_invert_rthdr(sk, (struct ipv6_rt_hdr*)(req->af.v6_req.pktopts->nh.raw+rxopt->srcrt));
	}

	if (dst == NULL) {
		fl.proto = IPPROTO_TCP;
		fl.nl_u.ip6_u.daddr = &req->af.v6_req.rmt_addr;
		if (opt && opt->srcrt) {
			struct rt0_hdr *rt0 = (struct rt0_hdr *) opt->srcrt;
			fl.nl_u.ip6_u.daddr = rt0->addr;
		}
		fl.nl_u.ip6_u.saddr = &req->af.v6_req.loc_addr;
		fl.fl6_flowlabel = 0;
		fl.oif = sk->bound_dev_if;
		fl.uli_u.ports.dport = req->rmt_port;
		fl.uli_u.ports.sport = sk->sport;

		dst = ip6_route_output(sk, &fl);
	}

	if (dst->error)
		goto out;

	sk->tp_pinfo.af_tcp.syn_backlog--;
	sk->ack_backlog++;

	newsk = tcp_create_openreq_child(sk, req, skb);
	if (newsk == NULL)
		goto out;

	ip6_dst_store(newsk, dst, NULL);

	newtp = &(newsk->tp_pinfo.af_tcp);

	np = &newsk->net_pinfo.af_inet6;
	ipv6_addr_copy(&np->daddr, &req->af.v6_req.rmt_addr);
	ipv6_addr_copy(&np->saddr, &req->af.v6_req.loc_addr);
	ipv6_addr_copy(&np->rcv_saddr, &req->af.v6_req.loc_addr);
	newsk->bound_dev_if = req->af.v6_req.iif;

	/* Now IPv6 options... 

	   First: no IPv4 options.
	 */
	newsk->opt = NULL;

	/* Clone RX bits */
	np->rxopt.all = sk->net_pinfo.af_inet6.rxopt.all;

	/* Clone pktoptions received with SYN */
	np->pktoptions = req->af.v6_req.pktopts;
	if (np->pktoptions)
		atomic_inc(&np->pktoptions->users);
	np->opt = NULL;

	/* Clone native IPv6 options from listening socket (if any)

	   Yes, keeping reference count would be much more clever,
	   but we make one more one thing there: reattach optmem
	   to newsk.
	 */
	if (opt) {
		np->opt = ipv6_dup_options(newsk, opt);
		if (opt != sk->net_pinfo.af_inet6.opt)
			sock_kfree_s(sk, opt, opt->tot_len);
	}

	newtp->ext_header_len = 0;
	if (np->opt)
		newtp->ext_header_len = np->opt->opt_nflen + np->opt->opt_flen;

	tcp_sync_mss(newsk, dst->pmtu);
	newtp->rcv_mss = newtp->mss_clamp;

	newsk->daddr	= LOOPBACK4_IPV6;
	newsk->saddr	= LOOPBACK4_IPV6;
	newsk->rcv_saddr= LOOPBACK4_IPV6;

	newsk->prot->hash(newsk);
	tcp_inherit_port(sk, newsk);
	add_to_prot_sklist(newsk);
	sk->data_ready(sk, 0); /* Deliver SIGIO */ 

	return newsk;

out:
	if (opt && opt != sk->net_pinfo.af_inet6.opt)
		sock_kfree_s(sk, opt, opt->tot_len);
	dst_release(dst);
	return NULL;
}

static void tcp_v6_send_reset(struct sk_buff *skb)
{
	struct tcphdr *th = skb->h.th, *t1; 
	struct sk_buff *buff;
	struct flowi fl;

	if (th->rst)
		return;

	if (ipv6_addr_is_multicast(&skb->nh.ipv6h->daddr))
		return; 

	/*
	 * We need to grab some memory, and put together an RST,
	 * and then put it into the queue to be sent.
	 */

	buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr), GFP_ATOMIC);
	if (buff == NULL) 
	  	return;

	skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr));

	t1 = (struct tcphdr *) skb_push(buff,sizeof(struct tcphdr));

	/* Swap the send and the receive. */
	memset(t1, 0, sizeof(*t1));
	t1->dest = th->source;
	t1->source = th->dest;
	t1->doff = sizeof(*t1)/4;
	t1->rst = 1;
  
	if(th->ack) {
	  	t1->seq = th->ack_seq;
	} else {
		t1->ack = 1;
	  	if(!th->syn)
			t1->ack_seq = th->seq;
		else
			t1->ack_seq = htonl(ntohl(th->seq)+1);
	}

	buff->csum = csum_partial((char *)t1, sizeof(*t1), 0);

	fl.nl_u.ip6_u.daddr = &skb->nh.ipv6h->saddr;
	fl.nl_u.ip6_u.saddr = &skb->nh.ipv6h->daddr;
	fl.fl6_flowlabel = 0;

	t1->check = csum_ipv6_magic(fl.nl_u.ip6_u.saddr,
				    fl.nl_u.ip6_u.daddr, 
				    sizeof(*t1), IPPROTO_TCP,
				    buff->csum);

	fl.proto = IPPROTO_TCP;
	fl.oif = tcp_v6_iif(skb);
	fl.uli_u.ports.dport = t1->dest;
	fl.uli_u.ports.sport = t1->source;

	/* sk = NULL, but it is safe for now. RST socket required. */
	buff->dst = ip6_route_output(NULL, &fl);

	if (buff->dst->error == 0) {
		ip6_xmit(NULL, buff, &fl, NULL);
		tcp_statistics.TcpOutSegs++;
		tcp_statistics.TcpOutRsts++;
		return;
	}

	kfree_skb(buff);
}

static void tcp_v6_send_ack(struct sk_buff *skb, __u32 seq, __u32 ack, __u16 window)
{
	struct tcphdr *th = skb->h.th, *t1; 
	struct sk_buff *buff;
	struct flowi fl;

	buff = alloc_skb(MAX_HEADER + sizeof(struct ipv6hdr), GFP_ATOMIC);
	if (buff == NULL) 
	  	return;

	skb_reserve(buff, MAX_HEADER + sizeof(struct ipv6hdr));

	t1 = (struct tcphdr *) skb_push(buff,sizeof(struct tcphdr));

	/* Swap the send and the receive. */
	memset(t1, 0, sizeof(*t1));
	t1->dest = th->source;
	t1->source = th->dest;
	t1->doff = sizeof(*t1)/4;
	t1->ack = 1;
	t1->seq = seq;
	t1->ack_seq = ack; 

	t1->window = htons(window);

	buff->csum = csum_partial((char *)t1, sizeof(*t1), 0);

	fl.nl_u.ip6_u.daddr = &skb->nh.ipv6h->saddr;
	fl.nl_u.ip6_u.saddr = &skb->nh.ipv6h->daddr;
	fl.fl6_flowlabel = 0;

	t1->check = csum_ipv6_magic(fl.nl_u.ip6_u.saddr,
				    fl.nl_u.ip6_u.daddr, 
				    sizeof(*t1), IPPROTO_TCP,
				    buff->csum);

	fl.proto = IPPROTO_TCP;