tcp_ipv4.c

GNU Hurd 源代码

next:
	if(firstpass--) {
		struct tcp_bind_bucket *tb = tcp_bhash[tcp_bhashfn(hpnum)];
		for( ; (tb && tb->port != hpnum); tb = tb->next)
			;
		if(tb) {
			s = tb->owners;
			goto pass2;
		}
	}
gotit:
	return result;
}
#endif /* CONFIG_IP_TRANSPARENT_PROXY */

static inline __u32 tcp_v4_init_sequence(struct sock *sk, struct sk_buff *skb)
{
	return secure_tcp_sequence_number(skb->nh.iph->daddr,
					  skb->nh.iph->saddr,
					  skb->h.th->dest,
					  skb->h.th->source);
}

/* Check that a TCP address is unique, don't allow multiple
 * connects to/from the same address.  Actually we can optimize
 * quite a bit, since the socket about to connect is still
 * in TCP_CLOSE, a tcp_bind_bucket for the local port he will
 * use will exist, with a NULL owners list.  So check for that.
 * The good_socknum and verify_bind scheme we use makes this
 * work.
 */
static int tcp_v4_unique_address(struct sock *sk)
{
	struct tcp_bind_bucket *tb;
	unsigned short snum = sk->num;
	int retval = 1;

	/* Freeze the hash while we snoop around. */
	SOCKHASH_LOCK();
	tb = tcp_bhash[tcp_bhashfn(snum)];
	for(; tb; tb = tb->next) {
		if(tb->port == snum && tb->owners != NULL) {
			/* Almost certainly the re-use port case, search the real hashes
			 * so it actually scales.
			 */
			sk = __tcp_v4_lookup(NULL, sk->daddr, sk->dport,
					     sk->rcv_saddr, htons(snum),
					     sk->bound_dev_if);
			if((sk != NULL) && (sk->state != TCP_LISTEN))
				retval = 0;
			break;
		}
	}
	SOCKHASH_UNLOCK();
	return retval;
}

/* This will initiate an outgoing connection. */
int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len)
{
	struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
	struct sockaddr_in *usin = (struct sockaddr_in *) uaddr;
	struct sk_buff *buff;
	struct rtable *rt;
	u32 daddr, nexthop;
	int tmp;

	if (sk->state != TCP_CLOSE) 
		return(-EISCONN);

	/* Don't allow a double connect. */
	if (sk->daddr)
		return -EINVAL;

	if (addr_len < sizeof(struct sockaddr_in))
		return(-EINVAL);

	if (usin->sin_family != AF_INET) {
		static int complained;
		if (usin->sin_family)
			return(-EAFNOSUPPORT);
		if (!complained++)
			printk(KERN_DEBUG "%s forgot to set AF_INET in %s\n", current->comm, __FUNCTION__);
	}

	nexthop = daddr = usin->sin_addr.s_addr;
	if (sk->opt && sk->opt->srr) {
		if (daddr == 0)
			return -EINVAL;
		nexthop = sk->opt->faddr;
	}

	tmp = ip_route_connect(&rt, nexthop, sk->saddr,
			       RT_TOS(sk->ip_tos)|RTO_CONN|sk->localroute, sk->bound_dev_if);
	if (tmp < 0)
		return tmp;

	if (rt->rt_flags&(RTCF_MULTICAST|RTCF_BROADCAST)) {
		ip_rt_put(rt);
		return -ENETUNREACH;
	}

	dst_release(xchg(&sk->dst_cache, rt));

	buff = sock_wmalloc(sk, (MAX_HEADER + sk->prot->max_header),
			    0, GFP_KERNEL);

	if (buff == NULL)
		return -ENOBUFS;

	/* Socket has no identity, so lock_sock() is useless.  Also
	 * since state==TCP_CLOSE (checked above) the socket cannot
	 * possibly be in the hashes.  TCP hash locking is only
	 * needed while checking quickly for a unique address.
	 * However, the socket does need to be (and is) locked
	 * in tcp_connect().
	 * Perhaps this addresses all of ANK's concerns. 8-)  -DaveM
	 */
	sk->dport = usin->sin_port;
	sk->daddr = rt->rt_dst;
	if (sk->opt && sk->opt->srr)
		sk->daddr = daddr;
	if (!sk->saddr)
		sk->saddr = rt->rt_src;
	sk->rcv_saddr = sk->saddr;

	if (!tcp_v4_unique_address(sk)) {
		kfree_skb(buff);
		sk->daddr = 0;
		return -EADDRNOTAVAIL;
	}

	tp->write_seq = secure_tcp_sequence_number(sk->saddr, sk->daddr,
						   sk->sport, usin->sin_port);

	tp->ext_header_len = 0;
	if (sk->opt)
		tp->ext_header_len = sk->opt->optlen;

	/* Reset mss clamp */
	tp->mss_clamp = ~0;

	if (!ip_dont_fragment(sk, &rt->u.dst) &&
	    rt->u.dst.pmtu > 576 && rt->rt_dst != rt->rt_gateway) {
		/* Clamp mss at maximum of 536 and user_mss.
		   Probably, user ordered to override tiny segment size
		   in gatewayed case.
		 */
		tp->mss_clamp = max(tp->user_mss, 536);
	}

	tcp_connect(sk, buff, rt->u.dst.pmtu);
	return 0;
}

static int tcp_v4_sendmsg(struct sock *sk, struct msghdr *msg, int len)
{
	int retval = -EINVAL;

	/* Do sanity checking for sendmsg/sendto/send. */
	if (msg->msg_flags & ~(MSG_OOB|MSG_DONTROUTE|MSG_DONTWAIT|MSG_NOSIGNAL))
		goto out;
	if (msg->msg_name) {
		struct sockaddr_in *addr=(struct sockaddr_in *)msg->msg_name;

		if (msg->msg_namelen < sizeof(*addr))
			goto out;
		if (addr->sin_family && addr->sin_family != AF_INET)
			goto out;
		retval = -ENOTCONN;
		if(sk->state == TCP_CLOSE)
			goto out;
		retval = -EISCONN;
		if (addr->sin_port != sk->dport)
			goto out;
		if (addr->sin_addr.s_addr != sk->daddr)
			goto out;
	}
	retval = tcp_do_sendmsg(sk, msg);

out:
	return retval;
}


/*
 * Do a linear search in the socket open_request list. 
 * This should be replaced with a global hash table.
 */
static struct open_request *tcp_v4_search_req(struct tcp_opt *tp, 
					      struct iphdr *iph,
					      struct tcphdr *th,
					      struct open_request **prevp)
{
	struct open_request *req, *prev;  
	__u16 rport = th->source; 

	/*	assumption: the socket is not in use.
	 *	as we checked the user count on tcp_rcv and we're
	 *	running from a soft interrupt.
	 */
	prev = (struct open_request *) (&tp->syn_wait_queue); 
	for (req = prev->dl_next; req; req = req->dl_next) {
		if (req->af.v4_req.rmt_addr == iph->saddr &&
		    req->af.v4_req.loc_addr == iph->daddr &&
		    req->rmt_port == rport
#ifdef CONFIG_IP_TRANSPARENT_PROXY
		    && req->lcl_port == th->dest
#endif
		    ) {
			*prevp = prev; 
			return req; 
		}
		prev = req; 
	}
	return NULL; 
}


/* 
 * This routine does path mtu discovery as defined in RFC1191.
 */
static inline void do_pmtu_discovery(struct sock *sk, struct iphdr *ip, unsigned mtu)
{
	struct tcp_opt *tp = &sk->tp_pinfo.af_tcp;

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

	/* Don't interested in TCP_LISTEN and open_requests (SYN-ACKs
	 * send out by Linux are always <576bytes so they should go through
	 * unfragmented).
	 */
	if (sk->state == TCP_LISTEN)
		return; 

	/* We don't check in the destentry if pmtu discovery is forbidden
	 * on this route. We just assume that no packet_to_big packets
	 * are send back when pmtu discovery is not active.
     	 * There is a small race when the user changes this flag in the
	 * route, but I think that's acceptable.
	 */
	if (sk->dst_cache == NULL)
		return;
	ip_rt_update_pmtu(sk->dst_cache, mtu);
	if (sk->ip_pmtudisc != IP_PMTUDISC_DONT &&
	    tp->pmtu_cookie > sk->dst_cache->pmtu) {
		tcp_sync_mss(sk, sk->dst_cache->pmtu);

		/* Resend the TCP packet because it's  
		 * clear that the old packet has been
		 * dropped. This is the new "fast" path mtu
		 * discovery.
		 */
		tcp_simple_retransmit(sk);
	} /* else let the usual retransmit timer handle it */
}

/*
 * This routine is called by the ICMP module when it gets some
 * sort of error condition.  If err < 0 then the socket should
 * be closed and the error returned to the user.  If err > 0
 * it's just the icmp type << 8 | icmp code.  After adjustment
 * header points to the first 8 bytes of the tcp header.  We need
 * to find the appropriate port.
 *
 * The locking strategy used here is very "optimistic". When
 * someone else accesses the socket the ICMP is just dropped
 * and for some paths there is no check at all.
 * A more general error queue to queue errors for later handling
 * is probably better.
 *
 * sk->err and sk->err_soft should be atomic_t.
 */

void tcp_v4_err(struct sk_buff *skb, unsigned char *dp, int len)
{
	struct iphdr *iph = (struct iphdr*)dp;
	struct tcphdr *th; 
	struct tcp_opt *tp;
	int type = skb->h.icmph->type;
	int code = skb->h.icmph->code;
#if ICMP_MIN_LENGTH < 14
	int no_flags = 0;
#else
#define no_flags 0
#endif
	struct sock *sk;
	__u32 seq;
	int err;

	if (len < (iph->ihl << 2) + ICMP_MIN_LENGTH) { 
		icmp_statistics.IcmpInErrors++; 
		return;
	}
#if ICMP_MIN_LENGTH < 14
	if (len < (iph->ihl << 2) + 14)
		no_flags = 1;
#endif

	th = (struct tcphdr*)(dp+(iph->ihl<<2));

	sk = tcp_v4_lookup(iph->daddr, th->dest, iph->saddr, th->source, skb->dev->ifindex);
	if (sk == NULL || sk->state == TCP_TIME_WAIT) {
		icmp_statistics.IcmpInErrors++;
		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; 
	}

	switch (type) {
	case ICMP_SOURCE_QUENCH:
#ifndef OLD_SOURCE_QUENCH /* This is deprecated */
		tp->snd_ssthresh = tcp_recalc_ssthresh(tp);
		tp->snd_cwnd = tp->snd_ssthresh;
		tp->snd_cwnd_cnt = 0;
		tp->high_seq = tp->snd_nxt;
#endif
		return;
	case ICMP_PARAMETERPROB:
		err = EPROTO;
		break; 
	case ICMP_DEST_UNREACH:
		if (code > NR_ICMP_UNREACH)
			return;

		if (code == ICMP_FRAG_NEEDED) { /* PMTU discovery (RFC1191) */
			do_pmtu_discovery(sk, iph, ntohs(skb->h.icmph->un.frag.mtu));
			return;
		}

		err = icmp_err_convert[code].errno;
		break;
	case ICMP_TIME_EXCEEDED:
		err = EHOSTUNREACH;
		break;
	default:
		return;
	}

	switch (sk->state) {
		struct open_request *req, *prev;
	case TCP_LISTEN:
		/* Prevent race conditions with accept() - 
		 * ICMP is unreliable. 
		 */
		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;
		}

		/* The final ACK of the handshake should be already 
		 * handled in the new socket context, not here.
		 * Strictly speaking - an ICMP error for the final
		 * ACK should set the opening flag, but that is too
		 * complicated right now. 
		 */ 
		if (!no_flags && !th->syn && !th->ack)
			return;

		req = tcp_v4_search_req(tp, iph, th, &prev); 
		if (!req)
			return;
		if (seq != req->snt_isn) {
			net_statistics.OutOfWindowIcmps++;
			return;
		}
		if (req->sk) {	
			/* 
			 * Already in ESTABLISHED and a big socket is created,
			 * set error code there.
			 * The error will _not_ be reported in the accept(),
			 * but only with the next operation on the socket after
			 * accept. 
			 */
			sk = req->sk;
		} else {
			/* 
			 * Still in SYN_RECV, just remove it silently.
			 * There is no good way to pass the error to the newly
			 * created socket, and POSIX does not want network
			 * errors returned from accept(). 
			 */ 
			tp->syn_backlog--;
			tcp_synq_unlink(tp, req, prev);
			req->class->destructor(req);
			tcp_openreq_free(req);
			return; 
		}
		break;
	case TCP_SYN_SENT:
	case TCP_SYN_RECV:  /* Cannot happen */ 
		if (!no_flags && !th->syn)
			return;
		tcp_statistics.TcpAttemptFails++;
		sk->err = err;
		sk->zapped = 1;
		mb();
		sk->error_report(sk);
		return;
	}

	/* If we've already connected we will keep trying
	 * until we time out, or the user gives up.
	 *
	 * rfc1122 4.2.3.9 allows to consider as hard errors
	 * only PROTO_UNREACH and PORT_UNREACH (well, FRAG_FAILED too,
	 * but it is obsoleted by pmtu discovery).
	 *
	 * Note, that in modern internet, where routing is unreliable
	 * and in each dark corner broken firewalls sit, sending random
	 * errors ordered by their masters even this two messages finally lose
	 * their original sense (even Linux sends invalid PORT_UNREACHs)
	 *
	 * Now we are in compliance with RFCs.
	 *							--ANK (980905)
	 */

	if (sk->ip_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);		/* Wake people up to see the error (see connect in sock.c) */
	} else	{ /* Only an error on timeout */
		sk->err_soft = err;
		mb(); 
	}
}

/* This routine computes an IPv4 TCP checksum. */
void tcp_v4_send_check(struct sock *sk, struct tcphdr *th, int len, 
		       struct sk_buff *skb)
{
	th->check = 0;
	th->check = tcp_v4_check(th, len, sk->saddr, sk->daddr,
				 csum_partial((char *)th, th->doff<<2, skb->csum));
}

/*
 *	This routine will send an RST to the other tcp.
 *
 *	Someone asks: why I NEVER use socket parameters (TOS, TTL etc.)
 *		      for reset.
 *	Answer: if a packet caused RST, it is not for a socket
 *		existing in our system, if it is matched to a socket,
 *		it is just duplicate segment or bug in other side's TCP.
 *		So that we build reply only basing on parameters
 *		arrived with segment.
 *	Exception: precedence violation. We do not implement it in any case.
 */

static void tcp_v4_send_reset(struct sk_buff *skb)
{
	struct tcphdr *th = skb->h.th;
	struct tcphdr rth;
	struct ip_reply_arg arg;

	/* Never send a reset in response to a reset. */
	if (th->rst)
		return;

	if (((struct rtable*)skb->dst)->rt_type != RTN_LOCAL) {
#ifdef CONFIG_IP_TRANSPARENT_PROXY
		if (((struct rtable*)skb->dst)->rt_type == RTN_UNICAST)
			icmp_send(skb, ICMP_DEST_UNREACH,
				  ICMP_PORT_UNREACH, 0);
#endif
		return;
	}

	/* Swap the send and the receive. */
	memset(&rth, 0, sizeof(struct tcphdr)); 
	rth.dest = th->source;
	rth.source = th->dest; 
	rth.doff = sizeof(struct tcphdr)/4;
	rth.rst = 1;

	if (th->ack) {
		rth.seq = th->ack_seq;
	} else {
		rth.ack = 1;
		rth.ack_seq = th->syn ? htonl(ntohl(th->seq)+1) : th->seq;
	}

	memset(&arg, 0, sizeof arg); 
	arg.iov[0].iov_base = (unsigned char *)&rth; 
	arg.iov[0].iov_len  = sizeof rth;
	arg.csum = csum_tcpudp_nofold(skb->nh.iph->daddr, 
				      skb->nh.iph->saddr, /*XXX*/
				      sizeof(struct tcphdr),
				      IPPROTO_TCP,
				      0); 
	arg.n_iov = 1;
	arg.csumoffset = offsetof(struct tcphdr, check) / 2; 

	ip_send_reply(tcp_socket->sk, skb, &arg, sizeof rth);

	tcp_statistics.TcpOutSegs++;
	tcp_statistics.TcpOutRsts++;
}

/*