tcp_ipv4.c

GNU Hurd 源代码

 *	Send an ACK for a socket less packet (needed for time wait) 
 *
 *  FIXME: Does not echo timestamps yet.
 *
 *  Assumes that the caller did basic address and flag checks.
 */
static void tcp_v4_send_ack(struct sk_buff *skb, __u32 seq, __u32 ack, __u16 window)
{
	struct tcphdr *th = skb->h.th;
	struct tcphdr rth;
	struct ip_reply_arg arg;

	/* 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.seq = seq;
	rth.ack_seq = ack; 
	rth.ack = 1;

	rth.window = htons(window);

	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++;
}


#ifdef CONFIG_IP_TRANSPARENT_PROXY

/*
   Seems, I never wrote nothing more stupid.
   I hope Gods will forgive me, but I cannot forgive myself 8)
                                                --ANK (981001)
 */

static struct sock *tcp_v4_search_proxy_openreq(struct sk_buff *skb)
{
	struct iphdr *iph = skb->nh.iph;
	struct tcphdr *th = (struct tcphdr *)(skb->nh.raw + iph->ihl*4);
	struct sock *sk;
	int i;

	for (i=0; i<TCP_LHTABLE_SIZE; i++) {
		for(sk = tcp_listening_hash[i]; sk; sk = sk->next) {
			struct open_request *dummy;
			if (tcp_v4_search_req(&sk->tp_pinfo.af_tcp, iph,
					      th, &dummy) &&
			    (!sk->bound_dev_if ||
			     sk->bound_dev_if == skb->dev->ifindex))
				return sk;
		}
	}
	return NULL;
}

/*
 *	Check whether a received TCP packet might be for one of our
 *	connections.
 */

int tcp_chkaddr(struct sk_buff *skb)
{
	struct iphdr *iph = skb->nh.iph;
	struct tcphdr *th = (struct tcphdr *)(skb->nh.raw + iph->ihl*4);
	struct sock *sk;

	sk = tcp_v4_lookup(iph->saddr, th->source, iph->daddr,
			   th->dest, skb->dev->ifindex);

	if (!sk)
		return tcp_v4_search_proxy_openreq(skb) != NULL;

	if (sk->state == TCP_LISTEN) {
		struct open_request *dummy;
		if (tcp_v4_search_req(&sk->tp_pinfo.af_tcp, skb->nh.iph,
				      th, &dummy) &&
		    (!sk->bound_dev_if ||
		     sk->bound_dev_if == skb->dev->ifindex))
			return 1;
	}

	/* 0 means accept all LOCAL addresses here, not all the world... */

	if (sk->rcv_saddr == 0)
		return 0;

	return 1;
}
#endif

/*
 *	Send a SYN-ACK after having received an ACK. 
 *	This still operates on a open_request only, not on a big
 *	socket.
 */ 
static void tcp_v4_send_synack(struct sock *sk, struct open_request *req)
{
	struct rtable *rt;
	struct ip_options *opt;
	struct sk_buff * skb;
	int mss;

	/* First, grab a route. */
	opt = req->af.v4_req.opt;
	if(ip_route_output(&rt, ((opt && opt->srr) ?
				 opt->faddr :
				 req->af.v4_req.rmt_addr),
			   req->af.v4_req.loc_addr,
			   RT_TOS(sk->ip_tos) | RTO_CONN | sk->localroute,
			   sk->bound_dev_if)) {
		ip_statistics.IpOutNoRoutes++;
		return;
	}
	if(opt && opt->is_strictroute && rt->rt_dst != rt->rt_gateway) {
		ip_rt_put(rt);
		ip_statistics.IpOutNoRoutes++;
		return;
	}

	mss = rt->u.dst.pmtu - sizeof(struct iphdr) - sizeof(struct tcphdr);

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

#ifdef CONFIG_IP_TRANSPARENT_PROXY
		th->source = req->lcl_port; /* LVE */
#endif

		th->check = tcp_v4_check(th, skb->len,
					 req->af.v4_req.loc_addr, req->af.v4_req.rmt_addr,
					 csum_partial((char *)th, skb->len, skb->csum));

		ip_build_and_send_pkt(skb, sk, req->af.v4_req.loc_addr,
				      req->af.v4_req.rmt_addr, req->af.v4_req.opt);
	}
	ip_rt_put(rt);
}

/*
 *	IPv4 open_request destructor.
 */ 
static void tcp_v4_or_free(struct open_request *req)
{
	if(!req->sk && req->af.v4_req.opt)
		kfree_s(req->af.v4_req.opt, optlength(req->af.v4_req.opt));
}

static inline void syn_flood_warning(struct sk_buff *skb)
{
	static unsigned long warntime;
	
	if (jiffies - warntime > HZ*60) {
		warntime = jiffies;
		printk(KERN_INFO 
		       "possible SYN flooding on port %d. Sending cookies.\n",  
		       ntohs(skb->h.th->dest));
	}
}

/* 
 * Save and compile IPv4 options into the open_request if needed. 
 */
static inline struct ip_options * 
tcp_v4_save_options(struct sock *sk, struct sk_buff *skb)
{
	struct ip_options *opt = &(IPCB(skb)->opt);
	struct ip_options *dopt = NULL; 

	if (opt && opt->optlen) {
		int opt_size = optlength(opt); 
		dopt = kmalloc(opt_size, GFP_ATOMIC);
		if (dopt) {
			if (ip_options_echo(dopt, skb)) {
				kfree_s(dopt, opt_size);
				dopt = NULL;
			}
		}
	}
	return dopt;
}

/* 
 * Maximum number of SYN_RECV sockets in queue per LISTEN socket.
 * One SYN_RECV socket costs about 80bytes on a 32bit machine.
 * It would be better to replace it with a global counter for all sockets
 * but then some measure against one socket starving all other sockets
 * would be needed.
 */
int sysctl_max_syn_backlog = 128; 

struct or_calltable or_ipv4 = {
	tcp_v4_send_synack,
	tcp_v4_or_free,
	tcp_v4_send_reset
};

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

int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb, __u32 isn)
{
	struct tcp_opt tp;
	struct open_request *req;
	struct tcphdr *th = skb->h.th;
	__u32 saddr = skb->nh.iph->saddr;
	__u32 daddr = skb->nh.iph->daddr;
#ifdef CONFIG_SYN_COOKIES
	int want_cookie = 0;
#else
#define want_cookie 0 /* Argh, why doesn't gcc optimize this :( */
#endif

	/* If the socket is dead, don't accept the connection.	*/
	if (sk->dead) 
		goto dead; 

	/* Never answer to SYNs send to broadcast or multicast */
	if (((struct rtable *)skb->dst)->rt_flags & 
	    (RTCF_BROADCAST|RTCF_MULTICAST))
		goto drop; 

	/* XXX: Check against a global syn pool counter. */
	if (BACKLOG(sk) > BACKLOGMAX(sk)) {
#ifdef CONFIG_SYN_COOKIES
		if (sysctl_tcp_syncookies) {
			syn_flood_warning(skb);
			want_cookie = 1; 
		} else
#endif
		goto drop;
	} else { 
		if (isn == 0)
			isn = tcp_v4_init_sequence(sk, skb);
		BACKLOG(sk)++;
	}

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

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

	req->rcv_isn = TCP_SKB_CB(skb)->seq;
 	tp.tstamp_ok = tp.sack_ok = tp.wscale_ok = tp.snd_wscale = 0;

	tp.mss_clamp = 65535;
	tcp_parse_options(NULL, th, &tp, want_cookie);
	if (tp.mss_clamp == 65535)
		tp.mss_clamp = 576 - sizeof(struct iphdr) - 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 = th->source;
#ifdef CONFIG_IP_TRANSPARENT_PROXY
	req->lcl_port = th->dest ; /* LVE */
#endif
	req->af.v4_req.loc_addr = daddr;
	req->af.v4_req.rmt_addr = saddr;

	/* Note that we ignore the isn passed from the TIME_WAIT
	 * state here. That's the price we pay for cookies.
	 */
	if (want_cookie)
		isn = cookie_v4_init_sequence(sk, skb, &req->mss);

	req->snt_isn = isn;

	req->af.v4_req.opt = tcp_v4_save_options(sk, skb);

	req->class = &or_ipv4;
	req->retrans = 0;
	req->sk = NULL;

	tcp_v4_send_synack(sk, req);

	if (want_cookie) {
		if (req->af.v4_req.opt)
			kfree(req->af.v4_req.opt);
		tcp_v4_or_free(req); 
	   	tcp_openreq_free(req); 
	} else {
		req->expires = jiffies + TCP_TIMEOUT_INIT;
		tcp_inc_slow_timer(TCP_SLT_SYNACK);
		tcp_synq_queue(&sk->tp_pinfo.af_tcp, req);
	}

	return 0;

dead:
	SOCK_DEBUG(sk, "Reset on %p: Connect on dead socket.\n",sk);
	tcp_statistics.TcpAttemptFails++;
	return -ENOTCONN; /* send reset */

dropbacklog:
	if (!want_cookie) 
		BACKLOG(sk)--;
drop:
	tcp_statistics.TcpAttemptFails++;
	return 0;
}

/* This is not only more efficient than what we used to do, it eliminates
 * a lot of code duplication between IPv4/IPv6 SYN recv processing. -DaveM
 *
 * This function wants to be moved to a common for IPv[46] file. --ANK
 */
struct sock *tcp_create_openreq_child(struct sock *sk, struct open_request *req, struct sk_buff *skb)
{
	struct sock *newsk = sk_alloc(PF_INET, GFP_ATOMIC, 0);

	if(newsk != NULL) {
		struct tcp_opt *newtp;
#ifdef CONFIG_FILTER
		struct sk_filter *filter;
#endif

		memcpy(newsk, sk, sizeof(*newsk));
		newsk->sklist_next = NULL;
		newsk->state = TCP_SYN_RECV;

		/* Clone the TCP header template */
		newsk->dport = req->rmt_port;

		atomic_set(&newsk->sock_readers, 0);
		atomic_set(&newsk->rmem_alloc, 0);
		skb_queue_head_init(&newsk->receive_queue);
		atomic_set(&newsk->wmem_alloc, 0);
		skb_queue_head_init(&newsk->write_queue);
		atomic_set(&newsk->omem_alloc, 0);

		newsk->done = 0;
		newsk->proc = 0;
		skb_queue_head_init(&newsk->back_log);
		skb_queue_head_init(&newsk->error_queue);
#ifdef CONFIG_FILTER
		if ((filter = newsk->filter) != NULL)
			sk_filter_charge(newsk, filter);
#endif

		/* Now setup tcp_opt */
		newtp = &(newsk->tp_pinfo.af_tcp);
		newtp->pred_flags = 0;
		newtp->rcv_nxt = req->rcv_isn + 1;
		newtp->snd_nxt = req->snt_isn + 1;
		newtp->snd_una = req->snt_isn + 1;
		newtp->srtt = 0;
		newtp->ato = 0;
		newtp->snd_wl1 = req->rcv_isn;
		newtp->snd_wl2 = req->snt_isn;

		/* RFC1323: The window in SYN & SYN/ACK segments
		 * is never scaled.
		 */
		newtp->snd_wnd = ntohs(skb->h.th->window);

		newtp->max_window = newtp->snd_wnd;
		newtp->pending = 0;
		newtp->retransmits = 0;
		newtp->last_ack_sent = req->rcv_isn + 1;
		newtp->backoff = 0;
		newtp->mdev = TCP_TIMEOUT_INIT;

		/* So many TCP implementations out there (incorrectly) count the
		 * initial SYN frame in their delayed-ACK and congestion control
		 * algorithms that we must have the following bandaid to talk
		 * efficiently to them.  -DaveM
		 */
		newtp->snd_cwnd = 2;

		newtp->rto = TCP_TIMEOUT_INIT;
		newtp->packets_out = 0;
		newtp->fackets_out = 0;
		newtp->retrans_out = 0;
		newtp->high_seq = 0;
		newtp->snd_ssthresh = 0x7fffffff;
		newtp->snd_cwnd_cnt = 0;
		newtp->dup_acks = 0;
		newtp->delayed_acks = 0;
		init_timer(&newtp->retransmit_timer);
		newtp->retransmit_timer.function = &tcp_retransmit_timer;
		newtp->retransmit_timer.data = (unsigned long) newsk;
		init_timer(&newtp->delack_timer);
		newtp->delack_timer.function = &tcp_delack_timer;
		newtp->delack_timer.data = (unsigned long) newsk;
		skb_queue_head_init(&newtp->out_of_order_queue);
		newtp->send_head = newtp->retrans_head = NULL;
		newtp->rcv_wup = req->rcv_isn + 1;
		newtp->write_seq = req->snt_isn + 1;
		newtp->copied_seq = req->rcv_isn + 1;

		newtp->saw_tstamp = 0;
		newtp->mss_clamp = req->mss;

		init_timer(&newtp->probe_timer);
		newtp->probe_timer.function = &tcp_probe_timer;
		newtp->probe_timer.data = (unsigned long) newsk;
		newtp->probes_out = 0;
		newtp->syn_seq = req->rcv_isn;
		newtp->fin_seq = req->rcv_isn;
		newtp->urg_data = 0;
		tcp_synq_init(newtp);
		newtp->syn_backlog = 0;
		if (skb->len >= 536)
			newtp->last_seg_size = skb->len; 

		/* Back to base struct sock members. */
		newsk->err = 0;
		newsk->ack_backlog = 0;
		newsk->max_ack_backlog = SOMAXCONN;
		newsk->priority = 0;

		/* IP layer stuff */
		newsk->timeout = 0;
		init_timer(&newsk->timer);
		newsk->timer.function = &net_timer;
		newsk->timer.data = (unsigned long) newsk;
		newsk->socket = NULL;

		newtp->tstamp_ok = req->tstamp_ok;
		if((newtp->sack_ok = req->sack_ok) != 0)
			newtp->num_sacks = 0;
		newtp->window_clamp = req->window_clamp;
		newtp->rcv_wnd = req->rcv_wnd;
		newtp->wscale_ok = req->wscale_ok;
		if (newtp->wscale_ok) {
			newtp->snd_wscale = req->snd_wscale;
			newtp->rcv_wscale = req->rcv_wscale;
		} else {
			newtp->snd_wscale = newtp->rcv_wscale = 0;
			newtp->window_clamp = min(newtp->window_clamp,65535);
		}
		if (newtp->tstamp_ok) {
			newtp->ts_recent = req->ts_recent;
			newtp->ts_recent_stamp = tcp_time_stamp;
			newtp->tcp_header_len = sizeof(struct tcphdr) + TCPOLEN_TSTAMP_ALIGNED;
		} else {
			newtp->tcp_header_len = sizeof(struct tcphdr);
		}
	}
	return newsk;
}

/* 
 * The three way handshake has completed - we got a valid synack - 
 * now create the new socket. 
 */
struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
				   struct open_request *req,
				   struct dst_entry *dst)
{
	struct ip_options *opt = req->af.v4_req.opt;
	struct tcp_opt *newtp;
	struct sock *newsk;

	if (sk->ack_backlog > sk->max_ack_backlog)
		goto exit; /* head drop */
	if (dst == NULL) { 
		struct rtable *rt;
		
		if (ip_route_output(&rt,
			opt && opt->srr ? opt->faddr : req->af.v4_req.rmt_addr,
			req->af.v4_req.loc_addr, sk->ip_tos|RTO_CONN, 0))
			return NULL;
	        dst = &rt->u.dst;
	}
#ifdef CONFIG_IP_TRANSPARENT_PROXY
	/* The new socket created for transparent proxy may fall
	 * into a non-existed bind bucket because sk->num != newsk->num.
	 * Ensure existance of the bucket now. The placement of the check
	 * later will require to destroy just created newsk in the case of fail.
	 * 1998/04/22 Andrey V. Savochkin <[e-mail]>
	 */
	if (__tcp_bucket_check(ntohs(skb->h.th->dest)))
		goto exit;
#endif

	newsk = tcp_create_openreq_child(sk, req, skb);
	if (!newsk) 
		goto exit;

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

	newsk->dst_cache = dst;

	newtp = &(newsk->tp_pinfo.af_tcp);
	newsk->daddr = req->af.v4_req.rmt_addr;
	newsk->saddr = req->af.v4_req.loc_addr;
	newsk->rcv_saddr = req->af.v4_req.loc_addr;