patch-2.3.41 linux/include/net/tcp.h

Next file: linux/ipc/shm.c
Previous file: linux/include/net/sock.h
Back to the patch index
Back to the overall index

diff -u --recursive --new-file v2.3.40/linux/include/net/tcp.h linux/include/net/tcp.h
@@ -19,6 +19,7 @@
 #define _TCP_H
 
 #define TCP_DEBUG 1
+#undef  TCP_FORMAL_WINDOW
 
 #include <linux/config.h>
 #include <linux/tcp.h>
@@ -130,27 +131,27 @@
 	struct sock		*bind_next;
 	struct sock		**bind_pprev;
 	unsigned char		state,
-				zapped;
+				substate; /* "zapped" is replaced with "substate" */
 	__u16			sport;
 	unsigned short		family;
 	unsigned char		reuse,
-				nonagle;
+				rcv_wscale; /* It is also TW bucket specific */
 	atomic_t		refcnt;
 
 	/* And these are ours. */
 	int			hashent;
+	int			timeout;
 	__u32			rcv_nxt;
 	__u32			snd_nxt;
+	__u32			rcv_wnd;
+	__u32			syn_seq;
         __u32			ts_recent;
         long			ts_recent_stamp;
+	unsigned long		ttd;
 	struct tcp_bind_bucket	*tb;
 	struct tcp_tw_bucket	*next_death;
 	struct tcp_tw_bucket	**pprev_death;
-	int			death_slot;
-#ifdef CONFIG_TCP_TW_RECYCLE
-	unsigned long		ttd;
-	int			rto;
-#endif
+
 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 	struct in6_addr		v6_daddr;
 	struct in6_addr		v6_rcv_saddr;
@@ -169,10 +170,11 @@
 	}
 }
 
-extern int tcp_tw_death_row_slot;
+extern atomic_t tcp_orphan_count;
+extern int  tcp_tw_count;
+extern void tcp_time_wait(struct sock *sk, int state, int timeo);
 extern void tcp_timewait_kill(struct tcp_tw_bucket *tw);
-extern void tcp_tw_schedule(struct tcp_tw_bucket *tw);
-extern void tcp_tw_reschedule(struct tcp_tw_bucket *tw);
+extern void tcp_tw_schedule(struct tcp_tw_bucket *tw, int timeo);
 extern void tcp_tw_deschedule(struct tcp_tw_bucket *tw);
 
 
@@ -224,67 +226,81 @@
 	return tcp_lhashfn(sk->num);
 }
 
-/* Note, that it is > than ipv6 header */
-#define NETHDR_SIZE	(sizeof(struct iphdr) + 40)
-
-/*
- * 40 is maximal IP options size
- * 20 is the maximum TCP options size we can currently construct on a SYN.
- * 40 is the maximum possible TCP options size.
- */
-
-#define MAX_SYN_SIZE	(NETHDR_SIZE + sizeof(struct tcphdr) + 20 + MAX_HEADER + 15)
-#define MAX_FIN_SIZE	(NETHDR_SIZE + sizeof(struct tcphdr) + MAX_HEADER + 15)
-#define BASE_ACK_SIZE	(NETHDR_SIZE + MAX_HEADER + 15)
-#define MAX_ACK_SIZE	(NETHDR_SIZE + sizeof(struct tcphdr) + MAX_HEADER + 15)
-#define MAX_RESET_SIZE	(NETHDR_SIZE + sizeof(struct tcphdr) + MAX_HEADER + 15)
-#define MAX_TCPHEADER_SIZE (NETHDR_SIZE + sizeof(struct tcphdr) + 20 + MAX_HEADER + 15)
+#define MAX_TCP_HEADER	(128 + MAX_HEADER)
 
 /* 
  * Never offer a window over 32767 without using window scaling. Some
  * poor stacks do signed 16bit maths! 
  */
-#define MAX_WINDOW	32767	
-#define MAX_DELAY_ACK	2
+#define MAX_TCP_WINDOW		32767
+
+/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
+#define TCP_MIN_MSS		88
+
+/* Minimal RCV_MSS. */
+#define TCP_MIN_RCVMSS		536
 
 /* 
  * How much of the receive buffer do we advertize 
  * (the rest is reserved for headers and driver packet overhead)
  * Use a power of 2.
  */
-#define WINDOW_ADVERTISE_DIVISOR 2
+#define TCP_WINDOW_ADVERTISE_DIVISOR 2
 
 /* urg_data states */
-#define URG_VALID	0x0100
-#define URG_NOTYET	0x0200
-#define URG_READ	0x0400
+#define TCP_URG_VALID	0x0100
+#define TCP_URG_NOTYET	0x0200
+#define TCP_URG_READ	0x0400
 
-#define TCP_RETR1	7	/*
+#define TCP_RETR1	3	/*
 				 * This is how many retries it does before it
 				 * tries to figure out if the gateway is
-				 * down.
+				 * down. Minimal RFC value is 3; it corresponds
+				 * to ~3sec-8min depending on RTO.
 				 */
 
 #define TCP_RETR2	15	/*
 				 * This should take at least
 				 * 90 minutes to time out.
+				 * RFC1122 says that the limit is 100 sec.
+				 * 15 is ~13-30min depending on RTO.
 				 */
 
-#define TCP_TIMEOUT_LEN	(15*60*HZ) /* should be about 15 mins		*/
-#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to successfully 
-				  * close the socket, about 60 seconds	*/
-#define TCP_FIN_TIMEOUT (3*60*HZ) /* BSD style FIN_WAIT2 deadlock breaker */
-
-#define TCP_ACK_TIME	(3*HZ)	/* time to delay before sending an ACK	*/
-#define TCP_WRITE_TIME	(30*HZ)	/* initial time to wait for an ACK,
-			         * after last transmit			*/
-#define TCP_TIMEOUT_INIT (3*HZ)	/* RFC 1122 initial timeout value	*/
-#define TCP_SYN_RETRIES	 10	/* number of times to retry opening a
-				 * connection 	(TCP_RETR2-....)	*/
-#define TCP_PROBEWAIT_LEN (1*HZ)/* time to wait between probes when
-				 * I've got something to write and
-				 * there is no window			*/
-#define TCP_KEEPALIVE_TIME (120*60*HZ)		/* two hours */
+#define TCP_SYN_RETRIES	 5	/* number of times to retry active opening a
+				 * connection: ~180sec is RFC minumum	*/
+
+#define TCP_SYNACK_RETRIES 5	/* number of times to retry passive opening a
+				 * connection: ~180sec is RFC minumum	*/
+
+
+#define TCP_ORPHAN_RETRIES 7	/* number of times to retry on an orphaned
+				 * socket. 7 is ~50sec-16min.
+				 */
+
+
+#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
+				  * state, about 60 seconds	*/
+#define TCP_FIN_TIMEOUT	TCP_TIMEWAIT_LEN
+                                 /* BSD style FIN_WAIT2 deadlock breaker.
+				  * It used to be 3min, new value is 60sec,
+				  * to combine FIN-WAIT-2 timeout with
+				  * TIME-WAIT timer.
+				  */
+
+#define TCP_DELACK_MAX	(HZ/2)	/* maximal time to delay before sending an ACK */
+#define TCP_DELACK_MIN	(2)	/* minimal time to delay before sending an ACK,
+				 * 2 scheduler ticks, not depending on HZ */
+#define TCP_ATO_MAX	((TCP_DELACK_MAX*4)/5) /* ATO producing TCP_DELACK_MAX */
+#define TCP_ATO_MIN	2
+#define TCP_RTO_MAX	(120*HZ)
+#define TCP_RTO_MIN	(HZ/5)
+#define TCP_TIMEOUT_INIT (3*HZ)	/* RFC 1122 initial RTO value	*/
+
+#define TCP_RESOURCE_PROBE_INTERVAL (HZ/2) /* Maximal interval between probes
+					    * for local resources.
+					    */
+
+#define TCP_KEEPALIVE_TIME	(120*60*HZ)	/* two hours */
 #define TCP_KEEPALIVE_PROBES	9		/* Max of 9 keepalive probes	*/
 #define TCP_KEEPALIVE_INTVL	(75*HZ)
 
@@ -293,14 +309,39 @@
 #define MAX_TCP_KEEPCNT		127
 #define MAX_TCP_SYNCNT		127
 
-#define TCP_SYNACK_PERIOD	(HZ/2) /* How often to run the synack slow timer */
-#define TCP_QUICK_TRIES		8  /* How often we try to retransmit, until
-				    * we tell the link layer that it is something
-				    * wrong (e.g. that it can expire redirects) */
-
 /* TIME_WAIT reaping mechanism. */
 #define TCP_TWKILL_SLOTS	8	/* Please keep this a power of 2. */
-#define TCP_TWKILL_PERIOD	((HZ*60)/TCP_TWKILL_SLOTS)
+#define TCP_TWKILL_PERIOD	(TCP_TIMEWAIT_LEN/TCP_TWKILL_SLOTS)
+
+#define TCP_SYNQ_INTERVAL	(HZ/5)	/* Period of SYNACK timer */
+#define TCP_SYNQ_HSIZE		64	/* Size of SYNACK hash table */
+
+#define TCP_PAWS_24DAYS	(60 * 60 * 24 * 24)
+#define TCP_PAWS_MSL	60		/* Per-host timestamps are invalidated
+					 * after this time. It should be equal
+					 * (or greater than) TCP_TIMEWAIT_LEN
+					 * to provide reliability equal to one
+					 * provided by timewait state.
+					 */
+#define TCP_PAWS_WINDOW	1		/* Replay window for per-host
+					 * timestamps. It must be less than
+					 * minimal timewait lifetime.
+					 */
+
+#define TCP_TW_RECYCLE_SLOTS_LOG	5
+#define TCP_TW_RECYCLE_SLOTS		(1<<TCP_TW_RECYCLE_SLOTS_LOG)
+
+/* If time > 4sec, it is "slow" path, no recycling is required,
+   so that we select tick to get range about 4 seconds.
+ */
+
+#if HZ == 100
+#define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
+#elif HZ == 1024
+#define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
+#else
+#error HZ != 100 && HZ != 1024.
+#endif
 
 /*
  *	TCP option
@@ -331,23 +372,40 @@
 #define TCPOLEN_SACK_BASE_ALIGNED	4
 #define TCPOLEN_SACK_PERBLOCK		8
 
-#define TIME_WRITE	1	/* Not yet used */
-#define TIME_RETRANS	2	/* Retransmit timer */
-#define TIME_DACK	3	/* Delayed ack timer */
-#define TIME_PROBE0	4
-#define TIME_KEEPOPEN	5
+#define TCP_TIME_RETRANS	1	/* Retransmit timer */
+#define TCP_TIME_DACK		2	/* Delayed ack timer */
+#define TCP_TIME_PROBE0		3	/* Zero window probe timer */
+#define TCP_TIME_KEEPOPEN	4	/* Keepalive timer */
 
 /* sysctl variables for tcp */
+extern int sysctl_max_syn_backlog;
+extern int sysctl_tcp_timestamps;
+extern int sysctl_tcp_window_scaling;
+extern int sysctl_tcp_sack;
+extern int sysctl_tcp_fin_timeout;
+extern int sysctl_tcp_tw_recycle;
 extern int sysctl_tcp_keepalive_time;
 extern int sysctl_tcp_keepalive_probes;
 extern int sysctl_tcp_keepalive_intvl;
 extern int sysctl_tcp_syn_retries;
+extern int sysctl_tcp_synack_retries;
+extern int sysctl_tcp_retries1;
+extern int sysctl_tcp_retries2;
+extern int sysctl_tcp_orphan_retries;
+extern int sysctl_tcp_syncookies;
+extern int sysctl_tcp_retrans_collapse;
+extern int sysctl_tcp_stdurg;
+extern int sysctl_tcp_rfc1337;
+extern int sysctl_tcp_tw_recycle;
+extern int sysctl_tcp_abort_on_overflow;
+extern int sysctl_tcp_max_orphans;
+extern int sysctl_tcp_max_tw_buckets;
 
 struct open_request;
 
 struct or_calltable {
 	int  family;
-	void (*rtx_syn_ack)	(struct sock *sk, struct open_request *req);
+	int  (*rtx_syn_ack)	(struct sock *sk, struct open_request *req, struct dst_entry*);
 	void (*send_ack)	(struct sk_buff *skb, struct open_request *req);
 	void (*destructor)	(struct open_request *req);
 	void (*send_reset)	(struct sk_buff *skb);
@@ -376,12 +434,14 @@
 	__u16			rmt_port;
 	__u16			mss;
 	__u8			retrans;
-	__u8			__pad;
-	unsigned snd_wscale : 4, 
+	__u8			index;
+	__u16	snd_wscale : 4, 
 		rcv_wscale : 4, 
 		tstamp_ok : 1,
 		sack_ok : 1,
-		wscale_ok : 1;
+		wscale_ok : 1,
+		ecn_ok : 1,
+		acked : 1;
 	/* The following two fields can be easily recomputed I think -AK */
 	__u32			window_clamp;	/* window clamp at creation time */
 	__u32			rcv_wnd;	/* rcv_wnd offered first time */
@@ -400,8 +460,14 @@
 /* SLAB cache for open requests. */
 extern kmem_cache_t *tcp_openreq_cachep;
 
-#define tcp_openreq_alloc()	kmem_cache_alloc(tcp_openreq_cachep, SLAB_ATOMIC)
-#define tcp_openreq_free(req)	kmem_cache_free(tcp_openreq_cachep, req)
+#define tcp_openreq_alloc()		kmem_cache_alloc(tcp_openreq_cachep, SLAB_ATOMIC)
+#define tcp_openreq_fastfree(req)	kmem_cache_free(tcp_openreq_cachep, req)
+
+extern __inline__ void tcp_openreq_free(struct open_request *req)
+{
+	req->class->destructor(req);
+	tcp_openreq_fastfree(req);
+}
 
 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
 #define TCP_INET_FAMILY(fam) ((fam) == AF_INET)
@@ -441,9 +507,9 @@
 	
 	int			(*hash_connecting)	(struct sock *sk);
 
-	__u16			net_header_len;
-
+	int			(*remember_stamp)	(struct sock *sk);
 
+	__u16			net_header_len;
 
 	int			(*setsockopt)		(struct sock *sk, 
 							 int level, 
@@ -506,7 +572,11 @@
 extern int			tcp_v4_rcv(struct sk_buff *skb,
 					   unsigned short len);
 
-extern int			tcp_do_sendmsg(struct sock *sk, struct msghdr *msg);
+extern int			tcp_v4_remember_stamp(struct sock *sk);
+
+extern int		    	tcp_v4_tw_remember_stamp(struct tcp_tw_bucket *tw);
+
+extern int			tcp_sendmsg(struct sock *sk, struct msghdr *msg, int size);
 
 extern int			tcp_ioctl(struct sock *sk, 
 					  int cmd, 
@@ -522,6 +592,23 @@
 						    struct tcphdr *th, 
 						    unsigned len);
 
+static __inline__ void tcp_dec_quickack_mode(struct tcp_opt *tp)
+{
+	if (tp->ack.quick && --tp->ack.quick == 0 && !tp->ack.pingpong) {
+		/* Leaving quickack mode we deflate ATO to give peer
+		 * a time to adapt to new worse(!) RTO. It is not required
+		 * in pingpong mode, when ACKs were delayed in any case.
+		 */
+		tp->ack.ato = TCP_ATO_MIN;
+	}
+}
+
+static __inline__ void tcp_delack_init(struct tcp_opt *tp)
+{
+	memset(&tp->ack, 0, sizeof(tp->ack));
+}
+
+
 enum tcp_tw_status
 {
 	TCP_TW_SUCCESS = 0,
@@ -530,6 +617,7 @@
 	TCP_TW_SYN = 3
 };
 
+
 extern enum tcp_tw_status	tcp_timewait_state_process(struct tcp_tw_bucket *tw,
 							   struct sk_buff *skb,
 							   struct tcphdr *th,
@@ -537,7 +625,10 @@
 
 extern struct sock *		tcp_check_req(struct sock *sk,struct sk_buff *skb,
 					      struct open_request *req,
-					      struct open_request *prev);
+					      struct open_request **prev);
+extern int			tcp_child_process(struct sock *parent,
+						  struct sock *child,
+						  struct sk_buff *skb);
 
 extern void			tcp_close(struct sock *sk, 
 					  long timeout);
@@ -557,6 +648,8 @@
 					    int len, int nonblock, 
 					    int flags, int *addr_len);
 
+extern int			tcp_listen_start(struct sock *sk);
+
 extern void			tcp_parse_options(struct sock *sk, struct tcphdr *th,
 						  struct tcp_opt *tp, int no_fancy);
 
@@ -614,9 +707,7 @@
 
 /* tcp_output.c */
 
-extern void tcp_read_wakeup(struct sock *);
-extern void tcp_write_xmit(struct sock *);
-extern void tcp_time_wait(struct sock *);
+extern int tcp_write_xmit(struct sock *);
 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
 extern void tcp_fack_retransmit(struct sock *);
 extern void tcp_xmit_retransmit_queue(struct sock *);
@@ -624,46 +715,22 @@
 
 extern void tcp_send_probe0(struct sock *);
 extern void tcp_send_partial(struct sock *);
-extern void tcp_write_wakeup(struct sock *);
+extern int  tcp_write_wakeup(struct sock *);
 extern void tcp_send_fin(struct sock *sk);
 extern void tcp_send_active_reset(struct sock *sk, int priority);
 extern int  tcp_send_synack(struct sock *);
-extern void tcp_transmit_skb(struct sock *, struct sk_buff *);
-extern void tcp_send_skb(struct sock *, struct sk_buff *, int force_queue);
+extern int  tcp_transmit_skb(struct sock *, struct sk_buff *);
+extern void tcp_send_skb(struct sock *, struct sk_buff *, int force_queue, unsigned mss_now);
 extern void tcp_send_ack(struct sock *sk);
-extern void tcp_send_delayed_ack(struct sock *sk, int max_timeout);
+extern void tcp_send_delayed_ack(struct sock *sk);
 
 /* tcp_timer.c */
 extern void tcp_reset_xmit_timer(struct sock *, int, unsigned long);
 extern void tcp_init_xmit_timers(struct sock *);
 extern void tcp_clear_xmit_timers(struct sock *);
 
-extern void tcp_retransmit_timer(unsigned long);
-extern void tcp_delack_timer(unsigned long);
-extern void tcp_probe_timer(unsigned long);
-
 extern void tcp_delete_keepalive_timer (struct sock *);
 extern void tcp_reset_keepalive_timer (struct sock *, unsigned long);
-extern void tcp_keepalive_timer (unsigned long);
-
-/*
- *	TCP slow timer
- */
-extern struct timer_list	tcp_slow_timer;
-
-struct tcp_sl_timer {
-	atomic_t	count;
-	unsigned long	period;
-	unsigned long	last;
-	void (*handler)	(unsigned long);
-};
-
-#define TCP_SLT_SYNACK		0
-#define TCP_SLT_TWKILL		1
-#define TCP_SLT_MAX		2
-
-extern struct tcp_sl_timer tcp_slt_array[TCP_SLT_MAX];
- 
 extern int tcp_sync_mss(struct sock *sk, u32 pmtu);
 
 /* Compute the current effective MSS, taking SACKs and IP options,
@@ -673,7 +740,7 @@
 static __inline__ unsigned int tcp_current_mss(struct sock *sk)
 {
 	struct tcp_opt *tp = &sk->tp_pinfo.af_tcp;
-	struct dst_entry *dst = sk->dst_cache;
+	struct dst_entry *dst = __sk_dst_get(sk);
 	int mss_now = tp->mss_cache; 
 
 	if (dst && dst->pmtu != tp->pmtu_cookie)
@@ -682,7 +749,7 @@
 	if(tp->sack_ok && tp->num_sacks)
 		mss_now -= (TCPOLEN_SACK_BASE_ALIGNED +
 			    (tp->num_sacks * TCPOLEN_SACK_PERBLOCK));
-	return mss_now > 8 ? mss_now : 8;
+	return mss_now;
 }
 
 /* Initialize RCV_MSS value.
@@ -704,9 +771,24 @@
 	else
 		mss = tp->mss_cache;
 
-	tp->rcv_mss = max(min(mss, 536), 8);
+	tp->ack.rcv_mss = max(min(mss, TCP_MIN_RCVMSS), TCP_MIN_MSS);
+}
+
+static __inline__ void __tcp_fast_path_on(struct tcp_opt *tp, u32 snd_wnd)
+{
+	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
+			       ntohl(TCP_FLAG_ACK) |
+			       snd_wnd);
+}
+
+static __inline__ void tcp_fast_path_on(struct tcp_opt *tp)
+{
+	__tcp_fast_path_on(tp, tp->snd_wnd>>tp->snd_wscale);
 }
 
+
+
+
 /* Compute the actual receive window we are currently advertising.
  * Rcv_nxt can be after the window if our peer push more data
  * than the offered window.
@@ -751,23 +833,26 @@
 	}
 
 	/* RFC1323 scaling applied */
-	return new_win >> tp->rcv_wscale;
-}
+	new_win >>= tp->rcv_wscale;
 
-/* See if we can advertise non-zero, and if so how much we
- * can increase our advertisement.  If it becomes more than
- * twice what we are talking about right now, return true.
- */
-extern __inline__ int tcp_raise_window(struct sock *sk)
-{
-	struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
-	u32 cur_win = tcp_receive_window(tp);
-	u32 new_win = __tcp_select_window(sk);
+#ifdef TCP_FORMAL_WINDOW
+	if (new_win == 0) {
+		/* If we advertise zero window, disable fast path. */
+		tp->pred_flags = 0;
+	} else if (cur_win == 0 && tp->pred_flags == 0 &&
+		   skb_queue_len(&tp->out_of_order_queue) == 0 &&
+		   !tp->urg_data) {
+		/* If we open zero window, enable fast path.
+		   Without this it will be open by the first data packet,
+		   it is too late to merge checksumming to copy.
+		 */
+		tcp_fast_path_on(tp);
+	}
+#endif
 
-	return (new_win && (new_win > (cur_win << 1)));
+	return new_win;
 }
 
-
 /* TCP timestamps are only 32-bits, this causes a slight
  * complication on 64-bit systems since we store a snapshot
  * of jiffies in the buffer control blocks below.  We decidely
@@ -804,6 +889,8 @@
 #define TCPCB_FLAG_PSH		0x08
 #define TCPCB_FLAG_ACK		0x10
 #define TCPCB_FLAG_URG		0x20
+#define TCPCB_FLAG_ECE		0x40
+#define TCPCB_FLAG_CWR		0x80
 
 	__u8		sacked;		/* State flags for SACK/FACK.	*/
 #define TCPCB_SACKED_ACKED	0x01	/* SKB ACK'd by a SACK block	*/
@@ -860,13 +947,91 @@
 	return max(min(FlightSize, tp->snd_cwnd) >> 1, 2);
 }
 
+/* Set slow start threshould and cwnd not falling to slow start */
+extern __inline__ void __tcp_enter_cong_avoid(struct tcp_opt *tp)
+{
+	tp->snd_ssthresh = tcp_recalc_ssthresh(tp);
+	if (tp->snd_ssthresh > tp->snd_cwnd_clamp)
+		tp->snd_ssthresh = tp->snd_cwnd_clamp;
+	tp->snd_cwnd = tp->snd_ssthresh;
+	tp->snd_cwnd_cnt = 0;
+	tp->high_seq = tp->snd_nxt;
+}
+
+extern __inline__ void tcp_enter_cong_avoid(struct tcp_opt *tp)
+{
+	if (!tp->high_seq || after(tp->snd_nxt, tp->high_seq))
+		__tcp_enter_cong_avoid(tp);
+}
+
+
+/* Increase initial CWND conservatively, i.e. only if estimated
+   RTT is low enough. It is not quite correct, we should use
+   POWER i.e. RTT*BANDWIDTH, but we still cannot estimate this.
+
+   Numbers are taken from RFC1414.
+ */
+static __inline__ __u32 tcp_init_cwnd(struct tcp_opt *tp)
+{
+	__u32 cwnd;
+
+	if (!tp->srtt || tp->srtt > ((HZ/50)<<3) || tp->mss_cache > 1460)
+		cwnd = 2;
+	else if (tp->mss_cache > 1095)
+		cwnd = 3;
+	else
+		cwnd = 4;
+
+	return min(cwnd, tp->snd_cwnd_clamp);
+}
+
+
+static __inline__ int tcp_minshall_check(struct tcp_opt *tp)
+{
+	return after(tp->snd_sml,tp->snd_una) &&
+		!after(tp->snd_sml, tp->snd_nxt);
+}
+
+static __inline__ void tcp_minshall_update(struct tcp_opt *tp, int mss, int len)
+{
+	if (len < mss)
+		tp->snd_sml = tp->snd_nxt;
+}
+
+/* Return 0, if packet can be sent now without violation Nagle's rules:
+   1. It is full sized.
+   2. Or it contains FIN or URG.
+   3. Or TCP_NODELAY was set.
+   4. Or TCP_CORK is not set, and all sent packets are ACKed.
+      With Minshall's modification: all sent small packets are ACKed.
+ */
+
+static __inline__ int tcp_nagle_check(struct tcp_opt *tp, struct sk_buff *skb, unsigned mss_now)
+{
+	return (skb->len < mss_now &&
+		!(TCP_SKB_CB(skb)->flags & (TCPCB_FLAG_URG|TCPCB_FLAG_FIN)) &&
+		(tp->nonagle == 2 ||
+		 (!tp->nonagle &&
+		  tp->packets_out &&
+		  tcp_minshall_check(tp))));
+}
+
 /* This checks if the data bearing packet SKB (usually tp->send_head)
  * should be put on the wire right now.
  */
-static __inline__ int tcp_snd_test(struct sock *sk, struct sk_buff *skb)
+static __inline__ int tcp_snd_test(struct tcp_opt *tp, struct sk_buff *skb,
+				   unsigned cur_mss, int tail)
 {
-	struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
-	int nagle_check = 1;
+	/*
+	 * Reset CWND after idle period longer RTO to "restart window".
+	 * It is "side" effect of the function, which is _not_ good
+	 * from viewpoint of clarity. But we have to make it before
+	 * checking congestion window below. Alternative is to prepend
+	 * all the calls with this test.
+	 */
+	if (tp->packets_out==0 &&
+	    (s32)(tcp_time_stamp - tp->lsndtime) > tp->rto)
+		tp->snd_cwnd = min(tp->snd_cwnd, tcp_init_cwnd(tp));
 
 	/*	RFC 1122 - section 4.2.3.4
 	 *
@@ -876,97 +1041,126 @@
 	 *	b) There are packets in flight and we have a small segment
 	 *	   [SWS avoidance and Nagle algorithm]
 	 *	   (part of SWS is done on packetization)
+	 *	   Minshall version sounds: there are no _small_
+	 *	   segments in flight. (tcp_nagle_check)
 	 *	c) We are retransmiting [Nagle]
 	 *	d) We have too many packets 'in flight'
 	 *
 	 * 	Don't use the nagle rule for urgent data (or
 	 *	for the final FIN -DaveM).
+	 *
+	 *	Also, Nagle rule does not apply to frames, which
+	 *	sit in the middle of queue (they have no chances
+	 *	to get new data) and if room at tail of skb is
+	 *	not enough to save something seriously (<32 for now).
 	 */
-	if ((sk->nonagle == 2 && (skb->len < tp->mss_cache)) ||
-	    (!sk->nonagle &&
-	     skb->len < (tp->mss_cache >> 1) &&
-	     tp->packets_out &&
-	     !(TCP_SKB_CB(skb)->flags & (TCPCB_FLAG_URG|TCPCB_FLAG_FIN))))
-		nagle_check = 0;
-
-	/*
-	 * Reset CWND after idle period longer rto. Actually, it would
-	 * be better to save last send time, but VJ in SIGCOMM'88 proposes
-	 * to use keepalive timestamp. Well, it is not good, certainly,
-	 * because SMTP is still broken, but it is better than nothing yet.
-	 */
-	if (tp->packets_out==0 && (s32)(tcp_time_stamp - tp->rcv_tstamp) > tp->rto)
-		tp->snd_cwnd = min(tp->snd_cwnd, 2);
 
 	/* Don't be strict about the congestion window for the
 	 * final FIN frame.  -DaveM
 	 */
-	return (nagle_check &&
+	return ((!tail || !tcp_nagle_check(tp, skb, cur_mss) ||
+		 skb_tailroom(skb) < 32) &&
 		((tcp_packets_in_flight(tp) < tp->snd_cwnd) ||
 		 (TCP_SKB_CB(skb)->flags & TCPCB_FLAG_FIN)) &&
 		!after(TCP_SKB_CB(skb)->end_seq, tp->snd_una + tp->snd_wnd) &&
 		tp->retransmits == 0);
 }
 
+static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_opt *tp)
+{
+	if (!tp->packets_out && !tp->probe_timer.prev)
+		tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto);
+}
+
+static __inline__ int tcp_skb_is_last(struct sock *sk, struct sk_buff *skb)
+{
+	return (skb->next == (struct sk_buff*)&sk->write_queue);
+}
+
 /* Push out any pending frames which were held back due to
  * TCP_CORK or attempt at coalescing tiny packets.
  * The socket must be locked by the caller.
  */
-static __inline__ void tcp_push_pending_frames(struct sock *sk, struct tcp_opt *tp)
-{
-	if(tp->send_head) {
-		if(tcp_snd_test(sk, tp->send_head))
-			tcp_write_xmit(sk);
-		else if(tp->packets_out == 0 && !tp->pending) {
-			/* We held off on this in tcp_send_skb() */
-			tp->pending = TIME_PROBE0;
-			tcp_reset_xmit_timer(sk, TIME_PROBE0, tp->rto);
-		}
+static __inline__ void __tcp_push_pending_frames(struct sock *sk,
+						 struct tcp_opt *tp,
+						 unsigned cur_mss)
+{
+	struct sk_buff *skb = tp->send_head;
+
+	if (skb) {
+		if (!tcp_snd_test(tp, skb, cur_mss, tcp_skb_is_last(sk, skb)) ||
+		    tcp_write_xmit(sk))
+			tcp_check_probe_timer(sk, tp);
 	}
 }
 
-/* This tells the input processing path that an ACK should go out
- * right now.
- */
-#define tcp_enter_quickack_mode(__tp)	((__tp)->ato |= (1<<31))
-#define tcp_exit_quickack_mode(__tp)	((__tp)->ato &= ~(1<<31))
-#define tcp_in_quickack_mode(__tp)	(((__tp)->ato & (1 << 31)) != 0)
+static __inline__ void tcp_push_pending_frames(struct sock *sk,
+					       struct tcp_opt *tp)
+{
+	__tcp_push_pending_frames(sk, tp, tcp_current_mss(sk));
+}
+
+extern void			tcp_destroy_sock(struct sock *sk);
+
 
 /*
- * List all states of a TCP socket that can be viewed as a "connected"
- * state.  This now includes TCP_SYN_RECV, although I am not yet fully
- * convinced that this is the solution for the 'getpeername(2)'
- * problem. Thanks to Stephen A. Wood <saw@cebaf.gov>  -FvK
+ * Calculate(/check) TCP checksum
  */
+static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
+				   unsigned long saddr, unsigned long daddr, 
+				   unsigned long base)
+{
+	return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
+}
 
-extern __inline const int tcp_connected(const int state)
+static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
 {
-	return ((1 << state) &
-	       	(TCPF_ESTABLISHED|TCPF_CLOSE_WAIT|TCPF_FIN_WAIT1|
-		 TCPF_FIN_WAIT2|TCPF_SYN_RECV));
+	return (unsigned short)csum_fold(csum_partial(skb->h.raw, skb->len, skb->csum));
 }
 
-extern __inline const int tcp_established(const int state)
+static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
 {
-	return ((1 << state) &
-	       	(TCPF_ESTABLISHED|TCPF_CLOSE_WAIT|TCPF_FIN_WAIT1|
-		 TCPF_FIN_WAIT2));
+	return skb->ip_summed != CHECKSUM_UNNECESSARY &&
+		__tcp_checksum_complete(skb);
 }
 
 
-extern void			tcp_destroy_sock(struct sock *sk);
+/* Prequeue for VJ style copy to user, combined with checksumming. */
 
+static __inline__ void tcp_prequeue_init(struct tcp_opt *tp)
+{
+	tp->ucopy.task = NULL;
+	tp->ucopy.len = 0;
+	tp->ucopy.memory = 0;
+	skb_queue_head_init(&tp->ucopy.prequeue);
+}
 
-/*
- * Calculate(/check) TCP checksum
+/* Packet is added to VJ-style prequeue for processing in process
+ * context, if a reader task is waiting. Apparently, this exciting
+ * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
+ * failed somewhere. Latency? Burstiness? Well, at least now we will
+ * see, why it failed. 8)8)				  --ANK
  */
-static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
-				   unsigned long saddr, unsigned long daddr, 
-				   unsigned long base)
+static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
 {
-	return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
+	struct tcp_opt *tp = &sk->tp_pinfo.af_tcp;
+
+	if (tp->ucopy.task) {
+		if ((tp->ucopy.memory += skb->truesize) <= (sk->rcvbuf<<1)) {
+			__skb_queue_tail(&tp->ucopy.prequeue, skb);
+			if (skb_queue_len(&tp->ucopy.prequeue) == 1)
+				wake_up_interruptible(sk->sleep);
+		} else {
+			NET_INC_STATS_BH(TCPPrequeueDropped);
+			tp->ucopy.memory -= skb->truesize;
+			kfree_skb(skb);
+		}
+		return 1;
+	}
+	return 0;
 }
 
+
 #undef STATE_TRACE
 
 #ifdef STATE_TRACE
@@ -1007,9 +1201,12 @@
 
 static __inline__ void tcp_done(struct sock *sk)
 {
+	tcp_set_state(sk, TCP_CLOSE);
+	tcp_clear_xmit_timers(sk);
+
 	sk->shutdown = SHUTDOWN_MASK;
 
-	if (!sk->dead) 
+	if (!sk->dead)
 		sk->state_change(sk);
 	else
 		tcp_destroy_sock(sk);
@@ -1106,7 +1303,7 @@
 	 * our initial window offering to 32k. There should also
 	 * be a sysctl option to stop being nice.
 	 */
-	(*rcv_wnd) = min(space, MAX_WINDOW);
+	(*rcv_wnd) = min(space, MAX_TCP_WINDOW);
 	(*rcv_wscale) = 0;
 	if (wscale_ok) {
 		/* See RFC1323 for an explanation of the limit to 14 */
@@ -1123,52 +1320,127 @@
 extern __inline__ int tcp_space(struct sock *sk)
 {
 	return (sk->rcvbuf - atomic_read(&sk->rmem_alloc)) / 
-		WINDOW_ADVERTISE_DIVISOR; 
+		TCP_WINDOW_ADVERTISE_DIVISOR; 
 } 
 
 extern __inline__ int tcp_full_space( struct sock *sk)
 {
-	return sk->rcvbuf / WINDOW_ADVERTISE_DIVISOR; 
+	return sk->rcvbuf / TCP_WINDOW_ADVERTISE_DIVISOR; 
+}
+
+extern __inline__ void tcp_init_buffer_space(struct sock *sk)
+{
+	struct tcp_opt *tp = &(sk->tp_pinfo.af_tcp);
+	int rcvbuf = tp->advmss+MAX_TCP_HEADER+16+sizeof(struct sk_buff);
+	int sndbuf = tp->mss_clamp+MAX_TCP_HEADER+16+sizeof(struct sk_buff);
+
+	if (sk->rcvbuf < 3*rcvbuf)
+		sk->rcvbuf = min (3*rcvbuf, sysctl_rmem_max);
+	if (sk->sndbuf < 3*sndbuf)
+		sk->sndbuf = min (3*sndbuf, sysctl_wmem_max);
 }
 
-extern __inline__ void tcp_synq_unlink(struct tcp_opt *tp, struct open_request *req, struct open_request *prev)
+extern __inline__ void tcp_acceptq_removed(struct sock *sk)
 {
-	if(!req->dl_next)
-		tp->syn_wait_last = (struct open_request **)prev;
-	prev->dl_next = req->dl_next;
+	sk->ack_backlog--;
 }
 
-extern __inline__ void tcp_synq_queue(struct tcp_opt *tp, struct open_request *req)
-{ 
-	req->dl_next = NULL;
-	*tp->syn_wait_last = req; 
-	tp->syn_wait_last = &req->dl_next;
+extern __inline__ void tcp_acceptq_added(struct sock *sk)
+{
+	sk->ack_backlog++;
 }
 
-extern __inline__ void tcp_synq_init(struct tcp_opt *tp)
+extern __inline__ int tcp_acceptq_is_full(struct sock *sk)
 {
-	tp->syn_wait_queue = NULL;
-	tp->syn_wait_last = &tp->syn_wait_queue;
+	return sk->ack_backlog > sk->max_ack_backlog;
 }
 
-extern void __tcp_inc_slow_timer(struct tcp_sl_timer *slt);
-extern __inline__ void tcp_inc_slow_timer(int timer)
+extern __inline__ void tcp_acceptq_queue(struct sock *sk, struct open_request *req,
+					 struct sock *child)
 {
-	struct tcp_sl_timer *slt = &tcp_slt_array[timer];
-	
-	if (atomic_read(&slt->count) == 0)
-	{
-		__tcp_inc_slow_timer(slt);
-	}
+	struct tcp_opt *tp = &sk->tp_pinfo.af_tcp;
 
-	atomic_inc(&slt->count);
+	req->sk = child;
+	tcp_acceptq_added(sk);
+
+	req->dl_next = tp->accept_queue;
+	tp->accept_queue = req;
 }
 
-extern __inline__ void tcp_dec_slow_timer(int timer)
+struct tcp_listen_opt
+{
+	u8			max_qlen_log;	/* log_2 of maximal queued SYNs */
+	int			qlen;
+	int			qlen_young;
+	int			clock_hand;
+	struct open_request	*syn_table[TCP_SYNQ_HSIZE];
+};
+
+extern __inline__ void
+tcp_synq_removed(struct sock *sk, struct open_request *req)
 {
-	struct tcp_sl_timer *slt = &tcp_slt_array[timer];
+	struct tcp_listen_opt *lopt = sk->tp_pinfo.af_tcp.listen_opt;
+
+	if (--lopt->qlen == 0)
+		tcp_delete_keepalive_timer(sk);
+	if (req->retrans == 0)
+		lopt->qlen_young--;
+}
 
-	atomic_dec(&slt->count);
+extern __inline__ void tcp_synq_added(struct sock *sk)
+{
+	struct tcp_listen_opt *lopt = sk->tp_pinfo.af_tcp.listen_opt;
+
+	if (lopt->qlen++ == 0)
+		tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT);
+	lopt->qlen_young++;
+}
+
+extern __inline__ int tcp_synq_len(struct sock *sk)
+{
+	return sk->tp_pinfo.af_tcp.listen_opt->qlen;
+}
+
+extern __inline__ int tcp_synq_young(struct sock *sk)
+{
+	return sk->tp_pinfo.af_tcp.listen_opt->qlen_young;
+}
+
+extern __inline__ int tcp_synq_is_full(struct sock *sk)
+{
+	return tcp_synq_len(sk)>>sk->tp_pinfo.af_tcp.listen_opt->max_qlen_log;
+}
+
+extern __inline__ void tcp_synq_unlink(struct tcp_opt *tp, struct open_request *req,
+				       struct open_request **prev)
+{
+	write_lock(&tp->syn_wait_lock);
+	*prev = req->dl_next;
+	write_unlock(&tp->syn_wait_lock);
+}
+
+extern __inline__ void tcp_synq_drop(struct sock *sk, struct open_request *req,
+				     struct open_request **prev)
+{
+	tcp_synq_unlink(&sk->tp_pinfo.af_tcp, req, prev);
+	tcp_synq_removed(sk, req);
+	tcp_openreq_free(req);
+}
+
+static __inline__ void tcp_openreq_init(struct open_request *req,
+					struct tcp_opt *tp,
+					struct sk_buff *skb)
+{
+	req->rcv_wnd = 0;		/* So that tcp_send_synack() knows! */
+	req->rcv_isn = TCP_SKB_CB(skb)->seq;
+	req->mss = tp->mss_clamp;
+	req->ts_recent = tp->saw_tstamp ? tp->rcv_tsval : 0;
+	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->acked = 0;
+	req->rmt_port = skb->h.th->source;
 }
 
 extern const char timer_bug_msg[];
@@ -1179,13 +1451,14 @@
 	struct timer_list *timer;
 	
 	switch (what) {
-	case TIME_RETRANS:
+	case TCP_TIME_RETRANS:
 		timer = &tp->retransmit_timer;
 		break;
-	case TIME_DACK:
+	case TCP_TIME_DACK:
+		tp->ack.blocked = 0;
 		timer = &tp->delack_timer;
 		break;
-	case TIME_PROBE0:
+	case TCP_TIME_PROBE0:
 		timer = &tp->probe_timer;
 		break;	
 	default:
@@ -1199,7 +1472,7 @@
 	spin_unlock_bh(&sk->timer_lock);
 }
 
-/* This function does not return reliable answer. You is only as advice.
+/* This function does not return reliable answer. Use it only as advice.
  */
 
 static inline int tcp_timer_is_set(struct sock *sk, int what)
@@ -1208,13 +1481,13 @@
 	int ret;
 
 	switch (what) {
-	case TIME_RETRANS:
+	case TCP_TIME_RETRANS:
 		ret = tp->retransmit_timer.prev != NULL;
 		break;
-	case TIME_DACK:
+	case TCP_TIME_DACK:
 		ret = tp->delack_timer.prev != NULL;
 		break;
-	case TIME_PROBE0:
+	case TCP_TIME_PROBE0:
 		ret = tp->probe_timer.prev != NULL;
 		break;	
 	default:
@@ -1248,18 +1521,46 @@
 
 static inline int keepalive_intvl_when(struct tcp_opt *tp)
 {
-	if (tp->keepalive_intvl)
-		return tp->keepalive_intvl;
-	else
-		return sysctl_tcp_keepalive_intvl;
+	return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
 }
 
 static inline int keepalive_time_when(struct tcp_opt *tp)
 {
-	if (tp->keepalive_time)
-		return tp->keepalive_time;
-	else
-		return sysctl_tcp_keepalive_time;
+	return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
 }
+
+static inline int tcp_fin_time(struct tcp_opt *tp)
+{
+	int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout;
+
+	if (fin_timeout < (tp->rto<<2) - (tp->rto>>1))
+		fin_timeout = (tp->rto<<2) - (tp->rto>>1);
+
+	return fin_timeout;
+}
+
+#if 0 /* TCP_DEBUG */
+#define TCP_CHECK_TIMER(sk) \
+do { 	struct tcp_opt *__tp = &sk->tp_pinfo.af_tcp; \
+	if (sk->state != TCP_CLOSE) { \
+		if (__tp->packets_out) { \
+			if (!tcp_timer_is_set(sk, TCP_TIME_RETRANS) && !timer_is_running(&__tp->retransmit_timer) && net_ratelimit()) \
+				printk(KERN_DEBUG "sk=%p RETRANS" __FUNCTION__ "(%d) %d\n", sk, __LINE__, sk->state); \
+		} else if (__tp->send_head) { \
+			if (!tcp_timer_is_set(sk, TCP_TIME_PROBE0) && !timer_is_running(&__tp->probe_timer) && net_ratelimit()) \
+				printk(KERN_DEBUG "sk=%p PROBE0" __FUNCTION__ "(%d) %d\n", sk, __LINE__, sk->state); \
+		} \
+	        if (__tp->ack.pending) { \
+			if (!tcp_timer_is_set(sk, TCP_TIME_DACK) && !timer_is_running(&__tp->delack_timer) && net_ratelimit()) \
+				printk(KERN_DEBUG "sk=%p DACK" __FUNCTION__ "(%d) %d\n", sk, __LINE__, sk->state); \
+		} \
+                if (__tp->packets_out > skb_queue_len(&sk->write_queue) || \
+		    (__tp->send_head && skb_queue_len(&sk->write_queue) == 0)) { \
+			 printk(KERN_DEBUG "sk=%p QUEUE" __FUNCTION__ "(%d) %d %d %d %p\n", sk, __LINE__, sk->state, __tp->packets_out, skb_queue_len(&sk->write_queue), __tp->send_head); \
+		} \
+	} } while (0)
+#else
+#define TCP_CHECK_TIMER(sk) do { } while (0);
+#endif
 
 #endif	/* _TCP_H */

FUNET's LINUX-ADM group, linux-adm@nic.funet.fi
TCL-scripts by Sam Shen (who was at: slshen@lbl.gov)