2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the TCP module.
8 * Version: @(#)tcp.h 1.0.5 05/23/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
22 #define FASTRETRANS_DEBUG 1
24 #include <linux/config.h>
25 #include <linux/list.h>
26 #include <linux/tcp.h>
27 #include <linux/slab.h>
28 #include <linux/cache.h>
29 #include <linux/percpu.h>
31 #include <net/inet_connection_sock.h>
32 #include <net/inet_timewait_sock.h>
33 #include <net/inet_hashtables.h>
34 #include <net/checksum.h>
35 #include <net/request_sock.h>
39 #include <net/tcp_states.h>
41 #include <linux/seq_file.h>
43 extern struct inet_hashinfo tcp_hashinfo;
45 extern atomic_t tcp_orphan_count;
46 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
47 extern void inet_twsk_deschedule(struct inet_timewait_sock *tw,
48 struct inet_timewait_death_row *twdr);
50 #define MAX_TCP_HEADER (128 + MAX_HEADER)
53 * Never offer a window over 32767 without using window scaling. Some
54 * poor stacks do signed 16bit maths!
56 #define MAX_TCP_WINDOW 32767U
58 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
59 #define TCP_MIN_MSS 88U
61 /* Minimal RCV_MSS. */
62 #define TCP_MIN_RCVMSS 536U
64 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
65 #define TCP_FASTRETRANS_THRESH 3
67 /* Maximal reordering. */
68 #define TCP_MAX_REORDERING 127
70 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
71 #define TCP_MAX_QUICKACKS 16U
74 #define TCP_URG_VALID 0x0100
75 #define TCP_URG_NOTYET 0x0200
76 #define TCP_URG_READ 0x0400
78 #define TCP_RETR1 3 /*
79 * This is how many retries it does before it
80 * tries to figure out if the gateway is
81 * down. Minimal RFC value is 3; it corresponds
82 * to ~3sec-8min depending on RTO.
85 #define TCP_RETR2 15 /*
86 * This should take at least
87 * 90 minutes to time out.
88 * RFC1122 says that the limit is 100 sec.
89 * 15 is ~13-30min depending on RTO.
92 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
93 * connection: ~180sec is RFC minumum */
95 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
96 * connection: ~180sec is RFC minumum */
99 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
100 * socket. 7 is ~50sec-16min.
104 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
105 * state, about 60 seconds */
106 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
107 /* BSD style FIN_WAIT2 deadlock breaker.
108 * It used to be 3min, new value is 60sec,
109 * to combine FIN-WAIT-2 timeout with
113 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
115 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
116 #define TCP_ATO_MIN ((unsigned)(HZ/25))
118 #define TCP_DELACK_MIN 4U
119 #define TCP_ATO_MIN 4U
121 #define TCP_RTO_MAX ((unsigned)(120*HZ))
122 #define TCP_RTO_MIN ((unsigned)(HZ/5))
123 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
125 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
126 * for local resources.
129 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
130 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
131 #define TCP_KEEPALIVE_INTVL (75*HZ)
133 #define MAX_TCP_KEEPIDLE 32767
134 #define MAX_TCP_KEEPINTVL 32767
135 #define MAX_TCP_KEEPCNT 127
136 #define MAX_TCP_SYNCNT 127
138 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
139 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
141 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
142 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
143 * after this time. It should be equal
144 * (or greater than) TCP_TIMEWAIT_LEN
145 * to provide reliability equal to one
146 * provided by timewait state.
148 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
149 * timestamps. It must be less than
150 * minimal timewait lifetime.
156 #define TCPOPT_NOP 1 /* Padding */
157 #define TCPOPT_EOL 0 /* End of options */
158 #define TCPOPT_MSS 2 /* Segment size negotiating */
159 #define TCPOPT_WINDOW 3 /* Window scaling */
160 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
161 #define TCPOPT_SACK 5 /* SACK Block */
162 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
168 #define TCPOLEN_MSS 4
169 #define TCPOLEN_WINDOW 3
170 #define TCPOLEN_SACK_PERM 2
171 #define TCPOLEN_TIMESTAMP 10
173 /* But this is what stacks really send out. */
174 #define TCPOLEN_TSTAMP_ALIGNED 12
175 #define TCPOLEN_WSCALE_ALIGNED 4
176 #define TCPOLEN_SACKPERM_ALIGNED 4
177 #define TCPOLEN_SACK_BASE 2
178 #define TCPOLEN_SACK_BASE_ALIGNED 4
179 #define TCPOLEN_SACK_PERBLOCK 8
181 /* Flags in tp->nonagle */
182 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
183 #define TCP_NAGLE_CORK 2 /* Socket is corked */
184 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
186 extern struct inet_timewait_death_row tcp_death_row;
188 /* sysctl variables for tcp */
189 extern int sysctl_tcp_timestamps;
190 extern int sysctl_tcp_window_scaling;
191 extern int sysctl_tcp_sack;
192 extern int sysctl_tcp_fin_timeout;
193 extern int sysctl_tcp_keepalive_time;
194 extern int sysctl_tcp_keepalive_probes;
195 extern int sysctl_tcp_keepalive_intvl;
196 extern int sysctl_tcp_syn_retries;
197 extern int sysctl_tcp_synack_retries;
198 extern int sysctl_tcp_retries1;
199 extern int sysctl_tcp_retries2;
200 extern int sysctl_tcp_orphan_retries;
201 extern int sysctl_tcp_syncookies;
202 extern int sysctl_tcp_retrans_collapse;
203 extern int sysctl_tcp_stdurg;
204 extern int sysctl_tcp_rfc1337;
205 extern int sysctl_tcp_abort_on_overflow;
206 extern int sysctl_tcp_max_orphans;
207 extern int sysctl_tcp_fack;
208 extern int sysctl_tcp_reordering;
209 extern int sysctl_tcp_ecn;
210 extern int sysctl_tcp_dsack;
211 extern int sysctl_tcp_mem[3];
212 extern int sysctl_tcp_wmem[3];
213 extern int sysctl_tcp_rmem[3];
214 extern int sysctl_tcp_app_win;
215 extern int sysctl_tcp_adv_win_scale;
216 extern int sysctl_tcp_tw_reuse;
217 extern int sysctl_tcp_frto;
218 extern int sysctl_tcp_low_latency;
219 extern int sysctl_tcp_nometrics_save;
220 extern int sysctl_tcp_moderate_rcvbuf;
221 extern int sysctl_tcp_tso_win_divisor;
223 extern atomic_t tcp_memory_allocated;
224 extern atomic_t tcp_sockets_allocated;
225 extern int tcp_memory_pressure;
228 * Pointers to address related TCP functions
229 * (i.e. things that depend on the address family)
233 int (*queue_xmit) (struct sk_buff *skb,
236 void (*send_check) (struct sock *sk,
239 struct sk_buff *skb);
241 int (*rebuild_header) (struct sock *sk);
243 int (*conn_request) (struct sock *sk,
244 struct sk_buff *skb);
246 struct sock * (*syn_recv_sock) (struct sock *sk,
248 struct request_sock *req,
249 struct dst_entry *dst);
251 int (*remember_stamp) (struct sock *sk);
253 __u16 net_header_len;
255 int (*setsockopt) (struct sock *sk,
261 int (*getsockopt) (struct sock *sk,
268 void (*addr2sockaddr) (struct sock *sk,
275 * The next routines deal with comparing 32 bit unsigned ints
276 * and worry about wraparound (automatic with unsigned arithmetic).
279 static inline int before(__u32 seq1, __u32 seq2)
281 return (__s32)(seq1-seq2) < 0;
284 static inline int after(__u32 seq1, __u32 seq2)
286 return (__s32)(seq2-seq1) < 0;
290 /* is s2<=s1<=s3 ? */
291 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
293 return seq3 - seq2 >= seq1 - seq2;
297 extern struct proto tcp_prot;
299 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
300 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
301 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
302 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
303 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
304 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
305 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
307 extern void tcp_v4_err(struct sk_buff *skb, u32);
309 extern void tcp_shutdown (struct sock *sk, int how);
311 extern int tcp_v4_rcv(struct sk_buff *skb);
313 extern int tcp_v4_remember_stamp(struct sock *sk);
315 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
317 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
318 struct msghdr *msg, size_t size);
319 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
321 extern int tcp_ioctl(struct sock *sk,
325 extern int tcp_rcv_state_process(struct sock *sk,
330 extern int tcp_rcv_established(struct sock *sk,
335 extern void tcp_rcv_space_adjust(struct sock *sk);
337 static inline void tcp_dec_quickack_mode(struct sock *sk,
338 const unsigned int pkts)
340 struct inet_connection_sock *icsk = inet_csk(sk);
342 if (icsk->icsk_ack.quick) {
343 if (pkts >= icsk->icsk_ack.quick) {
344 icsk->icsk_ack.quick = 0;
345 /* Leaving quickack mode we deflate ATO. */
346 icsk->icsk_ack.ato = TCP_ATO_MIN;
348 icsk->icsk_ack.quick -= pkts;
352 extern void tcp_enter_quickack_mode(struct sock *sk);
354 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
356 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
368 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
370 const struct tcphdr *th);
372 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
373 struct request_sock *req,
374 struct request_sock **prev);
375 extern int tcp_child_process(struct sock *parent,
377 struct sk_buff *skb);
378 extern void tcp_enter_frto(struct sock *sk);
379 extern void tcp_enter_loss(struct sock *sk, int how);
380 extern void tcp_clear_retrans(struct tcp_sock *tp);
381 extern void tcp_update_metrics(struct sock *sk);
383 extern void tcp_close(struct sock *sk,
385 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
387 extern int tcp_getsockopt(struct sock *sk, int level,
391 extern int tcp_setsockopt(struct sock *sk, int level,
392 int optname, char __user *optval,
394 extern void tcp_set_keepalive(struct sock *sk, int val);
395 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
397 size_t len, int nonblock,
398 int flags, int *addr_len);
400 extern void tcp_parse_options(struct sk_buff *skb,
401 struct tcp_options_received *opt_rx,
405 * TCP v4 functions exported for the inet6 API
408 extern void tcp_v4_send_check(struct sock *sk,
409 struct tcphdr *th, int len,
410 struct sk_buff *skb);
412 extern int tcp_v4_conn_request(struct sock *sk,
413 struct sk_buff *skb);
415 extern struct sock * tcp_create_openreq_child(struct sock *sk,
416 struct request_sock *req,
417 struct sk_buff *skb);
419 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
421 struct request_sock *req,
422 struct dst_entry *dst);
424 extern int tcp_v4_do_rcv(struct sock *sk,
425 struct sk_buff *skb);
427 extern int tcp_v4_connect(struct sock *sk,
428 struct sockaddr *uaddr,
431 extern int tcp_connect(struct sock *sk);
433 extern struct sk_buff * tcp_make_synack(struct sock *sk,
434 struct dst_entry *dst,
435 struct request_sock *req);
437 extern int tcp_disconnect(struct sock *sk, int flags);
439 extern void tcp_unhash(struct sock *sk);
441 extern int tcp_v4_hash_connecting(struct sock *sk);
444 /* From syncookies.c */
445 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
446 struct ip_options *opt);
447 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
452 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
453 unsigned int cur_mss, int nonagle);
454 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
455 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
456 extern void tcp_xmit_retransmit_queue(struct sock *);
457 extern void tcp_simple_retransmit(struct sock *);
458 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
460 extern void tcp_send_probe0(struct sock *);
461 extern void tcp_send_partial(struct sock *);
462 extern int tcp_write_wakeup(struct sock *);
463 extern void tcp_send_fin(struct sock *sk);
464 extern void tcp_send_active_reset(struct sock *sk,
465 unsigned int __nocast priority);
466 extern int tcp_send_synack(struct sock *);
467 extern void tcp_push_one(struct sock *, unsigned int mss_now);
468 extern void tcp_send_ack(struct sock *sk);
469 extern void tcp_send_delayed_ack(struct sock *sk);
472 extern void tcp_cwnd_application_limited(struct sock *sk);
475 extern void tcp_init_xmit_timers(struct sock *);
476 static inline void tcp_clear_xmit_timers(struct sock *sk)
478 inet_csk_clear_xmit_timers(sk);
481 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
482 extern unsigned int tcp_current_mss(struct sock *sk, int large);
485 extern void tcp_get_info(struct sock *, struct tcp_info *);
487 /* Read 'sendfile()'-style from a TCP socket */
488 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
489 unsigned int, size_t);
490 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
491 sk_read_actor_t recv_actor);
493 /* Initialize RCV_MSS value.
494 * RCV_MSS is an our guess about MSS used by the peer.
495 * We haven't any direct information about the MSS.
496 * It's better to underestimate the RCV_MSS rather than overestimate.
497 * Overestimations make us ACKing less frequently than needed.
498 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
501 static inline void tcp_initialize_rcv_mss(struct sock *sk)
503 struct tcp_sock *tp = tcp_sk(sk);
504 unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);
506 hint = min(hint, tp->rcv_wnd/2);
507 hint = min(hint, TCP_MIN_RCVMSS);
508 hint = max(hint, TCP_MIN_MSS);
510 inet_csk(sk)->icsk_ack.rcv_mss = hint;
513 static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
515 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
516 ntohl(TCP_FLAG_ACK) |
520 static __inline__ void tcp_fast_path_on(struct tcp_sock *tp)
522 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
525 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
527 if (skb_queue_empty(&tp->out_of_order_queue) &&
529 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
531 tcp_fast_path_on(tp);
534 /* Compute the actual receive window we are currently advertising.
535 * Rcv_nxt can be after the window if our peer push more data
536 * than the offered window.
538 static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp)
540 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
547 /* Choose a new window, without checks for shrinking, and without
548 * scaling applied to the result. The caller does these things
549 * if necessary. This is a "raw" window selection.
551 extern u32 __tcp_select_window(struct sock *sk);
553 /* TCP timestamps are only 32-bits, this causes a slight
554 * complication on 64-bit systems since we store a snapshot
555 * of jiffies in the buffer control blocks below. We decidely
556 * only use of the low 32-bits of jiffies and hide the ugly
557 * casts with the following macro.
559 #define tcp_time_stamp ((__u32)(jiffies))
561 /* This is what the send packet queueing engine uses to pass
562 * TCP per-packet control information to the transmission
563 * code. We also store the host-order sequence numbers in
564 * here too. This is 36 bytes on 32-bit architectures,
565 * 40 bytes on 64-bit machines, if this grows please adjust
566 * skbuff.h:skbuff->cb[xxx] size appropriately.
570 struct inet_skb_parm h4;
571 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
572 struct inet6_skb_parm h6;
574 } header; /* For incoming frames */
575 __u32 seq; /* Starting sequence number */
576 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
577 __u32 when; /* used to compute rtt's */
578 __u8 flags; /* TCP header flags. */
580 /* NOTE: These must match up to the flags byte in a
583 #define TCPCB_FLAG_FIN 0x01
584 #define TCPCB_FLAG_SYN 0x02
585 #define TCPCB_FLAG_RST 0x04
586 #define TCPCB_FLAG_PSH 0x08
587 #define TCPCB_FLAG_ACK 0x10
588 #define TCPCB_FLAG_URG 0x20
589 #define TCPCB_FLAG_ECE 0x40
590 #define TCPCB_FLAG_CWR 0x80
592 __u8 sacked; /* State flags for SACK/FACK. */
593 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
594 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
595 #define TCPCB_LOST 0x04 /* SKB is lost */
596 #define TCPCB_TAGBITS 0x07 /* All tag bits */
598 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
599 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
601 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
603 #define TCPCB_AT_TAIL (TCPCB_URG)
605 __u16 urg_ptr; /* Valid w/URG flags is set. */
606 __u32 ack_seq; /* Sequence number ACK'd */
609 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
611 #include <net/tcp_ecn.h>
613 /* Due to TSO, an SKB can be composed of multiple actual
614 * packets. To keep these tracked properly, we use this.
616 static inline int tcp_skb_pcount(const struct sk_buff *skb)
618 return skb_shinfo(skb)->tso_segs;
621 /* This is valid iff tcp_skb_pcount() > 1. */
622 static inline int tcp_skb_mss(const struct sk_buff *skb)
624 return skb_shinfo(skb)->tso_size;
627 static inline void tcp_dec_pcount_approx(__u32 *count,
628 const struct sk_buff *skb)
631 *count -= tcp_skb_pcount(skb);
637 static inline void tcp_packets_out_inc(struct sock *sk,
639 const struct sk_buff *skb)
641 int orig = tp->packets_out;
643 tp->packets_out += tcp_skb_pcount(skb);
645 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
646 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
649 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
650 const struct sk_buff *skb)
652 tp->packets_out -= tcp_skb_pcount(skb);
655 /* Events passed to congestion control interface */
657 CA_EVENT_TX_START, /* first transmit when no packets in flight */
658 CA_EVENT_CWND_RESTART, /* congestion window restart */
659 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
660 CA_EVENT_FRTO, /* fast recovery timeout */
661 CA_EVENT_LOSS, /* loss timeout */
662 CA_EVENT_FAST_ACK, /* in sequence ack */
663 CA_EVENT_SLOW_ACK, /* other ack */
667 * Interface for adding new TCP congestion control handlers
669 #define TCP_CA_NAME_MAX 16
670 struct tcp_congestion_ops {
671 struct list_head list;
673 /* initialize private data (optional) */
674 void (*init)(struct tcp_sock *tp);
675 /* cleanup private data (optional) */
676 void (*release)(struct tcp_sock *tp);
678 /* return slow start threshold (required) */
679 u32 (*ssthresh)(struct tcp_sock *tp);
680 /* lower bound for congestion window (optional) */
681 u32 (*min_cwnd)(struct tcp_sock *tp);
682 /* do new cwnd calculation (required) */
683 void (*cong_avoid)(struct tcp_sock *tp, u32 ack,
684 u32 rtt, u32 in_flight, int good_ack);
685 /* round trip time sample per acked packet (optional) */
686 void (*rtt_sample)(struct tcp_sock *tp, u32 usrtt);
687 /* call before changing ca_state (optional) */
688 void (*set_state)(struct tcp_sock *tp, u8 new_state);
689 /* call when cwnd event occurs (optional) */
690 void (*cwnd_event)(struct tcp_sock *tp, enum tcp_ca_event ev);
691 /* new value of cwnd after loss (optional) */
692 u32 (*undo_cwnd)(struct tcp_sock *tp);
693 /* hook for packet ack accounting (optional) */
694 void (*pkts_acked)(struct tcp_sock *tp, u32 num_acked);
695 /* get info for tcp_diag (optional) */
696 void (*get_info)(struct tcp_sock *tp, u32 ext, struct sk_buff *skb);
698 char name[TCP_CA_NAME_MAX];
699 struct module *owner;
702 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
703 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
705 extern void tcp_init_congestion_control(struct tcp_sock *tp);
706 extern void tcp_cleanup_congestion_control(struct tcp_sock *tp);
707 extern int tcp_set_default_congestion_control(const char *name);
708 extern void tcp_get_default_congestion_control(char *name);
709 extern int tcp_set_congestion_control(struct tcp_sock *tp, const char *name);
711 extern struct tcp_congestion_ops tcp_init_congestion_ops;
712 extern u32 tcp_reno_ssthresh(struct tcp_sock *tp);
713 extern void tcp_reno_cong_avoid(struct tcp_sock *tp, u32 ack,
714 u32 rtt, u32 in_flight, int flag);
715 extern u32 tcp_reno_min_cwnd(struct tcp_sock *tp);
716 extern struct tcp_congestion_ops tcp_reno;
718 static inline void tcp_set_ca_state(struct tcp_sock *tp, u8 ca_state)
720 if (tp->ca_ops->set_state)
721 tp->ca_ops->set_state(tp, ca_state);
722 tp->ca_state = ca_state;
725 static inline void tcp_ca_event(struct tcp_sock *tp, enum tcp_ca_event event)
727 if (tp->ca_ops->cwnd_event)
728 tp->ca_ops->cwnd_event(tp, event);
731 /* This determines how many packets are "in the network" to the best
732 * of our knowledge. In many cases it is conservative, but where
733 * detailed information is available from the receiver (via SACK
734 * blocks etc.) we can make more aggressive calculations.
736 * Use this for decisions involving congestion control, use just
737 * tp->packets_out to determine if the send queue is empty or not.
739 * Read this equation as:
741 * "Packets sent once on transmission queue" MINUS
742 * "Packets left network, but not honestly ACKed yet" PLUS
743 * "Packets fast retransmitted"
745 static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
747 return (tp->packets_out - tp->left_out + tp->retrans_out);
750 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
751 * The exception is rate halving phase, when cwnd is decreasing towards
754 static inline __u32 tcp_current_ssthresh(struct tcp_sock *tp)
756 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
757 return tp->snd_ssthresh;
759 return max(tp->snd_ssthresh,
760 ((tp->snd_cwnd >> 1) +
761 (tp->snd_cwnd >> 2)));
764 static inline void tcp_sync_left_out(struct tcp_sock *tp)
766 if (tp->rx_opt.sack_ok &&
767 (tp->sacked_out >= tp->packets_out - tp->lost_out))
768 tp->sacked_out = tp->packets_out - tp->lost_out;
769 tp->left_out = tp->sacked_out + tp->lost_out;
772 /* Set slow start threshold and cwnd not falling to slow start */
773 static inline void __tcp_enter_cwr(struct tcp_sock *tp)
776 tp->snd_ssthresh = tp->ca_ops->ssthresh(tp);
777 tp->snd_cwnd = min(tp->snd_cwnd,
778 tcp_packets_in_flight(tp) + 1U);
779 tp->snd_cwnd_cnt = 0;
780 tp->high_seq = tp->snd_nxt;
781 tp->snd_cwnd_stamp = tcp_time_stamp;
782 TCP_ECN_queue_cwr(tp);
785 static inline void tcp_enter_cwr(struct tcp_sock *tp)
787 tp->prior_ssthresh = 0;
788 if (tp->ca_state < TCP_CA_CWR) {
790 tcp_set_ca_state(tp, TCP_CA_CWR);
794 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
796 /* Slow start with delack produces 3 packets of burst, so that
797 * it is safe "de facto".
799 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
804 static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss,
805 const struct sk_buff *skb)
808 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
811 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
813 const struct inet_connection_sock *icsk = inet_csk(sk);
814 if (!tp->packets_out && !icsk->icsk_pending)
815 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
816 icsk->icsk_rto, TCP_RTO_MAX);
819 static __inline__ void tcp_push_pending_frames(struct sock *sk,
822 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
825 static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
830 static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
836 * Calculate(/check) TCP checksum
838 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
839 unsigned long saddr, unsigned long daddr,
842 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
845 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
847 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
850 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
852 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
853 __tcp_checksum_complete(skb);
856 /* Prequeue for VJ style copy to user, combined with checksumming. */
858 static __inline__ void tcp_prequeue_init(struct tcp_sock *tp)
860 tp->ucopy.task = NULL;
862 tp->ucopy.memory = 0;
863 skb_queue_head_init(&tp->ucopy.prequeue);
866 /* Packet is added to VJ-style prequeue for processing in process
867 * context, if a reader task is waiting. Apparently, this exciting
868 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
869 * failed somewhere. Latency? Burstiness? Well, at least now we will
870 * see, why it failed. 8)8) --ANK
872 * NOTE: is this not too big to inline?
874 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
876 struct tcp_sock *tp = tcp_sk(sk);
878 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
879 __skb_queue_tail(&tp->ucopy.prequeue, skb);
880 tp->ucopy.memory += skb->truesize;
881 if (tp->ucopy.memory > sk->sk_rcvbuf) {
882 struct sk_buff *skb1;
884 BUG_ON(sock_owned_by_user(sk));
886 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
887 sk->sk_backlog_rcv(sk, skb1);
888 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
891 tp->ucopy.memory = 0;
892 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
893 wake_up_interruptible(sk->sk_sleep);
894 if (!inet_csk_ack_scheduled(sk))
895 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
896 (3 * TCP_RTO_MIN) / 4,
908 static const char *statename[]={
909 "Unused","Established","Syn Sent","Syn Recv",
910 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
911 "Close Wait","Last ACK","Listen","Closing"
915 static __inline__ void tcp_set_state(struct sock *sk, int state)
917 int oldstate = sk->sk_state;
920 case TCP_ESTABLISHED:
921 if (oldstate != TCP_ESTABLISHED)
922 TCP_INC_STATS(TCP_MIB_CURRESTAB);
926 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
927 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
929 sk->sk_prot->unhash(sk);
930 if (inet_csk(sk)->icsk_bind_hash &&
931 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
932 inet_put_port(&tcp_hashinfo, sk);
935 if (oldstate==TCP_ESTABLISHED)
936 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
939 /* Change state AFTER socket is unhashed to avoid closed
940 * socket sitting in hash tables.
942 sk->sk_state = state;
945 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
949 static __inline__ void tcp_done(struct sock *sk)
951 tcp_set_state(sk, TCP_CLOSE);
952 tcp_clear_xmit_timers(sk);
954 sk->sk_shutdown = SHUTDOWN_MASK;
956 if (!sock_flag(sk, SOCK_DEAD))
957 sk->sk_state_change(sk);
959 inet_csk_destroy_sock(sk);
962 static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt)
965 rx_opt->eff_sacks = 0;
966 rx_opt->num_sacks = 0;
969 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp)
971 if (tp->rx_opt.tstamp_ok) {
972 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
974 (TCPOPT_TIMESTAMP << 8) |
976 *ptr++ = htonl(tstamp);
977 *ptr++ = htonl(tp->rx_opt.ts_recent);
979 if (tp->rx_opt.eff_sacks) {
980 struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks;
983 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
987 (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK)));
988 for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) {
989 *ptr++ = htonl(sp[this_sack].start_seq);
990 *ptr++ = htonl(sp[this_sack].end_seq);
992 if (tp->rx_opt.dsack) {
993 tp->rx_opt.dsack = 0;
994 tp->rx_opt.eff_sacks--;
999 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1000 * If this is every changed make sure to change the definition of
1001 * MAX_SYN_SIZE to match the new maximum number of options that you
1004 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1005 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1007 /* We always get an MSS option.
1008 * The option bytes which will be seen in normal data
1009 * packets should timestamps be used, must be in the MSS
1010 * advertised. But we subtract them from tp->mss_cache so
1011 * that calculations in tcp_sendmsg are simpler etc.
1012 * So account for this fact here if necessary. If we
1013 * don't do this correctly, as a receiver we won't
1014 * recognize data packets as being full sized when we
1015 * should, and thus we won't abide by the delayed ACK
1017 * SACKs don't matter, we never delay an ACK when we
1018 * have any of those going out.
1020 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1023 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1024 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1026 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1027 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1028 *ptr++ = htonl(tstamp); /* TSVAL */
1029 *ptr++ = htonl(ts_recent); /* TSECR */
1031 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1032 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1034 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1037 /* Determine a window scaling and initial window to offer. */
1038 extern void tcp_select_initial_window(int __space, __u32 mss,
1039 __u32 *rcv_wnd, __u32 *window_clamp,
1040 int wscale_ok, __u8 *rcv_wscale);
1042 static inline int tcp_win_from_space(int space)
1044 return sysctl_tcp_adv_win_scale<=0 ?
1045 (space>>(-sysctl_tcp_adv_win_scale)) :
1046 space - (space>>sysctl_tcp_adv_win_scale);
1049 /* Note: caller must be prepared to deal with negative returns */
1050 static inline int tcp_space(const struct sock *sk)
1052 return tcp_win_from_space(sk->sk_rcvbuf -
1053 atomic_read(&sk->sk_rmem_alloc));
1056 static inline int tcp_full_space(const struct sock *sk)
1058 return tcp_win_from_space(sk->sk_rcvbuf);
1061 static __inline__ void tcp_openreq_init(struct request_sock *req,
1062 struct tcp_options_received *rx_opt,
1063 struct sk_buff *skb)
1065 struct inet_request_sock *ireq = inet_rsk(req);
1067 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1068 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1069 req->mss = rx_opt->mss_clamp;
1070 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1071 ireq->tstamp_ok = rx_opt->tstamp_ok;
1072 ireq->sack_ok = rx_opt->sack_ok;
1073 ireq->snd_wscale = rx_opt->snd_wscale;
1074 ireq->wscale_ok = rx_opt->wscale_ok;
1077 ireq->rmt_port = skb->h.th->source;
1080 extern void tcp_enter_memory_pressure(void);
1082 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1084 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1087 static inline int keepalive_time_when(const struct tcp_sock *tp)
1089 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1092 static inline int tcp_fin_time(const struct sock *sk)
1094 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1095 const int rto = inet_csk(sk)->icsk_rto;
1097 if (fin_timeout < (rto << 2) - (rto >> 1))
1098 fin_timeout = (rto << 2) - (rto >> 1);
1103 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1105 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1107 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1110 /* RST segments are not recommended to carry timestamp,
1111 and, if they do, it is recommended to ignore PAWS because
1112 "their cleanup function should take precedence over timestamps."
1113 Certainly, it is mistake. It is necessary to understand the reasons
1114 of this constraint to relax it: if peer reboots, clock may go
1115 out-of-sync and half-open connections will not be reset.
1116 Actually, the problem would be not existing if all
1117 the implementations followed draft about maintaining clock
1118 via reboots. Linux-2.2 DOES NOT!
1120 However, we can relax time bounds for RST segments to MSL.
1122 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1127 #define TCP_CHECK_TIMER(sk) do { } while (0)
1129 static inline int tcp_use_frto(const struct sock *sk)
1131 const struct tcp_sock *tp = tcp_sk(sk);
1133 /* F-RTO must be activated in sysctl and there must be some
1134 * unsent new data, and the advertised window should allow
1137 return (sysctl_tcp_frto && sk->sk_send_head &&
1138 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1139 tp->snd_una + tp->snd_wnd));
1142 static inline void tcp_mib_init(void)
1145 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1146 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1147 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1148 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1152 enum tcp_seq_states {
1153 TCP_SEQ_STATE_LISTENING,
1154 TCP_SEQ_STATE_OPENREQ,
1155 TCP_SEQ_STATE_ESTABLISHED,
1156 TCP_SEQ_STATE_TIME_WAIT,
1159 struct tcp_seq_afinfo {
1160 struct module *owner;
1163 int (*seq_show) (struct seq_file *m, void *v);
1164 struct file_operations *seq_fops;
1167 struct tcp_iter_state {
1169 enum tcp_seq_states state;
1170 struct sock *syn_wait_sk;
1171 int bucket, sbucket, num, uid;
1172 struct seq_operations seq_ops;
1175 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1176 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);