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);
48 #define MAX_TCP_HEADER (128 + MAX_HEADER)
51 * Never offer a window over 32767 without using window scaling. Some
52 * poor stacks do signed 16bit maths!
54 #define MAX_TCP_WINDOW 32767U
56 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
57 #define TCP_MIN_MSS 88U
59 /* Minimal RCV_MSS. */
60 #define TCP_MIN_RCVMSS 536U
62 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
63 #define TCP_FASTRETRANS_THRESH 3
65 /* Maximal reordering. */
66 #define TCP_MAX_REORDERING 127
68 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
69 #define TCP_MAX_QUICKACKS 16U
72 #define TCP_URG_VALID 0x0100
73 #define TCP_URG_NOTYET 0x0200
74 #define TCP_URG_READ 0x0400
76 #define TCP_RETR1 3 /*
77 * This is how many retries it does before it
78 * tries to figure out if the gateway is
79 * down. Minimal RFC value is 3; it corresponds
80 * to ~3sec-8min depending on RTO.
83 #define TCP_RETR2 15 /*
84 * This should take at least
85 * 90 minutes to time out.
86 * RFC1122 says that the limit is 100 sec.
87 * 15 is ~13-30min depending on RTO.
90 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
91 * connection: ~180sec is RFC minumum */
93 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
94 * connection: ~180sec is RFC minumum */
97 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
98 * socket. 7 is ~50sec-16min.
102 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
103 * state, about 60 seconds */
104 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
105 /* BSD style FIN_WAIT2 deadlock breaker.
106 * It used to be 3min, new value is 60sec,
107 * to combine FIN-WAIT-2 timeout with
111 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
113 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
114 #define TCP_ATO_MIN ((unsigned)(HZ/25))
116 #define TCP_DELACK_MIN 4U
117 #define TCP_ATO_MIN 4U
119 #define TCP_RTO_MAX ((unsigned)(120*HZ))
120 #define TCP_RTO_MIN ((unsigned)(HZ/5))
121 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
123 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
124 * for local resources.
127 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
128 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
129 #define TCP_KEEPALIVE_INTVL (75*HZ)
131 #define MAX_TCP_KEEPIDLE 32767
132 #define MAX_TCP_KEEPINTVL 32767
133 #define MAX_TCP_KEEPCNT 127
134 #define MAX_TCP_SYNCNT 127
136 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
137 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
139 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
140 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
141 * after this time. It should be equal
142 * (or greater than) TCP_TIMEWAIT_LEN
143 * to provide reliability equal to one
144 * provided by timewait state.
146 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
147 * timestamps. It must be less than
148 * minimal timewait lifetime.
154 #define TCPOPT_NOP 1 /* Padding */
155 #define TCPOPT_EOL 0 /* End of options */
156 #define TCPOPT_MSS 2 /* Segment size negotiating */
157 #define TCPOPT_WINDOW 3 /* Window scaling */
158 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
159 #define TCPOPT_SACK 5 /* SACK Block */
160 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
166 #define TCPOLEN_MSS 4
167 #define TCPOLEN_WINDOW 3
168 #define TCPOLEN_SACK_PERM 2
169 #define TCPOLEN_TIMESTAMP 10
171 /* But this is what stacks really send out. */
172 #define TCPOLEN_TSTAMP_ALIGNED 12
173 #define TCPOLEN_WSCALE_ALIGNED 4
174 #define TCPOLEN_SACKPERM_ALIGNED 4
175 #define TCPOLEN_SACK_BASE 2
176 #define TCPOLEN_SACK_BASE_ALIGNED 4
177 #define TCPOLEN_SACK_PERBLOCK 8
179 /* Flags in tp->nonagle */
180 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
181 #define TCP_NAGLE_CORK 2 /* Socket is corked */
182 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
184 extern struct inet_timewait_death_row tcp_death_row;
186 /* sysctl variables for tcp */
187 extern int sysctl_tcp_timestamps;
188 extern int sysctl_tcp_window_scaling;
189 extern int sysctl_tcp_sack;
190 extern int sysctl_tcp_fin_timeout;
191 extern int sysctl_tcp_keepalive_time;
192 extern int sysctl_tcp_keepalive_probes;
193 extern int sysctl_tcp_keepalive_intvl;
194 extern int sysctl_tcp_syn_retries;
195 extern int sysctl_tcp_synack_retries;
196 extern int sysctl_tcp_retries1;
197 extern int sysctl_tcp_retries2;
198 extern int sysctl_tcp_orphan_retries;
199 extern int sysctl_tcp_syncookies;
200 extern int sysctl_tcp_retrans_collapse;
201 extern int sysctl_tcp_stdurg;
202 extern int sysctl_tcp_rfc1337;
203 extern int sysctl_tcp_abort_on_overflow;
204 extern int sysctl_tcp_max_orphans;
205 extern int sysctl_tcp_fack;
206 extern int sysctl_tcp_reordering;
207 extern int sysctl_tcp_ecn;
208 extern int sysctl_tcp_dsack;
209 extern int sysctl_tcp_mem[3];
210 extern int sysctl_tcp_wmem[3];
211 extern int sysctl_tcp_rmem[3];
212 extern int sysctl_tcp_app_win;
213 extern int sysctl_tcp_adv_win_scale;
214 extern int sysctl_tcp_tw_reuse;
215 extern int sysctl_tcp_frto;
216 extern int sysctl_tcp_low_latency;
217 extern int sysctl_tcp_nometrics_save;
218 extern int sysctl_tcp_moderate_rcvbuf;
219 extern int sysctl_tcp_tso_win_divisor;
221 extern atomic_t tcp_memory_allocated;
222 extern atomic_t tcp_sockets_allocated;
223 extern int tcp_memory_pressure;
226 * Pointers to address related TCP functions
227 * (i.e. things that depend on the address family)
231 int (*queue_xmit) (struct sk_buff *skb,
234 void (*send_check) (struct sock *sk,
237 struct sk_buff *skb);
239 int (*rebuild_header) (struct sock *sk);
241 int (*conn_request) (struct sock *sk,
242 struct sk_buff *skb);
244 struct sock * (*syn_recv_sock) (struct sock *sk,
246 struct request_sock *req,
247 struct dst_entry *dst);
249 int (*remember_stamp) (struct sock *sk);
251 __u16 net_header_len;
253 int (*setsockopt) (struct sock *sk,
259 int (*getsockopt) (struct sock *sk,
266 void (*addr2sockaddr) (struct sock *sk,
273 * The next routines deal with comparing 32 bit unsigned ints
274 * and worry about wraparound (automatic with unsigned arithmetic).
277 static inline int before(__u32 seq1, __u32 seq2)
279 return (__s32)(seq1-seq2) < 0;
282 static inline int after(__u32 seq1, __u32 seq2)
284 return (__s32)(seq2-seq1) < 0;
288 /* is s2<=s1<=s3 ? */
289 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
291 return seq3 - seq2 >= seq1 - seq2;
295 extern struct proto tcp_prot;
297 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
298 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
299 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
300 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
301 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
302 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
303 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
305 extern void tcp_v4_err(struct sk_buff *skb, u32);
307 extern void tcp_shutdown (struct sock *sk, int how);
309 extern int tcp_v4_rcv(struct sk_buff *skb);
311 extern int tcp_v4_remember_stamp(struct sock *sk);
313 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
315 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
316 struct msghdr *msg, size_t size);
317 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
319 extern int tcp_ioctl(struct sock *sk,
323 extern int tcp_rcv_state_process(struct sock *sk,
328 extern int tcp_rcv_established(struct sock *sk,
333 extern void tcp_rcv_space_adjust(struct sock *sk);
335 static inline void tcp_dec_quickack_mode(struct sock *sk,
336 const unsigned int pkts)
338 struct inet_connection_sock *icsk = inet_csk(sk);
340 if (icsk->icsk_ack.quick) {
341 if (pkts >= icsk->icsk_ack.quick) {
342 icsk->icsk_ack.quick = 0;
343 /* Leaving quickack mode we deflate ATO. */
344 icsk->icsk_ack.ato = TCP_ATO_MIN;
346 icsk->icsk_ack.quick -= pkts;
350 extern void tcp_enter_quickack_mode(struct sock *sk);
352 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
354 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
366 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
368 const struct tcphdr *th);
370 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
371 struct request_sock *req,
372 struct request_sock **prev);
373 extern int tcp_child_process(struct sock *parent,
375 struct sk_buff *skb);
376 extern void tcp_enter_frto(struct sock *sk);
377 extern void tcp_enter_loss(struct sock *sk, int how);
378 extern void tcp_clear_retrans(struct tcp_sock *tp);
379 extern void tcp_update_metrics(struct sock *sk);
381 extern void tcp_close(struct sock *sk,
383 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
385 extern int tcp_getsockopt(struct sock *sk, int level,
389 extern int tcp_setsockopt(struct sock *sk, int level,
390 int optname, char __user *optval,
392 extern void tcp_set_keepalive(struct sock *sk, int val);
393 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
395 size_t len, int nonblock,
396 int flags, int *addr_len);
398 extern void tcp_parse_options(struct sk_buff *skb,
399 struct tcp_options_received *opt_rx,
403 * TCP v4 functions exported for the inet6 API
406 extern void tcp_v4_send_check(struct sock *sk,
407 struct tcphdr *th, int len,
408 struct sk_buff *skb);
410 extern int tcp_v4_conn_request(struct sock *sk,
411 struct sk_buff *skb);
413 extern struct sock * tcp_create_openreq_child(struct sock *sk,
414 struct request_sock *req,
415 struct sk_buff *skb);
417 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
419 struct request_sock *req,
420 struct dst_entry *dst);
422 extern int tcp_v4_do_rcv(struct sock *sk,
423 struct sk_buff *skb);
425 extern int tcp_v4_connect(struct sock *sk,
426 struct sockaddr *uaddr,
429 extern int tcp_connect(struct sock *sk);
431 extern struct sk_buff * tcp_make_synack(struct sock *sk,
432 struct dst_entry *dst,
433 struct request_sock *req);
435 extern int tcp_disconnect(struct sock *sk, int flags);
437 extern void tcp_unhash(struct sock *sk);
439 extern int tcp_v4_hash_connecting(struct sock *sk);
442 /* From syncookies.c */
443 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
444 struct ip_options *opt);
445 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
450 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
451 unsigned int cur_mss, int nonagle);
452 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
453 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
454 extern void tcp_xmit_retransmit_queue(struct sock *);
455 extern void tcp_simple_retransmit(struct sock *);
456 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
458 extern void tcp_send_probe0(struct sock *);
459 extern void tcp_send_partial(struct sock *);
460 extern int tcp_write_wakeup(struct sock *);
461 extern void tcp_send_fin(struct sock *sk);
462 extern void tcp_send_active_reset(struct sock *sk,
463 unsigned int __nocast priority);
464 extern int tcp_send_synack(struct sock *);
465 extern void tcp_push_one(struct sock *, unsigned int mss_now);
466 extern void tcp_send_ack(struct sock *sk);
467 extern void tcp_send_delayed_ack(struct sock *sk);
470 extern void tcp_cwnd_application_limited(struct sock *sk);
473 extern void tcp_init_xmit_timers(struct sock *);
474 static inline void tcp_clear_xmit_timers(struct sock *sk)
476 inet_csk_clear_xmit_timers(sk);
479 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
480 extern unsigned int tcp_current_mss(struct sock *sk, int large);
483 extern void tcp_get_info(struct sock *, struct tcp_info *);
485 /* Read 'sendfile()'-style from a TCP socket */
486 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
487 unsigned int, size_t);
488 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
489 sk_read_actor_t recv_actor);
491 /* Initialize RCV_MSS value.
492 * RCV_MSS is an our guess about MSS used by the peer.
493 * We haven't any direct information about the MSS.
494 * It's better to underestimate the RCV_MSS rather than overestimate.
495 * Overestimations make us ACKing less frequently than needed.
496 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
499 static inline void tcp_initialize_rcv_mss(struct sock *sk)
501 struct tcp_sock *tp = tcp_sk(sk);
502 unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);
504 hint = min(hint, tp->rcv_wnd/2);
505 hint = min(hint, TCP_MIN_RCVMSS);
506 hint = max(hint, TCP_MIN_MSS);
508 inet_csk(sk)->icsk_ack.rcv_mss = hint;
511 static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
513 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
514 ntohl(TCP_FLAG_ACK) |
518 static __inline__ void tcp_fast_path_on(struct tcp_sock *tp)
520 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
523 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
525 if (skb_queue_empty(&tp->out_of_order_queue) &&
527 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
529 tcp_fast_path_on(tp);
532 /* Compute the actual receive window we are currently advertising.
533 * Rcv_nxt can be after the window if our peer push more data
534 * than the offered window.
536 static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp)
538 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
545 /* Choose a new window, without checks for shrinking, and without
546 * scaling applied to the result. The caller does these things
547 * if necessary. This is a "raw" window selection.
549 extern u32 __tcp_select_window(struct sock *sk);
551 /* TCP timestamps are only 32-bits, this causes a slight
552 * complication on 64-bit systems since we store a snapshot
553 * of jiffies in the buffer control blocks below. We decidely
554 * only use of the low 32-bits of jiffies and hide the ugly
555 * casts with the following macro.
557 #define tcp_time_stamp ((__u32)(jiffies))
559 /* This is what the send packet queueing engine uses to pass
560 * TCP per-packet control information to the transmission
561 * code. We also store the host-order sequence numbers in
562 * here too. This is 36 bytes on 32-bit architectures,
563 * 40 bytes on 64-bit machines, if this grows please adjust
564 * skbuff.h:skbuff->cb[xxx] size appropriately.
568 struct inet_skb_parm h4;
569 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
570 struct inet6_skb_parm h6;
572 } header; /* For incoming frames */
573 __u32 seq; /* Starting sequence number */
574 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
575 __u32 when; /* used to compute rtt's */
576 __u8 flags; /* TCP header flags. */
578 /* NOTE: These must match up to the flags byte in a
581 #define TCPCB_FLAG_FIN 0x01
582 #define TCPCB_FLAG_SYN 0x02
583 #define TCPCB_FLAG_RST 0x04
584 #define TCPCB_FLAG_PSH 0x08
585 #define TCPCB_FLAG_ACK 0x10
586 #define TCPCB_FLAG_URG 0x20
587 #define TCPCB_FLAG_ECE 0x40
588 #define TCPCB_FLAG_CWR 0x80
590 __u8 sacked; /* State flags for SACK/FACK. */
591 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
592 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
593 #define TCPCB_LOST 0x04 /* SKB is lost */
594 #define TCPCB_TAGBITS 0x07 /* All tag bits */
596 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
597 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
599 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
601 #define TCPCB_AT_TAIL (TCPCB_URG)
603 __u16 urg_ptr; /* Valid w/URG flags is set. */
604 __u32 ack_seq; /* Sequence number ACK'd */
607 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
609 #include <net/tcp_ecn.h>
611 /* Due to TSO, an SKB can be composed of multiple actual
612 * packets. To keep these tracked properly, we use this.
614 static inline int tcp_skb_pcount(const struct sk_buff *skb)
616 return skb_shinfo(skb)->tso_segs;
619 /* This is valid iff tcp_skb_pcount() > 1. */
620 static inline int tcp_skb_mss(const struct sk_buff *skb)
622 return skb_shinfo(skb)->tso_size;
625 static inline void tcp_dec_pcount_approx(__u32 *count,
626 const struct sk_buff *skb)
629 *count -= tcp_skb_pcount(skb);
635 static inline void tcp_packets_out_inc(struct sock *sk,
637 const struct sk_buff *skb)
639 int orig = tp->packets_out;
641 tp->packets_out += tcp_skb_pcount(skb);
643 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
644 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
647 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
648 const struct sk_buff *skb)
650 tp->packets_out -= tcp_skb_pcount(skb);
653 /* Events passed to congestion control interface */
655 CA_EVENT_TX_START, /* first transmit when no packets in flight */
656 CA_EVENT_CWND_RESTART, /* congestion window restart */
657 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
658 CA_EVENT_FRTO, /* fast recovery timeout */
659 CA_EVENT_LOSS, /* loss timeout */
660 CA_EVENT_FAST_ACK, /* in sequence ack */
661 CA_EVENT_SLOW_ACK, /* other ack */
665 * Interface for adding new TCP congestion control handlers
667 #define TCP_CA_NAME_MAX 16
668 struct tcp_congestion_ops {
669 struct list_head list;
671 /* initialize private data (optional) */
672 void (*init)(struct tcp_sock *tp);
673 /* cleanup private data (optional) */
674 void (*release)(struct tcp_sock *tp);
676 /* return slow start threshold (required) */
677 u32 (*ssthresh)(struct tcp_sock *tp);
678 /* lower bound for congestion window (optional) */
679 u32 (*min_cwnd)(struct tcp_sock *tp);
680 /* do new cwnd calculation (required) */
681 void (*cong_avoid)(struct tcp_sock *tp, u32 ack,
682 u32 rtt, u32 in_flight, int good_ack);
683 /* round trip time sample per acked packet (optional) */
684 void (*rtt_sample)(struct tcp_sock *tp, u32 usrtt);
685 /* call before changing ca_state (optional) */
686 void (*set_state)(struct tcp_sock *tp, u8 new_state);
687 /* call when cwnd event occurs (optional) */
688 void (*cwnd_event)(struct tcp_sock *tp, enum tcp_ca_event ev);
689 /* new value of cwnd after loss (optional) */
690 u32 (*undo_cwnd)(struct tcp_sock *tp);
691 /* hook for packet ack accounting (optional) */
692 void (*pkts_acked)(struct tcp_sock *tp, u32 num_acked);
693 /* get info for tcp_diag (optional) */
694 void (*get_info)(struct tcp_sock *tp, u32 ext, struct sk_buff *skb);
696 char name[TCP_CA_NAME_MAX];
697 struct module *owner;
700 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
701 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
703 extern void tcp_init_congestion_control(struct tcp_sock *tp);
704 extern void tcp_cleanup_congestion_control(struct tcp_sock *tp);
705 extern int tcp_set_default_congestion_control(const char *name);
706 extern void tcp_get_default_congestion_control(char *name);
707 extern int tcp_set_congestion_control(struct tcp_sock *tp, const char *name);
709 extern struct tcp_congestion_ops tcp_init_congestion_ops;
710 extern u32 tcp_reno_ssthresh(struct tcp_sock *tp);
711 extern void tcp_reno_cong_avoid(struct tcp_sock *tp, u32 ack,
712 u32 rtt, u32 in_flight, int flag);
713 extern u32 tcp_reno_min_cwnd(struct tcp_sock *tp);
714 extern struct tcp_congestion_ops tcp_reno;
716 static inline void tcp_set_ca_state(struct tcp_sock *tp, u8 ca_state)
718 if (tp->ca_ops->set_state)
719 tp->ca_ops->set_state(tp, ca_state);
720 tp->ca_state = ca_state;
723 static inline void tcp_ca_event(struct tcp_sock *tp, enum tcp_ca_event event)
725 if (tp->ca_ops->cwnd_event)
726 tp->ca_ops->cwnd_event(tp, event);
729 /* This determines how many packets are "in the network" to the best
730 * of our knowledge. In many cases it is conservative, but where
731 * detailed information is available from the receiver (via SACK
732 * blocks etc.) we can make more aggressive calculations.
734 * Use this for decisions involving congestion control, use just
735 * tp->packets_out to determine if the send queue is empty or not.
737 * Read this equation as:
739 * "Packets sent once on transmission queue" MINUS
740 * "Packets left network, but not honestly ACKed yet" PLUS
741 * "Packets fast retransmitted"
743 static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
745 return (tp->packets_out - tp->left_out + tp->retrans_out);
748 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
749 * The exception is rate halving phase, when cwnd is decreasing towards
752 static inline __u32 tcp_current_ssthresh(struct tcp_sock *tp)
754 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
755 return tp->snd_ssthresh;
757 return max(tp->snd_ssthresh,
758 ((tp->snd_cwnd >> 1) +
759 (tp->snd_cwnd >> 2)));
762 static inline void tcp_sync_left_out(struct tcp_sock *tp)
764 if (tp->rx_opt.sack_ok &&
765 (tp->sacked_out >= tp->packets_out - tp->lost_out))
766 tp->sacked_out = tp->packets_out - tp->lost_out;
767 tp->left_out = tp->sacked_out + tp->lost_out;
770 /* Set slow start threshold and cwnd not falling to slow start */
771 static inline void __tcp_enter_cwr(struct tcp_sock *tp)
774 tp->snd_ssthresh = tp->ca_ops->ssthresh(tp);
775 tp->snd_cwnd = min(tp->snd_cwnd,
776 tcp_packets_in_flight(tp) + 1U);
777 tp->snd_cwnd_cnt = 0;
778 tp->high_seq = tp->snd_nxt;
779 tp->snd_cwnd_stamp = tcp_time_stamp;
780 TCP_ECN_queue_cwr(tp);
783 static inline void tcp_enter_cwr(struct tcp_sock *tp)
785 tp->prior_ssthresh = 0;
786 if (tp->ca_state < TCP_CA_CWR) {
788 tcp_set_ca_state(tp, TCP_CA_CWR);
792 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
794 /* Slow start with delack produces 3 packets of burst, so that
795 * it is safe "de facto".
797 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
802 static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss,
803 const struct sk_buff *skb)
806 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
809 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
811 const struct inet_connection_sock *icsk = inet_csk(sk);
812 if (!tp->packets_out && !icsk->icsk_pending)
813 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
814 icsk->icsk_rto, TCP_RTO_MAX);
817 static __inline__ void tcp_push_pending_frames(struct sock *sk,
820 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
823 static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
828 static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
834 * Calculate(/check) TCP checksum
836 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
837 unsigned long saddr, unsigned long daddr,
840 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
843 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
845 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
848 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
850 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
851 __tcp_checksum_complete(skb);
854 /* Prequeue for VJ style copy to user, combined with checksumming. */
856 static __inline__ void tcp_prequeue_init(struct tcp_sock *tp)
858 tp->ucopy.task = NULL;
860 tp->ucopy.memory = 0;
861 skb_queue_head_init(&tp->ucopy.prequeue);
864 /* Packet is added to VJ-style prequeue for processing in process
865 * context, if a reader task is waiting. Apparently, this exciting
866 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
867 * failed somewhere. Latency? Burstiness? Well, at least now we will
868 * see, why it failed. 8)8) --ANK
870 * NOTE: is this not too big to inline?
872 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
874 struct tcp_sock *tp = tcp_sk(sk);
876 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
877 __skb_queue_tail(&tp->ucopy.prequeue, skb);
878 tp->ucopy.memory += skb->truesize;
879 if (tp->ucopy.memory > sk->sk_rcvbuf) {
880 struct sk_buff *skb1;
882 BUG_ON(sock_owned_by_user(sk));
884 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
885 sk->sk_backlog_rcv(sk, skb1);
886 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
889 tp->ucopy.memory = 0;
890 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
891 wake_up_interruptible(sk->sk_sleep);
892 if (!inet_csk_ack_scheduled(sk))
893 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
894 (3 * TCP_RTO_MIN) / 4,
906 static const char *statename[]={
907 "Unused","Established","Syn Sent","Syn Recv",
908 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
909 "Close Wait","Last ACK","Listen","Closing"
913 static __inline__ void tcp_set_state(struct sock *sk, int state)
915 int oldstate = sk->sk_state;
918 case TCP_ESTABLISHED:
919 if (oldstate != TCP_ESTABLISHED)
920 TCP_INC_STATS(TCP_MIB_CURRESTAB);
924 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
925 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
927 sk->sk_prot->unhash(sk);
928 if (inet_csk(sk)->icsk_bind_hash &&
929 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
930 inet_put_port(&tcp_hashinfo, sk);
933 if (oldstate==TCP_ESTABLISHED)
934 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
937 /* Change state AFTER socket is unhashed to avoid closed
938 * socket sitting in hash tables.
940 sk->sk_state = state;
943 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
947 static __inline__ void tcp_done(struct sock *sk)
949 tcp_set_state(sk, TCP_CLOSE);
950 tcp_clear_xmit_timers(sk);
952 sk->sk_shutdown = SHUTDOWN_MASK;
954 if (!sock_flag(sk, SOCK_DEAD))
955 sk->sk_state_change(sk);
957 inet_csk_destroy_sock(sk);
960 static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt)
963 rx_opt->eff_sacks = 0;
964 rx_opt->num_sacks = 0;
967 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp)
969 if (tp->rx_opt.tstamp_ok) {
970 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
972 (TCPOPT_TIMESTAMP << 8) |
974 *ptr++ = htonl(tstamp);
975 *ptr++ = htonl(tp->rx_opt.ts_recent);
977 if (tp->rx_opt.eff_sacks) {
978 struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks;
981 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
985 (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK)));
986 for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) {
987 *ptr++ = htonl(sp[this_sack].start_seq);
988 *ptr++ = htonl(sp[this_sack].end_seq);
990 if (tp->rx_opt.dsack) {
991 tp->rx_opt.dsack = 0;
992 tp->rx_opt.eff_sacks--;
997 /* Construct a tcp options header for a SYN or SYN_ACK packet.
998 * If this is every changed make sure to change the definition of
999 * MAX_SYN_SIZE to match the new maximum number of options that you
1002 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1003 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1005 /* We always get an MSS option.
1006 * The option bytes which will be seen in normal data
1007 * packets should timestamps be used, must be in the MSS
1008 * advertised. But we subtract them from tp->mss_cache so
1009 * that calculations in tcp_sendmsg are simpler etc.
1010 * So account for this fact here if necessary. If we
1011 * don't do this correctly, as a receiver we won't
1012 * recognize data packets as being full sized when we
1013 * should, and thus we won't abide by the delayed ACK
1015 * SACKs don't matter, we never delay an ACK when we
1016 * have any of those going out.
1018 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1021 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1022 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1024 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1025 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1026 *ptr++ = htonl(tstamp); /* TSVAL */
1027 *ptr++ = htonl(ts_recent); /* TSECR */
1029 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1030 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1032 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1035 /* Determine a window scaling and initial window to offer. */
1036 extern void tcp_select_initial_window(int __space, __u32 mss,
1037 __u32 *rcv_wnd, __u32 *window_clamp,
1038 int wscale_ok, __u8 *rcv_wscale);
1040 static inline int tcp_win_from_space(int space)
1042 return sysctl_tcp_adv_win_scale<=0 ?
1043 (space>>(-sysctl_tcp_adv_win_scale)) :
1044 space - (space>>sysctl_tcp_adv_win_scale);
1047 /* Note: caller must be prepared to deal with negative returns */
1048 static inline int tcp_space(const struct sock *sk)
1050 return tcp_win_from_space(sk->sk_rcvbuf -
1051 atomic_read(&sk->sk_rmem_alloc));
1054 static inline int tcp_full_space(const struct sock *sk)
1056 return tcp_win_from_space(sk->sk_rcvbuf);
1059 static __inline__ void tcp_openreq_init(struct request_sock *req,
1060 struct tcp_options_received *rx_opt,
1061 struct sk_buff *skb)
1063 struct inet_request_sock *ireq = inet_rsk(req);
1065 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1066 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1067 req->mss = rx_opt->mss_clamp;
1068 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1069 ireq->tstamp_ok = rx_opt->tstamp_ok;
1070 ireq->sack_ok = rx_opt->sack_ok;
1071 ireq->snd_wscale = rx_opt->snd_wscale;
1072 ireq->wscale_ok = rx_opt->wscale_ok;
1075 ireq->rmt_port = skb->h.th->source;
1078 extern void tcp_enter_memory_pressure(void);
1080 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1082 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1085 static inline int keepalive_time_when(const struct tcp_sock *tp)
1087 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1090 static inline int tcp_fin_time(const struct sock *sk)
1092 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1093 const int rto = inet_csk(sk)->icsk_rto;
1095 if (fin_timeout < (rto << 2) - (rto >> 1))
1096 fin_timeout = (rto << 2) - (rto >> 1);
1101 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1103 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1105 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1108 /* RST segments are not recommended to carry timestamp,
1109 and, if they do, it is recommended to ignore PAWS because
1110 "their cleanup function should take precedence over timestamps."
1111 Certainly, it is mistake. It is necessary to understand the reasons
1112 of this constraint to relax it: if peer reboots, clock may go
1113 out-of-sync and half-open connections will not be reset.
1114 Actually, the problem would be not existing if all
1115 the implementations followed draft about maintaining clock
1116 via reboots. Linux-2.2 DOES NOT!
1118 However, we can relax time bounds for RST segments to MSL.
1120 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1125 #define TCP_CHECK_TIMER(sk) do { } while (0)
1127 static inline int tcp_use_frto(const struct sock *sk)
1129 const struct tcp_sock *tp = tcp_sk(sk);
1131 /* F-RTO must be activated in sysctl and there must be some
1132 * unsent new data, and the advertised window should allow
1135 return (sysctl_tcp_frto && sk->sk_send_head &&
1136 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1137 tp->snd_una + tp->snd_wnd));
1140 static inline void tcp_mib_init(void)
1143 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1144 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1145 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1146 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1150 enum tcp_seq_states {
1151 TCP_SEQ_STATE_LISTENING,
1152 TCP_SEQ_STATE_OPENREQ,
1153 TCP_SEQ_STATE_ESTABLISHED,
1154 TCP_SEQ_STATE_TIME_WAIT,
1157 struct tcp_seq_afinfo {
1158 struct module *owner;
1161 int (*seq_show) (struct seq_file *m, void *v);
1162 struct file_operations *seq_fops;
1165 struct tcp_iter_state {
1167 enum tcp_seq_states state;
1168 struct sock *syn_wait_sk;
1169 int bucket, sbucket, num, uid;
1170 struct seq_operations seq_ops;
1173 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1174 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);