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_hashtables.h>
33 #include <net/checksum.h>
34 #include <net/request_sock.h>
38 #include <net/tcp_states.h>
40 #include <linux/seq_file.h>
42 extern struct inet_hashinfo tcp_hashinfo;
44 extern atomic_t tcp_orphan_count;
45 extern int tcp_tw_count;
46 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
47 extern void tcp_tw_deschedule(struct inet_timewait_sock *tw);
49 #define MAX_TCP_HEADER (128 + MAX_HEADER)
52 * Never offer a window over 32767 without using window scaling. Some
53 * poor stacks do signed 16bit maths!
55 #define MAX_TCP_WINDOW 32767U
57 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
58 #define TCP_MIN_MSS 88U
60 /* Minimal RCV_MSS. */
61 #define TCP_MIN_RCVMSS 536U
63 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
64 #define TCP_FASTRETRANS_THRESH 3
66 /* Maximal reordering. */
67 #define TCP_MAX_REORDERING 127
69 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
70 #define TCP_MAX_QUICKACKS 16U
73 #define TCP_URG_VALID 0x0100
74 #define TCP_URG_NOTYET 0x0200
75 #define TCP_URG_READ 0x0400
77 #define TCP_RETR1 3 /*
78 * This is how many retries it does before it
79 * tries to figure out if the gateway is
80 * down. Minimal RFC value is 3; it corresponds
81 * to ~3sec-8min depending on RTO.
84 #define TCP_RETR2 15 /*
85 * This should take at least
86 * 90 minutes to time out.
87 * RFC1122 says that the limit is 100 sec.
88 * 15 is ~13-30min depending on RTO.
91 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
92 * connection: ~180sec is RFC minumum */
94 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
95 * connection: ~180sec is RFC minumum */
98 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
99 * socket. 7 is ~50sec-16min.
103 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
104 * state, about 60 seconds */
105 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
106 /* BSD style FIN_WAIT2 deadlock breaker.
107 * It used to be 3min, new value is 60sec,
108 * to combine FIN-WAIT-2 timeout with
112 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
114 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
115 #define TCP_ATO_MIN ((unsigned)(HZ/25))
117 #define TCP_DELACK_MIN 4U
118 #define TCP_ATO_MIN 4U
120 #define TCP_RTO_MAX ((unsigned)(120*HZ))
121 #define TCP_RTO_MIN ((unsigned)(HZ/5))
122 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
124 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
125 * for local resources.
128 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
129 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
130 #define TCP_KEEPALIVE_INTVL (75*HZ)
132 #define MAX_TCP_KEEPIDLE 32767
133 #define MAX_TCP_KEEPINTVL 32767
134 #define MAX_TCP_KEEPCNT 127
135 #define MAX_TCP_SYNCNT 127
137 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
138 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
140 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
141 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
142 * after this time. It should be equal
143 * (or greater than) TCP_TIMEWAIT_LEN
144 * to provide reliability equal to one
145 * provided by timewait state.
147 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
148 * timestamps. It must be less than
149 * minimal timewait lifetime.
152 #define TCP_TW_RECYCLE_SLOTS_LOG 5
153 #define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG)
155 /* If time > 4sec, it is "slow" path, no recycling is required,
156 so that we select tick to get range about 4 seconds.
159 #if HZ <= 16 || HZ > 4096
160 # error Unsupported: HZ <= 16 or HZ > 4096
162 # define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
164 # define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
166 # define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
168 # define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
170 # define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
172 # define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
174 # define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
176 # define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
182 #define TCPOPT_NOP 1 /* Padding */
183 #define TCPOPT_EOL 0 /* End of options */
184 #define TCPOPT_MSS 2 /* Segment size negotiating */
185 #define TCPOPT_WINDOW 3 /* Window scaling */
186 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
187 #define TCPOPT_SACK 5 /* SACK Block */
188 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
194 #define TCPOLEN_MSS 4
195 #define TCPOLEN_WINDOW 3
196 #define TCPOLEN_SACK_PERM 2
197 #define TCPOLEN_TIMESTAMP 10
199 /* But this is what stacks really send out. */
200 #define TCPOLEN_TSTAMP_ALIGNED 12
201 #define TCPOLEN_WSCALE_ALIGNED 4
202 #define TCPOLEN_SACKPERM_ALIGNED 4
203 #define TCPOLEN_SACK_BASE 2
204 #define TCPOLEN_SACK_BASE_ALIGNED 4
205 #define TCPOLEN_SACK_PERBLOCK 8
207 /* Flags in tp->nonagle */
208 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
209 #define TCP_NAGLE_CORK 2 /* Socket is corked */
210 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
212 /* sysctl variables for tcp */
213 extern int sysctl_tcp_timestamps;
214 extern int sysctl_tcp_window_scaling;
215 extern int sysctl_tcp_sack;
216 extern int sysctl_tcp_fin_timeout;
217 extern int sysctl_tcp_tw_recycle;
218 extern int sysctl_tcp_keepalive_time;
219 extern int sysctl_tcp_keepalive_probes;
220 extern int sysctl_tcp_keepalive_intvl;
221 extern int sysctl_tcp_syn_retries;
222 extern int sysctl_tcp_synack_retries;
223 extern int sysctl_tcp_retries1;
224 extern int sysctl_tcp_retries2;
225 extern int sysctl_tcp_orphan_retries;
226 extern int sysctl_tcp_syncookies;
227 extern int sysctl_tcp_retrans_collapse;
228 extern int sysctl_tcp_stdurg;
229 extern int sysctl_tcp_rfc1337;
230 extern int sysctl_tcp_abort_on_overflow;
231 extern int sysctl_tcp_max_orphans;
232 extern int sysctl_tcp_max_tw_buckets;
233 extern int sysctl_tcp_fack;
234 extern int sysctl_tcp_reordering;
235 extern int sysctl_tcp_ecn;
236 extern int sysctl_tcp_dsack;
237 extern int sysctl_tcp_mem[3];
238 extern int sysctl_tcp_wmem[3];
239 extern int sysctl_tcp_rmem[3];
240 extern int sysctl_tcp_app_win;
241 extern int sysctl_tcp_adv_win_scale;
242 extern int sysctl_tcp_tw_reuse;
243 extern int sysctl_tcp_frto;
244 extern int sysctl_tcp_low_latency;
245 extern int sysctl_tcp_nometrics_save;
246 extern int sysctl_tcp_moderate_rcvbuf;
247 extern int sysctl_tcp_tso_win_divisor;
249 extern atomic_t tcp_memory_allocated;
250 extern atomic_t tcp_sockets_allocated;
251 extern int tcp_memory_pressure;
254 * Pointers to address related TCP functions
255 * (i.e. things that depend on the address family)
259 int (*queue_xmit) (struct sk_buff *skb,
262 void (*send_check) (struct sock *sk,
265 struct sk_buff *skb);
267 int (*rebuild_header) (struct sock *sk);
269 int (*conn_request) (struct sock *sk,
270 struct sk_buff *skb);
272 struct sock * (*syn_recv_sock) (struct sock *sk,
274 struct request_sock *req,
275 struct dst_entry *dst);
277 int (*remember_stamp) (struct sock *sk);
279 __u16 net_header_len;
281 int (*setsockopt) (struct sock *sk,
287 int (*getsockopt) (struct sock *sk,
294 void (*addr2sockaddr) (struct sock *sk,
301 * The next routines deal with comparing 32 bit unsigned ints
302 * and worry about wraparound (automatic with unsigned arithmetic).
305 static inline int before(__u32 seq1, __u32 seq2)
307 return (__s32)(seq1-seq2) < 0;
310 static inline int after(__u32 seq1, __u32 seq2)
312 return (__s32)(seq2-seq1) < 0;
316 /* is s2<=s1<=s3 ? */
317 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
319 return seq3 - seq2 >= seq1 - seq2;
323 extern struct proto tcp_prot;
325 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
326 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
327 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
328 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
329 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
330 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
331 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
333 extern void tcp_v4_err(struct sk_buff *skb, u32);
335 extern void tcp_shutdown (struct sock *sk, int how);
337 extern int tcp_v4_rcv(struct sk_buff *skb);
339 extern int tcp_v4_remember_stamp(struct sock *sk);
341 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
343 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
344 struct msghdr *msg, size_t size);
345 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
347 extern int tcp_ioctl(struct sock *sk,
351 extern int tcp_rcv_state_process(struct sock *sk,
356 extern int tcp_rcv_established(struct sock *sk,
361 extern void tcp_rcv_space_adjust(struct sock *sk);
363 static inline void tcp_dec_quickack_mode(struct sock *sk,
364 const unsigned int pkts)
366 struct inet_connection_sock *icsk = inet_csk(sk);
368 if (icsk->icsk_ack.quick) {
369 if (pkts >= icsk->icsk_ack.quick) {
370 icsk->icsk_ack.quick = 0;
371 /* Leaving quickack mode we deflate ATO. */
372 icsk->icsk_ack.ato = TCP_ATO_MIN;
374 icsk->icsk_ack.quick -= pkts;
378 extern void tcp_enter_quickack_mode(struct sock *sk);
380 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
382 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
394 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
396 const struct tcphdr *th);
398 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
399 struct request_sock *req,
400 struct request_sock **prev);
401 extern int tcp_child_process(struct sock *parent,
403 struct sk_buff *skb);
404 extern void tcp_enter_frto(struct sock *sk);
405 extern void tcp_enter_loss(struct sock *sk, int how);
406 extern void tcp_clear_retrans(struct tcp_sock *tp);
407 extern void tcp_update_metrics(struct sock *sk);
409 extern void tcp_close(struct sock *sk,
411 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
413 extern int tcp_getsockopt(struct sock *sk, int level,
417 extern int tcp_setsockopt(struct sock *sk, int level,
418 int optname, char __user *optval,
420 extern void tcp_set_keepalive(struct sock *sk, int val);
421 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
423 size_t len, int nonblock,
424 int flags, int *addr_len);
426 extern int tcp_listen_start(struct sock *sk);
428 extern void tcp_parse_options(struct sk_buff *skb,
429 struct tcp_options_received *opt_rx,
433 * TCP v4 functions exported for the inet6 API
436 extern void tcp_v4_send_check(struct sock *sk,
437 struct tcphdr *th, int len,
438 struct sk_buff *skb);
440 extern int tcp_v4_conn_request(struct sock *sk,
441 struct sk_buff *skb);
443 extern struct sock * tcp_create_openreq_child(struct sock *sk,
444 struct request_sock *req,
445 struct sk_buff *skb);
447 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
449 struct request_sock *req,
450 struct dst_entry *dst);
452 extern int tcp_v4_do_rcv(struct sock *sk,
453 struct sk_buff *skb);
455 extern int tcp_v4_connect(struct sock *sk,
456 struct sockaddr *uaddr,
459 extern int tcp_connect(struct sock *sk);
461 extern struct sk_buff * tcp_make_synack(struct sock *sk,
462 struct dst_entry *dst,
463 struct request_sock *req);
465 extern int tcp_disconnect(struct sock *sk, int flags);
467 extern void tcp_unhash(struct sock *sk);
469 extern int tcp_v4_hash_connecting(struct sock *sk);
472 /* From syncookies.c */
473 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
474 struct ip_options *opt);
475 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
480 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
481 unsigned int cur_mss, int nonagle);
482 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
483 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
484 extern void tcp_xmit_retransmit_queue(struct sock *);
485 extern void tcp_simple_retransmit(struct sock *);
486 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
488 extern void tcp_send_probe0(struct sock *);
489 extern void tcp_send_partial(struct sock *);
490 extern int tcp_write_wakeup(struct sock *);
491 extern void tcp_send_fin(struct sock *sk);
492 extern void tcp_send_active_reset(struct sock *sk,
493 unsigned int __nocast priority);
494 extern int tcp_send_synack(struct sock *);
495 extern void tcp_push_one(struct sock *, unsigned int mss_now);
496 extern void tcp_send_ack(struct sock *sk);
497 extern void tcp_send_delayed_ack(struct sock *sk);
500 extern void tcp_cwnd_application_limited(struct sock *sk);
503 extern void tcp_init_xmit_timers(struct sock *);
504 static inline void tcp_clear_xmit_timers(struct sock *sk)
506 inet_csk_clear_xmit_timers(sk);
509 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
510 extern unsigned int tcp_current_mss(struct sock *sk, int large);
513 extern void tcp_get_info(struct sock *, struct tcp_info *);
515 /* Read 'sendfile()'-style from a TCP socket */
516 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
517 unsigned int, size_t);
518 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
519 sk_read_actor_t recv_actor);
521 /* Initialize RCV_MSS value.
522 * RCV_MSS is an our guess about MSS used by the peer.
523 * We haven't any direct information about the MSS.
524 * It's better to underestimate the RCV_MSS rather than overestimate.
525 * Overestimations make us ACKing less frequently than needed.
526 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
529 static inline void tcp_initialize_rcv_mss(struct sock *sk)
531 struct tcp_sock *tp = tcp_sk(sk);
532 unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);
534 hint = min(hint, tp->rcv_wnd/2);
535 hint = min(hint, TCP_MIN_RCVMSS);
536 hint = max(hint, TCP_MIN_MSS);
538 inet_csk(sk)->icsk_ack.rcv_mss = hint;
541 static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
543 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
544 ntohl(TCP_FLAG_ACK) |
548 static __inline__ void tcp_fast_path_on(struct tcp_sock *tp)
550 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
553 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
555 if (skb_queue_empty(&tp->out_of_order_queue) &&
557 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
559 tcp_fast_path_on(tp);
562 /* Compute the actual receive window we are currently advertising.
563 * Rcv_nxt can be after the window if our peer push more data
564 * than the offered window.
566 static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp)
568 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
575 /* Choose a new window, without checks for shrinking, and without
576 * scaling applied to the result. The caller does these things
577 * if necessary. This is a "raw" window selection.
579 extern u32 __tcp_select_window(struct sock *sk);
581 /* TCP timestamps are only 32-bits, this causes a slight
582 * complication on 64-bit systems since we store a snapshot
583 * of jiffies in the buffer control blocks below. We decidely
584 * only use of the low 32-bits of jiffies and hide the ugly
585 * casts with the following macro.
587 #define tcp_time_stamp ((__u32)(jiffies))
589 /* This is what the send packet queueing engine uses to pass
590 * TCP per-packet control information to the transmission
591 * code. We also store the host-order sequence numbers in
592 * here too. This is 36 bytes on 32-bit architectures,
593 * 40 bytes on 64-bit machines, if this grows please adjust
594 * skbuff.h:skbuff->cb[xxx] size appropriately.
598 struct inet_skb_parm h4;
599 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
600 struct inet6_skb_parm h6;
602 } header; /* For incoming frames */
603 __u32 seq; /* Starting sequence number */
604 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
605 __u32 when; /* used to compute rtt's */
606 __u8 flags; /* TCP header flags. */
608 /* NOTE: These must match up to the flags byte in a
611 #define TCPCB_FLAG_FIN 0x01
612 #define TCPCB_FLAG_SYN 0x02
613 #define TCPCB_FLAG_RST 0x04
614 #define TCPCB_FLAG_PSH 0x08
615 #define TCPCB_FLAG_ACK 0x10
616 #define TCPCB_FLAG_URG 0x20
617 #define TCPCB_FLAG_ECE 0x40
618 #define TCPCB_FLAG_CWR 0x80
620 __u8 sacked; /* State flags for SACK/FACK. */
621 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
622 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
623 #define TCPCB_LOST 0x04 /* SKB is lost */
624 #define TCPCB_TAGBITS 0x07 /* All tag bits */
626 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
627 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
629 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
631 #define TCPCB_AT_TAIL (TCPCB_URG)
633 __u16 urg_ptr; /* Valid w/URG flags is set. */
634 __u32 ack_seq; /* Sequence number ACK'd */
637 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
639 #include <net/tcp_ecn.h>
641 /* Due to TSO, an SKB can be composed of multiple actual
642 * packets. To keep these tracked properly, we use this.
644 static inline int tcp_skb_pcount(const struct sk_buff *skb)
646 return skb_shinfo(skb)->tso_segs;
649 /* This is valid iff tcp_skb_pcount() > 1. */
650 static inline int tcp_skb_mss(const struct sk_buff *skb)
652 return skb_shinfo(skb)->tso_size;
655 static inline void tcp_dec_pcount_approx(__u32 *count,
656 const struct sk_buff *skb)
659 *count -= tcp_skb_pcount(skb);
665 static inline void tcp_packets_out_inc(struct sock *sk,
667 const struct sk_buff *skb)
669 int orig = tp->packets_out;
671 tp->packets_out += tcp_skb_pcount(skb);
673 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
674 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
677 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
678 const struct sk_buff *skb)
680 tp->packets_out -= tcp_skb_pcount(skb);
683 /* Events passed to congestion control interface */
685 CA_EVENT_TX_START, /* first transmit when no packets in flight */
686 CA_EVENT_CWND_RESTART, /* congestion window restart */
687 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
688 CA_EVENT_FRTO, /* fast recovery timeout */
689 CA_EVENT_LOSS, /* loss timeout */
690 CA_EVENT_FAST_ACK, /* in sequence ack */
691 CA_EVENT_SLOW_ACK, /* other ack */
695 * Interface for adding new TCP congestion control handlers
697 #define TCP_CA_NAME_MAX 16
698 struct tcp_congestion_ops {
699 struct list_head list;
701 /* initialize private data (optional) */
702 void (*init)(struct tcp_sock *tp);
703 /* cleanup private data (optional) */
704 void (*release)(struct tcp_sock *tp);
706 /* return slow start threshold (required) */
707 u32 (*ssthresh)(struct tcp_sock *tp);
708 /* lower bound for congestion window (optional) */
709 u32 (*min_cwnd)(struct tcp_sock *tp);
710 /* do new cwnd calculation (required) */
711 void (*cong_avoid)(struct tcp_sock *tp, u32 ack,
712 u32 rtt, u32 in_flight, int good_ack);
713 /* round trip time sample per acked packet (optional) */
714 void (*rtt_sample)(struct tcp_sock *tp, u32 usrtt);
715 /* call before changing ca_state (optional) */
716 void (*set_state)(struct tcp_sock *tp, u8 new_state);
717 /* call when cwnd event occurs (optional) */
718 void (*cwnd_event)(struct tcp_sock *tp, enum tcp_ca_event ev);
719 /* new value of cwnd after loss (optional) */
720 u32 (*undo_cwnd)(struct tcp_sock *tp);
721 /* hook for packet ack accounting (optional) */
722 void (*pkts_acked)(struct tcp_sock *tp, u32 num_acked);
723 /* get info for tcp_diag (optional) */
724 void (*get_info)(struct tcp_sock *tp, u32 ext, struct sk_buff *skb);
726 char name[TCP_CA_NAME_MAX];
727 struct module *owner;
730 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
731 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
733 extern void tcp_init_congestion_control(struct tcp_sock *tp);
734 extern void tcp_cleanup_congestion_control(struct tcp_sock *tp);
735 extern int tcp_set_default_congestion_control(const char *name);
736 extern void tcp_get_default_congestion_control(char *name);
737 extern int tcp_set_congestion_control(struct tcp_sock *tp, const char *name);
739 extern struct tcp_congestion_ops tcp_init_congestion_ops;
740 extern u32 tcp_reno_ssthresh(struct tcp_sock *tp);
741 extern void tcp_reno_cong_avoid(struct tcp_sock *tp, u32 ack,
742 u32 rtt, u32 in_flight, int flag);
743 extern u32 tcp_reno_min_cwnd(struct tcp_sock *tp);
744 extern struct tcp_congestion_ops tcp_reno;
746 static inline void tcp_set_ca_state(struct tcp_sock *tp, u8 ca_state)
748 if (tp->ca_ops->set_state)
749 tp->ca_ops->set_state(tp, ca_state);
750 tp->ca_state = ca_state;
753 static inline void tcp_ca_event(struct tcp_sock *tp, enum tcp_ca_event event)
755 if (tp->ca_ops->cwnd_event)
756 tp->ca_ops->cwnd_event(tp, event);
759 /* This determines how many packets are "in the network" to the best
760 * of our knowledge. In many cases it is conservative, but where
761 * detailed information is available from the receiver (via SACK
762 * blocks etc.) we can make more aggressive calculations.
764 * Use this for decisions involving congestion control, use just
765 * tp->packets_out to determine if the send queue is empty or not.
767 * Read this equation as:
769 * "Packets sent once on transmission queue" MINUS
770 * "Packets left network, but not honestly ACKed yet" PLUS
771 * "Packets fast retransmitted"
773 static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
775 return (tp->packets_out - tp->left_out + tp->retrans_out);
778 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
779 * The exception is rate halving phase, when cwnd is decreasing towards
782 static inline __u32 tcp_current_ssthresh(struct tcp_sock *tp)
784 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
785 return tp->snd_ssthresh;
787 return max(tp->snd_ssthresh,
788 ((tp->snd_cwnd >> 1) +
789 (tp->snd_cwnd >> 2)));
792 static inline void tcp_sync_left_out(struct tcp_sock *tp)
794 if (tp->rx_opt.sack_ok &&
795 (tp->sacked_out >= tp->packets_out - tp->lost_out))
796 tp->sacked_out = tp->packets_out - tp->lost_out;
797 tp->left_out = tp->sacked_out + tp->lost_out;
800 /* Set slow start threshold and cwnd not falling to slow start */
801 static inline void __tcp_enter_cwr(struct tcp_sock *tp)
804 tp->snd_ssthresh = tp->ca_ops->ssthresh(tp);
805 tp->snd_cwnd = min(tp->snd_cwnd,
806 tcp_packets_in_flight(tp) + 1U);
807 tp->snd_cwnd_cnt = 0;
808 tp->high_seq = tp->snd_nxt;
809 tp->snd_cwnd_stamp = tcp_time_stamp;
810 TCP_ECN_queue_cwr(tp);
813 static inline void tcp_enter_cwr(struct tcp_sock *tp)
815 tp->prior_ssthresh = 0;
816 if (tp->ca_state < TCP_CA_CWR) {
818 tcp_set_ca_state(tp, TCP_CA_CWR);
822 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
824 /* Slow start with delack produces 3 packets of burst, so that
825 * it is safe "de facto".
827 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
832 static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss,
833 const struct sk_buff *skb)
836 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
839 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
841 const struct inet_connection_sock *icsk = inet_csk(sk);
842 if (!tp->packets_out && !icsk->icsk_pending)
843 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
844 icsk->icsk_rto, TCP_RTO_MAX);
847 static __inline__ void tcp_push_pending_frames(struct sock *sk,
850 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
853 static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
858 static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
863 extern void inet_csk_destroy_sock(struct sock *sk);
867 * Calculate(/check) TCP checksum
869 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
870 unsigned long saddr, unsigned long daddr,
873 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
876 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
878 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
881 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
883 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
884 __tcp_checksum_complete(skb);
887 /* Prequeue for VJ style copy to user, combined with checksumming. */
889 static __inline__ void tcp_prequeue_init(struct tcp_sock *tp)
891 tp->ucopy.task = NULL;
893 tp->ucopy.memory = 0;
894 skb_queue_head_init(&tp->ucopy.prequeue);
897 /* Packet is added to VJ-style prequeue for processing in process
898 * context, if a reader task is waiting. Apparently, this exciting
899 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
900 * failed somewhere. Latency? Burstiness? Well, at least now we will
901 * see, why it failed. 8)8) --ANK
903 * NOTE: is this not too big to inline?
905 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
907 struct tcp_sock *tp = tcp_sk(sk);
909 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
910 __skb_queue_tail(&tp->ucopy.prequeue, skb);
911 tp->ucopy.memory += skb->truesize;
912 if (tp->ucopy.memory > sk->sk_rcvbuf) {
913 struct sk_buff *skb1;
915 BUG_ON(sock_owned_by_user(sk));
917 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
918 sk->sk_backlog_rcv(sk, skb1);
919 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
922 tp->ucopy.memory = 0;
923 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
924 wake_up_interruptible(sk->sk_sleep);
925 if (!inet_csk_ack_scheduled(sk))
926 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
927 (3 * TCP_RTO_MIN) / 4,
939 static const char *statename[]={
940 "Unused","Established","Syn Sent","Syn Recv",
941 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
942 "Close Wait","Last ACK","Listen","Closing"
946 static __inline__ void tcp_set_state(struct sock *sk, int state)
948 int oldstate = sk->sk_state;
951 case TCP_ESTABLISHED:
952 if (oldstate != TCP_ESTABLISHED)
953 TCP_INC_STATS(TCP_MIB_CURRESTAB);
957 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
958 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
960 sk->sk_prot->unhash(sk);
961 if (inet_csk(sk)->icsk_bind_hash &&
962 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
963 inet_put_port(&tcp_hashinfo, sk);
966 if (oldstate==TCP_ESTABLISHED)
967 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
970 /* Change state AFTER socket is unhashed to avoid closed
971 * socket sitting in hash tables.
973 sk->sk_state = state;
976 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
980 static __inline__ void tcp_done(struct sock *sk)
982 tcp_set_state(sk, TCP_CLOSE);
983 tcp_clear_xmit_timers(sk);
985 sk->sk_shutdown = SHUTDOWN_MASK;
987 if (!sock_flag(sk, SOCK_DEAD))
988 sk->sk_state_change(sk);
990 inet_csk_destroy_sock(sk);
993 static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt)
996 rx_opt->eff_sacks = 0;
997 rx_opt->num_sacks = 0;
1000 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp)
1002 if (tp->rx_opt.tstamp_ok) {
1003 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1004 (TCPOPT_NOP << 16) |
1005 (TCPOPT_TIMESTAMP << 8) |
1007 *ptr++ = htonl(tstamp);
1008 *ptr++ = htonl(tp->rx_opt.ts_recent);
1010 if (tp->rx_opt.eff_sacks) {
1011 struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks;
1014 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1015 (TCPOPT_NOP << 16) |
1016 (TCPOPT_SACK << 8) |
1017 (TCPOLEN_SACK_BASE +
1018 (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1019 for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) {
1020 *ptr++ = htonl(sp[this_sack].start_seq);
1021 *ptr++ = htonl(sp[this_sack].end_seq);
1023 if (tp->rx_opt.dsack) {
1024 tp->rx_opt.dsack = 0;
1025 tp->rx_opt.eff_sacks--;
1030 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1031 * If this is every changed make sure to change the definition of
1032 * MAX_SYN_SIZE to match the new maximum number of options that you
1035 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1036 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1038 /* We always get an MSS option.
1039 * The option bytes which will be seen in normal data
1040 * packets should timestamps be used, must be in the MSS
1041 * advertised. But we subtract them from tp->mss_cache so
1042 * that calculations in tcp_sendmsg are simpler etc.
1043 * So account for this fact here if necessary. If we
1044 * don't do this correctly, as a receiver we won't
1045 * recognize data packets as being full sized when we
1046 * should, and thus we won't abide by the delayed ACK
1048 * SACKs don't matter, we never delay an ACK when we
1049 * have any of those going out.
1051 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1054 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1055 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1057 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1058 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1059 *ptr++ = htonl(tstamp); /* TSVAL */
1060 *ptr++ = htonl(ts_recent); /* TSECR */
1062 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1063 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1065 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1068 /* Determine a window scaling and initial window to offer. */
1069 extern void tcp_select_initial_window(int __space, __u32 mss,
1070 __u32 *rcv_wnd, __u32 *window_clamp,
1071 int wscale_ok, __u8 *rcv_wscale);
1073 static inline int tcp_win_from_space(int space)
1075 return sysctl_tcp_adv_win_scale<=0 ?
1076 (space>>(-sysctl_tcp_adv_win_scale)) :
1077 space - (space>>sysctl_tcp_adv_win_scale);
1080 /* Note: caller must be prepared to deal with negative returns */
1081 static inline int tcp_space(const struct sock *sk)
1083 return tcp_win_from_space(sk->sk_rcvbuf -
1084 atomic_read(&sk->sk_rmem_alloc));
1087 static inline int tcp_full_space(const struct sock *sk)
1089 return tcp_win_from_space(sk->sk_rcvbuf);
1092 static __inline__ void tcp_openreq_init(struct request_sock *req,
1093 struct tcp_options_received *rx_opt,
1094 struct sk_buff *skb)
1096 struct inet_request_sock *ireq = inet_rsk(req);
1098 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1099 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1100 req->mss = rx_opt->mss_clamp;
1101 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1102 ireq->tstamp_ok = rx_opt->tstamp_ok;
1103 ireq->sack_ok = rx_opt->sack_ok;
1104 ireq->snd_wscale = rx_opt->snd_wscale;
1105 ireq->wscale_ok = rx_opt->wscale_ok;
1108 ireq->rmt_port = skb->h.th->source;
1111 extern void tcp_enter_memory_pressure(void);
1113 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1115 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1118 static inline int keepalive_time_when(const struct tcp_sock *tp)
1120 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1123 static inline int tcp_fin_time(const struct sock *sk)
1125 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1126 const int rto = inet_csk(sk)->icsk_rto;
1128 if (fin_timeout < (rto << 2) - (rto >> 1))
1129 fin_timeout = (rto << 2) - (rto >> 1);
1134 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1136 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1138 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1141 /* RST segments are not recommended to carry timestamp,
1142 and, if they do, it is recommended to ignore PAWS because
1143 "their cleanup function should take precedence over timestamps."
1144 Certainly, it is mistake. It is necessary to understand the reasons
1145 of this constraint to relax it: if peer reboots, clock may go
1146 out-of-sync and half-open connections will not be reset.
1147 Actually, the problem would be not existing if all
1148 the implementations followed draft about maintaining clock
1149 via reboots. Linux-2.2 DOES NOT!
1151 However, we can relax time bounds for RST segments to MSL.
1153 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1158 #define TCP_CHECK_TIMER(sk) do { } while (0)
1160 static inline int tcp_use_frto(const struct sock *sk)
1162 const struct tcp_sock *tp = tcp_sk(sk);
1164 /* F-RTO must be activated in sysctl and there must be some
1165 * unsent new data, and the advertised window should allow
1168 return (sysctl_tcp_frto && sk->sk_send_head &&
1169 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1170 tp->snd_una + tp->snd_wnd));
1173 static inline void tcp_mib_init(void)
1176 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1177 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1178 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1179 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1183 enum tcp_seq_states {
1184 TCP_SEQ_STATE_LISTENING,
1185 TCP_SEQ_STATE_OPENREQ,
1186 TCP_SEQ_STATE_ESTABLISHED,
1187 TCP_SEQ_STATE_TIME_WAIT,
1190 struct tcp_seq_afinfo {
1191 struct module *owner;
1194 int (*seq_show) (struct seq_file *m, void *v);
1195 struct file_operations *seq_fops;
1198 struct tcp_iter_state {
1200 enum tcp_seq_states state;
1201 struct sock *syn_wait_sk;
1202 int bucket, sbucket, num, uid;
1203 struct seq_operations seq_ops;
1206 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1207 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);