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>
30 #include <linux/skbuff.h>
32 #include <net/inet_connection_sock.h>
33 #include <net/inet_timewait_sock.h>
34 #include <net/inet_hashtables.h>
35 #include <net/checksum.h>
36 #include <net/request_sock.h>
40 #include <net/tcp_states.h>
42 #include <linux/seq_file.h>
44 extern struct inet_hashinfo tcp_hashinfo;
46 extern atomic_t tcp_orphan_count;
47 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
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 /* The least MTU to use for probing */
64 #define TCP_BASE_MSS 512
66 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
67 #define TCP_FASTRETRANS_THRESH 3
69 /* Maximal reordering. */
70 #define TCP_MAX_REORDERING 127
72 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
73 #define TCP_MAX_QUICKACKS 16U
76 #define TCP_URG_VALID 0x0100
77 #define TCP_URG_NOTYET 0x0200
78 #define TCP_URG_READ 0x0400
80 #define TCP_RETR1 3 /*
81 * This is how many retries it does before it
82 * tries to figure out if the gateway is
83 * down. Minimal RFC value is 3; it corresponds
84 * to ~3sec-8min depending on RTO.
87 #define TCP_RETR2 15 /*
88 * This should take at least
89 * 90 minutes to time out.
90 * RFC1122 says that the limit is 100 sec.
91 * 15 is ~13-30min depending on RTO.
94 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
95 * connection: ~180sec is RFC minimum */
97 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
98 * connection: ~180sec is RFC minimum */
101 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
102 * socket. 7 is ~50sec-16min.
106 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
107 * state, about 60 seconds */
108 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
109 /* BSD style FIN_WAIT2 deadlock breaker.
110 * It used to be 3min, new value is 60sec,
111 * to combine FIN-WAIT-2 timeout with
115 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
117 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
118 #define TCP_ATO_MIN ((unsigned)(HZ/25))
120 #define TCP_DELACK_MIN 4U
121 #define TCP_ATO_MIN 4U
123 #define TCP_RTO_MAX ((unsigned)(120*HZ))
124 #define TCP_RTO_MIN ((unsigned)(HZ/5))
125 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
127 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
128 * for local resources.
131 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
132 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
133 #define TCP_KEEPALIVE_INTVL (75*HZ)
135 #define MAX_TCP_KEEPIDLE 32767
136 #define MAX_TCP_KEEPINTVL 32767
137 #define MAX_TCP_KEEPCNT 127
138 #define MAX_TCP_SYNCNT 127
140 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
141 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
143 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
144 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
145 * after this time. It should be equal
146 * (or greater than) TCP_TIMEWAIT_LEN
147 * to provide reliability equal to one
148 * provided by timewait state.
150 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
151 * timestamps. It must be less than
152 * minimal timewait lifetime.
158 #define TCPOPT_NOP 1 /* Padding */
159 #define TCPOPT_EOL 0 /* End of options */
160 #define TCPOPT_MSS 2 /* Segment size negotiating */
161 #define TCPOPT_WINDOW 3 /* Window scaling */
162 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
163 #define TCPOPT_SACK 5 /* SACK Block */
164 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
170 #define TCPOLEN_MSS 4
171 #define TCPOLEN_WINDOW 3
172 #define TCPOLEN_SACK_PERM 2
173 #define TCPOLEN_TIMESTAMP 10
175 /* But this is what stacks really send out. */
176 #define TCPOLEN_TSTAMP_ALIGNED 12
177 #define TCPOLEN_WSCALE_ALIGNED 4
178 #define TCPOLEN_SACKPERM_ALIGNED 4
179 #define TCPOLEN_SACK_BASE 2
180 #define TCPOLEN_SACK_BASE_ALIGNED 4
181 #define TCPOLEN_SACK_PERBLOCK 8
183 /* Flags in tp->nonagle */
184 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
185 #define TCP_NAGLE_CORK 2 /* Socket is corked */
186 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
188 extern struct inet_timewait_death_row tcp_death_row;
190 /* sysctl variables for tcp */
191 extern int sysctl_tcp_timestamps;
192 extern int sysctl_tcp_window_scaling;
193 extern int sysctl_tcp_sack;
194 extern int sysctl_tcp_fin_timeout;
195 extern int sysctl_tcp_keepalive_time;
196 extern int sysctl_tcp_keepalive_probes;
197 extern int sysctl_tcp_keepalive_intvl;
198 extern int sysctl_tcp_syn_retries;
199 extern int sysctl_tcp_synack_retries;
200 extern int sysctl_tcp_retries1;
201 extern int sysctl_tcp_retries2;
202 extern int sysctl_tcp_orphan_retries;
203 extern int sysctl_tcp_syncookies;
204 extern int sysctl_tcp_retrans_collapse;
205 extern int sysctl_tcp_stdurg;
206 extern int sysctl_tcp_rfc1337;
207 extern int sysctl_tcp_abort_on_overflow;
208 extern int sysctl_tcp_max_orphans;
209 extern int sysctl_tcp_fack;
210 extern int sysctl_tcp_reordering;
211 extern int sysctl_tcp_ecn;
212 extern int sysctl_tcp_dsack;
213 extern int sysctl_tcp_mem[3];
214 extern int sysctl_tcp_wmem[3];
215 extern int sysctl_tcp_rmem[3];
216 extern int sysctl_tcp_app_win;
217 extern int sysctl_tcp_adv_win_scale;
218 extern int sysctl_tcp_tw_reuse;
219 extern int sysctl_tcp_frto;
220 extern int sysctl_tcp_low_latency;
221 extern int sysctl_tcp_nometrics_save;
222 extern int sysctl_tcp_moderate_rcvbuf;
223 extern int sysctl_tcp_tso_win_divisor;
224 extern int sysctl_tcp_abc;
225 extern int sysctl_tcp_mtu_probing;
226 extern int sysctl_tcp_base_mss;
228 extern atomic_t tcp_memory_allocated;
229 extern atomic_t tcp_sockets_allocated;
230 extern int tcp_memory_pressure;
233 * The next routines deal with comparing 32 bit unsigned ints
234 * and worry about wraparound (automatic with unsigned arithmetic).
237 static inline int before(__u32 seq1, __u32 seq2)
239 return (__s32)(seq1-seq2) < 0;
242 static inline int after(__u32 seq1, __u32 seq2)
244 return (__s32)(seq2-seq1) < 0;
248 /* is s2<=s1<=s3 ? */
249 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
251 return seq3 - seq2 >= seq1 - seq2;
255 extern struct proto tcp_prot;
257 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
258 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
259 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
260 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
261 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
262 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
263 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
265 extern void tcp_v4_err(struct sk_buff *skb, u32);
267 extern void tcp_shutdown (struct sock *sk, int how);
269 extern int tcp_v4_rcv(struct sk_buff *skb);
271 extern int tcp_v4_remember_stamp(struct sock *sk);
273 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
275 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
276 struct msghdr *msg, size_t size);
277 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
279 extern int tcp_ioctl(struct sock *sk,
283 extern int tcp_rcv_state_process(struct sock *sk,
288 extern int tcp_rcv_established(struct sock *sk,
293 extern void tcp_rcv_space_adjust(struct sock *sk);
295 extern int tcp_twsk_unique(struct sock *sk,
296 struct sock *sktw, void *twp);
298 static inline void tcp_dec_quickack_mode(struct sock *sk,
299 const unsigned int pkts)
301 struct inet_connection_sock *icsk = inet_csk(sk);
303 if (icsk->icsk_ack.quick) {
304 if (pkts >= icsk->icsk_ack.quick) {
305 icsk->icsk_ack.quick = 0;
306 /* Leaving quickack mode we deflate ATO. */
307 icsk->icsk_ack.ato = TCP_ATO_MIN;
309 icsk->icsk_ack.quick -= pkts;
313 extern void tcp_enter_quickack_mode(struct sock *sk);
315 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
317 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
329 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
331 const struct tcphdr *th);
333 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
334 struct request_sock *req,
335 struct request_sock **prev);
336 extern int tcp_child_process(struct sock *parent,
338 struct sk_buff *skb);
339 extern void tcp_enter_frto(struct sock *sk);
340 extern void tcp_enter_loss(struct sock *sk, int how);
341 extern void tcp_clear_retrans(struct tcp_sock *tp);
342 extern void tcp_update_metrics(struct sock *sk);
344 extern void tcp_close(struct sock *sk,
346 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
348 extern int tcp_getsockopt(struct sock *sk, int level,
352 extern int tcp_setsockopt(struct sock *sk, int level,
353 int optname, char __user *optval,
355 extern void tcp_set_keepalive(struct sock *sk, int val);
356 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
358 size_t len, int nonblock,
359 int flags, int *addr_len);
361 extern void tcp_parse_options(struct sk_buff *skb,
362 struct tcp_options_received *opt_rx,
366 * TCP v4 functions exported for the inet6 API
369 extern void tcp_v4_send_check(struct sock *sk, int len,
370 struct sk_buff *skb);
372 extern int tcp_v4_conn_request(struct sock *sk,
373 struct sk_buff *skb);
375 extern struct sock * tcp_create_openreq_child(struct sock *sk,
376 struct request_sock *req,
377 struct sk_buff *skb);
379 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
381 struct request_sock *req,
382 struct dst_entry *dst);
384 extern int tcp_v4_do_rcv(struct sock *sk,
385 struct sk_buff *skb);
387 extern int tcp_v4_connect(struct sock *sk,
388 struct sockaddr *uaddr,
391 extern int tcp_connect(struct sock *sk);
393 extern struct sk_buff * tcp_make_synack(struct sock *sk,
394 struct dst_entry *dst,
395 struct request_sock *req);
397 extern int tcp_disconnect(struct sock *sk, int flags);
399 extern void tcp_unhash(struct sock *sk);
401 extern int tcp_v4_hash_connecting(struct sock *sk);
404 /* From syncookies.c */
405 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
406 struct ip_options *opt);
407 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
412 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
413 unsigned int cur_mss, int nonagle);
414 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
415 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
416 extern void tcp_xmit_retransmit_queue(struct sock *);
417 extern void tcp_simple_retransmit(struct sock *);
418 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
419 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
421 extern void tcp_send_probe0(struct sock *);
422 extern void tcp_send_partial(struct sock *);
423 extern int tcp_write_wakeup(struct sock *);
424 extern void tcp_send_fin(struct sock *sk);
425 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
426 extern int tcp_send_synack(struct sock *);
427 extern void tcp_push_one(struct sock *, unsigned int mss_now);
428 extern void tcp_send_ack(struct sock *sk);
429 extern void tcp_send_delayed_ack(struct sock *sk);
432 extern void tcp_cwnd_application_limited(struct sock *sk);
435 extern void tcp_init_xmit_timers(struct sock *);
436 static inline void tcp_clear_xmit_timers(struct sock *sk)
438 inet_csk_clear_xmit_timers(sk);
441 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
442 extern unsigned int tcp_current_mss(struct sock *sk, int large);
445 extern void tcp_get_info(struct sock *, struct tcp_info *);
447 /* Read 'sendfile()'-style from a TCP socket */
448 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
449 unsigned int, size_t);
450 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
451 sk_read_actor_t recv_actor);
453 extern void tcp_initialize_rcv_mss(struct sock *sk);
455 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
456 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
457 extern void tcp_mtup_init(struct sock *sk);
459 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
461 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
462 ntohl(TCP_FLAG_ACK) |
466 static inline void tcp_fast_path_on(struct tcp_sock *tp)
468 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
471 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
473 if (skb_queue_empty(&tp->out_of_order_queue) &&
475 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
477 tcp_fast_path_on(tp);
480 /* Compute the actual receive window we are currently advertising.
481 * Rcv_nxt can be after the window if our peer push more data
482 * than the offered window.
484 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
486 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
493 /* Choose a new window, without checks for shrinking, and without
494 * scaling applied to the result. The caller does these things
495 * if necessary. This is a "raw" window selection.
497 extern u32 __tcp_select_window(struct sock *sk);
499 /* TCP timestamps are only 32-bits, this causes a slight
500 * complication on 64-bit systems since we store a snapshot
501 * of jiffies in the buffer control blocks below. We decided
502 * to use only the low 32-bits of jiffies and hide the ugly
503 * casts with the following macro.
505 #define tcp_time_stamp ((__u32)(jiffies))
507 /* This is what the send packet queuing engine uses to pass
508 * TCP per-packet control information to the transmission
509 * code. We also store the host-order sequence numbers in
510 * here too. This is 36 bytes on 32-bit architectures,
511 * 40 bytes on 64-bit machines, if this grows please adjust
512 * skbuff.h:skbuff->cb[xxx] size appropriately.
516 struct inet_skb_parm h4;
517 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
518 struct inet6_skb_parm h6;
520 } header; /* For incoming frames */
521 __u32 seq; /* Starting sequence number */
522 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
523 __u32 when; /* used to compute rtt's */
524 __u8 flags; /* TCP header flags. */
526 /* NOTE: These must match up to the flags byte in a
529 #define TCPCB_FLAG_FIN 0x01
530 #define TCPCB_FLAG_SYN 0x02
531 #define TCPCB_FLAG_RST 0x04
532 #define TCPCB_FLAG_PSH 0x08
533 #define TCPCB_FLAG_ACK 0x10
534 #define TCPCB_FLAG_URG 0x20
535 #define TCPCB_FLAG_ECE 0x40
536 #define TCPCB_FLAG_CWR 0x80
538 __u8 sacked; /* State flags for SACK/FACK. */
539 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
540 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
541 #define TCPCB_LOST 0x04 /* SKB is lost */
542 #define TCPCB_TAGBITS 0x07 /* All tag bits */
544 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
545 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
547 #define TCPCB_URG 0x20 /* Urgent pointer advanced here */
549 #define TCPCB_AT_TAIL (TCPCB_URG)
551 __u16 urg_ptr; /* Valid w/URG flags is set. */
552 __u32 ack_seq; /* Sequence number ACK'd */
555 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
557 #include <net/tcp_ecn.h>
559 /* Due to TSO, an SKB can be composed of multiple actual
560 * packets. To keep these tracked properly, we use this.
562 static inline int tcp_skb_pcount(const struct sk_buff *skb)
564 return skb_shinfo(skb)->tso_segs;
567 /* This is valid iff tcp_skb_pcount() > 1. */
568 static inline int tcp_skb_mss(const struct sk_buff *skb)
570 return skb_shinfo(skb)->tso_size;
573 static inline void tcp_dec_pcount_approx(__u32 *count,
574 const struct sk_buff *skb)
577 *count -= tcp_skb_pcount(skb);
583 static inline void tcp_packets_out_inc(struct sock *sk,
585 const struct sk_buff *skb)
587 int orig = tp->packets_out;
589 tp->packets_out += tcp_skb_pcount(skb);
591 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
592 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
595 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
596 const struct sk_buff *skb)
598 tp->packets_out -= tcp_skb_pcount(skb);
601 /* Events passed to congestion control interface */
603 CA_EVENT_TX_START, /* first transmit when no packets in flight */
604 CA_EVENT_CWND_RESTART, /* congestion window restart */
605 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
606 CA_EVENT_FRTO, /* fast recovery timeout */
607 CA_EVENT_LOSS, /* loss timeout */
608 CA_EVENT_FAST_ACK, /* in sequence ack */
609 CA_EVENT_SLOW_ACK, /* other ack */
613 * Interface for adding new TCP congestion control handlers
615 #define TCP_CA_NAME_MAX 16
616 struct tcp_congestion_ops {
617 struct list_head list;
619 /* initialize private data (optional) */
620 void (*init)(struct sock *sk);
621 /* cleanup private data (optional) */
622 void (*release)(struct sock *sk);
624 /* return slow start threshold (required) */
625 u32 (*ssthresh)(struct sock *sk);
626 /* lower bound for congestion window (optional) */
627 u32 (*min_cwnd)(struct sock *sk);
628 /* do new cwnd calculation (required) */
629 void (*cong_avoid)(struct sock *sk, u32 ack,
630 u32 rtt, u32 in_flight, int good_ack);
631 /* round trip time sample per acked packet (optional) */
632 void (*rtt_sample)(struct sock *sk, u32 usrtt);
633 /* call before changing ca_state (optional) */
634 void (*set_state)(struct sock *sk, u8 new_state);
635 /* call when cwnd event occurs (optional) */
636 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
637 /* new value of cwnd after loss (optional) */
638 u32 (*undo_cwnd)(struct sock *sk);
639 /* hook for packet ack accounting (optional) */
640 void (*pkts_acked)(struct sock *sk, u32 num_acked);
641 /* get info for inet_diag (optional) */
642 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
644 char name[TCP_CA_NAME_MAX];
645 struct module *owner;
648 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
649 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
651 extern void tcp_init_congestion_control(struct sock *sk);
652 extern void tcp_cleanup_congestion_control(struct sock *sk);
653 extern int tcp_set_default_congestion_control(const char *name);
654 extern void tcp_get_default_congestion_control(char *name);
655 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
656 extern void tcp_slow_start(struct tcp_sock *tp);
658 extern struct tcp_congestion_ops tcp_init_congestion_ops;
659 extern u32 tcp_reno_ssthresh(struct sock *sk);
660 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
661 u32 rtt, u32 in_flight, int flag);
662 extern u32 tcp_reno_min_cwnd(struct sock *sk);
663 extern struct tcp_congestion_ops tcp_reno;
665 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
667 struct inet_connection_sock *icsk = inet_csk(sk);
669 if (icsk->icsk_ca_ops->set_state)
670 icsk->icsk_ca_ops->set_state(sk, ca_state);
671 icsk->icsk_ca_state = ca_state;
674 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
676 const struct inet_connection_sock *icsk = inet_csk(sk);
678 if (icsk->icsk_ca_ops->cwnd_event)
679 icsk->icsk_ca_ops->cwnd_event(sk, event);
682 /* This determines how many packets are "in the network" to the best
683 * of our knowledge. In many cases it is conservative, but where
684 * detailed information is available from the receiver (via SACK
685 * blocks etc.) we can make more aggressive calculations.
687 * Use this for decisions involving congestion control, use just
688 * tp->packets_out to determine if the send queue is empty or not.
690 * Read this equation as:
692 * "Packets sent once on transmission queue" MINUS
693 * "Packets left network, but not honestly ACKed yet" PLUS
694 * "Packets fast retransmitted"
696 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
698 return (tp->packets_out - tp->left_out + tp->retrans_out);
701 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
702 * The exception is rate halving phase, when cwnd is decreasing towards
705 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
707 const struct tcp_sock *tp = tcp_sk(sk);
708 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
709 return tp->snd_ssthresh;
711 return max(tp->snd_ssthresh,
712 ((tp->snd_cwnd >> 1) +
713 (tp->snd_cwnd >> 2)));
716 static inline void tcp_sync_left_out(struct tcp_sock *tp)
718 if (tp->rx_opt.sack_ok &&
719 (tp->sacked_out >= tp->packets_out - tp->lost_out))
720 tp->sacked_out = tp->packets_out - tp->lost_out;
721 tp->left_out = tp->sacked_out + tp->lost_out;
724 extern void tcp_enter_cwr(struct sock *sk);
725 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
727 /* Slow start with delack produces 3 packets of burst, so that
728 * it is safe "de facto".
730 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
735 /* RFC2861 Check whether we are limited by application or congestion window
736 * This is the inverse of cwnd check in tcp_tso_should_defer
738 static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
740 const struct tcp_sock *tp = tcp_sk(sk);
743 if (in_flight >= tp->snd_cwnd)
746 if (!(sk->sk_route_caps & NETIF_F_TSO))
749 left = tp->snd_cwnd - in_flight;
750 if (sysctl_tcp_tso_win_divisor)
751 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
753 return left <= tcp_max_burst(tp);
756 static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
757 const struct sk_buff *skb)
760 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
763 static inline void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
765 const struct inet_connection_sock *icsk = inet_csk(sk);
766 if (!tp->packets_out && !icsk->icsk_pending)
767 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
768 icsk->icsk_rto, TCP_RTO_MAX);
771 static inline void tcp_push_pending_frames(struct sock *sk,
774 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
777 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
782 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
788 * Calculate(/check) TCP checksum
790 static inline u16 tcp_v4_check(struct tcphdr *th, int len,
791 unsigned long saddr, unsigned long daddr,
794 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
797 static inline int __tcp_checksum_complete(struct sk_buff *skb)
799 return __skb_checksum_complete(skb);
802 static inline int tcp_checksum_complete(struct sk_buff *skb)
804 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
805 __tcp_checksum_complete(skb);
808 /* Prequeue for VJ style copy to user, combined with checksumming. */
810 static inline void tcp_prequeue_init(struct tcp_sock *tp)
812 tp->ucopy.task = NULL;
814 tp->ucopy.memory = 0;
815 skb_queue_head_init(&tp->ucopy.prequeue);
818 /* Packet is added to VJ-style prequeue for processing in process
819 * context, if a reader task is waiting. Apparently, this exciting
820 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
821 * failed somewhere. Latency? Burstiness? Well, at least now we will
822 * see, why it failed. 8)8) --ANK
824 * NOTE: is this not too big to inline?
826 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
828 struct tcp_sock *tp = tcp_sk(sk);
830 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
831 __skb_queue_tail(&tp->ucopy.prequeue, skb);
832 tp->ucopy.memory += skb->truesize;
833 if (tp->ucopy.memory > sk->sk_rcvbuf) {
834 struct sk_buff *skb1;
836 BUG_ON(sock_owned_by_user(sk));
838 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
839 sk->sk_backlog_rcv(sk, skb1);
840 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
843 tp->ucopy.memory = 0;
844 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
845 wake_up_interruptible(sk->sk_sleep);
846 if (!inet_csk_ack_scheduled(sk))
847 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
848 (3 * TCP_RTO_MIN) / 4,
860 static const char *statename[]={
861 "Unused","Established","Syn Sent","Syn Recv",
862 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
863 "Close Wait","Last ACK","Listen","Closing"
867 static inline void tcp_set_state(struct sock *sk, int state)
869 int oldstate = sk->sk_state;
872 case TCP_ESTABLISHED:
873 if (oldstate != TCP_ESTABLISHED)
874 TCP_INC_STATS(TCP_MIB_CURRESTAB);
878 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
879 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
881 sk->sk_prot->unhash(sk);
882 if (inet_csk(sk)->icsk_bind_hash &&
883 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
884 inet_put_port(&tcp_hashinfo, sk);
887 if (oldstate==TCP_ESTABLISHED)
888 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
891 /* Change state AFTER socket is unhashed to avoid closed
892 * socket sitting in hash tables.
894 sk->sk_state = state;
897 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
901 static inline void tcp_done(struct sock *sk)
903 tcp_set_state(sk, TCP_CLOSE);
904 tcp_clear_xmit_timers(sk);
906 sk->sk_shutdown = SHUTDOWN_MASK;
908 if (!sock_flag(sk, SOCK_DEAD))
909 sk->sk_state_change(sk);
911 inet_csk_destroy_sock(sk);
914 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
917 rx_opt->eff_sacks = 0;
918 rx_opt->num_sacks = 0;
921 /* Determine a window scaling and initial window to offer. */
922 extern void tcp_select_initial_window(int __space, __u32 mss,
923 __u32 *rcv_wnd, __u32 *window_clamp,
924 int wscale_ok, __u8 *rcv_wscale);
926 static inline int tcp_win_from_space(int space)
928 return sysctl_tcp_adv_win_scale<=0 ?
929 (space>>(-sysctl_tcp_adv_win_scale)) :
930 space - (space>>sysctl_tcp_adv_win_scale);
933 /* Note: caller must be prepared to deal with negative returns */
934 static inline int tcp_space(const struct sock *sk)
936 return tcp_win_from_space(sk->sk_rcvbuf -
937 atomic_read(&sk->sk_rmem_alloc));
940 static inline int tcp_full_space(const struct sock *sk)
942 return tcp_win_from_space(sk->sk_rcvbuf);
945 static inline void tcp_openreq_init(struct request_sock *req,
946 struct tcp_options_received *rx_opt,
949 struct inet_request_sock *ireq = inet_rsk(req);
951 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
952 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
953 req->mss = rx_opt->mss_clamp;
954 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
955 ireq->tstamp_ok = rx_opt->tstamp_ok;
956 ireq->sack_ok = rx_opt->sack_ok;
957 ireq->snd_wscale = rx_opt->snd_wscale;
958 ireq->wscale_ok = rx_opt->wscale_ok;
961 ireq->rmt_port = skb->h.th->source;
964 extern void tcp_enter_memory_pressure(void);
966 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
968 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
971 static inline int keepalive_time_when(const struct tcp_sock *tp)
973 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
976 static inline int tcp_fin_time(const struct sock *sk)
978 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
979 const int rto = inet_csk(sk)->icsk_rto;
981 if (fin_timeout < (rto << 2) - (rto >> 1))
982 fin_timeout = (rto << 2) - (rto >> 1);
987 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
989 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
991 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
994 /* RST segments are not recommended to carry timestamp,
995 and, if they do, it is recommended to ignore PAWS because
996 "their cleanup function should take precedence over timestamps."
997 Certainly, it is mistake. It is necessary to understand the reasons
998 of this constraint to relax it: if peer reboots, clock may go
999 out-of-sync and half-open connections will not be reset.
1000 Actually, the problem would be not existing if all
1001 the implementations followed draft about maintaining clock
1002 via reboots. Linux-2.2 DOES NOT!
1004 However, we can relax time bounds for RST segments to MSL.
1006 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1011 #define TCP_CHECK_TIMER(sk) do { } while (0)
1013 static inline int tcp_use_frto(const struct sock *sk)
1015 const struct tcp_sock *tp = tcp_sk(sk);
1017 /* F-RTO must be activated in sysctl and there must be some
1018 * unsent new data, and the advertised window should allow
1021 return (sysctl_tcp_frto && sk->sk_send_head &&
1022 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1023 tp->snd_una + tp->snd_wnd));
1026 static inline void tcp_mib_init(void)
1029 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1030 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1031 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1032 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1036 static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1037 tp->lost_skb_hint = NULL;
1038 tp->scoreboard_skb_hint = NULL;
1039 tp->retransmit_skb_hint = NULL;
1040 tp->forward_skb_hint = NULL;
1041 tp->fastpath_skb_hint = NULL;
1045 enum tcp_seq_states {
1046 TCP_SEQ_STATE_LISTENING,
1047 TCP_SEQ_STATE_OPENREQ,
1048 TCP_SEQ_STATE_ESTABLISHED,
1049 TCP_SEQ_STATE_TIME_WAIT,
1052 struct tcp_seq_afinfo {
1053 struct module *owner;
1056 int (*seq_show) (struct seq_file *m, void *v);
1057 struct file_operations *seq_fops;
1060 struct tcp_iter_state {
1062 enum tcp_seq_states state;
1063 struct sock *syn_wait_sk;
1064 int bucket, sbucket, num, uid;
1065 struct seq_operations seq_ops;
1068 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1069 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1071 extern struct request_sock_ops tcp_request_sock_ops;
1073 extern int tcp_v4_destroy_sock(struct sock *sk);
1075 #ifdef CONFIG_PROC_FS
1076 extern int tcp4_proc_init(void);
1077 extern void tcp4_proc_exit(void);
1080 extern void tcp_v4_init(struct net_proto_family *ops);
1081 extern void tcp_init(void);