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 inet_csk_listen_start(struct sock *sk,
427 const int nr_table_entries);
429 extern void tcp_parse_options(struct sk_buff *skb,
430 struct tcp_options_received *opt_rx,
434 * TCP v4 functions exported for the inet6 API
437 extern void tcp_v4_send_check(struct sock *sk,
438 struct tcphdr *th, int len,
439 struct sk_buff *skb);
441 extern int tcp_v4_conn_request(struct sock *sk,
442 struct sk_buff *skb);
444 extern struct sock * tcp_create_openreq_child(struct sock *sk,
445 struct request_sock *req,
446 struct sk_buff *skb);
448 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
450 struct request_sock *req,
451 struct dst_entry *dst);
453 extern int tcp_v4_do_rcv(struct sock *sk,
454 struct sk_buff *skb);
456 extern int tcp_v4_connect(struct sock *sk,
457 struct sockaddr *uaddr,
460 extern int tcp_connect(struct sock *sk);
462 extern struct sk_buff * tcp_make_synack(struct sock *sk,
463 struct dst_entry *dst,
464 struct request_sock *req);
466 extern int tcp_disconnect(struct sock *sk, int flags);
468 extern void tcp_unhash(struct sock *sk);
470 extern int tcp_v4_hash_connecting(struct sock *sk);
473 /* From syncookies.c */
474 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
475 struct ip_options *opt);
476 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
481 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
482 unsigned int cur_mss, int nonagle);
483 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
484 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
485 extern void tcp_xmit_retransmit_queue(struct sock *);
486 extern void tcp_simple_retransmit(struct sock *);
487 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
489 extern void tcp_send_probe0(struct sock *);
490 extern void tcp_send_partial(struct sock *);
491 extern int tcp_write_wakeup(struct sock *);
492 extern void tcp_send_fin(struct sock *sk);
493 extern void tcp_send_active_reset(struct sock *sk,
494 unsigned int __nocast priority);
495 extern int tcp_send_synack(struct sock *);
496 extern void tcp_push_one(struct sock *, unsigned int mss_now);
497 extern void tcp_send_ack(struct sock *sk);
498 extern void tcp_send_delayed_ack(struct sock *sk);
501 extern void tcp_cwnd_application_limited(struct sock *sk);
504 extern void tcp_init_xmit_timers(struct sock *);
505 static inline void tcp_clear_xmit_timers(struct sock *sk)
507 inet_csk_clear_xmit_timers(sk);
510 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
511 extern unsigned int tcp_current_mss(struct sock *sk, int large);
514 extern void tcp_get_info(struct sock *, struct tcp_info *);
516 /* Read 'sendfile()'-style from a TCP socket */
517 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
518 unsigned int, size_t);
519 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
520 sk_read_actor_t recv_actor);
522 /* Initialize RCV_MSS value.
523 * RCV_MSS is an our guess about MSS used by the peer.
524 * We haven't any direct information about the MSS.
525 * It's better to underestimate the RCV_MSS rather than overestimate.
526 * Overestimations make us ACKing less frequently than needed.
527 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
530 static inline void tcp_initialize_rcv_mss(struct sock *sk)
532 struct tcp_sock *tp = tcp_sk(sk);
533 unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);
535 hint = min(hint, tp->rcv_wnd/2);
536 hint = min(hint, TCP_MIN_RCVMSS);
537 hint = max(hint, TCP_MIN_MSS);
539 inet_csk(sk)->icsk_ack.rcv_mss = hint;
542 static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
544 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
545 ntohl(TCP_FLAG_ACK) |
549 static __inline__ void tcp_fast_path_on(struct tcp_sock *tp)
551 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
554 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
556 if (skb_queue_empty(&tp->out_of_order_queue) &&
558 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
560 tcp_fast_path_on(tp);
563 /* Compute the actual receive window we are currently advertising.
564 * Rcv_nxt can be after the window if our peer push more data
565 * than the offered window.
567 static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp)
569 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
576 /* Choose a new window, without checks for shrinking, and without
577 * scaling applied to the result. The caller does these things
578 * if necessary. This is a "raw" window selection.
580 extern u32 __tcp_select_window(struct sock *sk);
582 /* TCP timestamps are only 32-bits, this causes a slight
583 * complication on 64-bit systems since we store a snapshot
584 * of jiffies in the buffer control blocks below. We decidely
585 * only use of the low 32-bits of jiffies and hide the ugly
586 * casts with the following macro.
588 #define tcp_time_stamp ((__u32)(jiffies))
590 /* This is what the send packet queueing engine uses to pass
591 * TCP per-packet control information to the transmission
592 * code. We also store the host-order sequence numbers in
593 * here too. This is 36 bytes on 32-bit architectures,
594 * 40 bytes on 64-bit machines, if this grows please adjust
595 * skbuff.h:skbuff->cb[xxx] size appropriately.
599 struct inet_skb_parm h4;
600 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
601 struct inet6_skb_parm h6;
603 } header; /* For incoming frames */
604 __u32 seq; /* Starting sequence number */
605 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
606 __u32 when; /* used to compute rtt's */
607 __u8 flags; /* TCP header flags. */
609 /* NOTE: These must match up to the flags byte in a
612 #define TCPCB_FLAG_FIN 0x01
613 #define TCPCB_FLAG_SYN 0x02
614 #define TCPCB_FLAG_RST 0x04
615 #define TCPCB_FLAG_PSH 0x08
616 #define TCPCB_FLAG_ACK 0x10
617 #define TCPCB_FLAG_URG 0x20
618 #define TCPCB_FLAG_ECE 0x40
619 #define TCPCB_FLAG_CWR 0x80
621 __u8 sacked; /* State flags for SACK/FACK. */
622 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
623 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
624 #define TCPCB_LOST 0x04 /* SKB is lost */
625 #define TCPCB_TAGBITS 0x07 /* All tag bits */
627 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
628 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
630 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
632 #define TCPCB_AT_TAIL (TCPCB_URG)
634 __u16 urg_ptr; /* Valid w/URG flags is set. */
635 __u32 ack_seq; /* Sequence number ACK'd */
638 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
640 #include <net/tcp_ecn.h>
642 /* Due to TSO, an SKB can be composed of multiple actual
643 * packets. To keep these tracked properly, we use this.
645 static inline int tcp_skb_pcount(const struct sk_buff *skb)
647 return skb_shinfo(skb)->tso_segs;
650 /* This is valid iff tcp_skb_pcount() > 1. */
651 static inline int tcp_skb_mss(const struct sk_buff *skb)
653 return skb_shinfo(skb)->tso_size;
656 static inline void tcp_dec_pcount_approx(__u32 *count,
657 const struct sk_buff *skb)
660 *count -= tcp_skb_pcount(skb);
666 static inline void tcp_packets_out_inc(struct sock *sk,
668 const struct sk_buff *skb)
670 int orig = tp->packets_out;
672 tp->packets_out += tcp_skb_pcount(skb);
674 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
675 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
678 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
679 const struct sk_buff *skb)
681 tp->packets_out -= tcp_skb_pcount(skb);
684 /* Events passed to congestion control interface */
686 CA_EVENT_TX_START, /* first transmit when no packets in flight */
687 CA_EVENT_CWND_RESTART, /* congestion window restart */
688 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
689 CA_EVENT_FRTO, /* fast recovery timeout */
690 CA_EVENT_LOSS, /* loss timeout */
691 CA_EVENT_FAST_ACK, /* in sequence ack */
692 CA_EVENT_SLOW_ACK, /* other ack */
696 * Interface for adding new TCP congestion control handlers
698 #define TCP_CA_NAME_MAX 16
699 struct tcp_congestion_ops {
700 struct list_head list;
702 /* initialize private data (optional) */
703 void (*init)(struct tcp_sock *tp);
704 /* cleanup private data (optional) */
705 void (*release)(struct tcp_sock *tp);
707 /* return slow start threshold (required) */
708 u32 (*ssthresh)(struct tcp_sock *tp);
709 /* lower bound for congestion window (optional) */
710 u32 (*min_cwnd)(struct tcp_sock *tp);
711 /* do new cwnd calculation (required) */
712 void (*cong_avoid)(struct tcp_sock *tp, u32 ack,
713 u32 rtt, u32 in_flight, int good_ack);
714 /* round trip time sample per acked packet (optional) */
715 void (*rtt_sample)(struct tcp_sock *tp, u32 usrtt);
716 /* call before changing ca_state (optional) */
717 void (*set_state)(struct tcp_sock *tp, u8 new_state);
718 /* call when cwnd event occurs (optional) */
719 void (*cwnd_event)(struct tcp_sock *tp, enum tcp_ca_event ev);
720 /* new value of cwnd after loss (optional) */
721 u32 (*undo_cwnd)(struct tcp_sock *tp);
722 /* hook for packet ack accounting (optional) */
723 void (*pkts_acked)(struct tcp_sock *tp, u32 num_acked);
724 /* get info for tcp_diag (optional) */
725 void (*get_info)(struct tcp_sock *tp, u32 ext, struct sk_buff *skb);
727 char name[TCP_CA_NAME_MAX];
728 struct module *owner;
731 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
732 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
734 extern void tcp_init_congestion_control(struct tcp_sock *tp);
735 extern void tcp_cleanup_congestion_control(struct tcp_sock *tp);
736 extern int tcp_set_default_congestion_control(const char *name);
737 extern void tcp_get_default_congestion_control(char *name);
738 extern int tcp_set_congestion_control(struct tcp_sock *tp, const char *name);
740 extern struct tcp_congestion_ops tcp_init_congestion_ops;
741 extern u32 tcp_reno_ssthresh(struct tcp_sock *tp);
742 extern void tcp_reno_cong_avoid(struct tcp_sock *tp, u32 ack,
743 u32 rtt, u32 in_flight, int flag);
744 extern u32 tcp_reno_min_cwnd(struct tcp_sock *tp);
745 extern struct tcp_congestion_ops tcp_reno;
747 static inline void tcp_set_ca_state(struct tcp_sock *tp, u8 ca_state)
749 if (tp->ca_ops->set_state)
750 tp->ca_ops->set_state(tp, ca_state);
751 tp->ca_state = ca_state;
754 static inline void tcp_ca_event(struct tcp_sock *tp, enum tcp_ca_event event)
756 if (tp->ca_ops->cwnd_event)
757 tp->ca_ops->cwnd_event(tp, event);
760 /* This determines how many packets are "in the network" to the best
761 * of our knowledge. In many cases it is conservative, but where
762 * detailed information is available from the receiver (via SACK
763 * blocks etc.) we can make more aggressive calculations.
765 * Use this for decisions involving congestion control, use just
766 * tp->packets_out to determine if the send queue is empty or not.
768 * Read this equation as:
770 * "Packets sent once on transmission queue" MINUS
771 * "Packets left network, but not honestly ACKed yet" PLUS
772 * "Packets fast retransmitted"
774 static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
776 return (tp->packets_out - tp->left_out + tp->retrans_out);
779 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
780 * The exception is rate halving phase, when cwnd is decreasing towards
783 static inline __u32 tcp_current_ssthresh(struct tcp_sock *tp)
785 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
786 return tp->snd_ssthresh;
788 return max(tp->snd_ssthresh,
789 ((tp->snd_cwnd >> 1) +
790 (tp->snd_cwnd >> 2)));
793 static inline void tcp_sync_left_out(struct tcp_sock *tp)
795 if (tp->rx_opt.sack_ok &&
796 (tp->sacked_out >= tp->packets_out - tp->lost_out))
797 tp->sacked_out = tp->packets_out - tp->lost_out;
798 tp->left_out = tp->sacked_out + tp->lost_out;
801 /* Set slow start threshold and cwnd not falling to slow start */
802 static inline void __tcp_enter_cwr(struct tcp_sock *tp)
805 tp->snd_ssthresh = tp->ca_ops->ssthresh(tp);
806 tp->snd_cwnd = min(tp->snd_cwnd,
807 tcp_packets_in_flight(tp) + 1U);
808 tp->snd_cwnd_cnt = 0;
809 tp->high_seq = tp->snd_nxt;
810 tp->snd_cwnd_stamp = tcp_time_stamp;
811 TCP_ECN_queue_cwr(tp);
814 static inline void tcp_enter_cwr(struct tcp_sock *tp)
816 tp->prior_ssthresh = 0;
817 if (tp->ca_state < TCP_CA_CWR) {
819 tcp_set_ca_state(tp, TCP_CA_CWR);
823 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
825 /* Slow start with delack produces 3 packets of burst, so that
826 * it is safe "de facto".
828 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
833 static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss,
834 const struct sk_buff *skb)
837 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
840 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
842 const struct inet_connection_sock *icsk = inet_csk(sk);
843 if (!tp->packets_out && !icsk->icsk_pending)
844 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
845 icsk->icsk_rto, TCP_RTO_MAX);
848 static __inline__ void tcp_push_pending_frames(struct sock *sk,
851 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
854 static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
859 static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
864 extern void inet_csk_destroy_sock(struct sock *sk);
868 * Calculate(/check) TCP checksum
870 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
871 unsigned long saddr, unsigned long daddr,
874 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
877 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
879 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
882 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
884 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
885 __tcp_checksum_complete(skb);
888 /* Prequeue for VJ style copy to user, combined with checksumming. */
890 static __inline__ void tcp_prequeue_init(struct tcp_sock *tp)
892 tp->ucopy.task = NULL;
894 tp->ucopy.memory = 0;
895 skb_queue_head_init(&tp->ucopy.prequeue);
898 /* Packet is added to VJ-style prequeue for processing in process
899 * context, if a reader task is waiting. Apparently, this exciting
900 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
901 * failed somewhere. Latency? Burstiness? Well, at least now we will
902 * see, why it failed. 8)8) --ANK
904 * NOTE: is this not too big to inline?
906 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
908 struct tcp_sock *tp = tcp_sk(sk);
910 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
911 __skb_queue_tail(&tp->ucopy.prequeue, skb);
912 tp->ucopy.memory += skb->truesize;
913 if (tp->ucopy.memory > sk->sk_rcvbuf) {
914 struct sk_buff *skb1;
916 BUG_ON(sock_owned_by_user(sk));
918 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
919 sk->sk_backlog_rcv(sk, skb1);
920 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
923 tp->ucopy.memory = 0;
924 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
925 wake_up_interruptible(sk->sk_sleep);
926 if (!inet_csk_ack_scheduled(sk))
927 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
928 (3 * TCP_RTO_MIN) / 4,
940 static const char *statename[]={
941 "Unused","Established","Syn Sent","Syn Recv",
942 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
943 "Close Wait","Last ACK","Listen","Closing"
947 static __inline__ void tcp_set_state(struct sock *sk, int state)
949 int oldstate = sk->sk_state;
952 case TCP_ESTABLISHED:
953 if (oldstate != TCP_ESTABLISHED)
954 TCP_INC_STATS(TCP_MIB_CURRESTAB);
958 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
959 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
961 sk->sk_prot->unhash(sk);
962 if (inet_csk(sk)->icsk_bind_hash &&
963 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
964 inet_put_port(&tcp_hashinfo, sk);
967 if (oldstate==TCP_ESTABLISHED)
968 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
971 /* Change state AFTER socket is unhashed to avoid closed
972 * socket sitting in hash tables.
974 sk->sk_state = state;
977 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
981 static __inline__ void tcp_done(struct sock *sk)
983 tcp_set_state(sk, TCP_CLOSE);
984 tcp_clear_xmit_timers(sk);
986 sk->sk_shutdown = SHUTDOWN_MASK;
988 if (!sock_flag(sk, SOCK_DEAD))
989 sk->sk_state_change(sk);
991 inet_csk_destroy_sock(sk);
994 static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt)
997 rx_opt->eff_sacks = 0;
998 rx_opt->num_sacks = 0;
1001 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp)
1003 if (tp->rx_opt.tstamp_ok) {
1004 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1005 (TCPOPT_NOP << 16) |
1006 (TCPOPT_TIMESTAMP << 8) |
1008 *ptr++ = htonl(tstamp);
1009 *ptr++ = htonl(tp->rx_opt.ts_recent);
1011 if (tp->rx_opt.eff_sacks) {
1012 struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks;
1015 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1016 (TCPOPT_NOP << 16) |
1017 (TCPOPT_SACK << 8) |
1018 (TCPOLEN_SACK_BASE +
1019 (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1020 for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) {
1021 *ptr++ = htonl(sp[this_sack].start_seq);
1022 *ptr++ = htonl(sp[this_sack].end_seq);
1024 if (tp->rx_opt.dsack) {
1025 tp->rx_opt.dsack = 0;
1026 tp->rx_opt.eff_sacks--;
1031 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1032 * If this is every changed make sure to change the definition of
1033 * MAX_SYN_SIZE to match the new maximum number of options that you
1036 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1037 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1039 /* We always get an MSS option.
1040 * The option bytes which will be seen in normal data
1041 * packets should timestamps be used, must be in the MSS
1042 * advertised. But we subtract them from tp->mss_cache so
1043 * that calculations in tcp_sendmsg are simpler etc.
1044 * So account for this fact here if necessary. If we
1045 * don't do this correctly, as a receiver we won't
1046 * recognize data packets as being full sized when we
1047 * should, and thus we won't abide by the delayed ACK
1049 * SACKs don't matter, we never delay an ACK when we
1050 * have any of those going out.
1052 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1055 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1056 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1058 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1059 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1060 *ptr++ = htonl(tstamp); /* TSVAL */
1061 *ptr++ = htonl(ts_recent); /* TSECR */
1063 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1064 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1066 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1069 /* Determine a window scaling and initial window to offer. */
1070 extern void tcp_select_initial_window(int __space, __u32 mss,
1071 __u32 *rcv_wnd, __u32 *window_clamp,
1072 int wscale_ok, __u8 *rcv_wscale);
1074 static inline int tcp_win_from_space(int space)
1076 return sysctl_tcp_adv_win_scale<=0 ?
1077 (space>>(-sysctl_tcp_adv_win_scale)) :
1078 space - (space>>sysctl_tcp_adv_win_scale);
1081 /* Note: caller must be prepared to deal with negative returns */
1082 static inline int tcp_space(const struct sock *sk)
1084 return tcp_win_from_space(sk->sk_rcvbuf -
1085 atomic_read(&sk->sk_rmem_alloc));
1088 static inline int tcp_full_space(const struct sock *sk)
1090 return tcp_win_from_space(sk->sk_rcvbuf);
1093 static __inline__ void tcp_openreq_init(struct request_sock *req,
1094 struct tcp_options_received *rx_opt,
1095 struct sk_buff *skb)
1097 struct inet_request_sock *ireq = inet_rsk(req);
1099 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1100 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1101 req->mss = rx_opt->mss_clamp;
1102 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1103 ireq->tstamp_ok = rx_opt->tstamp_ok;
1104 ireq->sack_ok = rx_opt->sack_ok;
1105 ireq->snd_wscale = rx_opt->snd_wscale;
1106 ireq->wscale_ok = rx_opt->wscale_ok;
1109 ireq->rmt_port = skb->h.th->source;
1112 extern void tcp_enter_memory_pressure(void);
1114 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1116 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1119 static inline int keepalive_time_when(const struct tcp_sock *tp)
1121 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1124 static inline int tcp_fin_time(const struct sock *sk)
1126 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1127 const int rto = inet_csk(sk)->icsk_rto;
1129 if (fin_timeout < (rto << 2) - (rto >> 1))
1130 fin_timeout = (rto << 2) - (rto >> 1);
1135 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1137 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1139 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1142 /* RST segments are not recommended to carry timestamp,
1143 and, if they do, it is recommended to ignore PAWS because
1144 "their cleanup function should take precedence over timestamps."
1145 Certainly, it is mistake. It is necessary to understand the reasons
1146 of this constraint to relax it: if peer reboots, clock may go
1147 out-of-sync and half-open connections will not be reset.
1148 Actually, the problem would be not existing if all
1149 the implementations followed draft about maintaining clock
1150 via reboots. Linux-2.2 DOES NOT!
1152 However, we can relax time bounds for RST segments to MSL.
1154 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1159 #define TCP_CHECK_TIMER(sk) do { } while (0)
1161 static inline int tcp_use_frto(const struct sock *sk)
1163 const struct tcp_sock *tp = tcp_sk(sk);
1165 /* F-RTO must be activated in sysctl and there must be some
1166 * unsent new data, and the advertised window should allow
1169 return (sysctl_tcp_frto && sk->sk_send_head &&
1170 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1171 tp->snd_una + tp->snd_wnd));
1174 static inline void tcp_mib_init(void)
1177 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1178 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1179 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1180 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1184 enum tcp_seq_states {
1185 TCP_SEQ_STATE_LISTENING,
1186 TCP_SEQ_STATE_OPENREQ,
1187 TCP_SEQ_STATE_ESTABLISHED,
1188 TCP_SEQ_STATE_TIME_WAIT,
1191 struct tcp_seq_afinfo {
1192 struct module *owner;
1195 int (*seq_show) (struct seq_file *m, void *v);
1196 struct file_operations *seq_fops;
1199 struct tcp_iter_state {
1201 enum tcp_seq_states state;
1202 struct sock *syn_wait_sk;
1203 int bucket, sbucket, num, uid;
1204 struct seq_operations seq_ops;
1207 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1208 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);