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/list.h>
25 #include <linux/tcp.h>
26 #include <linux/slab.h>
27 #include <linux/cache.h>
28 #include <linux/percpu.h>
29 #include <linux/skbuff.h>
30 #include <linux/dmaengine.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_dma_copybreak;
222 extern int sysctl_tcp_nometrics_save;
223 extern int sysctl_tcp_moderate_rcvbuf;
224 extern int sysctl_tcp_tso_win_divisor;
225 extern int sysctl_tcp_abc;
226 extern int sysctl_tcp_mtu_probing;
227 extern int sysctl_tcp_base_mss;
228 extern int sysctl_tcp_workaround_signed_windows;
229 extern int sysctl_tcp_slow_start_after_idle;
231 extern atomic_t tcp_memory_allocated;
232 extern atomic_t tcp_sockets_allocated;
233 extern int tcp_memory_pressure;
236 * The next routines deal with comparing 32 bit unsigned ints
237 * and worry about wraparound (automatic with unsigned arithmetic).
240 static inline int before(__u32 seq1, __u32 seq2)
242 return (__s32)(seq1-seq2) < 0;
245 static inline int after(__u32 seq1, __u32 seq2)
247 return (__s32)(seq2-seq1) < 0;
251 /* is s2<=s1<=s3 ? */
252 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
254 return seq3 - seq2 >= seq1 - seq2;
258 extern struct proto tcp_prot;
260 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
261 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
262 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
263 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
264 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
265 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
266 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
268 extern void tcp_v4_err(struct sk_buff *skb, u32);
270 extern void tcp_shutdown (struct sock *sk, int how);
272 extern int tcp_v4_rcv(struct sk_buff *skb);
274 extern struct sock * tcp_v4_lookup_listener(u32 daddr, unsigned short hnum, int dif);
276 extern int tcp_v4_remember_stamp(struct sock *sk);
278 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
280 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
281 struct msghdr *msg, size_t size);
282 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
284 extern int tcp_ioctl(struct sock *sk,
288 extern int tcp_rcv_state_process(struct sock *sk,
293 extern int tcp_rcv_established(struct sock *sk,
298 extern void tcp_rcv_space_adjust(struct sock *sk);
300 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
302 extern int tcp_twsk_unique(struct sock *sk,
303 struct sock *sktw, void *twp);
305 static inline void tcp_dec_quickack_mode(struct sock *sk,
306 const unsigned int pkts)
308 struct inet_connection_sock *icsk = inet_csk(sk);
310 if (icsk->icsk_ack.quick) {
311 if (pkts >= icsk->icsk_ack.quick) {
312 icsk->icsk_ack.quick = 0;
313 /* Leaving quickack mode we deflate ATO. */
314 icsk->icsk_ack.ato = TCP_ATO_MIN;
316 icsk->icsk_ack.quick -= pkts;
320 extern void tcp_enter_quickack_mode(struct sock *sk);
322 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
324 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
336 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
338 const struct tcphdr *th);
340 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
341 struct request_sock *req,
342 struct request_sock **prev);
343 extern int tcp_child_process(struct sock *parent,
345 struct sk_buff *skb);
346 extern void tcp_enter_frto(struct sock *sk);
347 extern void tcp_enter_loss(struct sock *sk, int how);
348 extern void tcp_clear_retrans(struct tcp_sock *tp);
349 extern void tcp_update_metrics(struct sock *sk);
351 extern void tcp_close(struct sock *sk,
353 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
355 extern int tcp_getsockopt(struct sock *sk, int level,
359 extern int tcp_setsockopt(struct sock *sk, int level,
360 int optname, char __user *optval,
362 extern int compat_tcp_getsockopt(struct sock *sk,
363 int level, int optname,
364 char __user *optval, int __user *optlen);
365 extern int compat_tcp_setsockopt(struct sock *sk,
366 int level, int optname,
367 char __user *optval, int optlen);
368 extern void tcp_set_keepalive(struct sock *sk, int val);
369 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
371 size_t len, int nonblock,
372 int flags, int *addr_len);
374 extern void tcp_parse_options(struct sk_buff *skb,
375 struct tcp_options_received *opt_rx,
379 * TCP v4 functions exported for the inet6 API
382 extern void tcp_v4_send_check(struct sock *sk, int len,
383 struct sk_buff *skb);
385 extern int tcp_v4_conn_request(struct sock *sk,
386 struct sk_buff *skb);
388 extern struct sock * tcp_create_openreq_child(struct sock *sk,
389 struct request_sock *req,
390 struct sk_buff *skb);
392 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
394 struct request_sock *req,
395 struct dst_entry *dst);
397 extern int tcp_v4_do_rcv(struct sock *sk,
398 struct sk_buff *skb);
400 extern int tcp_v4_connect(struct sock *sk,
401 struct sockaddr *uaddr,
404 extern int tcp_connect(struct sock *sk);
406 extern struct sk_buff * tcp_make_synack(struct sock *sk,
407 struct dst_entry *dst,
408 struct request_sock *req);
410 extern int tcp_disconnect(struct sock *sk, int flags);
412 extern void tcp_unhash(struct sock *sk);
414 /* From syncookies.c */
415 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
416 struct ip_options *opt);
417 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
422 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
423 unsigned int cur_mss, int nonagle);
424 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
425 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
426 extern void tcp_xmit_retransmit_queue(struct sock *);
427 extern void tcp_simple_retransmit(struct sock *);
428 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
429 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
431 extern void tcp_send_probe0(struct sock *);
432 extern void tcp_send_partial(struct sock *);
433 extern int tcp_write_wakeup(struct sock *);
434 extern void tcp_send_fin(struct sock *sk);
435 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
436 extern int tcp_send_synack(struct sock *);
437 extern void tcp_push_one(struct sock *, unsigned int mss_now);
438 extern void tcp_send_ack(struct sock *sk);
439 extern void tcp_send_delayed_ack(struct sock *sk);
440 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
443 extern void tcp_cwnd_application_limited(struct sock *sk);
446 extern void tcp_init_xmit_timers(struct sock *);
447 static inline void tcp_clear_xmit_timers(struct sock *sk)
449 inet_csk_clear_xmit_timers(sk);
452 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
453 extern unsigned int tcp_current_mss(struct sock *sk, int large);
456 extern void tcp_get_info(struct sock *, struct tcp_info *);
458 /* Read 'sendfile()'-style from a TCP socket */
459 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
460 unsigned int, size_t);
461 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
462 sk_read_actor_t recv_actor);
464 extern void tcp_initialize_rcv_mss(struct sock *sk);
466 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
467 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
468 extern void tcp_mtup_init(struct sock *sk);
470 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
472 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
473 ntohl(TCP_FLAG_ACK) |
477 static inline void tcp_fast_path_on(struct tcp_sock *tp)
479 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
482 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
484 if (skb_queue_empty(&tp->out_of_order_queue) &&
486 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
488 tcp_fast_path_on(tp);
491 /* Compute the actual receive window we are currently advertising.
492 * Rcv_nxt can be after the window if our peer push more data
493 * than the offered window.
495 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
497 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
504 /* Choose a new window, without checks for shrinking, and without
505 * scaling applied to the result. The caller does these things
506 * if necessary. This is a "raw" window selection.
508 extern u32 __tcp_select_window(struct sock *sk);
510 /* TCP timestamps are only 32-bits, this causes a slight
511 * complication on 64-bit systems since we store a snapshot
512 * of jiffies in the buffer control blocks below. We decided
513 * to use only the low 32-bits of jiffies and hide the ugly
514 * casts with the following macro.
516 #define tcp_time_stamp ((__u32)(jiffies))
518 /* This is what the send packet queuing engine uses to pass
519 * TCP per-packet control information to the transmission
520 * code. We also store the host-order sequence numbers in
521 * here too. This is 36 bytes on 32-bit architectures,
522 * 40 bytes on 64-bit machines, if this grows please adjust
523 * skbuff.h:skbuff->cb[xxx] size appropriately.
527 struct inet_skb_parm h4;
528 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
529 struct inet6_skb_parm h6;
531 } header; /* For incoming frames */
532 __u32 seq; /* Starting sequence number */
533 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
534 __u32 when; /* used to compute rtt's */
535 __u8 flags; /* TCP header flags. */
537 /* NOTE: These must match up to the flags byte in a
540 #define TCPCB_FLAG_FIN 0x01
541 #define TCPCB_FLAG_SYN 0x02
542 #define TCPCB_FLAG_RST 0x04
543 #define TCPCB_FLAG_PSH 0x08
544 #define TCPCB_FLAG_ACK 0x10
545 #define TCPCB_FLAG_URG 0x20
546 #define TCPCB_FLAG_ECE 0x40
547 #define TCPCB_FLAG_CWR 0x80
549 __u8 sacked; /* State flags for SACK/FACK. */
550 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
551 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
552 #define TCPCB_LOST 0x04 /* SKB is lost */
553 #define TCPCB_TAGBITS 0x07 /* All tag bits */
555 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
556 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
558 #define TCPCB_URG 0x20 /* Urgent pointer advanced here */
560 #define TCPCB_AT_TAIL (TCPCB_URG)
562 __u16 urg_ptr; /* Valid w/URG flags is set. */
563 __u32 ack_seq; /* Sequence number ACK'd */
566 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
568 #include <net/tcp_ecn.h>
570 /* Due to TSO, an SKB can be composed of multiple actual
571 * packets. To keep these tracked properly, we use this.
573 static inline int tcp_skb_pcount(const struct sk_buff *skb)
575 return skb_shinfo(skb)->gso_segs;
578 /* This is valid iff tcp_skb_pcount() > 1. */
579 static inline int tcp_skb_mss(const struct sk_buff *skb)
581 return skb_shinfo(skb)->gso_size;
584 static inline void tcp_dec_pcount_approx(__u32 *count,
585 const struct sk_buff *skb)
588 *count -= tcp_skb_pcount(skb);
594 static inline void tcp_packets_out_inc(struct sock *sk,
596 const struct sk_buff *skb)
598 int orig = tp->packets_out;
600 tp->packets_out += tcp_skb_pcount(skb);
602 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
603 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
606 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
607 const struct sk_buff *skb)
609 tp->packets_out -= tcp_skb_pcount(skb);
612 /* Events passed to congestion control interface */
614 CA_EVENT_TX_START, /* first transmit when no packets in flight */
615 CA_EVENT_CWND_RESTART, /* congestion window restart */
616 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
617 CA_EVENT_FRTO, /* fast recovery timeout */
618 CA_EVENT_LOSS, /* loss timeout */
619 CA_EVENT_FAST_ACK, /* in sequence ack */
620 CA_EVENT_SLOW_ACK, /* other ack */
624 * Interface for adding new TCP congestion control handlers
626 #define TCP_CA_NAME_MAX 16
627 struct tcp_congestion_ops {
628 struct list_head list;
630 /* initialize private data (optional) */
631 void (*init)(struct sock *sk);
632 /* cleanup private data (optional) */
633 void (*release)(struct sock *sk);
635 /* return slow start threshold (required) */
636 u32 (*ssthresh)(struct sock *sk);
637 /* lower bound for congestion window (optional) */
638 u32 (*min_cwnd)(const struct sock *sk);
639 /* do new cwnd calculation (required) */
640 void (*cong_avoid)(struct sock *sk, u32 ack,
641 u32 rtt, u32 in_flight, int good_ack);
642 /* round trip time sample per acked packet (optional) */
643 void (*rtt_sample)(struct sock *sk, u32 usrtt);
644 /* call before changing ca_state (optional) */
645 void (*set_state)(struct sock *sk, u8 new_state);
646 /* call when cwnd event occurs (optional) */
647 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
648 /* new value of cwnd after loss (optional) */
649 u32 (*undo_cwnd)(struct sock *sk);
650 /* hook for packet ack accounting (optional) */
651 void (*pkts_acked)(struct sock *sk, u32 num_acked);
652 /* get info for inet_diag (optional) */
653 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
655 char name[TCP_CA_NAME_MAX];
656 struct module *owner;
659 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
660 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
662 extern void tcp_init_congestion_control(struct sock *sk);
663 extern void tcp_cleanup_congestion_control(struct sock *sk);
664 extern int tcp_set_default_congestion_control(const char *name);
665 extern void tcp_get_default_congestion_control(char *name);
666 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
667 extern void tcp_slow_start(struct tcp_sock *tp);
669 extern struct tcp_congestion_ops tcp_init_congestion_ops;
670 extern u32 tcp_reno_ssthresh(struct sock *sk);
671 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
672 u32 rtt, u32 in_flight, int flag);
673 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
674 extern struct tcp_congestion_ops tcp_reno;
676 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
678 struct inet_connection_sock *icsk = inet_csk(sk);
680 if (icsk->icsk_ca_ops->set_state)
681 icsk->icsk_ca_ops->set_state(sk, ca_state);
682 icsk->icsk_ca_state = ca_state;
685 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
687 const struct inet_connection_sock *icsk = inet_csk(sk);
689 if (icsk->icsk_ca_ops->cwnd_event)
690 icsk->icsk_ca_ops->cwnd_event(sk, event);
693 /* This determines how many packets are "in the network" to the best
694 * of our knowledge. In many cases it is conservative, but where
695 * detailed information is available from the receiver (via SACK
696 * blocks etc.) we can make more aggressive calculations.
698 * Use this for decisions involving congestion control, use just
699 * tp->packets_out to determine if the send queue is empty or not.
701 * Read this equation as:
703 * "Packets sent once on transmission queue" MINUS
704 * "Packets left network, but not honestly ACKed yet" PLUS
705 * "Packets fast retransmitted"
707 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
709 return (tp->packets_out - tp->left_out + tp->retrans_out);
712 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
713 * The exception is rate halving phase, when cwnd is decreasing towards
716 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
718 const struct tcp_sock *tp = tcp_sk(sk);
719 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
720 return tp->snd_ssthresh;
722 return max(tp->snd_ssthresh,
723 ((tp->snd_cwnd >> 1) +
724 (tp->snd_cwnd >> 2)));
727 static inline void tcp_sync_left_out(struct tcp_sock *tp)
729 if (tp->rx_opt.sack_ok &&
730 (tp->sacked_out >= tp->packets_out - tp->lost_out))
731 tp->sacked_out = tp->packets_out - tp->lost_out;
732 tp->left_out = tp->sacked_out + tp->lost_out;
735 extern void tcp_enter_cwr(struct sock *sk);
736 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
738 /* Slow start with delack produces 3 packets of burst, so that
739 * it is safe "de facto".
741 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
746 /* RFC2861 Check whether we are limited by application or congestion window
747 * This is the inverse of cwnd check in tcp_tso_should_defer
749 static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
751 const struct tcp_sock *tp = tcp_sk(sk);
754 if (in_flight >= tp->snd_cwnd)
760 left = tp->snd_cwnd - in_flight;
761 if (sysctl_tcp_tso_win_divisor)
762 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
764 return left <= tcp_max_burst(tp);
767 static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
768 const struct sk_buff *skb)
771 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
774 static inline void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
776 const struct inet_connection_sock *icsk = inet_csk(sk);
777 if (!tp->packets_out && !icsk->icsk_pending)
778 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
779 icsk->icsk_rto, TCP_RTO_MAX);
782 static inline void tcp_push_pending_frames(struct sock *sk,
785 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
788 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
793 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
799 * Calculate(/check) TCP checksum
801 static inline u16 tcp_v4_check(struct tcphdr *th, int len,
802 unsigned long saddr, unsigned long daddr,
805 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
808 static inline int __tcp_checksum_complete(struct sk_buff *skb)
810 return __skb_checksum_complete(skb);
813 static inline int tcp_checksum_complete(struct sk_buff *skb)
815 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
816 __tcp_checksum_complete(skb);
819 /* Prequeue for VJ style copy to user, combined with checksumming. */
821 static inline void tcp_prequeue_init(struct tcp_sock *tp)
823 tp->ucopy.task = NULL;
825 tp->ucopy.memory = 0;
826 skb_queue_head_init(&tp->ucopy.prequeue);
827 #ifdef CONFIG_NET_DMA
828 tp->ucopy.dma_chan = NULL;
829 tp->ucopy.wakeup = 0;
830 tp->ucopy.pinned_list = NULL;
831 tp->ucopy.dma_cookie = 0;
835 /* Packet is added to VJ-style prequeue for processing in process
836 * context, if a reader task is waiting. Apparently, this exciting
837 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
838 * failed somewhere. Latency? Burstiness? Well, at least now we will
839 * see, why it failed. 8)8) --ANK
841 * NOTE: is this not too big to inline?
843 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
845 struct tcp_sock *tp = tcp_sk(sk);
847 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
848 __skb_queue_tail(&tp->ucopy.prequeue, skb);
849 tp->ucopy.memory += skb->truesize;
850 if (tp->ucopy.memory > sk->sk_rcvbuf) {
851 struct sk_buff *skb1;
853 BUG_ON(sock_owned_by_user(sk));
855 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
856 sk->sk_backlog_rcv(sk, skb1);
857 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
860 tp->ucopy.memory = 0;
861 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
862 wake_up_interruptible(sk->sk_sleep);
863 if (!inet_csk_ack_scheduled(sk))
864 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
865 (3 * TCP_RTO_MIN) / 4,
877 static const char *statename[]={
878 "Unused","Established","Syn Sent","Syn Recv",
879 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
880 "Close Wait","Last ACK","Listen","Closing"
884 static inline void tcp_set_state(struct sock *sk, int state)
886 int oldstate = sk->sk_state;
889 case TCP_ESTABLISHED:
890 if (oldstate != TCP_ESTABLISHED)
891 TCP_INC_STATS(TCP_MIB_CURRESTAB);
895 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
896 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
898 sk->sk_prot->unhash(sk);
899 if (inet_csk(sk)->icsk_bind_hash &&
900 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
901 inet_put_port(&tcp_hashinfo, sk);
904 if (oldstate==TCP_ESTABLISHED)
905 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
908 /* Change state AFTER socket is unhashed to avoid closed
909 * socket sitting in hash tables.
911 sk->sk_state = state;
914 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
918 static inline void tcp_done(struct sock *sk)
920 if(sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
921 TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
923 tcp_set_state(sk, TCP_CLOSE);
924 tcp_clear_xmit_timers(sk);
926 sk->sk_shutdown = SHUTDOWN_MASK;
928 if (!sock_flag(sk, SOCK_DEAD))
929 sk->sk_state_change(sk);
931 inet_csk_destroy_sock(sk);
934 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
937 rx_opt->eff_sacks = 0;
938 rx_opt->num_sacks = 0;
941 /* Determine a window scaling and initial window to offer. */
942 extern void tcp_select_initial_window(int __space, __u32 mss,
943 __u32 *rcv_wnd, __u32 *window_clamp,
944 int wscale_ok, __u8 *rcv_wscale);
946 static inline int tcp_win_from_space(int space)
948 return sysctl_tcp_adv_win_scale<=0 ?
949 (space>>(-sysctl_tcp_adv_win_scale)) :
950 space - (space>>sysctl_tcp_adv_win_scale);
953 /* Note: caller must be prepared to deal with negative returns */
954 static inline int tcp_space(const struct sock *sk)
956 return tcp_win_from_space(sk->sk_rcvbuf -
957 atomic_read(&sk->sk_rmem_alloc));
960 static inline int tcp_full_space(const struct sock *sk)
962 return tcp_win_from_space(sk->sk_rcvbuf);
965 static inline void tcp_openreq_init(struct request_sock *req,
966 struct tcp_options_received *rx_opt,
969 struct inet_request_sock *ireq = inet_rsk(req);
971 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
972 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
973 req->mss = rx_opt->mss_clamp;
974 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
975 ireq->tstamp_ok = rx_opt->tstamp_ok;
976 ireq->sack_ok = rx_opt->sack_ok;
977 ireq->snd_wscale = rx_opt->snd_wscale;
978 ireq->wscale_ok = rx_opt->wscale_ok;
981 ireq->rmt_port = skb->h.th->source;
984 extern void tcp_enter_memory_pressure(void);
986 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
988 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
991 static inline int keepalive_time_when(const struct tcp_sock *tp)
993 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
996 static inline int tcp_fin_time(const struct sock *sk)
998 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
999 const int rto = inet_csk(sk)->icsk_rto;
1001 if (fin_timeout < (rto << 2) - (rto >> 1))
1002 fin_timeout = (rto << 2) - (rto >> 1);
1007 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1009 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1011 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1014 /* RST segments are not recommended to carry timestamp,
1015 and, if they do, it is recommended to ignore PAWS because
1016 "their cleanup function should take precedence over timestamps."
1017 Certainly, it is mistake. It is necessary to understand the reasons
1018 of this constraint to relax it: if peer reboots, clock may go
1019 out-of-sync and half-open connections will not be reset.
1020 Actually, the problem would be not existing if all
1021 the implementations followed draft about maintaining clock
1022 via reboots. Linux-2.2 DOES NOT!
1024 However, we can relax time bounds for RST segments to MSL.
1026 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1031 #define TCP_CHECK_TIMER(sk) do { } while (0)
1033 static inline int tcp_use_frto(const struct sock *sk)
1035 const struct tcp_sock *tp = tcp_sk(sk);
1037 /* F-RTO must be activated in sysctl and there must be some
1038 * unsent new data, and the advertised window should allow
1041 return (sysctl_tcp_frto && sk->sk_send_head &&
1042 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1043 tp->snd_una + tp->snd_wnd));
1046 static inline void tcp_mib_init(void)
1049 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1050 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1051 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1052 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1056 static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1057 tp->lost_skb_hint = NULL;
1058 tp->scoreboard_skb_hint = NULL;
1059 tp->retransmit_skb_hint = NULL;
1060 tp->forward_skb_hint = NULL;
1061 tp->fastpath_skb_hint = NULL;
1065 enum tcp_seq_states {
1066 TCP_SEQ_STATE_LISTENING,
1067 TCP_SEQ_STATE_OPENREQ,
1068 TCP_SEQ_STATE_ESTABLISHED,
1069 TCP_SEQ_STATE_TIME_WAIT,
1072 struct tcp_seq_afinfo {
1073 struct module *owner;
1076 int (*seq_show) (struct seq_file *m, void *v);
1077 struct file_operations *seq_fops;
1080 struct tcp_iter_state {
1082 enum tcp_seq_states state;
1083 struct sock *syn_wait_sk;
1084 int bucket, sbucket, num, uid;
1085 struct seq_operations seq_ops;
1088 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1089 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1091 extern struct request_sock_ops tcp_request_sock_ops;
1093 extern int tcp_v4_destroy_sock(struct sock *sk);
1095 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1096 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1098 #ifdef CONFIG_PROC_FS
1099 extern int tcp4_proc_init(void);
1100 extern void tcp4_proc_exit(void);
1103 extern void tcp_v4_init(struct net_proto_family *ops);
1104 extern void tcp_init(void);