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>
31 #include <linux/dmaengine.h>
33 #include <net/inet_connection_sock.h>
34 #include <net/inet_timewait_sock.h>
35 #include <net/inet_hashtables.h>
36 #include <net/checksum.h>
37 #include <net/request_sock.h>
41 #include <net/tcp_states.h>
43 #include <linux/seq_file.h>
45 extern struct inet_hashinfo tcp_hashinfo;
47 extern atomic_t tcp_orphan_count;
48 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
50 #define MAX_TCP_HEADER (128 + MAX_HEADER)
53 * Never offer a window over 32767 without using window scaling. Some
54 * poor stacks do signed 16bit maths!
56 #define MAX_TCP_WINDOW 32767U
58 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
59 #define TCP_MIN_MSS 88U
61 /* Minimal RCV_MSS. */
62 #define TCP_MIN_RCVMSS 536U
64 /* The least MTU to use for probing */
65 #define TCP_BASE_MSS 512
67 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
68 #define TCP_FASTRETRANS_THRESH 3
70 /* Maximal reordering. */
71 #define TCP_MAX_REORDERING 127
73 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
74 #define TCP_MAX_QUICKACKS 16U
77 #define TCP_URG_VALID 0x0100
78 #define TCP_URG_NOTYET 0x0200
79 #define TCP_URG_READ 0x0400
81 #define TCP_RETR1 3 /*
82 * This is how many retries it does before it
83 * tries to figure out if the gateway is
84 * down. Minimal RFC value is 3; it corresponds
85 * to ~3sec-8min depending on RTO.
88 #define TCP_RETR2 15 /*
89 * This should take at least
90 * 90 minutes to time out.
91 * RFC1122 says that the limit is 100 sec.
92 * 15 is ~13-30min depending on RTO.
95 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
96 * connection: ~180sec is RFC minimum */
98 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
99 * connection: ~180sec is RFC minimum */
102 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
103 * socket. 7 is ~50sec-16min.
107 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
108 * state, about 60 seconds */
109 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
110 /* BSD style FIN_WAIT2 deadlock breaker.
111 * It used to be 3min, new value is 60sec,
112 * to combine FIN-WAIT-2 timeout with
116 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
118 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
119 #define TCP_ATO_MIN ((unsigned)(HZ/25))
121 #define TCP_DELACK_MIN 4U
122 #define TCP_ATO_MIN 4U
124 #define TCP_RTO_MAX ((unsigned)(120*HZ))
125 #define TCP_RTO_MIN ((unsigned)(HZ/5))
126 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
128 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
129 * for local resources.
132 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
133 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
134 #define TCP_KEEPALIVE_INTVL (75*HZ)
136 #define MAX_TCP_KEEPIDLE 32767
137 #define MAX_TCP_KEEPINTVL 32767
138 #define MAX_TCP_KEEPCNT 127
139 #define MAX_TCP_SYNCNT 127
141 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
142 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
144 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
145 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
146 * after this time. It should be equal
147 * (or greater than) TCP_TIMEWAIT_LEN
148 * to provide reliability equal to one
149 * provided by timewait state.
151 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
152 * timestamps. It must be less than
153 * minimal timewait lifetime.
159 #define TCPOPT_NOP 1 /* Padding */
160 #define TCPOPT_EOL 0 /* End of options */
161 #define TCPOPT_MSS 2 /* Segment size negotiating */
162 #define TCPOPT_WINDOW 3 /* Window scaling */
163 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
164 #define TCPOPT_SACK 5 /* SACK Block */
165 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
171 #define TCPOLEN_MSS 4
172 #define TCPOLEN_WINDOW 3
173 #define TCPOLEN_SACK_PERM 2
174 #define TCPOLEN_TIMESTAMP 10
176 /* But this is what stacks really send out. */
177 #define TCPOLEN_TSTAMP_ALIGNED 12
178 #define TCPOLEN_WSCALE_ALIGNED 4
179 #define TCPOLEN_SACKPERM_ALIGNED 4
180 #define TCPOLEN_SACK_BASE 2
181 #define TCPOLEN_SACK_BASE_ALIGNED 4
182 #define TCPOLEN_SACK_PERBLOCK 8
184 /* Flags in tp->nonagle */
185 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
186 #define TCP_NAGLE_CORK 2 /* Socket is corked */
187 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
189 extern struct inet_timewait_death_row tcp_death_row;
191 /* sysctl variables for tcp */
192 extern int sysctl_tcp_timestamps;
193 extern int sysctl_tcp_window_scaling;
194 extern int sysctl_tcp_sack;
195 extern int sysctl_tcp_fin_timeout;
196 extern int sysctl_tcp_keepalive_time;
197 extern int sysctl_tcp_keepalive_probes;
198 extern int sysctl_tcp_keepalive_intvl;
199 extern int sysctl_tcp_syn_retries;
200 extern int sysctl_tcp_synack_retries;
201 extern int sysctl_tcp_retries1;
202 extern int sysctl_tcp_retries2;
203 extern int sysctl_tcp_orphan_retries;
204 extern int sysctl_tcp_syncookies;
205 extern int sysctl_tcp_retrans_collapse;
206 extern int sysctl_tcp_stdurg;
207 extern int sysctl_tcp_rfc1337;
208 extern int sysctl_tcp_abort_on_overflow;
209 extern int sysctl_tcp_max_orphans;
210 extern int sysctl_tcp_fack;
211 extern int sysctl_tcp_reordering;
212 extern int sysctl_tcp_ecn;
213 extern int sysctl_tcp_dsack;
214 extern int sysctl_tcp_mem[3];
215 extern int sysctl_tcp_wmem[3];
216 extern int sysctl_tcp_rmem[3];
217 extern int sysctl_tcp_app_win;
218 extern int sysctl_tcp_adv_win_scale;
219 extern int sysctl_tcp_tw_reuse;
220 extern int sysctl_tcp_frto;
221 extern int sysctl_tcp_low_latency;
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;
230 extern atomic_t tcp_memory_allocated;
231 extern atomic_t tcp_sockets_allocated;
232 extern int tcp_memory_pressure;
235 * The next routines deal with comparing 32 bit unsigned ints
236 * and worry about wraparound (automatic with unsigned arithmetic).
239 static inline int before(__u32 seq1, __u32 seq2)
241 return (__s32)(seq1-seq2) < 0;
244 static inline int after(__u32 seq1, __u32 seq2)
246 return (__s32)(seq2-seq1) < 0;
250 /* is s2<=s1<=s3 ? */
251 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
253 return seq3 - seq2 >= seq1 - seq2;
257 extern struct proto tcp_prot;
259 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
260 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
261 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
262 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
263 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
264 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
265 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
267 extern void tcp_v4_err(struct sk_buff *skb, u32);
269 extern void tcp_shutdown (struct sock *sk, int how);
271 extern int tcp_v4_rcv(struct sk_buff *skb);
273 extern int tcp_v4_remember_stamp(struct sock *sk);
275 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
277 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
278 struct msghdr *msg, size_t size);
279 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
281 extern int tcp_ioctl(struct sock *sk,
285 extern int tcp_rcv_state_process(struct sock *sk,
290 extern int tcp_rcv_established(struct sock *sk,
295 extern void tcp_rcv_space_adjust(struct sock *sk);
297 extern int tcp_twsk_unique(struct sock *sk,
298 struct sock *sktw, void *twp);
300 static inline void tcp_dec_quickack_mode(struct sock *sk,
301 const unsigned int pkts)
303 struct inet_connection_sock *icsk = inet_csk(sk);
305 if (icsk->icsk_ack.quick) {
306 if (pkts >= icsk->icsk_ack.quick) {
307 icsk->icsk_ack.quick = 0;
308 /* Leaving quickack mode we deflate ATO. */
309 icsk->icsk_ack.ato = TCP_ATO_MIN;
311 icsk->icsk_ack.quick -= pkts;
315 extern void tcp_enter_quickack_mode(struct sock *sk);
317 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
319 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
331 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
333 const struct tcphdr *th);
335 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
336 struct request_sock *req,
337 struct request_sock **prev);
338 extern int tcp_child_process(struct sock *parent,
340 struct sk_buff *skb);
341 extern void tcp_enter_frto(struct sock *sk);
342 extern void tcp_enter_loss(struct sock *sk, int how);
343 extern void tcp_clear_retrans(struct tcp_sock *tp);
344 extern void tcp_update_metrics(struct sock *sk);
346 extern void tcp_close(struct sock *sk,
348 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
350 extern int tcp_getsockopt(struct sock *sk, int level,
354 extern int tcp_setsockopt(struct sock *sk, int level,
355 int optname, char __user *optval,
357 extern int compat_tcp_getsockopt(struct sock *sk,
358 int level, int optname,
359 char __user *optval, int __user *optlen);
360 extern int compat_tcp_setsockopt(struct sock *sk,
361 int level, int optname,
362 char __user *optval, int optlen);
363 extern void tcp_set_keepalive(struct sock *sk, int val);
364 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
366 size_t len, int nonblock,
367 int flags, int *addr_len);
369 extern void tcp_parse_options(struct sk_buff *skb,
370 struct tcp_options_received *opt_rx,
374 * TCP v4 functions exported for the inet6 API
377 extern void tcp_v4_send_check(struct sock *sk, int len,
378 struct sk_buff *skb);
380 extern int tcp_v4_conn_request(struct sock *sk,
381 struct sk_buff *skb);
383 extern struct sock * tcp_create_openreq_child(struct sock *sk,
384 struct request_sock *req,
385 struct sk_buff *skb);
387 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
389 struct request_sock *req,
390 struct dst_entry *dst);
392 extern int tcp_v4_do_rcv(struct sock *sk,
393 struct sk_buff *skb);
395 extern int tcp_v4_connect(struct sock *sk,
396 struct sockaddr *uaddr,
399 extern int tcp_connect(struct sock *sk);
401 extern struct sk_buff * tcp_make_synack(struct sock *sk,
402 struct dst_entry *dst,
403 struct request_sock *req);
405 extern int tcp_disconnect(struct sock *sk, int flags);
407 extern void tcp_unhash(struct sock *sk);
409 /* From syncookies.c */
410 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
411 struct ip_options *opt);
412 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
417 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
418 unsigned int cur_mss, int nonagle);
419 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
420 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
421 extern void tcp_xmit_retransmit_queue(struct sock *);
422 extern void tcp_simple_retransmit(struct sock *);
423 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
424 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
426 extern void tcp_send_probe0(struct sock *);
427 extern void tcp_send_partial(struct sock *);
428 extern int tcp_write_wakeup(struct sock *);
429 extern void tcp_send_fin(struct sock *sk);
430 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
431 extern int tcp_send_synack(struct sock *);
432 extern void tcp_push_one(struct sock *, unsigned int mss_now);
433 extern void tcp_send_ack(struct sock *sk);
434 extern void tcp_send_delayed_ack(struct sock *sk);
437 extern void tcp_cwnd_application_limited(struct sock *sk);
440 extern void tcp_init_xmit_timers(struct sock *);
441 static inline void tcp_clear_xmit_timers(struct sock *sk)
443 inet_csk_clear_xmit_timers(sk);
446 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
447 extern unsigned int tcp_current_mss(struct sock *sk, int large);
450 extern void tcp_get_info(struct sock *, struct tcp_info *);
452 /* Read 'sendfile()'-style from a TCP socket */
453 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
454 unsigned int, size_t);
455 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
456 sk_read_actor_t recv_actor);
458 extern void tcp_initialize_rcv_mss(struct sock *sk);
460 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
461 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
462 extern void tcp_mtup_init(struct sock *sk);
464 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
466 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
467 ntohl(TCP_FLAG_ACK) |
471 static inline void tcp_fast_path_on(struct tcp_sock *tp)
473 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
476 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
478 if (skb_queue_empty(&tp->out_of_order_queue) &&
480 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
482 tcp_fast_path_on(tp);
485 /* Compute the actual receive window we are currently advertising.
486 * Rcv_nxt can be after the window if our peer push more data
487 * than the offered window.
489 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
491 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
498 /* Choose a new window, without checks for shrinking, and without
499 * scaling applied to the result. The caller does these things
500 * if necessary. This is a "raw" window selection.
502 extern u32 __tcp_select_window(struct sock *sk);
504 /* TCP timestamps are only 32-bits, this causes a slight
505 * complication on 64-bit systems since we store a snapshot
506 * of jiffies in the buffer control blocks below. We decided
507 * to use only the low 32-bits of jiffies and hide the ugly
508 * casts with the following macro.
510 #define tcp_time_stamp ((__u32)(jiffies))
512 /* This is what the send packet queuing engine uses to pass
513 * TCP per-packet control information to the transmission
514 * code. We also store the host-order sequence numbers in
515 * here too. This is 36 bytes on 32-bit architectures,
516 * 40 bytes on 64-bit machines, if this grows please adjust
517 * skbuff.h:skbuff->cb[xxx] size appropriately.
521 struct inet_skb_parm h4;
522 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
523 struct inet6_skb_parm h6;
525 } header; /* For incoming frames */
526 __u32 seq; /* Starting sequence number */
527 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
528 __u32 when; /* used to compute rtt's */
529 __u8 flags; /* TCP header flags. */
531 /* NOTE: These must match up to the flags byte in a
534 #define TCPCB_FLAG_FIN 0x01
535 #define TCPCB_FLAG_SYN 0x02
536 #define TCPCB_FLAG_RST 0x04
537 #define TCPCB_FLAG_PSH 0x08
538 #define TCPCB_FLAG_ACK 0x10
539 #define TCPCB_FLAG_URG 0x20
540 #define TCPCB_FLAG_ECE 0x40
541 #define TCPCB_FLAG_CWR 0x80
543 __u8 sacked; /* State flags for SACK/FACK. */
544 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
545 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
546 #define TCPCB_LOST 0x04 /* SKB is lost */
547 #define TCPCB_TAGBITS 0x07 /* All tag bits */
549 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
550 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
552 #define TCPCB_URG 0x20 /* Urgent pointer advanced here */
554 #define TCPCB_AT_TAIL (TCPCB_URG)
556 __u16 urg_ptr; /* Valid w/URG flags is set. */
557 __u32 ack_seq; /* Sequence number ACK'd */
560 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
562 #include <net/tcp_ecn.h>
564 /* Due to TSO, an SKB can be composed of multiple actual
565 * packets. To keep these tracked properly, we use this.
567 static inline int tcp_skb_pcount(const struct sk_buff *skb)
569 return skb_shinfo(skb)->tso_segs;
572 /* This is valid iff tcp_skb_pcount() > 1. */
573 static inline int tcp_skb_mss(const struct sk_buff *skb)
575 return skb_shinfo(skb)->tso_size;
578 static inline void tcp_dec_pcount_approx(__u32 *count,
579 const struct sk_buff *skb)
582 *count -= tcp_skb_pcount(skb);
588 static inline void tcp_packets_out_inc(struct sock *sk,
590 const struct sk_buff *skb)
592 int orig = tp->packets_out;
594 tp->packets_out += tcp_skb_pcount(skb);
596 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
597 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
600 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
601 const struct sk_buff *skb)
603 tp->packets_out -= tcp_skb_pcount(skb);
606 /* Events passed to congestion control interface */
608 CA_EVENT_TX_START, /* first transmit when no packets in flight */
609 CA_EVENT_CWND_RESTART, /* congestion window restart */
610 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
611 CA_EVENT_FRTO, /* fast recovery timeout */
612 CA_EVENT_LOSS, /* loss timeout */
613 CA_EVENT_FAST_ACK, /* in sequence ack */
614 CA_EVENT_SLOW_ACK, /* other ack */
618 * Interface for adding new TCP congestion control handlers
620 #define TCP_CA_NAME_MAX 16
621 struct tcp_congestion_ops {
622 struct list_head list;
624 /* initialize private data (optional) */
625 void (*init)(struct sock *sk);
626 /* cleanup private data (optional) */
627 void (*release)(struct sock *sk);
629 /* return slow start threshold (required) */
630 u32 (*ssthresh)(struct sock *sk);
631 /* lower bound for congestion window (optional) */
632 u32 (*min_cwnd)(struct sock *sk);
633 /* do new cwnd calculation (required) */
634 void (*cong_avoid)(struct sock *sk, u32 ack,
635 u32 rtt, u32 in_flight, int good_ack);
636 /* round trip time sample per acked packet (optional) */
637 void (*rtt_sample)(struct sock *sk, u32 usrtt);
638 /* call before changing ca_state (optional) */
639 void (*set_state)(struct sock *sk, u8 new_state);
640 /* call when cwnd event occurs (optional) */
641 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
642 /* new value of cwnd after loss (optional) */
643 u32 (*undo_cwnd)(struct sock *sk);
644 /* hook for packet ack accounting (optional) */
645 void (*pkts_acked)(struct sock *sk, u32 num_acked);
646 /* get info for inet_diag (optional) */
647 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
649 char name[TCP_CA_NAME_MAX];
650 struct module *owner;
653 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
654 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
656 extern void tcp_init_congestion_control(struct sock *sk);
657 extern void tcp_cleanup_congestion_control(struct sock *sk);
658 extern int tcp_set_default_congestion_control(const char *name);
659 extern void tcp_get_default_congestion_control(char *name);
660 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
661 extern void tcp_slow_start(struct tcp_sock *tp);
663 extern struct tcp_congestion_ops tcp_init_congestion_ops;
664 extern u32 tcp_reno_ssthresh(struct sock *sk);
665 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
666 u32 rtt, u32 in_flight, int flag);
667 extern u32 tcp_reno_min_cwnd(struct sock *sk);
668 extern struct tcp_congestion_ops tcp_reno;
670 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
672 struct inet_connection_sock *icsk = inet_csk(sk);
674 if (icsk->icsk_ca_ops->set_state)
675 icsk->icsk_ca_ops->set_state(sk, ca_state);
676 icsk->icsk_ca_state = ca_state;
679 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
681 const struct inet_connection_sock *icsk = inet_csk(sk);
683 if (icsk->icsk_ca_ops->cwnd_event)
684 icsk->icsk_ca_ops->cwnd_event(sk, event);
687 /* This determines how many packets are "in the network" to the best
688 * of our knowledge. In many cases it is conservative, but where
689 * detailed information is available from the receiver (via SACK
690 * blocks etc.) we can make more aggressive calculations.
692 * Use this for decisions involving congestion control, use just
693 * tp->packets_out to determine if the send queue is empty or not.
695 * Read this equation as:
697 * "Packets sent once on transmission queue" MINUS
698 * "Packets left network, but not honestly ACKed yet" PLUS
699 * "Packets fast retransmitted"
701 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
703 return (tp->packets_out - tp->left_out + tp->retrans_out);
706 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
707 * The exception is rate halving phase, when cwnd is decreasing towards
710 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
712 const struct tcp_sock *tp = tcp_sk(sk);
713 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
714 return tp->snd_ssthresh;
716 return max(tp->snd_ssthresh,
717 ((tp->snd_cwnd >> 1) +
718 (tp->snd_cwnd >> 2)));
721 static inline void tcp_sync_left_out(struct tcp_sock *tp)
723 if (tp->rx_opt.sack_ok &&
724 (tp->sacked_out >= tp->packets_out - tp->lost_out))
725 tp->sacked_out = tp->packets_out - tp->lost_out;
726 tp->left_out = tp->sacked_out + tp->lost_out;
729 extern void tcp_enter_cwr(struct sock *sk);
730 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
732 /* Slow start with delack produces 3 packets of burst, so that
733 * it is safe "de facto".
735 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
740 /* RFC2861 Check whether we are limited by application or congestion window
741 * This is the inverse of cwnd check in tcp_tso_should_defer
743 static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
745 const struct tcp_sock *tp = tcp_sk(sk);
748 if (in_flight >= tp->snd_cwnd)
751 if (!(sk->sk_route_caps & NETIF_F_TSO))
754 left = tp->snd_cwnd - in_flight;
755 if (sysctl_tcp_tso_win_divisor)
756 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
758 return left <= tcp_max_burst(tp);
761 static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
762 const struct sk_buff *skb)
765 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
768 static inline void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
770 const struct inet_connection_sock *icsk = inet_csk(sk);
771 if (!tp->packets_out && !icsk->icsk_pending)
772 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
773 icsk->icsk_rto, TCP_RTO_MAX);
776 static inline void tcp_push_pending_frames(struct sock *sk,
779 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
782 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
787 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
793 * Calculate(/check) TCP checksum
795 static inline u16 tcp_v4_check(struct tcphdr *th, int len,
796 unsigned long saddr, unsigned long daddr,
799 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
802 static inline int __tcp_checksum_complete(struct sk_buff *skb)
804 return __skb_checksum_complete(skb);
807 static inline int tcp_checksum_complete(struct sk_buff *skb)
809 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
810 __tcp_checksum_complete(skb);
813 /* Prequeue for VJ style copy to user, combined with checksumming. */
815 static inline void tcp_prequeue_init(struct tcp_sock *tp)
817 tp->ucopy.task = NULL;
819 tp->ucopy.memory = 0;
820 skb_queue_head_init(&tp->ucopy.prequeue);
821 #ifdef CONFIG_NET_DMA
822 tp->ucopy.dma_chan = NULL;
823 tp->ucopy.wakeup = 0;
824 tp->ucopy.pinned_list = NULL;
825 tp->ucopy.dma_cookie = 0;
829 /* Packet is added to VJ-style prequeue for processing in process
830 * context, if a reader task is waiting. Apparently, this exciting
831 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
832 * failed somewhere. Latency? Burstiness? Well, at least now we will
833 * see, why it failed. 8)8) --ANK
835 * NOTE: is this not too big to inline?
837 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
839 struct tcp_sock *tp = tcp_sk(sk);
841 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
842 __skb_queue_tail(&tp->ucopy.prequeue, skb);
843 tp->ucopy.memory += skb->truesize;
844 if (tp->ucopy.memory > sk->sk_rcvbuf) {
845 struct sk_buff *skb1;
847 BUG_ON(sock_owned_by_user(sk));
849 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
850 sk->sk_backlog_rcv(sk, skb1);
851 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
854 tp->ucopy.memory = 0;
855 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
856 wake_up_interruptible(sk->sk_sleep);
857 if (!inet_csk_ack_scheduled(sk))
858 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
859 (3 * TCP_RTO_MIN) / 4,
871 static const char *statename[]={
872 "Unused","Established","Syn Sent","Syn Recv",
873 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
874 "Close Wait","Last ACK","Listen","Closing"
878 static inline void tcp_set_state(struct sock *sk, int state)
880 int oldstate = sk->sk_state;
883 case TCP_ESTABLISHED:
884 if (oldstate != TCP_ESTABLISHED)
885 TCP_INC_STATS(TCP_MIB_CURRESTAB);
889 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
890 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
892 sk->sk_prot->unhash(sk);
893 if (inet_csk(sk)->icsk_bind_hash &&
894 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
895 inet_put_port(&tcp_hashinfo, sk);
898 if (oldstate==TCP_ESTABLISHED)
899 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
902 /* Change state AFTER socket is unhashed to avoid closed
903 * socket sitting in hash tables.
905 sk->sk_state = state;
908 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
912 static inline void tcp_done(struct sock *sk)
914 tcp_set_state(sk, TCP_CLOSE);
915 tcp_clear_xmit_timers(sk);
917 sk->sk_shutdown = SHUTDOWN_MASK;
919 if (!sock_flag(sk, SOCK_DEAD))
920 sk->sk_state_change(sk);
922 inet_csk_destroy_sock(sk);
925 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
928 rx_opt->eff_sacks = 0;
929 rx_opt->num_sacks = 0;
932 /* Determine a window scaling and initial window to offer. */
933 extern void tcp_select_initial_window(int __space, __u32 mss,
934 __u32 *rcv_wnd, __u32 *window_clamp,
935 int wscale_ok, __u8 *rcv_wscale);
937 static inline int tcp_win_from_space(int space)
939 return sysctl_tcp_adv_win_scale<=0 ?
940 (space>>(-sysctl_tcp_adv_win_scale)) :
941 space - (space>>sysctl_tcp_adv_win_scale);
944 /* Note: caller must be prepared to deal with negative returns */
945 static inline int tcp_space(const struct sock *sk)
947 return tcp_win_from_space(sk->sk_rcvbuf -
948 atomic_read(&sk->sk_rmem_alloc));
951 static inline int tcp_full_space(const struct sock *sk)
953 return tcp_win_from_space(sk->sk_rcvbuf);
956 static inline void tcp_openreq_init(struct request_sock *req,
957 struct tcp_options_received *rx_opt,
960 struct inet_request_sock *ireq = inet_rsk(req);
962 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
963 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
964 req->mss = rx_opt->mss_clamp;
965 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
966 ireq->tstamp_ok = rx_opt->tstamp_ok;
967 ireq->sack_ok = rx_opt->sack_ok;
968 ireq->snd_wscale = rx_opt->snd_wscale;
969 ireq->wscale_ok = rx_opt->wscale_ok;
972 ireq->rmt_port = skb->h.th->source;
975 extern void tcp_enter_memory_pressure(void);
977 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
979 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
982 static inline int keepalive_time_when(const struct tcp_sock *tp)
984 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
987 static inline int tcp_fin_time(const struct sock *sk)
989 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
990 const int rto = inet_csk(sk)->icsk_rto;
992 if (fin_timeout < (rto << 2) - (rto >> 1))
993 fin_timeout = (rto << 2) - (rto >> 1);
998 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1000 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1002 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1005 /* RST segments are not recommended to carry timestamp,
1006 and, if they do, it is recommended to ignore PAWS because
1007 "their cleanup function should take precedence over timestamps."
1008 Certainly, it is mistake. It is necessary to understand the reasons
1009 of this constraint to relax it: if peer reboots, clock may go
1010 out-of-sync and half-open connections will not be reset.
1011 Actually, the problem would be not existing if all
1012 the implementations followed draft about maintaining clock
1013 via reboots. Linux-2.2 DOES NOT!
1015 However, we can relax time bounds for RST segments to MSL.
1017 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1022 #define TCP_CHECK_TIMER(sk) do { } while (0)
1024 static inline int tcp_use_frto(const struct sock *sk)
1026 const struct tcp_sock *tp = tcp_sk(sk);
1028 /* F-RTO must be activated in sysctl and there must be some
1029 * unsent new data, and the advertised window should allow
1032 return (sysctl_tcp_frto && sk->sk_send_head &&
1033 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1034 tp->snd_una + tp->snd_wnd));
1037 static inline void tcp_mib_init(void)
1040 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1041 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1042 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1043 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1047 static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1048 tp->lost_skb_hint = NULL;
1049 tp->scoreboard_skb_hint = NULL;
1050 tp->retransmit_skb_hint = NULL;
1051 tp->forward_skb_hint = NULL;
1052 tp->fastpath_skb_hint = NULL;
1056 enum tcp_seq_states {
1057 TCP_SEQ_STATE_LISTENING,
1058 TCP_SEQ_STATE_OPENREQ,
1059 TCP_SEQ_STATE_ESTABLISHED,
1060 TCP_SEQ_STATE_TIME_WAIT,
1063 struct tcp_seq_afinfo {
1064 struct module *owner;
1067 int (*seq_show) (struct seq_file *m, void *v);
1068 struct file_operations *seq_fops;
1071 struct tcp_iter_state {
1073 enum tcp_seq_states state;
1074 struct sock *syn_wait_sk;
1075 int bucket, sbucket, num, uid;
1076 struct seq_operations seq_ops;
1079 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1080 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1082 extern struct request_sock_ops tcp_request_sock_ops;
1084 extern int tcp_v4_destroy_sock(struct sock *sk);
1086 #ifdef CONFIG_PROC_FS
1087 extern int tcp4_proc_init(void);
1088 extern void tcp4_proc_exit(void);
1091 extern void tcp_v4_init(struct net_proto_family *ops);
1092 extern void tcp_init(void);