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>
31 #include <linux/crypto.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 */
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 */
165 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
171 #define TCPOLEN_MSS 4
172 #define TCPOLEN_WINDOW 3
173 #define TCPOLEN_SACK_PERM 2
174 #define TCPOLEN_TIMESTAMP 10
175 #define TCPOLEN_MD5SIG 18
177 /* But this is what stacks really send out. */
178 #define TCPOLEN_TSTAMP_ALIGNED 12
179 #define TCPOLEN_WSCALE_ALIGNED 4
180 #define TCPOLEN_SACKPERM_ALIGNED 4
181 #define TCPOLEN_SACK_BASE 2
182 #define TCPOLEN_SACK_BASE_ALIGNED 4
183 #define TCPOLEN_SACK_PERBLOCK 8
184 #define TCPOLEN_MD5SIG_ALIGNED 20
186 /* Flags in tp->nonagle */
187 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
188 #define TCP_NAGLE_CORK 2 /* Socket is corked */
189 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
191 extern struct inet_timewait_death_row tcp_death_row;
193 /* sysctl variables for tcp */
194 extern int sysctl_tcp_timestamps;
195 extern int sysctl_tcp_window_scaling;
196 extern int sysctl_tcp_sack;
197 extern int sysctl_tcp_fin_timeout;
198 extern int sysctl_tcp_keepalive_time;
199 extern int sysctl_tcp_keepalive_probes;
200 extern int sysctl_tcp_keepalive_intvl;
201 extern int sysctl_tcp_syn_retries;
202 extern int sysctl_tcp_synack_retries;
203 extern int sysctl_tcp_retries1;
204 extern int sysctl_tcp_retries2;
205 extern int sysctl_tcp_orphan_retries;
206 extern int sysctl_tcp_syncookies;
207 extern int sysctl_tcp_retrans_collapse;
208 extern int sysctl_tcp_stdurg;
209 extern int sysctl_tcp_rfc1337;
210 extern int sysctl_tcp_abort_on_overflow;
211 extern int sysctl_tcp_max_orphans;
212 extern int sysctl_tcp_fack;
213 extern int sysctl_tcp_reordering;
214 extern int sysctl_tcp_ecn;
215 extern int sysctl_tcp_dsack;
216 extern int sysctl_tcp_mem[3];
217 extern int sysctl_tcp_wmem[3];
218 extern int sysctl_tcp_rmem[3];
219 extern int sysctl_tcp_app_win;
220 extern int sysctl_tcp_adv_win_scale;
221 extern int sysctl_tcp_tw_reuse;
222 extern int sysctl_tcp_frto;
223 extern int sysctl_tcp_low_latency;
224 extern int sysctl_tcp_dma_copybreak;
225 extern int sysctl_tcp_nometrics_save;
226 extern int sysctl_tcp_moderate_rcvbuf;
227 extern int sysctl_tcp_tso_win_divisor;
228 extern int sysctl_tcp_abc;
229 extern int sysctl_tcp_mtu_probing;
230 extern int sysctl_tcp_base_mss;
231 extern int sysctl_tcp_workaround_signed_windows;
232 extern int sysctl_tcp_slow_start_after_idle;
234 extern atomic_t tcp_memory_allocated;
235 extern atomic_t tcp_sockets_allocated;
236 extern int tcp_memory_pressure;
239 * The next routines deal with comparing 32 bit unsigned ints
240 * and worry about wraparound (automatic with unsigned arithmetic).
243 static inline int before(__u32 seq1, __u32 seq2)
245 return (__s32)(seq1-seq2) < 0;
247 #define after(seq2, seq1) before(seq1, seq2)
249 /* is s2<=s1<=s3 ? */
250 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
252 return seq3 - seq2 >= seq1 - seq2;
256 extern struct proto tcp_prot;
258 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
259 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
260 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
261 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
262 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
263 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
264 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
266 extern void tcp_v4_err(struct sk_buff *skb, u32);
268 extern void tcp_shutdown (struct sock *sk, int how);
270 extern int tcp_v4_rcv(struct sk_buff *skb);
272 extern int tcp_v4_remember_stamp(struct sock *sk);
274 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
276 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
277 struct msghdr *msg, size_t size);
278 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
280 extern int tcp_ioctl(struct sock *sk,
284 extern int tcp_rcv_state_process(struct sock *sk,
289 extern int tcp_rcv_established(struct sock *sk,
294 extern void tcp_rcv_space_adjust(struct sock *sk);
296 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
298 extern int tcp_twsk_unique(struct sock *sk,
299 struct sock *sktw, void *twp);
301 extern void tcp_twsk_destructor(struct sock *sk);
303 static inline void tcp_dec_quickack_mode(struct sock *sk,
304 const unsigned int pkts)
306 struct inet_connection_sock *icsk = inet_csk(sk);
308 if (icsk->icsk_ack.quick) {
309 if (pkts >= icsk->icsk_ack.quick) {
310 icsk->icsk_ack.quick = 0;
311 /* Leaving quickack mode we deflate ATO. */
312 icsk->icsk_ack.ato = TCP_ATO_MIN;
314 icsk->icsk_ack.quick -= pkts;
318 extern void tcp_enter_quickack_mode(struct sock *sk);
320 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
322 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
334 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
336 const struct tcphdr *th);
338 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
339 struct request_sock *req,
340 struct request_sock **prev);
341 extern int tcp_child_process(struct sock *parent,
343 struct sk_buff *skb);
344 extern int tcp_use_frto(const struct sock *sk);
345 extern void tcp_enter_frto(struct sock *sk);
346 extern void tcp_enter_loss(struct sock *sk, int how);
347 extern void tcp_clear_retrans(struct tcp_sock *tp);
348 extern void tcp_update_metrics(struct sock *sk);
350 extern void tcp_close(struct sock *sk,
352 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
354 extern int tcp_getsockopt(struct sock *sk, int level,
358 extern int tcp_setsockopt(struct sock *sk, int level,
359 int optname, char __user *optval,
361 extern int compat_tcp_getsockopt(struct sock *sk,
362 int level, int optname,
363 char __user *optval, int __user *optlen);
364 extern int compat_tcp_setsockopt(struct sock *sk,
365 int level, int optname,
366 char __user *optval, int optlen);
367 extern void tcp_set_keepalive(struct sock *sk, int val);
368 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
370 size_t len, int nonblock,
371 int flags, int *addr_len);
373 extern void tcp_parse_options(struct sk_buff *skb,
374 struct tcp_options_received *opt_rx,
378 * TCP v4 functions exported for the inet6 API
381 extern void tcp_v4_send_check(struct sock *sk, int len,
382 struct sk_buff *skb);
384 extern int tcp_v4_conn_request(struct sock *sk,
385 struct sk_buff *skb);
387 extern struct sock * tcp_create_openreq_child(struct sock *sk,
388 struct request_sock *req,
389 struct sk_buff *skb);
391 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
393 struct request_sock *req,
394 struct dst_entry *dst);
396 extern int tcp_v4_do_rcv(struct sock *sk,
397 struct sk_buff *skb);
399 extern int tcp_v4_connect(struct sock *sk,
400 struct sockaddr *uaddr,
403 extern int tcp_connect(struct sock *sk);
405 extern struct sk_buff * tcp_make_synack(struct sock *sk,
406 struct dst_entry *dst,
407 struct request_sock *req);
409 extern int tcp_disconnect(struct sock *sk, int flags);
411 extern void tcp_unhash(struct sock *sk);
413 /* From syncookies.c */
414 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
415 struct ip_options *opt);
416 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
421 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
422 unsigned int cur_mss, int nonagle);
423 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
424 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
425 extern void tcp_xmit_retransmit_queue(struct sock *);
426 extern void tcp_simple_retransmit(struct sock *);
427 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
428 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
430 extern void tcp_send_probe0(struct sock *);
431 extern void tcp_send_partial(struct sock *);
432 extern int tcp_write_wakeup(struct sock *);
433 extern void tcp_send_fin(struct sock *sk);
434 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
435 extern int tcp_send_synack(struct sock *);
436 extern void tcp_push_one(struct sock *, unsigned int mss_now);
437 extern void tcp_send_ack(struct sock *sk);
438 extern void tcp_send_delayed_ack(struct sock *sk);
441 extern void tcp_cwnd_application_limited(struct sock *sk);
444 extern void tcp_init_xmit_timers(struct sock *);
445 static inline void tcp_clear_xmit_timers(struct sock *sk)
447 inet_csk_clear_xmit_timers(sk);
450 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
451 extern unsigned int tcp_current_mss(struct sock *sk, int large);
454 extern void tcp_get_info(struct sock *, struct tcp_info *);
456 /* Read 'sendfile()'-style from a TCP socket */
457 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
458 unsigned int, size_t);
459 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
460 sk_read_actor_t recv_actor);
462 extern void tcp_initialize_rcv_mss(struct sock *sk);
464 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
465 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
466 extern void tcp_mtup_init(struct sock *sk);
468 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
470 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
471 ntohl(TCP_FLAG_ACK) |
475 static inline void tcp_fast_path_on(struct tcp_sock *tp)
477 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
480 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
482 if (skb_queue_empty(&tp->out_of_order_queue) &&
484 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
486 tcp_fast_path_on(tp);
489 /* Compute the actual receive window we are currently advertising.
490 * Rcv_nxt can be after the window if our peer push more data
491 * than the offered window.
493 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
495 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
502 /* Choose a new window, without checks for shrinking, and without
503 * scaling applied to the result. The caller does these things
504 * if necessary. This is a "raw" window selection.
506 extern u32 __tcp_select_window(struct sock *sk);
508 /* TCP timestamps are only 32-bits, this causes a slight
509 * complication on 64-bit systems since we store a snapshot
510 * of jiffies in the buffer control blocks below. We decided
511 * to use only the low 32-bits of jiffies and hide the ugly
512 * casts with the following macro.
514 #define tcp_time_stamp ((__u32)(jiffies))
516 /* This is what the send packet queuing engine uses to pass
517 * TCP per-packet control information to the transmission
518 * code. We also store the host-order sequence numbers in
519 * here too. This is 36 bytes on 32-bit architectures,
520 * 40 bytes on 64-bit machines, if this grows please adjust
521 * skbuff.h:skbuff->cb[xxx] size appropriately.
525 struct inet_skb_parm h4;
526 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
527 struct inet6_skb_parm h6;
529 } header; /* For incoming frames */
530 __u32 seq; /* Starting sequence number */
531 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
532 __u32 when; /* used to compute rtt's */
533 __u8 flags; /* TCP header flags. */
535 /* NOTE: These must match up to the flags byte in a
538 #define TCPCB_FLAG_FIN 0x01
539 #define TCPCB_FLAG_SYN 0x02
540 #define TCPCB_FLAG_RST 0x04
541 #define TCPCB_FLAG_PSH 0x08
542 #define TCPCB_FLAG_ACK 0x10
543 #define TCPCB_FLAG_URG 0x20
544 #define TCPCB_FLAG_ECE 0x40
545 #define TCPCB_FLAG_CWR 0x80
547 __u8 sacked; /* State flags for SACK/FACK. */
548 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
549 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
550 #define TCPCB_LOST 0x04 /* SKB is lost */
551 #define TCPCB_TAGBITS 0x07 /* All tag bits */
553 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
554 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
556 #define TCPCB_URG 0x20 /* Urgent pointer advanced here */
558 #define TCPCB_AT_TAIL (TCPCB_URG)
560 __u16 urg_ptr; /* Valid w/URG flags is set. */
561 __u32 ack_seq; /* Sequence number ACK'd */
564 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
566 #include <net/tcp_ecn.h>
568 /* Due to TSO, an SKB can be composed of multiple actual
569 * packets. To keep these tracked properly, we use this.
571 static inline int tcp_skb_pcount(const struct sk_buff *skb)
573 return skb_shinfo(skb)->gso_segs;
576 /* This is valid iff tcp_skb_pcount() > 1. */
577 static inline int tcp_skb_mss(const struct sk_buff *skb)
579 return skb_shinfo(skb)->gso_size;
582 static inline void tcp_dec_pcount_approx(__u32 *count,
583 const struct sk_buff *skb)
586 *count -= tcp_skb_pcount(skb);
592 static inline void tcp_packets_out_inc(struct sock *sk,
594 const struct sk_buff *skb)
596 int orig = tp->packets_out;
598 tp->packets_out += tcp_skb_pcount(skb);
600 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
601 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
604 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
605 const struct sk_buff *skb)
607 tp->packets_out -= tcp_skb_pcount(skb);
610 /* Events passed to congestion control interface */
612 CA_EVENT_TX_START, /* first transmit when no packets in flight */
613 CA_EVENT_CWND_RESTART, /* congestion window restart */
614 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
615 CA_EVENT_FRTO, /* fast recovery timeout */
616 CA_EVENT_LOSS, /* loss timeout */
617 CA_EVENT_FAST_ACK, /* in sequence ack */
618 CA_EVENT_SLOW_ACK, /* other ack */
622 * Interface for adding new TCP congestion control handlers
624 #define TCP_CA_NAME_MAX 16
625 #define TCP_CA_MAX 128
626 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
628 struct tcp_congestion_ops {
629 struct list_head list;
632 /* initialize private data (optional) */
633 void (*init)(struct sock *sk);
634 /* cleanup private data (optional) */
635 void (*release)(struct sock *sk);
637 /* return slow start threshold (required) */
638 u32 (*ssthresh)(struct sock *sk);
639 /* lower bound for congestion window (optional) */
640 u32 (*min_cwnd)(const struct sock *sk);
641 /* do new cwnd calculation (required) */
642 void (*cong_avoid)(struct sock *sk, u32 ack,
643 u32 rtt, u32 in_flight, int good_ack);
644 /* round trip time sample per acked packet (optional) */
645 void (*rtt_sample)(struct sock *sk, u32 usrtt);
646 /* call before changing ca_state (optional) */
647 void (*set_state)(struct sock *sk, u8 new_state);
648 /* call when cwnd event occurs (optional) */
649 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
650 /* new value of cwnd after loss (optional) */
651 u32 (*undo_cwnd)(struct sock *sk);
652 /* hook for packet ack accounting (optional) */
653 void (*pkts_acked)(struct sock *sk, u32 num_acked);
654 /* get info for inet_diag (optional) */
655 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
657 char name[TCP_CA_NAME_MAX];
658 struct module *owner;
661 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
662 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
664 extern void tcp_init_congestion_control(struct sock *sk);
665 extern void tcp_cleanup_congestion_control(struct sock *sk);
666 extern int tcp_set_default_congestion_control(const char *name);
667 extern void tcp_get_default_congestion_control(char *name);
668 extern void tcp_get_available_congestion_control(char *buf, size_t len);
669 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
670 extern int tcp_set_allowed_congestion_control(char *allowed);
671 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
672 extern void tcp_slow_start(struct tcp_sock *tp);
674 extern struct tcp_congestion_ops tcp_init_congestion_ops;
675 extern u32 tcp_reno_ssthresh(struct sock *sk);
676 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
677 u32 rtt, u32 in_flight, int flag);
678 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
679 extern struct tcp_congestion_ops tcp_reno;
681 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
683 struct inet_connection_sock *icsk = inet_csk(sk);
685 if (icsk->icsk_ca_ops->set_state)
686 icsk->icsk_ca_ops->set_state(sk, ca_state);
687 icsk->icsk_ca_state = ca_state;
690 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
692 const struct inet_connection_sock *icsk = inet_csk(sk);
694 if (icsk->icsk_ca_ops->cwnd_event)
695 icsk->icsk_ca_ops->cwnd_event(sk, event);
698 /* This determines how many packets are "in the network" to the best
699 * of our knowledge. In many cases it is conservative, but where
700 * detailed information is available from the receiver (via SACK
701 * blocks etc.) we can make more aggressive calculations.
703 * Use this for decisions involving congestion control, use just
704 * tp->packets_out to determine if the send queue is empty or not.
706 * Read this equation as:
708 * "Packets sent once on transmission queue" MINUS
709 * "Packets left network, but not honestly ACKed yet" PLUS
710 * "Packets fast retransmitted"
712 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
714 return (tp->packets_out - tp->left_out + tp->retrans_out);
717 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
718 * The exception is rate halving phase, when cwnd is decreasing towards
721 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
723 const struct tcp_sock *tp = tcp_sk(sk);
724 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
725 return tp->snd_ssthresh;
727 return max(tp->snd_ssthresh,
728 ((tp->snd_cwnd >> 1) +
729 (tp->snd_cwnd >> 2)));
732 static inline void tcp_sync_left_out(struct tcp_sock *tp)
734 if (tp->rx_opt.sack_ok &&
735 (tp->sacked_out >= tp->packets_out - tp->lost_out))
736 tp->sacked_out = tp->packets_out - tp->lost_out;
737 tp->left_out = tp->sacked_out + tp->lost_out;
740 extern void tcp_enter_cwr(struct sock *sk);
741 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
743 /* Slow start with delack produces 3 packets of burst, so that
744 * it is safe "de facto".
746 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
751 /* RFC2861 Check whether we are limited by application or congestion window
752 * This is the inverse of cwnd check in tcp_tso_should_defer
754 static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
756 const struct tcp_sock *tp = tcp_sk(sk);
759 if (in_flight >= tp->snd_cwnd)
765 left = tp->snd_cwnd - in_flight;
766 if (sysctl_tcp_tso_win_divisor)
767 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
769 return left <= tcp_max_burst(tp);
772 static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
773 const struct sk_buff *skb)
776 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
779 static inline void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
781 const struct inet_connection_sock *icsk = inet_csk(sk);
782 if (!tp->packets_out && !icsk->icsk_pending)
783 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
784 icsk->icsk_rto, TCP_RTO_MAX);
787 static inline void tcp_push_pending_frames(struct sock *sk,
790 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
793 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
798 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
804 * Calculate(/check) TCP checksum
806 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
807 __be32 daddr, __wsum base)
809 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
812 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
814 return __skb_checksum_complete(skb);
817 static inline int tcp_checksum_complete(struct sk_buff *skb)
819 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
820 __tcp_checksum_complete(skb);
823 /* Prequeue for VJ style copy to user, combined with checksumming. */
825 static inline void tcp_prequeue_init(struct tcp_sock *tp)
827 tp->ucopy.task = NULL;
829 tp->ucopy.memory = 0;
830 skb_queue_head_init(&tp->ucopy.prequeue);
831 #ifdef CONFIG_NET_DMA
832 tp->ucopy.dma_chan = NULL;
833 tp->ucopy.wakeup = 0;
834 tp->ucopy.pinned_list = NULL;
835 tp->ucopy.dma_cookie = 0;
839 /* Packet is added to VJ-style prequeue for processing in process
840 * context, if a reader task is waiting. Apparently, this exciting
841 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
842 * failed somewhere. Latency? Burstiness? Well, at least now we will
843 * see, why it failed. 8)8) --ANK
845 * NOTE: is this not too big to inline?
847 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
849 struct tcp_sock *tp = tcp_sk(sk);
851 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
852 __skb_queue_tail(&tp->ucopy.prequeue, skb);
853 tp->ucopy.memory += skb->truesize;
854 if (tp->ucopy.memory > sk->sk_rcvbuf) {
855 struct sk_buff *skb1;
857 BUG_ON(sock_owned_by_user(sk));
859 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
860 sk->sk_backlog_rcv(sk, skb1);
861 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
864 tp->ucopy.memory = 0;
865 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
866 wake_up_interruptible(sk->sk_sleep);
867 if (!inet_csk_ack_scheduled(sk))
868 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
869 (3 * TCP_RTO_MIN) / 4,
881 static const char *statename[]={
882 "Unused","Established","Syn Sent","Syn Recv",
883 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
884 "Close Wait","Last ACK","Listen","Closing"
888 static inline void tcp_set_state(struct sock *sk, int state)
890 int oldstate = sk->sk_state;
893 case TCP_ESTABLISHED:
894 if (oldstate != TCP_ESTABLISHED)
895 TCP_INC_STATS(TCP_MIB_CURRESTAB);
899 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
900 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
902 sk->sk_prot->unhash(sk);
903 if (inet_csk(sk)->icsk_bind_hash &&
904 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
905 inet_put_port(&tcp_hashinfo, sk);
908 if (oldstate==TCP_ESTABLISHED)
909 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
912 /* Change state AFTER socket is unhashed to avoid closed
913 * socket sitting in hash tables.
915 sk->sk_state = state;
918 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
922 static inline void tcp_done(struct sock *sk)
924 if(sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
925 TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
927 tcp_set_state(sk, TCP_CLOSE);
928 tcp_clear_xmit_timers(sk);
930 sk->sk_shutdown = SHUTDOWN_MASK;
932 if (!sock_flag(sk, SOCK_DEAD))
933 sk->sk_state_change(sk);
935 inet_csk_destroy_sock(sk);
938 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
941 rx_opt->eff_sacks = 0;
942 rx_opt->num_sacks = 0;
945 /* Determine a window scaling and initial window to offer. */
946 extern void tcp_select_initial_window(int __space, __u32 mss,
947 __u32 *rcv_wnd, __u32 *window_clamp,
948 int wscale_ok, __u8 *rcv_wscale);
950 static inline int tcp_win_from_space(int space)
952 return sysctl_tcp_adv_win_scale<=0 ?
953 (space>>(-sysctl_tcp_adv_win_scale)) :
954 space - (space>>sysctl_tcp_adv_win_scale);
957 /* Note: caller must be prepared to deal with negative returns */
958 static inline int tcp_space(const struct sock *sk)
960 return tcp_win_from_space(sk->sk_rcvbuf -
961 atomic_read(&sk->sk_rmem_alloc));
964 static inline int tcp_full_space(const struct sock *sk)
966 return tcp_win_from_space(sk->sk_rcvbuf);
969 static inline void tcp_openreq_init(struct request_sock *req,
970 struct tcp_options_received *rx_opt,
973 struct inet_request_sock *ireq = inet_rsk(req);
975 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
976 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
977 req->mss = rx_opt->mss_clamp;
978 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
979 ireq->tstamp_ok = rx_opt->tstamp_ok;
980 ireq->sack_ok = rx_opt->sack_ok;
981 ireq->snd_wscale = rx_opt->snd_wscale;
982 ireq->wscale_ok = rx_opt->wscale_ok;
985 ireq->rmt_port = skb->h.th->source;
988 extern void tcp_enter_memory_pressure(void);
990 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
992 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
995 static inline int keepalive_time_when(const struct tcp_sock *tp)
997 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1000 static inline int tcp_fin_time(const struct sock *sk)
1002 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1003 const int rto = inet_csk(sk)->icsk_rto;
1005 if (fin_timeout < (rto << 2) - (rto >> 1))
1006 fin_timeout = (rto << 2) - (rto >> 1);
1011 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1013 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1015 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1018 /* RST segments are not recommended to carry timestamp,
1019 and, if they do, it is recommended to ignore PAWS because
1020 "their cleanup function should take precedence over timestamps."
1021 Certainly, it is mistake. It is necessary to understand the reasons
1022 of this constraint to relax it: if peer reboots, clock may go
1023 out-of-sync and half-open connections will not be reset.
1024 Actually, the problem would be not existing if all
1025 the implementations followed draft about maintaining clock
1026 via reboots. Linux-2.2 DOES NOT!
1028 However, we can relax time bounds for RST segments to MSL.
1030 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1035 #define TCP_CHECK_TIMER(sk) do { } while (0)
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;
1058 /* - key database */
1059 struct tcp_md5sig_key {
1064 struct tcp4_md5sig_key {
1070 struct tcp6_md5sig_key {
1074 u32 scope_id; /* XXX */
1076 struct in6_addr addr;
1080 struct tcp_md5sig_info {
1081 struct tcp4_md5sig_key *keys4;
1082 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1083 struct tcp6_md5sig_key *keys6;
1091 /* - pseudo header */
1092 struct tcp4_pseudohdr {
1100 struct tcp6_pseudohdr {
1101 struct in6_addr saddr;
1102 struct in6_addr daddr;
1104 __be32 protocol; /* including padding */
1107 union tcp_md5sum_block {
1108 struct tcp4_pseudohdr ip4;
1109 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1110 struct tcp6_pseudohdr ip6;
1114 /* - pool: digest algorithm, hash description and scratch buffer */
1115 struct tcp_md5sig_pool {
1116 struct hash_desc md5_desc;
1117 union tcp_md5sum_block md5_blk;
1120 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1123 extern int tcp_v4_calc_md5_hash(char *md5_hash,
1124 struct tcp_md5sig_key *key,
1126 struct dst_entry *dst,
1127 struct request_sock *req,
1129 int protocol, int tcplen);
1130 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1131 struct sock *addr_sk);
1133 extern int tcp_v4_md5_do_add(struct sock *sk,
1138 extern int tcp_v4_md5_do_del(struct sock *sk,
1141 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1142 extern void tcp_free_md5sig_pool(void);
1144 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1145 extern void __tcp_put_md5sig_pool(void);
1148 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1150 int cpu = get_cpu();
1151 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1157 static inline void tcp_put_md5sig_pool(void)
1159 __tcp_put_md5sig_pool();
1164 enum tcp_seq_states {
1165 TCP_SEQ_STATE_LISTENING,
1166 TCP_SEQ_STATE_OPENREQ,
1167 TCP_SEQ_STATE_ESTABLISHED,
1168 TCP_SEQ_STATE_TIME_WAIT,
1171 struct tcp_seq_afinfo {
1172 struct module *owner;
1175 int (*seq_show) (struct seq_file *m, void *v);
1176 struct file_operations *seq_fops;
1179 struct tcp_iter_state {
1181 enum tcp_seq_states state;
1182 struct sock *syn_wait_sk;
1183 int bucket, sbucket, num, uid;
1184 struct seq_operations seq_ops;
1187 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1188 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1190 extern struct request_sock_ops tcp_request_sock_ops;
1192 extern int tcp_v4_destroy_sock(struct sock *sk);
1194 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1195 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1197 #ifdef CONFIG_PROC_FS
1198 extern int tcp4_proc_init(void);
1199 extern void tcp4_proc_exit(void);
1202 /* TCP af-specific functions */
1203 struct tcp_sock_af_ops {
1204 #ifdef CONFIG_TCP_MD5SIG
1205 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1206 struct sock *addr_sk);
1207 int (*calc_md5_hash) (char *location,
1208 struct tcp_md5sig_key *md5,
1210 struct dst_entry *dst,
1211 struct request_sock *req,
1213 int protocol, int len);
1214 int (*md5_add) (struct sock *sk,
1215 struct sock *addr_sk,
1218 int (*md5_parse) (struct sock *sk,
1219 char __user *optval,
1224 struct tcp_request_sock_ops {
1225 #ifdef CONFIG_TCP_MD5SIG
1226 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1227 struct request_sock *req);
1231 extern void tcp_v4_init(struct net_proto_family *ops);
1232 extern void tcp_init(void);