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;
233 extern int sysctl_tcp_max_ssthresh;
235 extern atomic_t tcp_memory_allocated;
236 extern atomic_t tcp_sockets_allocated;
237 extern int tcp_memory_pressure;
240 * The next routines deal with comparing 32 bit unsigned ints
241 * and worry about wraparound (automatic with unsigned arithmetic).
244 static inline int before(__u32 seq1, __u32 seq2)
246 return (__s32)(seq1-seq2) < 0;
248 #define after(seq2, seq1) before(seq1, seq2)
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 void tcp_cleanup_rbuf(struct sock *sk, int copied);
299 extern int tcp_twsk_unique(struct sock *sk,
300 struct sock *sktw, void *twp);
302 extern void tcp_twsk_destructor(struct sock *sk);
304 static inline void tcp_dec_quickack_mode(struct sock *sk,
305 const unsigned int pkts)
307 struct inet_connection_sock *icsk = inet_csk(sk);
309 if (icsk->icsk_ack.quick) {
310 if (pkts >= icsk->icsk_ack.quick) {
311 icsk->icsk_ack.quick = 0;
312 /* Leaving quickack mode we deflate ATO. */
313 icsk->icsk_ack.ato = TCP_ATO_MIN;
315 icsk->icsk_ack.quick -= pkts;
319 extern void tcp_enter_quickack_mode(struct sock *sk);
321 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
323 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
335 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
337 const struct tcphdr *th);
339 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
340 struct request_sock *req,
341 struct request_sock **prev);
342 extern int tcp_child_process(struct sock *parent,
344 struct sk_buff *skb);
345 extern int tcp_use_frto(struct sock *sk);
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);
442 extern void tcp_cwnd_application_limited(struct sock *sk);
445 extern void tcp_init_xmit_timers(struct sock *);
446 static inline void tcp_clear_xmit_timers(struct sock *sk)
448 inet_csk_clear_xmit_timers(sk);
451 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
452 extern unsigned int tcp_current_mss(struct sock *sk, int large);
455 extern void tcp_get_info(struct sock *, struct tcp_info *);
457 /* Read 'sendfile()'-style from a TCP socket */
458 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
459 unsigned int, size_t);
460 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
461 sk_read_actor_t recv_actor);
463 extern void tcp_initialize_rcv_mss(struct sock *sk);
465 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
466 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
467 extern void tcp_mtup_init(struct sock *sk);
469 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
471 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
472 ntohl(TCP_FLAG_ACK) |
476 static inline void tcp_fast_path_on(struct tcp_sock *tp)
478 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
481 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
483 if (skb_queue_empty(&tp->out_of_order_queue) &&
485 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
487 tcp_fast_path_on(tp);
490 /* Compute the actual receive window we are currently advertising.
491 * Rcv_nxt can be after the window if our peer push more data
492 * than the offered window.
494 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
496 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
503 /* Choose a new window, without checks for shrinking, and without
504 * scaling applied to the result. The caller does these things
505 * if necessary. This is a "raw" window selection.
507 extern u32 __tcp_select_window(struct sock *sk);
509 /* TCP timestamps are only 32-bits, this causes a slight
510 * complication on 64-bit systems since we store a snapshot
511 * of jiffies in the buffer control blocks below. We decided
512 * to use only the low 32-bits of jiffies and hide the ugly
513 * casts with the following macro.
515 #define tcp_time_stamp ((__u32)(jiffies))
517 /* This is what the send packet queuing engine uses to pass
518 * TCP per-packet control information to the transmission
519 * code. We also store the host-order sequence numbers in
520 * here too. This is 36 bytes on 32-bit architectures,
521 * 40 bytes on 64-bit machines, if this grows please adjust
522 * skbuff.h:skbuff->cb[xxx] size appropriately.
526 struct inet_skb_parm h4;
527 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
528 struct inet6_skb_parm h6;
530 } header; /* For incoming frames */
531 __u32 seq; /* Starting sequence number */
532 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
533 __u32 when; /* used to compute rtt's */
534 __u8 flags; /* TCP header flags. */
536 /* NOTE: These must match up to the flags byte in a
539 #define TCPCB_FLAG_FIN 0x01
540 #define TCPCB_FLAG_SYN 0x02
541 #define TCPCB_FLAG_RST 0x04
542 #define TCPCB_FLAG_PSH 0x08
543 #define TCPCB_FLAG_ACK 0x10
544 #define TCPCB_FLAG_URG 0x20
545 #define TCPCB_FLAG_ECE 0x40
546 #define TCPCB_FLAG_CWR 0x80
548 __u8 sacked; /* State flags for SACK/FACK. */
549 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
550 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
551 #define TCPCB_LOST 0x04 /* SKB is lost */
552 #define TCPCB_TAGBITS 0x07 /* All tag bits */
554 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
555 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
557 #define TCPCB_URG 0x20 /* Urgent pointer advanced here */
559 #define TCPCB_AT_TAIL (TCPCB_URG)
561 __u16 urg_ptr; /* Valid w/URG flags is set. */
562 __u32 ack_seq; /* Sequence number ACK'd */
565 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
567 #include <net/tcp_ecn.h>
569 /* Due to TSO, an SKB can be composed of multiple actual
570 * packets. To keep these tracked properly, we use this.
572 static inline int tcp_skb_pcount(const struct sk_buff *skb)
574 return skb_shinfo(skb)->gso_segs;
577 /* This is valid iff tcp_skb_pcount() > 1. */
578 static inline int tcp_skb_mss(const struct sk_buff *skb)
580 return skb_shinfo(skb)->gso_size;
583 static inline void tcp_dec_pcount_approx(__u32 *count,
584 const struct sk_buff *skb)
587 *count -= tcp_skb_pcount(skb);
593 static inline void tcp_packets_out_inc(struct sock *sk,
595 const struct sk_buff *skb)
597 int orig = tp->packets_out;
599 tp->packets_out += tcp_skb_pcount(skb);
601 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
602 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
605 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
606 const struct sk_buff *skb)
608 tp->packets_out -= tcp_skb_pcount(skb);
611 /* Events passed to congestion control interface */
613 CA_EVENT_TX_START, /* first transmit when no packets in flight */
614 CA_EVENT_CWND_RESTART, /* congestion window restart */
615 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
616 CA_EVENT_FRTO, /* fast recovery timeout */
617 CA_EVENT_LOSS, /* loss timeout */
618 CA_EVENT_FAST_ACK, /* in sequence ack */
619 CA_EVENT_SLOW_ACK, /* other ack */
623 * Interface for adding new TCP congestion control handlers
625 #define TCP_CA_NAME_MAX 16
626 #define TCP_CA_MAX 128
627 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
629 struct tcp_congestion_ops {
630 struct list_head list;
633 /* initialize private data (optional) */
634 void (*init)(struct sock *sk);
635 /* cleanup private data (optional) */
636 void (*release)(struct sock *sk);
638 /* return slow start threshold (required) */
639 u32 (*ssthresh)(struct sock *sk);
640 /* lower bound for congestion window (optional) */
641 u32 (*min_cwnd)(const struct sock *sk);
642 /* do new cwnd calculation (required) */
643 void (*cong_avoid)(struct sock *sk, u32 ack,
644 u32 rtt, u32 in_flight, int good_ack);
645 /* round trip time sample per acked packet (optional) */
646 void (*rtt_sample)(struct sock *sk, u32 usrtt);
647 /* call before changing ca_state (optional) */
648 void (*set_state)(struct sock *sk, u8 new_state);
649 /* call when cwnd event occurs (optional) */
650 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
651 /* new value of cwnd after loss (optional) */
652 u32 (*undo_cwnd)(struct sock *sk);
653 /* hook for packet ack accounting (optional) */
654 void (*pkts_acked)(struct sock *sk, u32 num_acked);
655 /* get info for inet_diag (optional) */
656 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
658 char name[TCP_CA_NAME_MAX];
659 struct module *owner;
662 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
663 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
665 extern void tcp_init_congestion_control(struct sock *sk);
666 extern void tcp_cleanup_congestion_control(struct sock *sk);
667 extern int tcp_set_default_congestion_control(const char *name);
668 extern void tcp_get_default_congestion_control(char *name);
669 extern void tcp_get_available_congestion_control(char *buf, size_t len);
670 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
671 extern int tcp_set_allowed_congestion_control(char *allowed);
672 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
673 extern void tcp_slow_start(struct tcp_sock *tp);
675 extern struct tcp_congestion_ops tcp_init_congestion_ops;
676 extern u32 tcp_reno_ssthresh(struct sock *sk);
677 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
678 u32 rtt, u32 in_flight, int flag);
679 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
680 extern struct tcp_congestion_ops tcp_reno;
682 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
684 struct inet_connection_sock *icsk = inet_csk(sk);
686 if (icsk->icsk_ca_ops->set_state)
687 icsk->icsk_ca_ops->set_state(sk, ca_state);
688 icsk->icsk_ca_state = ca_state;
691 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
693 const struct inet_connection_sock *icsk = inet_csk(sk);
695 if (icsk->icsk_ca_ops->cwnd_event)
696 icsk->icsk_ca_ops->cwnd_event(sk, event);
699 /* This determines how many packets are "in the network" to the best
700 * of our knowledge. In many cases it is conservative, but where
701 * detailed information is available from the receiver (via SACK
702 * blocks etc.) we can make more aggressive calculations.
704 * Use this for decisions involving congestion control, use just
705 * tp->packets_out to determine if the send queue is empty or not.
707 * Read this equation as:
709 * "Packets sent once on transmission queue" MINUS
710 * "Packets left network, but not honestly ACKed yet" PLUS
711 * "Packets fast retransmitted"
713 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
715 return (tp->packets_out - tp->left_out + tp->retrans_out);
718 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
719 * The exception is rate halving phase, when cwnd is decreasing towards
722 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
724 const struct tcp_sock *tp = tcp_sk(sk);
725 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
726 return tp->snd_ssthresh;
728 return max(tp->snd_ssthresh,
729 ((tp->snd_cwnd >> 1) +
730 (tp->snd_cwnd >> 2)));
733 static inline void tcp_sync_left_out(struct tcp_sock *tp)
735 if (tp->rx_opt.sack_ok &&
736 (tp->sacked_out >= tp->packets_out - tp->lost_out))
737 tp->sacked_out = tp->packets_out - tp->lost_out;
738 tp->left_out = tp->sacked_out + tp->lost_out;
741 extern void tcp_enter_cwr(struct sock *sk);
742 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
744 /* Slow start with delack produces 3 packets of burst, so that
745 * it is safe "de facto".
747 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
752 /* RFC2861 Check whether we are limited by application or congestion window
753 * This is the inverse of cwnd check in tcp_tso_should_defer
755 static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
757 const struct tcp_sock *tp = tcp_sk(sk);
760 if (in_flight >= tp->snd_cwnd)
766 left = tp->snd_cwnd - in_flight;
767 if (sysctl_tcp_tso_win_divisor)
768 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
770 return left <= tcp_max_burst(tp);
773 static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
774 const struct sk_buff *skb)
777 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
780 static inline void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
782 const struct inet_connection_sock *icsk = inet_csk(sk);
783 if (!tp->packets_out && !icsk->icsk_pending)
784 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
785 icsk->icsk_rto, TCP_RTO_MAX);
788 static inline void tcp_push_pending_frames(struct sock *sk,
791 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
794 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
799 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
805 * Calculate(/check) TCP checksum
807 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
808 __be32 daddr, __wsum base)
810 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
813 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
815 return __skb_checksum_complete(skb);
818 static inline int tcp_checksum_complete(struct sk_buff *skb)
820 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
821 __tcp_checksum_complete(skb);
824 /* Prequeue for VJ style copy to user, combined with checksumming. */
826 static inline void tcp_prequeue_init(struct tcp_sock *tp)
828 tp->ucopy.task = NULL;
830 tp->ucopy.memory = 0;
831 skb_queue_head_init(&tp->ucopy.prequeue);
832 #ifdef CONFIG_NET_DMA
833 tp->ucopy.dma_chan = NULL;
834 tp->ucopy.wakeup = 0;
835 tp->ucopy.pinned_list = NULL;
836 tp->ucopy.dma_cookie = 0;
840 /* Packet is added to VJ-style prequeue for processing in process
841 * context, if a reader task is waiting. Apparently, this exciting
842 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
843 * failed somewhere. Latency? Burstiness? Well, at least now we will
844 * see, why it failed. 8)8) --ANK
846 * NOTE: is this not too big to inline?
848 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
850 struct tcp_sock *tp = tcp_sk(sk);
852 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
853 __skb_queue_tail(&tp->ucopy.prequeue, skb);
854 tp->ucopy.memory += skb->truesize;
855 if (tp->ucopy.memory > sk->sk_rcvbuf) {
856 struct sk_buff *skb1;
858 BUG_ON(sock_owned_by_user(sk));
860 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
861 sk->sk_backlog_rcv(sk, skb1);
862 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
865 tp->ucopy.memory = 0;
866 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
867 wake_up_interruptible(sk->sk_sleep);
868 if (!inet_csk_ack_scheduled(sk))
869 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
870 (3 * TCP_RTO_MIN) / 4,
882 static const char *statename[]={
883 "Unused","Established","Syn Sent","Syn Recv",
884 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
885 "Close Wait","Last ACK","Listen","Closing"
889 static inline void tcp_set_state(struct sock *sk, int state)
891 int oldstate = sk->sk_state;
894 case TCP_ESTABLISHED:
895 if (oldstate != TCP_ESTABLISHED)
896 TCP_INC_STATS(TCP_MIB_CURRESTAB);
900 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
901 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
903 sk->sk_prot->unhash(sk);
904 if (inet_csk(sk)->icsk_bind_hash &&
905 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
906 inet_put_port(&tcp_hashinfo, sk);
909 if (oldstate==TCP_ESTABLISHED)
910 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
913 /* Change state AFTER socket is unhashed to avoid closed
914 * socket sitting in hash tables.
916 sk->sk_state = state;
919 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
923 static inline void tcp_done(struct sock *sk)
925 if(sk->sk_state == TCP_SYN_SENT || sk->sk_state == TCP_SYN_RECV)
926 TCP_INC_STATS_BH(TCP_MIB_ATTEMPTFAILS);
928 tcp_set_state(sk, TCP_CLOSE);
929 tcp_clear_xmit_timers(sk);
931 sk->sk_shutdown = SHUTDOWN_MASK;
933 if (!sock_flag(sk, SOCK_DEAD))
934 sk->sk_state_change(sk);
936 inet_csk_destroy_sock(sk);
939 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
942 rx_opt->eff_sacks = 0;
943 rx_opt->num_sacks = 0;
946 /* Determine a window scaling and initial window to offer. */
947 extern void tcp_select_initial_window(int __space, __u32 mss,
948 __u32 *rcv_wnd, __u32 *window_clamp,
949 int wscale_ok, __u8 *rcv_wscale);
951 static inline int tcp_win_from_space(int space)
953 return sysctl_tcp_adv_win_scale<=0 ?
954 (space>>(-sysctl_tcp_adv_win_scale)) :
955 space - (space>>sysctl_tcp_adv_win_scale);
958 /* Note: caller must be prepared to deal with negative returns */
959 static inline int tcp_space(const struct sock *sk)
961 return tcp_win_from_space(sk->sk_rcvbuf -
962 atomic_read(&sk->sk_rmem_alloc));
965 static inline int tcp_full_space(const struct sock *sk)
967 return tcp_win_from_space(sk->sk_rcvbuf);
970 static inline void tcp_openreq_init(struct request_sock *req,
971 struct tcp_options_received *rx_opt,
974 struct inet_request_sock *ireq = inet_rsk(req);
976 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
977 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
978 req->mss = rx_opt->mss_clamp;
979 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
980 ireq->tstamp_ok = rx_opt->tstamp_ok;
981 ireq->sack_ok = rx_opt->sack_ok;
982 ireq->snd_wscale = rx_opt->snd_wscale;
983 ireq->wscale_ok = rx_opt->wscale_ok;
986 ireq->rmt_port = skb->h.th->source;
989 extern void tcp_enter_memory_pressure(void);
991 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
993 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
996 static inline int keepalive_time_when(const struct tcp_sock *tp)
998 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1001 static inline int tcp_fin_time(const struct sock *sk)
1003 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1004 const int rto = inet_csk(sk)->icsk_rto;
1006 if (fin_timeout < (rto << 2) - (rto >> 1))
1007 fin_timeout = (rto << 2) - (rto >> 1);
1012 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1014 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1016 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1019 /* RST segments are not recommended to carry timestamp,
1020 and, if they do, it is recommended to ignore PAWS because
1021 "their cleanup function should take precedence over timestamps."
1022 Certainly, it is mistake. It is necessary to understand the reasons
1023 of this constraint to relax it: if peer reboots, clock may go
1024 out-of-sync and half-open connections will not be reset.
1025 Actually, the problem would be not existing if all
1026 the implementations followed draft about maintaining clock
1027 via reboots. Linux-2.2 DOES NOT!
1029 However, we can relax time bounds for RST segments to MSL.
1031 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1036 #define TCP_CHECK_TIMER(sk) do { } while (0)
1038 static inline void tcp_mib_init(void)
1041 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1042 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1043 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1044 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1048 static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1049 tp->lost_skb_hint = NULL;
1050 tp->scoreboard_skb_hint = NULL;
1051 tp->retransmit_skb_hint = NULL;
1052 tp->forward_skb_hint = NULL;
1053 tp->fastpath_skb_hint = NULL;
1059 /* - key database */
1060 struct tcp_md5sig_key {
1065 struct tcp4_md5sig_key {
1071 struct tcp6_md5sig_key {
1075 u32 scope_id; /* XXX */
1077 struct in6_addr addr;
1081 struct tcp_md5sig_info {
1082 struct tcp4_md5sig_key *keys4;
1083 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1084 struct tcp6_md5sig_key *keys6;
1092 /* - pseudo header */
1093 struct tcp4_pseudohdr {
1101 struct tcp6_pseudohdr {
1102 struct in6_addr saddr;
1103 struct in6_addr daddr;
1105 __be32 protocol; /* including padding */
1108 union tcp_md5sum_block {
1109 struct tcp4_pseudohdr ip4;
1110 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1111 struct tcp6_pseudohdr ip6;
1115 /* - pool: digest algorithm, hash description and scratch buffer */
1116 struct tcp_md5sig_pool {
1117 struct hash_desc md5_desc;
1118 union tcp_md5sum_block md5_blk;
1121 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1124 extern int tcp_v4_calc_md5_hash(char *md5_hash,
1125 struct tcp_md5sig_key *key,
1127 struct dst_entry *dst,
1128 struct request_sock *req,
1130 int protocol, int tcplen);
1131 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1132 struct sock *addr_sk);
1134 extern int tcp_v4_md5_do_add(struct sock *sk,
1139 extern int tcp_v4_md5_do_del(struct sock *sk,
1142 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1143 extern void tcp_free_md5sig_pool(void);
1145 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1146 extern void __tcp_put_md5sig_pool(void);
1149 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1151 int cpu = get_cpu();
1152 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1158 static inline void tcp_put_md5sig_pool(void)
1160 __tcp_put_md5sig_pool();
1165 enum tcp_seq_states {
1166 TCP_SEQ_STATE_LISTENING,
1167 TCP_SEQ_STATE_OPENREQ,
1168 TCP_SEQ_STATE_ESTABLISHED,
1169 TCP_SEQ_STATE_TIME_WAIT,
1172 struct tcp_seq_afinfo {
1173 struct module *owner;
1176 int (*seq_show) (struct seq_file *m, void *v);
1177 struct file_operations *seq_fops;
1180 struct tcp_iter_state {
1182 enum tcp_seq_states state;
1183 struct sock *syn_wait_sk;
1184 int bucket, sbucket, num, uid;
1185 struct seq_operations seq_ops;
1188 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1189 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1191 extern struct request_sock_ops tcp_request_sock_ops;
1193 extern int tcp_v4_destroy_sock(struct sock *sk);
1195 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1196 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1198 #ifdef CONFIG_PROC_FS
1199 extern int tcp4_proc_init(void);
1200 extern void tcp4_proc_exit(void);
1203 /* TCP af-specific functions */
1204 struct tcp_sock_af_ops {
1205 #ifdef CONFIG_TCP_MD5SIG
1206 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1207 struct sock *addr_sk);
1208 int (*calc_md5_hash) (char *location,
1209 struct tcp_md5sig_key *md5,
1211 struct dst_entry *dst,
1212 struct request_sock *req,
1214 int protocol, int len);
1215 int (*md5_add) (struct sock *sk,
1216 struct sock *addr_sk,
1219 int (*md5_parse) (struct sock *sk,
1220 char __user *optval,
1225 struct tcp_request_sock_ops {
1226 #ifdef CONFIG_TCP_MD5SIG
1227 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1228 struct request_sock *req);
1232 extern void tcp_v4_init(struct net_proto_family *ops);
1233 extern void tcp_init(void);