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_frto_response;
224 extern int sysctl_tcp_low_latency;
225 extern int sysctl_tcp_dma_copybreak;
226 extern int sysctl_tcp_nometrics_save;
227 extern int sysctl_tcp_moderate_rcvbuf;
228 extern int sysctl_tcp_tso_win_divisor;
229 extern int sysctl_tcp_abc;
230 extern int sysctl_tcp_mtu_probing;
231 extern int sysctl_tcp_base_mss;
232 extern int sysctl_tcp_workaround_signed_windows;
233 extern int sysctl_tcp_slow_start_after_idle;
234 extern int sysctl_tcp_max_ssthresh;
236 extern atomic_t tcp_memory_allocated;
237 extern atomic_t tcp_sockets_allocated;
238 extern int tcp_memory_pressure;
241 * The next routines deal with comparing 32 bit unsigned ints
242 * and worry about wraparound (automatic with unsigned arithmetic).
245 static inline int before(__u32 seq1, __u32 seq2)
247 return (__s32)(seq1-seq2) < 0;
249 #define after(seq2, seq1) before(seq1, seq2)
251 /* is s2<=s1<=s3 ? */
252 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
254 return seq3 - seq2 >= seq1 - seq2;
257 static inline int tcp_too_many_orphans(struct sock *sk, int num)
259 return (num > sysctl_tcp_max_orphans) ||
260 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
261 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
264 extern struct proto tcp_prot;
266 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
267 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
268 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
269 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
270 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
271 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
272 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
274 extern void tcp_v4_err(struct sk_buff *skb, u32);
276 extern void tcp_shutdown (struct sock *sk, int how);
278 extern int tcp_v4_rcv(struct sk_buff *skb);
280 extern int tcp_v4_remember_stamp(struct sock *sk);
282 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
284 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
285 struct msghdr *msg, size_t size);
286 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
288 extern int tcp_ioctl(struct sock *sk,
292 extern int tcp_rcv_state_process(struct sock *sk,
297 extern int tcp_rcv_established(struct sock *sk,
302 extern void tcp_rcv_space_adjust(struct sock *sk);
304 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
306 extern int tcp_twsk_unique(struct sock *sk,
307 struct sock *sktw, void *twp);
309 extern void tcp_twsk_destructor(struct sock *sk);
311 static inline void tcp_dec_quickack_mode(struct sock *sk,
312 const unsigned int pkts)
314 struct inet_connection_sock *icsk = inet_csk(sk);
316 if (icsk->icsk_ack.quick) {
317 if (pkts >= icsk->icsk_ack.quick) {
318 icsk->icsk_ack.quick = 0;
319 /* Leaving quickack mode we deflate ATO. */
320 icsk->icsk_ack.ato = TCP_ATO_MIN;
322 icsk->icsk_ack.quick -= pkts;
326 extern void tcp_enter_quickack_mode(struct sock *sk);
328 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
330 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
342 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
344 const struct tcphdr *th);
346 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
347 struct request_sock *req,
348 struct request_sock **prev);
349 extern int tcp_child_process(struct sock *parent,
351 struct sk_buff *skb);
352 extern int tcp_use_frto(struct sock *sk);
353 extern void tcp_enter_frto(struct sock *sk);
354 extern void tcp_enter_loss(struct sock *sk, int how);
355 extern void tcp_clear_retrans(struct tcp_sock *tp);
356 extern void tcp_update_metrics(struct sock *sk);
358 extern void tcp_close(struct sock *sk,
360 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
362 extern int tcp_getsockopt(struct sock *sk, int level,
366 extern int tcp_setsockopt(struct sock *sk, int level,
367 int optname, char __user *optval,
369 extern int compat_tcp_getsockopt(struct sock *sk,
370 int level, int optname,
371 char __user *optval, int __user *optlen);
372 extern int compat_tcp_setsockopt(struct sock *sk,
373 int level, int optname,
374 char __user *optval, int optlen);
375 extern void tcp_set_keepalive(struct sock *sk, int val);
376 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
378 size_t len, int nonblock,
379 int flags, int *addr_len);
381 extern void tcp_parse_options(struct sk_buff *skb,
382 struct tcp_options_received *opt_rx,
386 * TCP v4 functions exported for the inet6 API
389 extern void tcp_v4_send_check(struct sock *sk, int len,
390 struct sk_buff *skb);
392 extern int tcp_v4_conn_request(struct sock *sk,
393 struct sk_buff *skb);
395 extern struct sock * tcp_create_openreq_child(struct sock *sk,
396 struct request_sock *req,
397 struct sk_buff *skb);
399 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
401 struct request_sock *req,
402 struct dst_entry *dst);
404 extern int tcp_v4_do_rcv(struct sock *sk,
405 struct sk_buff *skb);
407 extern int tcp_v4_connect(struct sock *sk,
408 struct sockaddr *uaddr,
411 extern int tcp_connect(struct sock *sk);
413 extern struct sk_buff * tcp_make_synack(struct sock *sk,
414 struct dst_entry *dst,
415 struct request_sock *req);
417 extern int tcp_disconnect(struct sock *sk, int flags);
419 extern void tcp_unhash(struct sock *sk);
421 /* From syncookies.c */
422 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
423 struct ip_options *opt);
424 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
429 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
431 extern int tcp_may_send_now(struct sock *sk);
432 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
433 extern void tcp_xmit_retransmit_queue(struct sock *);
434 extern void tcp_simple_retransmit(struct sock *);
435 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
436 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
438 extern void tcp_send_probe0(struct sock *);
439 extern void tcp_send_partial(struct sock *);
440 extern int tcp_write_wakeup(struct sock *);
441 extern void tcp_send_fin(struct sock *sk);
442 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
443 extern int tcp_send_synack(struct sock *);
444 extern void tcp_push_one(struct sock *, unsigned int mss_now);
445 extern void tcp_send_ack(struct sock *sk);
446 extern void tcp_send_delayed_ack(struct sock *sk);
449 extern void tcp_cwnd_application_limited(struct sock *sk);
452 extern void tcp_init_xmit_timers(struct sock *);
453 static inline void tcp_clear_xmit_timers(struct sock *sk)
455 inet_csk_clear_xmit_timers(sk);
458 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
459 extern unsigned int tcp_current_mss(struct sock *sk, int large);
462 extern void tcp_get_info(struct sock *, struct tcp_info *);
464 /* Read 'sendfile()'-style from a TCP socket */
465 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
466 unsigned int, size_t);
467 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
468 sk_read_actor_t recv_actor);
470 extern void tcp_initialize_rcv_mss(struct sock *sk);
472 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
473 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
474 extern void tcp_mtup_init(struct sock *sk);
476 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
478 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
479 ntohl(TCP_FLAG_ACK) |
483 static inline void tcp_fast_path_on(struct tcp_sock *tp)
485 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
488 static inline void tcp_fast_path_check(struct sock *sk)
490 struct tcp_sock *tp = tcp_sk(sk);
492 if (skb_queue_empty(&tp->out_of_order_queue) &&
494 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
496 tcp_fast_path_on(tp);
499 /* Compute the actual receive window we are currently advertising.
500 * Rcv_nxt can be after the window if our peer push more data
501 * than the offered window.
503 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
505 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
512 /* Choose a new window, without checks for shrinking, and without
513 * scaling applied to the result. The caller does these things
514 * if necessary. This is a "raw" window selection.
516 extern u32 __tcp_select_window(struct sock *sk);
518 /* TCP timestamps are only 32-bits, this causes a slight
519 * complication on 64-bit systems since we store a snapshot
520 * of jiffies in the buffer control blocks below. We decided
521 * to use only the low 32-bits of jiffies and hide the ugly
522 * casts with the following macro.
524 #define tcp_time_stamp ((__u32)(jiffies))
526 /* This is what the send packet queuing engine uses to pass
527 * TCP per-packet control information to the transmission
528 * code. We also store the host-order sequence numbers in
529 * here too. This is 36 bytes on 32-bit architectures,
530 * 40 bytes on 64-bit machines, if this grows please adjust
531 * skbuff.h:skbuff->cb[xxx] size appropriately.
535 struct inet_skb_parm h4;
536 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
537 struct inet6_skb_parm h6;
539 } header; /* For incoming frames */
540 __u32 seq; /* Starting sequence number */
541 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
542 __u32 when; /* used to compute rtt's */
543 __u8 flags; /* TCP header flags. */
545 /* NOTE: These must match up to the flags byte in a
548 #define TCPCB_FLAG_FIN 0x01
549 #define TCPCB_FLAG_SYN 0x02
550 #define TCPCB_FLAG_RST 0x04
551 #define TCPCB_FLAG_PSH 0x08
552 #define TCPCB_FLAG_ACK 0x10
553 #define TCPCB_FLAG_URG 0x20
554 #define TCPCB_FLAG_ECE 0x40
555 #define TCPCB_FLAG_CWR 0x80
557 __u8 sacked; /* State flags for SACK/FACK. */
558 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
559 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
560 #define TCPCB_LOST 0x04 /* SKB is lost */
561 #define TCPCB_TAGBITS 0x07 /* All tag bits */
563 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
564 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
566 #define TCPCB_URG 0x20 /* Urgent pointer advanced here */
568 #define TCPCB_AT_TAIL (TCPCB_URG)
570 __u16 urg_ptr; /* Valid w/URG flags is set. */
571 __u32 ack_seq; /* Sequence number ACK'd */
574 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
576 #include <net/tcp_ecn.h>
578 /* Due to TSO, an SKB can be composed of multiple actual
579 * packets. To keep these tracked properly, we use this.
581 static inline int tcp_skb_pcount(const struct sk_buff *skb)
583 return skb_shinfo(skb)->gso_segs;
586 /* This is valid iff tcp_skb_pcount() > 1. */
587 static inline int tcp_skb_mss(const struct sk_buff *skb)
589 return skb_shinfo(skb)->gso_size;
592 static inline void tcp_dec_pcount_approx(__u32 *count,
593 const struct sk_buff *skb)
596 *count -= tcp_skb_pcount(skb);
602 static inline void tcp_packets_out_inc(struct sock *sk,
603 const struct sk_buff *skb)
605 struct tcp_sock *tp = tcp_sk(sk);
606 int orig = tp->packets_out;
608 tp->packets_out += tcp_skb_pcount(skb);
610 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
611 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
614 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
615 const struct sk_buff *skb)
617 tp->packets_out -= tcp_skb_pcount(skb);
620 /* Events passed to congestion control interface */
622 CA_EVENT_TX_START, /* first transmit when no packets in flight */
623 CA_EVENT_CWND_RESTART, /* congestion window restart */
624 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
625 CA_EVENT_FRTO, /* fast recovery timeout */
626 CA_EVENT_LOSS, /* loss timeout */
627 CA_EVENT_FAST_ACK, /* in sequence ack */
628 CA_EVENT_SLOW_ACK, /* other ack */
632 * Interface for adding new TCP congestion control handlers
634 #define TCP_CA_NAME_MAX 16
635 #define TCP_CA_MAX 128
636 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
638 #define TCP_CONG_NON_RESTRICTED 0x1
639 #define TCP_CONG_RTT_STAMP 0x2
641 struct tcp_congestion_ops {
642 struct list_head list;
645 /* initialize private data (optional) */
646 void (*init)(struct sock *sk);
647 /* cleanup private data (optional) */
648 void (*release)(struct sock *sk);
650 /* return slow start threshold (required) */
651 u32 (*ssthresh)(struct sock *sk);
652 /* lower bound for congestion window (optional) */
653 u32 (*min_cwnd)(const struct sock *sk);
654 /* do new cwnd calculation (required) */
655 void (*cong_avoid)(struct sock *sk, u32 ack,
656 u32 rtt, u32 in_flight, int good_ack);
657 /* call before changing ca_state (optional) */
658 void (*set_state)(struct sock *sk, u8 new_state);
659 /* call when cwnd event occurs (optional) */
660 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
661 /* new value of cwnd after loss (optional) */
662 u32 (*undo_cwnd)(struct sock *sk);
663 /* hook for packet ack accounting (optional) */
664 void (*pkts_acked)(struct sock *sk, u32 num_acked, ktime_t last);
665 /* get info for inet_diag (optional) */
666 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
668 char name[TCP_CA_NAME_MAX];
669 struct module *owner;
672 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
673 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
675 extern void tcp_init_congestion_control(struct sock *sk);
676 extern void tcp_cleanup_congestion_control(struct sock *sk);
677 extern int tcp_set_default_congestion_control(const char *name);
678 extern void tcp_get_default_congestion_control(char *name);
679 extern void tcp_get_available_congestion_control(char *buf, size_t len);
680 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
681 extern int tcp_set_allowed_congestion_control(char *allowed);
682 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
683 extern void tcp_slow_start(struct tcp_sock *tp);
685 extern struct tcp_congestion_ops tcp_init_congestion_ops;
686 extern u32 tcp_reno_ssthresh(struct sock *sk);
687 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
688 u32 rtt, u32 in_flight, int flag);
689 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
690 extern struct tcp_congestion_ops tcp_reno;
692 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
694 struct inet_connection_sock *icsk = inet_csk(sk);
696 if (icsk->icsk_ca_ops->set_state)
697 icsk->icsk_ca_ops->set_state(sk, ca_state);
698 icsk->icsk_ca_state = ca_state;
701 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
703 const struct inet_connection_sock *icsk = inet_csk(sk);
705 if (icsk->icsk_ca_ops->cwnd_event)
706 icsk->icsk_ca_ops->cwnd_event(sk, event);
709 /* This determines how many packets are "in the network" to the best
710 * of our knowledge. In many cases it is conservative, but where
711 * detailed information is available from the receiver (via SACK
712 * blocks etc.) we can make more aggressive calculations.
714 * Use this for decisions involving congestion control, use just
715 * tp->packets_out to determine if the send queue is empty or not.
717 * Read this equation as:
719 * "Packets sent once on transmission queue" MINUS
720 * "Packets left network, but not honestly ACKed yet" PLUS
721 * "Packets fast retransmitted"
723 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
725 return (tp->packets_out - tp->left_out + tp->retrans_out);
728 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
729 * The exception is rate halving phase, when cwnd is decreasing towards
732 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
734 const struct tcp_sock *tp = tcp_sk(sk);
735 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
736 return tp->snd_ssthresh;
738 return max(tp->snd_ssthresh,
739 ((tp->snd_cwnd >> 1) +
740 (tp->snd_cwnd >> 2)));
743 static inline void tcp_sync_left_out(struct tcp_sock *tp)
745 BUG_ON(tp->rx_opt.sack_ok &&
746 (tp->sacked_out + tp->lost_out > tp->packets_out));
747 tp->left_out = tp->sacked_out + tp->lost_out;
750 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
751 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
753 /* Slow start with delack produces 3 packets of burst, so that
754 * it is safe "de facto".
756 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
761 /* RFC2861 Check whether we are limited by application or congestion window
762 * This is the inverse of cwnd check in tcp_tso_should_defer
764 static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
766 const struct tcp_sock *tp = tcp_sk(sk);
769 if (in_flight >= tp->snd_cwnd)
775 left = tp->snd_cwnd - in_flight;
776 if (sysctl_tcp_tso_win_divisor)
777 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
779 return left <= tcp_max_burst(tp);
782 static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
783 const struct sk_buff *skb)
786 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
789 static inline void tcp_check_probe_timer(struct sock *sk)
791 struct tcp_sock *tp = tcp_sk(sk);
792 const struct inet_connection_sock *icsk = inet_csk(sk);
794 if (!tp->packets_out && !icsk->icsk_pending)
795 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
796 icsk->icsk_rto, TCP_RTO_MAX);
799 static inline void tcp_push_pending_frames(struct sock *sk)
801 struct tcp_sock *tp = tcp_sk(sk);
803 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle);
806 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
811 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
817 * Calculate(/check) TCP checksum
819 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
820 __be32 daddr, __wsum base)
822 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
825 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
827 return __skb_checksum_complete(skb);
830 static inline int tcp_checksum_complete(struct sk_buff *skb)
832 return !skb_csum_unnecessary(skb) &&
833 __tcp_checksum_complete(skb);
836 /* Prequeue for VJ style copy to user, combined with checksumming. */
838 static inline void tcp_prequeue_init(struct tcp_sock *tp)
840 tp->ucopy.task = NULL;
842 tp->ucopy.memory = 0;
843 skb_queue_head_init(&tp->ucopy.prequeue);
844 #ifdef CONFIG_NET_DMA
845 tp->ucopy.dma_chan = NULL;
846 tp->ucopy.wakeup = 0;
847 tp->ucopy.pinned_list = NULL;
848 tp->ucopy.dma_cookie = 0;
852 /* Packet is added to VJ-style prequeue for processing in process
853 * context, if a reader task is waiting. Apparently, this exciting
854 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
855 * failed somewhere. Latency? Burstiness? Well, at least now we will
856 * see, why it failed. 8)8) --ANK
858 * NOTE: is this not too big to inline?
860 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
862 struct tcp_sock *tp = tcp_sk(sk);
864 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
865 __skb_queue_tail(&tp->ucopy.prequeue, skb);
866 tp->ucopy.memory += skb->truesize;
867 if (tp->ucopy.memory > sk->sk_rcvbuf) {
868 struct sk_buff *skb1;
870 BUG_ON(sock_owned_by_user(sk));
872 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
873 sk->sk_backlog_rcv(sk, skb1);
874 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
877 tp->ucopy.memory = 0;
878 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
879 wake_up_interruptible(sk->sk_sleep);
880 if (!inet_csk_ack_scheduled(sk))
881 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
882 (3 * TCP_RTO_MIN) / 4,
894 static const char *statename[]={
895 "Unused","Established","Syn Sent","Syn Recv",
896 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
897 "Close Wait","Last ACK","Listen","Closing"
901 static inline void tcp_set_state(struct sock *sk, int state)
903 int oldstate = sk->sk_state;
906 case TCP_ESTABLISHED:
907 if (oldstate != TCP_ESTABLISHED)
908 TCP_INC_STATS(TCP_MIB_CURRESTAB);
912 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
913 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
915 sk->sk_prot->unhash(sk);
916 if (inet_csk(sk)->icsk_bind_hash &&
917 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
918 inet_put_port(&tcp_hashinfo, sk);
921 if (oldstate==TCP_ESTABLISHED)
922 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
925 /* Change state AFTER socket is unhashed to avoid closed
926 * socket sitting in hash tables.
928 sk->sk_state = state;
931 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
935 extern void tcp_done(struct sock *sk);
937 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
940 rx_opt->eff_sacks = 0;
941 rx_opt->num_sacks = 0;
944 /* Determine a window scaling and initial window to offer. */
945 extern void tcp_select_initial_window(int __space, __u32 mss,
946 __u32 *rcv_wnd, __u32 *window_clamp,
947 int wscale_ok, __u8 *rcv_wscale);
949 static inline int tcp_win_from_space(int space)
951 return sysctl_tcp_adv_win_scale<=0 ?
952 (space>>(-sysctl_tcp_adv_win_scale)) :
953 space - (space>>sysctl_tcp_adv_win_scale);
956 /* Note: caller must be prepared to deal with negative returns */
957 static inline int tcp_space(const struct sock *sk)
959 return tcp_win_from_space(sk->sk_rcvbuf -
960 atomic_read(&sk->sk_rmem_alloc));
963 static inline int tcp_full_space(const struct sock *sk)
965 return tcp_win_from_space(sk->sk_rcvbuf);
968 static inline void tcp_openreq_init(struct request_sock *req,
969 struct tcp_options_received *rx_opt,
972 struct inet_request_sock *ireq = inet_rsk(req);
974 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
975 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
976 req->mss = rx_opt->mss_clamp;
977 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
978 ireq->tstamp_ok = rx_opt->tstamp_ok;
979 ireq->sack_ok = rx_opt->sack_ok;
980 ireq->snd_wscale = rx_opt->snd_wscale;
981 ireq->wscale_ok = rx_opt->wscale_ok;
984 ireq->rmt_port = tcp_hdr(skb)->source;
987 extern void tcp_enter_memory_pressure(void);
989 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
991 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
994 static inline int keepalive_time_when(const struct tcp_sock *tp)
996 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
999 static inline int tcp_fin_time(const struct sock *sk)
1001 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1002 const int rto = inet_csk(sk)->icsk_rto;
1004 if (fin_timeout < (rto << 2) - (rto >> 1))
1005 fin_timeout = (rto << 2) - (rto >> 1);
1010 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1012 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1014 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1017 /* RST segments are not recommended to carry timestamp,
1018 and, if they do, it is recommended to ignore PAWS because
1019 "their cleanup function should take precedence over timestamps."
1020 Certainly, it is mistake. It is necessary to understand the reasons
1021 of this constraint to relax it: if peer reboots, clock may go
1022 out-of-sync and half-open connections will not be reset.
1023 Actually, the problem would be not existing if all
1024 the implementations followed draft about maintaining clock
1025 via reboots. Linux-2.2 DOES NOT!
1027 However, we can relax time bounds for RST segments to MSL.
1029 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1034 #define TCP_CHECK_TIMER(sk) do { } while (0)
1036 static inline void tcp_mib_init(void)
1039 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1040 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1041 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1042 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1046 static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1047 tp->lost_skb_hint = NULL;
1048 tp->scoreboard_skb_hint = NULL;
1049 tp->retransmit_skb_hint = NULL;
1050 tp->forward_skb_hint = NULL;
1051 tp->fastpath_skb_hint = NULL;
1057 /* - key database */
1058 struct tcp_md5sig_key {
1063 struct tcp4_md5sig_key {
1069 struct tcp6_md5sig_key {
1073 u32 scope_id; /* XXX */
1075 struct in6_addr addr;
1079 struct tcp_md5sig_info {
1080 struct tcp4_md5sig_key *keys4;
1081 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1082 struct tcp6_md5sig_key *keys6;
1090 /* - pseudo header */
1091 struct tcp4_pseudohdr {
1099 struct tcp6_pseudohdr {
1100 struct in6_addr saddr;
1101 struct in6_addr daddr;
1103 __be32 protocol; /* including padding */
1106 union tcp_md5sum_block {
1107 struct tcp4_pseudohdr ip4;
1108 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1109 struct tcp6_pseudohdr ip6;
1113 /* - pool: digest algorithm, hash description and scratch buffer */
1114 struct tcp_md5sig_pool {
1115 struct hash_desc md5_desc;
1116 union tcp_md5sum_block md5_blk;
1119 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1122 extern int tcp_v4_calc_md5_hash(char *md5_hash,
1123 struct tcp_md5sig_key *key,
1125 struct dst_entry *dst,
1126 struct request_sock *req,
1128 int protocol, int tcplen);
1129 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1130 struct sock *addr_sk);
1132 extern int tcp_v4_md5_do_add(struct sock *sk,
1137 extern int tcp_v4_md5_do_del(struct sock *sk,
1140 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1141 extern void tcp_free_md5sig_pool(void);
1143 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1144 extern void __tcp_put_md5sig_pool(void);
1147 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1149 int cpu = get_cpu();
1150 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1156 static inline void tcp_put_md5sig_pool(void)
1158 __tcp_put_md5sig_pool();
1162 /* write queue abstraction */
1163 static inline void tcp_write_queue_purge(struct sock *sk)
1165 struct sk_buff *skb;
1167 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1168 sk_stream_free_skb(sk, skb);
1169 sk_stream_mem_reclaim(sk);
1172 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1174 struct sk_buff *skb = sk->sk_write_queue.next;
1175 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1180 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1182 struct sk_buff *skb = sk->sk_write_queue.prev;
1183 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1188 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1193 #define tcp_for_write_queue(skb, sk) \
1194 for (skb = (sk)->sk_write_queue.next; \
1195 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1198 #define tcp_for_write_queue_from(skb, sk) \
1199 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\
1202 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1204 return sk->sk_send_head;
1207 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1209 struct tcp_sock *tp = tcp_sk(sk);
1211 sk->sk_send_head = skb->next;
1212 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue)
1213 sk->sk_send_head = NULL;
1214 /* Don't override Nagle indefinately with F-RTO */
1215 if (tp->frto_counter == 2)
1216 tp->frto_counter = 3;
1219 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1221 if (sk->sk_send_head == skb_unlinked)
1222 sk->sk_send_head = NULL;
1225 static inline void tcp_init_send_head(struct sock *sk)
1227 sk->sk_send_head = NULL;
1230 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1232 __skb_queue_tail(&sk->sk_write_queue, skb);
1235 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1237 __tcp_add_write_queue_tail(sk, skb);
1239 /* Queue it, remembering where we must start sending. */
1240 if (sk->sk_send_head == NULL)
1241 sk->sk_send_head = skb;
1244 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1246 __skb_queue_head(&sk->sk_write_queue, skb);
1249 /* Insert buff after skb on the write queue of sk. */
1250 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1251 struct sk_buff *buff,
1254 __skb_append(skb, buff, &sk->sk_write_queue);
1257 /* Insert skb between prev and next on the write queue of sk. */
1258 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1259 struct sk_buff *skb,
1262 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue);
1265 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1267 __skb_unlink(skb, &sk->sk_write_queue);
1270 static inline int tcp_skb_is_last(const struct sock *sk,
1271 const struct sk_buff *skb)
1273 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1276 static inline int tcp_write_queue_empty(struct sock *sk)
1278 return skb_queue_empty(&sk->sk_write_queue);
1282 enum tcp_seq_states {
1283 TCP_SEQ_STATE_LISTENING,
1284 TCP_SEQ_STATE_OPENREQ,
1285 TCP_SEQ_STATE_ESTABLISHED,
1286 TCP_SEQ_STATE_TIME_WAIT,
1289 struct tcp_seq_afinfo {
1290 struct module *owner;
1293 int (*seq_show) (struct seq_file *m, void *v);
1294 struct file_operations *seq_fops;
1297 struct tcp_iter_state {
1299 enum tcp_seq_states state;
1300 struct sock *syn_wait_sk;
1301 int bucket, sbucket, num, uid;
1302 struct seq_operations seq_ops;
1305 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1306 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1308 extern struct request_sock_ops tcp_request_sock_ops;
1310 extern int tcp_v4_destroy_sock(struct sock *sk);
1312 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1313 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1315 #ifdef CONFIG_PROC_FS
1316 extern int tcp4_proc_init(void);
1317 extern void tcp4_proc_exit(void);
1320 /* TCP af-specific functions */
1321 struct tcp_sock_af_ops {
1322 #ifdef CONFIG_TCP_MD5SIG
1323 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1324 struct sock *addr_sk);
1325 int (*calc_md5_hash) (char *location,
1326 struct tcp_md5sig_key *md5,
1328 struct dst_entry *dst,
1329 struct request_sock *req,
1331 int protocol, int len);
1332 int (*md5_add) (struct sock *sk,
1333 struct sock *addr_sk,
1336 int (*md5_parse) (struct sock *sk,
1337 char __user *optval,
1342 struct tcp_request_sock_ops {
1343 #ifdef CONFIG_TCP_MD5SIG
1344 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1345 struct request_sock *req);
1349 extern void tcp_v4_init(struct net_proto_family *ops);
1350 extern void tcp_init(void);