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>
32 #include <linux/cryptohash.h>
34 #include <net/inet_connection_sock.h>
35 #include <net/inet_timewait_sock.h>
36 #include <net/inet_hashtables.h>
37 #include <net/checksum.h>
38 #include <net/request_sock.h>
42 #include <net/tcp_states.h>
43 #include <net/inet_ecn.h>
45 #include <linux/seq_file.h>
47 extern struct inet_hashinfo tcp_hashinfo;
49 extern atomic_t tcp_orphan_count;
50 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
52 #define MAX_TCP_HEADER (128 + MAX_HEADER)
55 * Never offer a window over 32767 without using window scaling. Some
56 * poor stacks do signed 16bit maths!
58 #define MAX_TCP_WINDOW 32767U
60 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
61 #define TCP_MIN_MSS 88U
63 /* Minimal RCV_MSS. */
64 #define TCP_MIN_RCVMSS 536U
66 /* The least MTU to use for probing */
67 #define TCP_BASE_MSS 512
69 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
70 #define TCP_FASTRETRANS_THRESH 3
72 /* Maximal reordering. */
73 #define TCP_MAX_REORDERING 127
75 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
76 #define TCP_MAX_QUICKACKS 16U
79 #define TCP_URG_VALID 0x0100
80 #define TCP_URG_NOTYET 0x0200
81 #define TCP_URG_READ 0x0400
83 #define TCP_RETR1 3 /*
84 * This is how many retries it does before it
85 * tries to figure out if the gateway is
86 * down. Minimal RFC value is 3; it corresponds
87 * to ~3sec-8min depending on RTO.
90 #define TCP_RETR2 15 /*
91 * This should take at least
92 * 90 minutes to time out.
93 * RFC1122 says that the limit is 100 sec.
94 * 15 is ~13-30min depending on RTO.
97 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
98 * connection: ~180sec is RFC minimum */
100 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
101 * connection: ~180sec is RFC minimum */
104 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
105 * socket. 7 is ~50sec-16min.
109 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
110 * state, about 60 seconds */
111 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
112 /* BSD style FIN_WAIT2 deadlock breaker.
113 * It used to be 3min, new value is 60sec,
114 * to combine FIN-WAIT-2 timeout with
118 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
120 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
121 #define TCP_ATO_MIN ((unsigned)(HZ/25))
123 #define TCP_DELACK_MIN 4U
124 #define TCP_ATO_MIN 4U
126 #define TCP_RTO_MAX ((unsigned)(120*HZ))
127 #define TCP_RTO_MIN ((unsigned)(HZ/5))
128 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
130 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
131 * for local resources.
134 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
135 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
136 #define TCP_KEEPALIVE_INTVL (75*HZ)
138 #define MAX_TCP_KEEPIDLE 32767
139 #define MAX_TCP_KEEPINTVL 32767
140 #define MAX_TCP_KEEPCNT 127
141 #define MAX_TCP_SYNCNT 127
143 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
145 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
146 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
147 * after this time. It should be equal
148 * (or greater than) TCP_TIMEWAIT_LEN
149 * to provide reliability equal to one
150 * provided by timewait state.
152 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
153 * timestamps. It must be less than
154 * minimal timewait lifetime.
160 #define TCPOPT_NOP 1 /* Padding */
161 #define TCPOPT_EOL 0 /* End of options */
162 #define TCPOPT_MSS 2 /* Segment size negotiating */
163 #define TCPOPT_WINDOW 3 /* Window scaling */
164 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
165 #define TCPOPT_SACK 5 /* SACK Block */
166 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
167 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
173 #define TCPOLEN_MSS 4
174 #define TCPOLEN_WINDOW 3
175 #define TCPOLEN_SACK_PERM 2
176 #define TCPOLEN_TIMESTAMP 10
177 #define TCPOLEN_MD5SIG 18
179 /* But this is what stacks really send out. */
180 #define TCPOLEN_TSTAMP_ALIGNED 12
181 #define TCPOLEN_WSCALE_ALIGNED 4
182 #define TCPOLEN_SACKPERM_ALIGNED 4
183 #define TCPOLEN_SACK_BASE 2
184 #define TCPOLEN_SACK_BASE_ALIGNED 4
185 #define TCPOLEN_SACK_PERBLOCK 8
186 #define TCPOLEN_MD5SIG_ALIGNED 20
188 /* Flags in tp->nonagle */
189 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
190 #define TCP_NAGLE_CORK 2 /* Socket is corked */
191 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
193 extern struct inet_timewait_death_row tcp_death_row;
195 /* sysctl variables for tcp */
196 extern int sysctl_tcp_timestamps;
197 extern int sysctl_tcp_window_scaling;
198 extern int sysctl_tcp_sack;
199 extern int sysctl_tcp_fin_timeout;
200 extern int sysctl_tcp_keepalive_time;
201 extern int sysctl_tcp_keepalive_probes;
202 extern int sysctl_tcp_keepalive_intvl;
203 extern int sysctl_tcp_syn_retries;
204 extern int sysctl_tcp_synack_retries;
205 extern int sysctl_tcp_retries1;
206 extern int sysctl_tcp_retries2;
207 extern int sysctl_tcp_orphan_retries;
208 extern int sysctl_tcp_syncookies;
209 extern int sysctl_tcp_retrans_collapse;
210 extern int sysctl_tcp_stdurg;
211 extern int sysctl_tcp_rfc1337;
212 extern int sysctl_tcp_abort_on_overflow;
213 extern int sysctl_tcp_max_orphans;
214 extern int sysctl_tcp_fack;
215 extern int sysctl_tcp_reordering;
216 extern int sysctl_tcp_ecn;
217 extern int sysctl_tcp_dsack;
218 extern int sysctl_tcp_mem[3];
219 extern int sysctl_tcp_wmem[3];
220 extern int sysctl_tcp_rmem[3];
221 extern int sysctl_tcp_app_win;
222 extern int sysctl_tcp_adv_win_scale;
223 extern int sysctl_tcp_tw_reuse;
224 extern int sysctl_tcp_frto;
225 extern int sysctl_tcp_frto_response;
226 extern int sysctl_tcp_low_latency;
227 extern int sysctl_tcp_dma_copybreak;
228 extern int sysctl_tcp_nometrics_save;
229 extern int sysctl_tcp_moderate_rcvbuf;
230 extern int sysctl_tcp_tso_win_divisor;
231 extern int sysctl_tcp_abc;
232 extern int sysctl_tcp_mtu_probing;
233 extern int sysctl_tcp_base_mss;
234 extern int sysctl_tcp_workaround_signed_windows;
235 extern int sysctl_tcp_slow_start_after_idle;
236 extern int sysctl_tcp_max_ssthresh;
238 extern atomic_t tcp_memory_allocated;
239 extern atomic_t tcp_sockets_allocated;
240 extern int tcp_memory_pressure;
243 * The next routines deal with comparing 32 bit unsigned ints
244 * and worry about wraparound (automatic with unsigned arithmetic).
247 static inline int before(__u32 seq1, __u32 seq2)
249 return (__s32)(seq1-seq2) < 0;
251 #define after(seq2, seq1) before(seq1, seq2)
253 /* is s2<=s1<=s3 ? */
254 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
256 return seq3 - seq2 >= seq1 - seq2;
259 static inline int tcp_too_many_orphans(struct sock *sk, int num)
261 return (num > sysctl_tcp_max_orphans) ||
262 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
263 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
266 extern struct proto tcp_prot;
268 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
269 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
270 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
271 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
272 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
273 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
274 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
276 extern void tcp_v4_err(struct sk_buff *skb, u32);
278 extern void tcp_shutdown (struct sock *sk, int how);
280 extern int tcp_v4_rcv(struct sk_buff *skb);
282 extern int tcp_v4_remember_stamp(struct sock *sk);
284 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
286 extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
287 struct msghdr *msg, size_t size);
288 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
290 extern int tcp_ioctl(struct sock *sk,
294 extern int tcp_rcv_state_process(struct sock *sk,
299 extern int tcp_rcv_established(struct sock *sk,
304 extern void tcp_rcv_space_adjust(struct sock *sk);
306 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
308 extern int tcp_twsk_unique(struct sock *sk,
309 struct sock *sktw, void *twp);
311 extern void tcp_twsk_destructor(struct sock *sk);
313 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
314 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
316 static inline void tcp_dec_quickack_mode(struct sock *sk,
317 const unsigned int pkts)
319 struct inet_connection_sock *icsk = inet_csk(sk);
321 if (icsk->icsk_ack.quick) {
322 if (pkts >= icsk->icsk_ack.quick) {
323 icsk->icsk_ack.quick = 0;
324 /* Leaving quickack mode we deflate ATO. */
325 icsk->icsk_ack.ato = TCP_ATO_MIN;
327 icsk->icsk_ack.quick -= pkts;
331 extern void tcp_enter_quickack_mode(struct sock *sk);
333 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
335 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
339 #define TCP_ECN_QUEUE_CWR 2
340 #define TCP_ECN_DEMAND_CWR 4
342 static __inline__ void
343 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
345 if (sysctl_tcp_ecn && th->ece && th->cwr)
346 inet_rsk(req)->ecn_ok = 1;
358 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
360 const struct tcphdr *th);
362 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
363 struct request_sock *req,
364 struct request_sock **prev);
365 extern int tcp_child_process(struct sock *parent,
367 struct sk_buff *skb);
368 extern int tcp_use_frto(struct sock *sk);
369 extern void tcp_enter_frto(struct sock *sk);
370 extern void tcp_enter_loss(struct sock *sk, int how);
371 extern void tcp_clear_retrans(struct tcp_sock *tp);
372 extern void tcp_update_metrics(struct sock *sk);
374 extern void tcp_close(struct sock *sk,
376 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
378 extern int tcp_getsockopt(struct sock *sk, int level,
382 extern int tcp_setsockopt(struct sock *sk, int level,
383 int optname, char __user *optval,
385 extern int compat_tcp_getsockopt(struct sock *sk,
386 int level, int optname,
387 char __user *optval, int __user *optlen);
388 extern int compat_tcp_setsockopt(struct sock *sk,
389 int level, int optname,
390 char __user *optval, int optlen);
391 extern void tcp_set_keepalive(struct sock *sk, int val);
392 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
394 size_t len, int nonblock,
395 int flags, int *addr_len);
397 extern void tcp_parse_options(struct sk_buff *skb,
398 struct tcp_options_received *opt_rx,
402 * TCP v4 functions exported for the inet6 API
405 extern void tcp_v4_send_check(struct sock *sk, int len,
406 struct sk_buff *skb);
408 extern int tcp_v4_conn_request(struct sock *sk,
409 struct sk_buff *skb);
411 extern struct sock * tcp_create_openreq_child(struct sock *sk,
412 struct request_sock *req,
413 struct sk_buff *skb);
415 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
417 struct request_sock *req,
418 struct dst_entry *dst);
420 extern int tcp_v4_do_rcv(struct sock *sk,
421 struct sk_buff *skb);
423 extern int tcp_v4_connect(struct sock *sk,
424 struct sockaddr *uaddr,
427 extern int tcp_connect(struct sock *sk);
429 extern struct sk_buff * tcp_make_synack(struct sock *sk,
430 struct dst_entry *dst,
431 struct request_sock *req);
433 extern int tcp_disconnect(struct sock *sk, int flags);
435 extern void tcp_unhash(struct sock *sk);
437 /* From syncookies.c */
438 extern __u32 syncookie_secret[2][16-3+SHA_DIGEST_WORDS];
439 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
440 struct ip_options *opt);
441 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
444 /* From net/ipv6/syncookies.c */
445 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
446 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
451 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
453 extern int tcp_may_send_now(struct sock *sk);
454 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
455 extern void tcp_xmit_retransmit_queue(struct sock *);
456 extern void tcp_simple_retransmit(struct sock *);
457 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
458 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
460 extern void tcp_send_probe0(struct sock *);
461 extern void tcp_send_partial(struct sock *);
462 extern int tcp_write_wakeup(struct sock *);
463 extern void tcp_send_fin(struct sock *sk);
464 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
465 extern int tcp_send_synack(struct sock *);
466 extern void tcp_push_one(struct sock *, unsigned int mss_now);
467 extern void tcp_send_ack(struct sock *sk);
468 extern void tcp_send_delayed_ack(struct sock *sk);
471 extern void tcp_cwnd_application_limited(struct sock *sk);
474 extern void tcp_init_xmit_timers(struct sock *);
475 static inline void tcp_clear_xmit_timers(struct sock *sk)
477 inet_csk_clear_xmit_timers(sk);
480 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
481 extern unsigned int tcp_current_mss(struct sock *sk, int large);
484 extern void tcp_get_info(struct sock *, struct tcp_info *);
486 /* Read 'sendfile()'-style from a TCP socket */
487 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
488 unsigned int, size_t);
489 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
490 sk_read_actor_t recv_actor);
492 extern void tcp_initialize_rcv_mss(struct sock *sk);
494 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
495 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
496 extern void tcp_mtup_init(struct sock *sk);
498 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
500 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
501 ntohl(TCP_FLAG_ACK) |
505 static inline void tcp_fast_path_on(struct tcp_sock *tp)
507 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
510 static inline void tcp_fast_path_check(struct sock *sk)
512 struct tcp_sock *tp = tcp_sk(sk);
514 if (skb_queue_empty(&tp->out_of_order_queue) &&
516 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
518 tcp_fast_path_on(tp);
521 /* Compute the actual receive window we are currently advertising.
522 * Rcv_nxt can be after the window if our peer push more data
523 * than the offered window.
525 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
527 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
534 /* Choose a new window, without checks for shrinking, and without
535 * scaling applied to the result. The caller does these things
536 * if necessary. This is a "raw" window selection.
538 extern u32 __tcp_select_window(struct sock *sk);
540 /* TCP timestamps are only 32-bits, this causes a slight
541 * complication on 64-bit systems since we store a snapshot
542 * of jiffies in the buffer control blocks below. We decided
543 * to use only the low 32-bits of jiffies and hide the ugly
544 * casts with the following macro.
546 #define tcp_time_stamp ((__u32)(jiffies))
548 /* This is what the send packet queuing engine uses to pass
549 * TCP per-packet control information to the transmission
550 * code. We also store the host-order sequence numbers in
551 * here too. This is 36 bytes on 32-bit architectures,
552 * 40 bytes on 64-bit machines, if this grows please adjust
553 * skbuff.h:skbuff->cb[xxx] size appropriately.
557 struct inet_skb_parm h4;
558 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
559 struct inet6_skb_parm h6;
561 } header; /* For incoming frames */
562 __u32 seq; /* Starting sequence number */
563 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
564 __u32 when; /* used to compute rtt's */
565 __u8 flags; /* TCP header flags. */
567 /* NOTE: These must match up to the flags byte in a
570 #define TCPCB_FLAG_FIN 0x01
571 #define TCPCB_FLAG_SYN 0x02
572 #define TCPCB_FLAG_RST 0x04
573 #define TCPCB_FLAG_PSH 0x08
574 #define TCPCB_FLAG_ACK 0x10
575 #define TCPCB_FLAG_URG 0x20
576 #define TCPCB_FLAG_ECE 0x40
577 #define TCPCB_FLAG_CWR 0x80
579 __u8 sacked; /* State flags for SACK/FACK. */
580 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
581 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
582 #define TCPCB_LOST 0x04 /* SKB is lost */
583 #define TCPCB_TAGBITS 0x07 /* All tag bits */
585 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
586 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
588 __u16 urg_ptr; /* Valid w/URG flags is set. */
589 __u32 ack_seq; /* Sequence number ACK'd */
592 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
594 /* Due to TSO, an SKB can be composed of multiple actual
595 * packets. To keep these tracked properly, we use this.
597 static inline int tcp_skb_pcount(const struct sk_buff *skb)
599 return skb_shinfo(skb)->gso_segs;
602 /* This is valid iff tcp_skb_pcount() > 1. */
603 static inline int tcp_skb_mss(const struct sk_buff *skb)
605 return skb_shinfo(skb)->gso_size;
608 static inline void tcp_dec_pcount_approx_int(__u32 *count, const int decr)
617 static inline void tcp_dec_pcount_approx(__u32 *count,
618 const struct sk_buff *skb)
620 tcp_dec_pcount_approx_int(count, tcp_skb_pcount(skb));
623 /* Events passed to congestion control interface */
625 CA_EVENT_TX_START, /* first transmit when no packets in flight */
626 CA_EVENT_CWND_RESTART, /* congestion window restart */
627 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
628 CA_EVENT_FRTO, /* fast recovery timeout */
629 CA_EVENT_LOSS, /* loss timeout */
630 CA_EVENT_FAST_ACK, /* in sequence ack */
631 CA_EVENT_SLOW_ACK, /* other ack */
635 * Interface for adding new TCP congestion control handlers
637 #define TCP_CA_NAME_MAX 16
638 #define TCP_CA_MAX 128
639 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
641 #define TCP_CONG_NON_RESTRICTED 0x1
642 #define TCP_CONG_RTT_STAMP 0x2
644 struct tcp_congestion_ops {
645 struct list_head list;
648 /* initialize private data (optional) */
649 void (*init)(struct sock *sk);
650 /* cleanup private data (optional) */
651 void (*release)(struct sock *sk);
653 /* return slow start threshold (required) */
654 u32 (*ssthresh)(struct sock *sk);
655 /* lower bound for congestion window (optional) */
656 u32 (*min_cwnd)(const struct sock *sk);
657 /* do new cwnd calculation (required) */
658 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
659 /* call before changing ca_state (optional) */
660 void (*set_state)(struct sock *sk, u8 new_state);
661 /* call when cwnd event occurs (optional) */
662 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
663 /* new value of cwnd after loss (optional) */
664 u32 (*undo_cwnd)(struct sock *sk);
665 /* hook for packet ack accounting (optional) */
666 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
667 /* get info for inet_diag (optional) */
668 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
670 char name[TCP_CA_NAME_MAX];
671 struct module *owner;
674 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
675 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
677 extern void tcp_init_congestion_control(struct sock *sk);
678 extern void tcp_cleanup_congestion_control(struct sock *sk);
679 extern int tcp_set_default_congestion_control(const char *name);
680 extern void tcp_get_default_congestion_control(char *name);
681 extern void tcp_get_available_congestion_control(char *buf, size_t len);
682 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
683 extern int tcp_set_allowed_congestion_control(char *allowed);
684 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
685 extern void tcp_slow_start(struct tcp_sock *tp);
687 extern struct tcp_congestion_ops tcp_init_congestion_ops;
688 extern u32 tcp_reno_ssthresh(struct sock *sk);
689 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
690 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
691 extern struct tcp_congestion_ops tcp_reno;
693 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
695 struct inet_connection_sock *icsk = inet_csk(sk);
697 if (icsk->icsk_ca_ops->set_state)
698 icsk->icsk_ca_ops->set_state(sk, ca_state);
699 icsk->icsk_ca_state = ca_state;
702 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
704 const struct inet_connection_sock *icsk = inet_csk(sk);
706 if (icsk->icsk_ca_ops->cwnd_event)
707 icsk->icsk_ca_ops->cwnd_event(sk, event);
710 /* These functions determine how the current flow behaves in respect of SACK
711 * handling. SACK is negotiated with the peer, and therefore it can vary
712 * between different flows.
714 * tcp_is_sack - SACK enabled
715 * tcp_is_reno - No SACK
716 * tcp_is_fack - FACK enabled, implies SACK enabled
718 static inline int tcp_is_sack(const struct tcp_sock *tp)
720 return tp->rx_opt.sack_ok;
723 static inline int tcp_is_reno(const struct tcp_sock *tp)
725 return !tcp_is_sack(tp);
728 static inline int tcp_is_fack(const struct tcp_sock *tp)
730 return tp->rx_opt.sack_ok & 2;
733 static inline void tcp_enable_fack(struct tcp_sock *tp)
735 tp->rx_opt.sack_ok |= 2;
738 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
740 return tp->sacked_out + tp->lost_out;
743 /* This determines how many packets are "in the network" to the best
744 * of our knowledge. In many cases it is conservative, but where
745 * detailed information is available from the receiver (via SACK
746 * blocks etc.) we can make more aggressive calculations.
748 * Use this for decisions involving congestion control, use just
749 * tp->packets_out to determine if the send queue is empty or not.
751 * Read this equation as:
753 * "Packets sent once on transmission queue" MINUS
754 * "Packets left network, but not honestly ACKed yet" PLUS
755 * "Packets fast retransmitted"
757 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
759 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
762 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
763 * The exception is rate halving phase, when cwnd is decreasing towards
766 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
768 const struct tcp_sock *tp = tcp_sk(sk);
769 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
770 return tp->snd_ssthresh;
772 return max(tp->snd_ssthresh,
773 ((tp->snd_cwnd >> 1) +
774 (tp->snd_cwnd >> 2)));
777 /* Use define here intentionally to get WARN_ON location shown at the caller */
778 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
780 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
781 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
783 /* Slow start with delack produces 3 packets of burst, so that
784 * it is safe "de facto".
786 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
791 /* Returns end sequence number of the receiver's advertised window */
792 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
794 return tp->snd_una + tp->snd_wnd;
796 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
798 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
799 const struct sk_buff *skb)
802 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
805 static inline void tcp_check_probe_timer(struct sock *sk)
807 struct tcp_sock *tp = tcp_sk(sk);
808 const struct inet_connection_sock *icsk = inet_csk(sk);
810 if (!tp->packets_out && !icsk->icsk_pending)
811 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
812 icsk->icsk_rto, TCP_RTO_MAX);
815 static inline void tcp_push_pending_frames(struct sock *sk)
817 struct tcp_sock *tp = tcp_sk(sk);
819 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle);
822 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
827 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
833 * Calculate(/check) TCP checksum
835 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
836 __be32 daddr, __wsum base)
838 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
841 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
843 return __skb_checksum_complete(skb);
846 static inline int tcp_checksum_complete(struct sk_buff *skb)
848 return !skb_csum_unnecessary(skb) &&
849 __tcp_checksum_complete(skb);
852 /* Prequeue for VJ style copy to user, combined with checksumming. */
854 static inline void tcp_prequeue_init(struct tcp_sock *tp)
856 tp->ucopy.task = NULL;
858 tp->ucopy.memory = 0;
859 skb_queue_head_init(&tp->ucopy.prequeue);
860 #ifdef CONFIG_NET_DMA
861 tp->ucopy.dma_chan = NULL;
862 tp->ucopy.wakeup = 0;
863 tp->ucopy.pinned_list = NULL;
864 tp->ucopy.dma_cookie = 0;
868 /* Packet is added to VJ-style prequeue for processing in process
869 * context, if a reader task is waiting. Apparently, this exciting
870 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
871 * failed somewhere. Latency? Burstiness? Well, at least now we will
872 * see, why it failed. 8)8) --ANK
874 * NOTE: is this not too big to inline?
876 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
878 struct tcp_sock *tp = tcp_sk(sk);
880 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
881 __skb_queue_tail(&tp->ucopy.prequeue, skb);
882 tp->ucopy.memory += skb->truesize;
883 if (tp->ucopy.memory > sk->sk_rcvbuf) {
884 struct sk_buff *skb1;
886 BUG_ON(sock_owned_by_user(sk));
888 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
889 sk->sk_backlog_rcv(sk, skb1);
890 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
893 tp->ucopy.memory = 0;
894 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
895 wake_up_interruptible(sk->sk_sleep);
896 if (!inet_csk_ack_scheduled(sk))
897 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
898 (3 * TCP_RTO_MIN) / 4,
910 static const char *statename[]={
911 "Unused","Established","Syn Sent","Syn Recv",
912 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
913 "Close Wait","Last ACK","Listen","Closing"
916 extern void tcp_set_state(struct sock *sk, int state);
918 extern void tcp_done(struct sock *sk);
920 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
923 rx_opt->eff_sacks = 0;
924 rx_opt->num_sacks = 0;
927 /* Determine a window scaling and initial window to offer. */
928 extern void tcp_select_initial_window(int __space, __u32 mss,
929 __u32 *rcv_wnd, __u32 *window_clamp,
930 int wscale_ok, __u8 *rcv_wscale);
932 static inline int tcp_win_from_space(int space)
934 return sysctl_tcp_adv_win_scale<=0 ?
935 (space>>(-sysctl_tcp_adv_win_scale)) :
936 space - (space>>sysctl_tcp_adv_win_scale);
939 /* Note: caller must be prepared to deal with negative returns */
940 static inline int tcp_space(const struct sock *sk)
942 return tcp_win_from_space(sk->sk_rcvbuf -
943 atomic_read(&sk->sk_rmem_alloc));
946 static inline int tcp_full_space(const struct sock *sk)
948 return tcp_win_from_space(sk->sk_rcvbuf);
951 static inline void tcp_openreq_init(struct request_sock *req,
952 struct tcp_options_received *rx_opt,
955 struct inet_request_sock *ireq = inet_rsk(req);
957 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
958 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
959 req->mss = rx_opt->mss_clamp;
960 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
961 ireq->tstamp_ok = rx_opt->tstamp_ok;
962 ireq->sack_ok = rx_opt->sack_ok;
963 ireq->snd_wscale = rx_opt->snd_wscale;
964 ireq->wscale_ok = rx_opt->wscale_ok;
967 ireq->rmt_port = tcp_hdr(skb)->source;
970 extern void tcp_enter_memory_pressure(void);
972 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
974 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
977 static inline int keepalive_time_when(const struct tcp_sock *tp)
979 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
982 static inline int tcp_fin_time(const struct sock *sk)
984 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
985 const int rto = inet_csk(sk)->icsk_rto;
987 if (fin_timeout < (rto << 2) - (rto >> 1))
988 fin_timeout = (rto << 2) - (rto >> 1);
993 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
995 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
997 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1000 /* RST segments are not recommended to carry timestamp,
1001 and, if they do, it is recommended to ignore PAWS because
1002 "their cleanup function should take precedence over timestamps."
1003 Certainly, it is mistake. It is necessary to understand the reasons
1004 of this constraint to relax it: if peer reboots, clock may go
1005 out-of-sync and half-open connections will not be reset.
1006 Actually, the problem would be not existing if all
1007 the implementations followed draft about maintaining clock
1008 via reboots. Linux-2.2 DOES NOT!
1010 However, we can relax time bounds for RST segments to MSL.
1012 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1017 #define TCP_CHECK_TIMER(sk) do { } while (0)
1019 static inline void tcp_mib_init(void)
1022 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1023 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1024 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1025 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1029 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1031 tp->lost_skb_hint = NULL;
1032 tp->scoreboard_skb_hint = NULL;
1033 tp->retransmit_skb_hint = NULL;
1034 tp->forward_skb_hint = NULL;
1037 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1039 tcp_clear_retrans_hints_partial(tp);
1045 /* - key database */
1046 struct tcp_md5sig_key {
1051 struct tcp4_md5sig_key {
1052 struct tcp_md5sig_key base;
1056 struct tcp6_md5sig_key {
1057 struct tcp_md5sig_key base;
1059 u32 scope_id; /* XXX */
1061 struct in6_addr addr;
1065 struct tcp_md5sig_info {
1066 struct tcp4_md5sig_key *keys4;
1067 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1068 struct tcp6_md5sig_key *keys6;
1076 /* - pseudo header */
1077 struct tcp4_pseudohdr {
1085 struct tcp6_pseudohdr {
1086 struct in6_addr saddr;
1087 struct in6_addr daddr;
1089 __be32 protocol; /* including padding */
1092 union tcp_md5sum_block {
1093 struct tcp4_pseudohdr ip4;
1094 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1095 struct tcp6_pseudohdr ip6;
1099 /* - pool: digest algorithm, hash description and scratch buffer */
1100 struct tcp_md5sig_pool {
1101 struct hash_desc md5_desc;
1102 union tcp_md5sum_block md5_blk;
1105 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1108 extern int tcp_v4_calc_md5_hash(char *md5_hash,
1109 struct tcp_md5sig_key *key,
1111 struct dst_entry *dst,
1112 struct request_sock *req,
1115 unsigned int tcplen);
1116 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1117 struct sock *addr_sk);
1119 extern int tcp_v4_md5_do_add(struct sock *sk,
1124 extern int tcp_v4_md5_do_del(struct sock *sk,
1127 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1128 extern void tcp_free_md5sig_pool(void);
1130 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1131 extern void __tcp_put_md5sig_pool(void);
1134 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1136 int cpu = get_cpu();
1137 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1143 static inline void tcp_put_md5sig_pool(void)
1145 __tcp_put_md5sig_pool();
1149 /* write queue abstraction */
1150 static inline void tcp_write_queue_purge(struct sock *sk)
1152 struct sk_buff *skb;
1154 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1155 sk_wmem_free_skb(sk, skb);
1159 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1161 struct sk_buff *skb = sk->sk_write_queue.next;
1162 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1167 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1169 struct sk_buff *skb = sk->sk_write_queue.prev;
1170 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1175 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1180 #define tcp_for_write_queue(skb, sk) \
1181 for (skb = (sk)->sk_write_queue.next; \
1182 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1185 #define tcp_for_write_queue_from(skb, sk) \
1186 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\
1189 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1190 for (tmp = skb->next; \
1191 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1192 skb = tmp, tmp = skb->next)
1194 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1196 return sk->sk_send_head;
1199 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1201 sk->sk_send_head = skb->next;
1202 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue)
1203 sk->sk_send_head = NULL;
1206 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1208 if (sk->sk_send_head == skb_unlinked)
1209 sk->sk_send_head = NULL;
1212 static inline void tcp_init_send_head(struct sock *sk)
1214 sk->sk_send_head = NULL;
1217 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1219 __skb_queue_tail(&sk->sk_write_queue, skb);
1222 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1224 __tcp_add_write_queue_tail(sk, skb);
1226 /* Queue it, remembering where we must start sending. */
1227 if (sk->sk_send_head == NULL) {
1228 sk->sk_send_head = skb;
1230 if (tcp_sk(sk)->highest_sack == NULL)
1231 tcp_sk(sk)->highest_sack = skb;
1235 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1237 __skb_queue_head(&sk->sk_write_queue, skb);
1240 /* Insert buff after skb on the write queue of sk. */
1241 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1242 struct sk_buff *buff,
1245 __skb_append(skb, buff, &sk->sk_write_queue);
1248 /* Insert skb between prev and next on the write queue of sk. */
1249 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1250 struct sk_buff *skb,
1253 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue);
1255 if (sk->sk_send_head == skb)
1256 sk->sk_send_head = new;
1259 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1261 __skb_unlink(skb, &sk->sk_write_queue);
1264 static inline int tcp_skb_is_last(const struct sock *sk,
1265 const struct sk_buff *skb)
1267 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1270 static inline int tcp_write_queue_empty(struct sock *sk)
1272 return skb_queue_empty(&sk->sk_write_queue);
1275 /* Start sequence of the highest skb with SACKed bit, valid only if
1276 * sacked > 0 or when the caller has ensured validity by itself.
1278 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1280 if (!tp->sacked_out)
1283 if (tp->highest_sack == NULL)
1286 return TCP_SKB_CB(tp->highest_sack)->seq;
1289 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1291 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1292 tcp_write_queue_next(sk, skb);
1295 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1297 return tcp_sk(sk)->highest_sack;
1300 static inline void tcp_highest_sack_reset(struct sock *sk)
1302 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1305 /* Called when old skb is about to be deleted (to be combined with new skb) */
1306 static inline void tcp_highest_sack_combine(struct sock *sk,
1307 struct sk_buff *old,
1308 struct sk_buff *new)
1310 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1311 tcp_sk(sk)->highest_sack = new;
1315 enum tcp_seq_states {
1316 TCP_SEQ_STATE_LISTENING,
1317 TCP_SEQ_STATE_OPENREQ,
1318 TCP_SEQ_STATE_ESTABLISHED,
1319 TCP_SEQ_STATE_TIME_WAIT,
1322 struct tcp_seq_afinfo {
1323 struct module *owner;
1326 int (*seq_show) (struct seq_file *m, void *v);
1327 struct file_operations *seq_fops;
1330 struct tcp_iter_state {
1332 enum tcp_seq_states state;
1333 struct sock *syn_wait_sk;
1334 int bucket, sbucket, num, uid;
1335 struct seq_operations seq_ops;
1338 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1339 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1341 extern struct request_sock_ops tcp_request_sock_ops;
1342 extern struct request_sock_ops tcp6_request_sock_ops;
1344 extern int tcp_v4_destroy_sock(struct sock *sk);
1346 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1347 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1349 #ifdef CONFIG_PROC_FS
1350 extern int tcp4_proc_init(void);
1351 extern void tcp4_proc_exit(void);
1354 /* TCP af-specific functions */
1355 struct tcp_sock_af_ops {
1356 #ifdef CONFIG_TCP_MD5SIG
1357 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1358 struct sock *addr_sk);
1359 int (*calc_md5_hash) (char *location,
1360 struct tcp_md5sig_key *md5,
1362 struct dst_entry *dst,
1363 struct request_sock *req,
1367 int (*md5_add) (struct sock *sk,
1368 struct sock *addr_sk,
1371 int (*md5_parse) (struct sock *sk,
1372 char __user *optval,
1377 struct tcp_request_sock_ops {
1378 #ifdef CONFIG_TCP_MD5SIG
1379 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1380 struct request_sock *req);
1384 extern void tcp_v4_init(void);
1385 extern void tcp_init(void);