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 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
269 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
270 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
271 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
273 extern void tcp_v4_err(struct sk_buff *skb, u32);
275 extern void tcp_shutdown (struct sock *sk, int how);
277 extern int tcp_v4_rcv(struct sk_buff *skb);
279 extern int tcp_v4_remember_stamp(struct sock *sk);
281 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
283 extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
284 struct msghdr *msg, size_t size);
285 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
287 extern int tcp_ioctl(struct sock *sk,
291 extern int tcp_rcv_state_process(struct sock *sk,
296 extern int tcp_rcv_established(struct sock *sk,
301 extern void tcp_rcv_space_adjust(struct sock *sk);
303 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
305 extern int tcp_twsk_unique(struct sock *sk,
306 struct sock *sktw, void *twp);
308 extern void tcp_twsk_destructor(struct sock *sk);
310 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
311 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
313 static inline void tcp_dec_quickack_mode(struct sock *sk,
314 const unsigned int pkts)
316 struct inet_connection_sock *icsk = inet_csk(sk);
318 if (icsk->icsk_ack.quick) {
319 if (pkts >= icsk->icsk_ack.quick) {
320 icsk->icsk_ack.quick = 0;
321 /* Leaving quickack mode we deflate ATO. */
322 icsk->icsk_ack.ato = TCP_ATO_MIN;
324 icsk->icsk_ack.quick -= pkts;
328 extern void tcp_enter_quickack_mode(struct sock *sk);
330 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
332 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
336 #define TCP_ECN_QUEUE_CWR 2
337 #define TCP_ECN_DEMAND_CWR 4
339 static __inline__ void
340 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
342 if (sysctl_tcp_ecn && th->ece && th->cwr)
343 inet_rsk(req)->ecn_ok = 1;
355 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
357 const struct tcphdr *th);
359 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
360 struct request_sock *req,
361 struct request_sock **prev);
362 extern int tcp_child_process(struct sock *parent,
364 struct sk_buff *skb);
365 extern int tcp_use_frto(struct sock *sk);
366 extern void tcp_enter_frto(struct sock *sk);
367 extern void tcp_enter_loss(struct sock *sk, int how);
368 extern void tcp_clear_retrans(struct tcp_sock *tp);
369 extern void tcp_update_metrics(struct sock *sk);
371 extern void tcp_close(struct sock *sk,
373 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
375 extern int tcp_getsockopt(struct sock *sk, int level,
379 extern int tcp_setsockopt(struct sock *sk, int level,
380 int optname, char __user *optval,
382 extern int compat_tcp_getsockopt(struct sock *sk,
383 int level, int optname,
384 char __user *optval, int __user *optlen);
385 extern int compat_tcp_setsockopt(struct sock *sk,
386 int level, int optname,
387 char __user *optval, int optlen);
388 extern void tcp_set_keepalive(struct sock *sk, int val);
389 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
391 size_t len, int nonblock,
392 int flags, int *addr_len);
394 extern void tcp_parse_options(struct sk_buff *skb,
395 struct tcp_options_received *opt_rx,
398 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
401 * TCP v4 functions exported for the inet6 API
404 extern void tcp_v4_send_check(struct sock *sk, int len,
405 struct sk_buff *skb);
407 extern int tcp_v4_conn_request(struct sock *sk,
408 struct sk_buff *skb);
410 extern struct sock * tcp_create_openreq_child(struct sock *sk,
411 struct request_sock *req,
412 struct sk_buff *skb);
414 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
416 struct request_sock *req,
417 struct dst_entry *dst);
419 extern int tcp_v4_do_rcv(struct sock *sk,
420 struct sk_buff *skb);
422 extern int tcp_v4_connect(struct sock *sk,
423 struct sockaddr *uaddr,
426 extern int tcp_connect(struct sock *sk);
428 extern struct sk_buff * tcp_make_synack(struct sock *sk,
429 struct dst_entry *dst,
430 struct request_sock *req);
432 extern int tcp_disconnect(struct sock *sk, int flags);
435 /* From syncookies.c */
436 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
437 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
438 struct ip_options *opt);
439 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
442 extern __u32 cookie_init_timestamp(struct request_sock *req);
443 extern void cookie_check_timestamp(struct tcp_options_received *tcp_opt);
445 /* From net/ipv6/syncookies.c */
446 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
447 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
452 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
454 extern int tcp_may_send_now(struct sock *sk);
455 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
456 extern void tcp_xmit_retransmit_queue(struct sock *);
457 extern void tcp_simple_retransmit(struct sock *);
458 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
459 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
461 extern void tcp_send_probe0(struct sock *);
462 extern void tcp_send_partial(struct sock *);
463 extern int tcp_write_wakeup(struct sock *);
464 extern void tcp_send_fin(struct sock *sk);
465 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
466 extern int tcp_send_synack(struct sock *);
467 extern void tcp_push_one(struct sock *, unsigned int mss_now);
468 extern void tcp_send_ack(struct sock *sk);
469 extern void tcp_send_delayed_ack(struct sock *sk);
472 extern void tcp_cwnd_application_limited(struct sock *sk);
475 extern void tcp_init_xmit_timers(struct sock *);
476 static inline void tcp_clear_xmit_timers(struct sock *sk)
478 inet_csk_clear_xmit_timers(sk);
481 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
482 extern unsigned int tcp_current_mss(struct sock *sk, int large);
485 extern void tcp_get_info(struct sock *, struct tcp_info *);
487 /* Read 'sendfile()'-style from a TCP socket */
488 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
489 unsigned int, size_t);
490 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
491 sk_read_actor_t recv_actor);
493 extern void tcp_initialize_rcv_mss(struct sock *sk);
495 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
496 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
497 extern void tcp_mtup_init(struct sock *sk);
499 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
501 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
502 ntohl(TCP_FLAG_ACK) |
506 static inline void tcp_fast_path_on(struct tcp_sock *tp)
508 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
511 static inline void tcp_fast_path_check(struct sock *sk)
513 struct tcp_sock *tp = tcp_sk(sk);
515 if (skb_queue_empty(&tp->out_of_order_queue) &&
517 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
519 tcp_fast_path_on(tp);
522 /* Compute the actual receive window we are currently advertising.
523 * Rcv_nxt can be after the window if our peer push more data
524 * than the offered window.
526 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
528 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
535 /* Choose a new window, without checks for shrinking, and without
536 * scaling applied to the result. The caller does these things
537 * if necessary. This is a "raw" window selection.
539 extern u32 __tcp_select_window(struct sock *sk);
541 /* TCP timestamps are only 32-bits, this causes a slight
542 * complication on 64-bit systems since we store a snapshot
543 * of jiffies in the buffer control blocks below. We decided
544 * to use only the low 32-bits of jiffies and hide the ugly
545 * casts with the following macro.
547 #define tcp_time_stamp ((__u32)(jiffies))
549 /* This is what the send packet queuing engine uses to pass
550 * TCP per-packet control information to the transmission
551 * code. We also store the host-order sequence numbers in
552 * here too. This is 36 bytes on 32-bit architectures,
553 * 40 bytes on 64-bit machines, if this grows please adjust
554 * skbuff.h:skbuff->cb[xxx] size appropriately.
558 struct inet_skb_parm h4;
559 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
560 struct inet6_skb_parm h6;
562 } header; /* For incoming frames */
563 __u32 seq; /* Starting sequence number */
564 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
565 __u32 when; /* used to compute rtt's */
566 __u8 flags; /* TCP header flags. */
568 /* NOTE: These must match up to the flags byte in a
571 #define TCPCB_FLAG_FIN 0x01
572 #define TCPCB_FLAG_SYN 0x02
573 #define TCPCB_FLAG_RST 0x04
574 #define TCPCB_FLAG_PSH 0x08
575 #define TCPCB_FLAG_ACK 0x10
576 #define TCPCB_FLAG_URG 0x20
577 #define TCPCB_FLAG_ECE 0x40
578 #define TCPCB_FLAG_CWR 0x80
580 __u8 sacked; /* State flags for SACK/FACK. */
581 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
582 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
583 #define TCPCB_LOST 0x04 /* SKB is lost */
584 #define TCPCB_TAGBITS 0x07 /* All tag bits */
586 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
587 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
589 __u16 urg_ptr; /* Valid w/URG flags is set. */
590 __u32 ack_seq; /* Sequence number ACK'd */
593 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
595 /* Due to TSO, an SKB can be composed of multiple actual
596 * packets. To keep these tracked properly, we use this.
598 static inline int tcp_skb_pcount(const struct sk_buff *skb)
600 return skb_shinfo(skb)->gso_segs;
603 /* This is valid iff tcp_skb_pcount() > 1. */
604 static inline int tcp_skb_mss(const struct sk_buff *skb)
606 return skb_shinfo(skb)->gso_size;
609 static inline void tcp_dec_pcount_approx_int(__u32 *count, const int decr)
618 static inline void tcp_dec_pcount_approx(__u32 *count,
619 const struct sk_buff *skb)
621 tcp_dec_pcount_approx_int(count, tcp_skb_pcount(skb));
624 /* Events passed to congestion control interface */
626 CA_EVENT_TX_START, /* first transmit when no packets in flight */
627 CA_EVENT_CWND_RESTART, /* congestion window restart */
628 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
629 CA_EVENT_FRTO, /* fast recovery timeout */
630 CA_EVENT_LOSS, /* loss timeout */
631 CA_EVENT_FAST_ACK, /* in sequence ack */
632 CA_EVENT_SLOW_ACK, /* other ack */
636 * Interface for adding new TCP congestion control handlers
638 #define TCP_CA_NAME_MAX 16
639 #define TCP_CA_MAX 128
640 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
642 #define TCP_CONG_NON_RESTRICTED 0x1
643 #define TCP_CONG_RTT_STAMP 0x2
645 struct tcp_congestion_ops {
646 struct list_head list;
649 /* initialize private data (optional) */
650 void (*init)(struct sock *sk);
651 /* cleanup private data (optional) */
652 void (*release)(struct sock *sk);
654 /* return slow start threshold (required) */
655 u32 (*ssthresh)(struct sock *sk);
656 /* lower bound for congestion window (optional) */
657 u32 (*min_cwnd)(const struct sock *sk);
658 /* do new cwnd calculation (required) */
659 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
660 /* call before changing ca_state (optional) */
661 void (*set_state)(struct sock *sk, u8 new_state);
662 /* call when cwnd event occurs (optional) */
663 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
664 /* new value of cwnd after loss (optional) */
665 u32 (*undo_cwnd)(struct sock *sk);
666 /* hook for packet ack accounting (optional) */
667 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
668 /* get info for inet_diag (optional) */
669 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
671 char name[TCP_CA_NAME_MAX];
672 struct module *owner;
675 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
676 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
678 extern void tcp_init_congestion_control(struct sock *sk);
679 extern void tcp_cleanup_congestion_control(struct sock *sk);
680 extern int tcp_set_default_congestion_control(const char *name);
681 extern void tcp_get_default_congestion_control(char *name);
682 extern void tcp_get_available_congestion_control(char *buf, size_t len);
683 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
684 extern int tcp_set_allowed_congestion_control(char *allowed);
685 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
686 extern void tcp_slow_start(struct tcp_sock *tp);
688 extern struct tcp_congestion_ops tcp_init_congestion_ops;
689 extern u32 tcp_reno_ssthresh(struct sock *sk);
690 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
691 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
692 extern struct tcp_congestion_ops tcp_reno;
694 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
696 struct inet_connection_sock *icsk = inet_csk(sk);
698 if (icsk->icsk_ca_ops->set_state)
699 icsk->icsk_ca_ops->set_state(sk, ca_state);
700 icsk->icsk_ca_state = ca_state;
703 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
705 const struct inet_connection_sock *icsk = inet_csk(sk);
707 if (icsk->icsk_ca_ops->cwnd_event)
708 icsk->icsk_ca_ops->cwnd_event(sk, event);
711 /* These functions determine how the current flow behaves in respect of SACK
712 * handling. SACK is negotiated with the peer, and therefore it can vary
713 * between different flows.
715 * tcp_is_sack - SACK enabled
716 * tcp_is_reno - No SACK
717 * tcp_is_fack - FACK enabled, implies SACK enabled
719 static inline int tcp_is_sack(const struct tcp_sock *tp)
721 return tp->rx_opt.sack_ok;
724 static inline int tcp_is_reno(const struct tcp_sock *tp)
726 return !tcp_is_sack(tp);
729 static inline int tcp_is_fack(const struct tcp_sock *tp)
731 return tp->rx_opt.sack_ok & 2;
734 static inline void tcp_enable_fack(struct tcp_sock *tp)
736 tp->rx_opt.sack_ok |= 2;
739 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
741 return tp->sacked_out + tp->lost_out;
744 /* This determines how many packets are "in the network" to the best
745 * of our knowledge. In many cases it is conservative, but where
746 * detailed information is available from the receiver (via SACK
747 * blocks etc.) we can make more aggressive calculations.
749 * Use this for decisions involving congestion control, use just
750 * tp->packets_out to determine if the send queue is empty or not.
752 * Read this equation as:
754 * "Packets sent once on transmission queue" MINUS
755 * "Packets left network, but not honestly ACKed yet" PLUS
756 * "Packets fast retransmitted"
758 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
760 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
763 extern int tcp_limit_reno_sacked(struct tcp_sock *tp);
765 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
766 * The exception is rate halving phase, when cwnd is decreasing towards
769 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
771 const struct tcp_sock *tp = tcp_sk(sk);
772 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
773 return tp->snd_ssthresh;
775 return max(tp->snd_ssthresh,
776 ((tp->snd_cwnd >> 1) +
777 (tp->snd_cwnd >> 2)));
780 /* Use define here intentionally to get WARN_ON location shown at the caller */
781 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
783 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
784 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
786 /* Slow start with delack produces 3 packets of burst, so that
787 * it is safe "de facto". This will be the default - same as
788 * the default reordering threshold - but if reordering increases,
789 * we must be able to allow cwnd to burst at least this much in order
790 * to not pull it back when holes are filled.
792 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
794 return tp->reordering;
797 /* Returns end sequence number of the receiver's advertised window */
798 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
800 return tp->snd_una + tp->snd_wnd;
802 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
804 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
805 const struct sk_buff *skb)
808 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
811 static inline void tcp_check_probe_timer(struct sock *sk)
813 struct tcp_sock *tp = tcp_sk(sk);
814 const struct inet_connection_sock *icsk = inet_csk(sk);
816 if (!tp->packets_out && !icsk->icsk_pending)
817 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
818 icsk->icsk_rto, TCP_RTO_MAX);
821 static inline void tcp_push_pending_frames(struct sock *sk)
823 struct tcp_sock *tp = tcp_sk(sk);
825 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle);
828 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
833 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
839 * Calculate(/check) TCP checksum
841 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
842 __be32 daddr, __wsum base)
844 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
847 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
849 return __skb_checksum_complete(skb);
852 static inline int tcp_checksum_complete(struct sk_buff *skb)
854 return !skb_csum_unnecessary(skb) &&
855 __tcp_checksum_complete(skb);
858 /* Prequeue for VJ style copy to user, combined with checksumming. */
860 static inline void tcp_prequeue_init(struct tcp_sock *tp)
862 tp->ucopy.task = NULL;
864 tp->ucopy.memory = 0;
865 skb_queue_head_init(&tp->ucopy.prequeue);
866 #ifdef CONFIG_NET_DMA
867 tp->ucopy.dma_chan = NULL;
868 tp->ucopy.wakeup = 0;
869 tp->ucopy.pinned_list = NULL;
870 tp->ucopy.dma_cookie = 0;
874 /* Packet is added to VJ-style prequeue for processing in process
875 * context, if a reader task is waiting. Apparently, this exciting
876 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
877 * failed somewhere. Latency? Burstiness? Well, at least now we will
878 * see, why it failed. 8)8) --ANK
880 * NOTE: is this not too big to inline?
882 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
884 struct tcp_sock *tp = tcp_sk(sk);
886 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
887 __skb_queue_tail(&tp->ucopy.prequeue, skb);
888 tp->ucopy.memory += skb->truesize;
889 if (tp->ucopy.memory > sk->sk_rcvbuf) {
890 struct sk_buff *skb1;
892 BUG_ON(sock_owned_by_user(sk));
894 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
895 sk->sk_backlog_rcv(sk, skb1);
896 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPPREQUEUEDROPPED);
899 tp->ucopy.memory = 0;
900 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
901 wake_up_interruptible(sk->sk_sleep);
902 if (!inet_csk_ack_scheduled(sk))
903 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
904 (3 * TCP_RTO_MIN) / 4,
916 static const char *statename[]={
917 "Unused","Established","Syn Sent","Syn Recv",
918 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
919 "Close Wait","Last ACK","Listen","Closing"
922 extern void tcp_set_state(struct sock *sk, int state);
924 extern void tcp_done(struct sock *sk);
926 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
929 rx_opt->eff_sacks = 0;
930 rx_opt->num_sacks = 0;
933 /* Determine a window scaling and initial window to offer. */
934 extern void tcp_select_initial_window(int __space, __u32 mss,
935 __u32 *rcv_wnd, __u32 *window_clamp,
936 int wscale_ok, __u8 *rcv_wscale);
938 static inline int tcp_win_from_space(int space)
940 return sysctl_tcp_adv_win_scale<=0 ?
941 (space>>(-sysctl_tcp_adv_win_scale)) :
942 space - (space>>sysctl_tcp_adv_win_scale);
945 /* Note: caller must be prepared to deal with negative returns */
946 static inline int tcp_space(const struct sock *sk)
948 return tcp_win_from_space(sk->sk_rcvbuf -
949 atomic_read(&sk->sk_rmem_alloc));
952 static inline int tcp_full_space(const struct sock *sk)
954 return tcp_win_from_space(sk->sk_rcvbuf);
957 static inline void tcp_openreq_init(struct request_sock *req,
958 struct tcp_options_received *rx_opt,
961 struct inet_request_sock *ireq = inet_rsk(req);
963 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
965 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
966 req->mss = rx_opt->mss_clamp;
967 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
968 ireq->tstamp_ok = rx_opt->tstamp_ok;
969 ireq->sack_ok = rx_opt->sack_ok;
970 ireq->snd_wscale = rx_opt->snd_wscale;
971 ireq->wscale_ok = rx_opt->wscale_ok;
974 ireq->rmt_port = tcp_hdr(skb)->source;
977 extern void tcp_enter_memory_pressure(struct sock *sk);
979 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
981 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
984 static inline int keepalive_time_when(const struct tcp_sock *tp)
986 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
989 static inline int tcp_fin_time(const struct sock *sk)
991 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
992 const int rto = inet_csk(sk)->icsk_rto;
994 if (fin_timeout < (rto << 2) - (rto >> 1))
995 fin_timeout = (rto << 2) - (rto >> 1);
1000 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1002 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1004 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1007 /* RST segments are not recommended to carry timestamp,
1008 and, if they do, it is recommended to ignore PAWS because
1009 "their cleanup function should take precedence over timestamps."
1010 Certainly, it is mistake. It is necessary to understand the reasons
1011 of this constraint to relax it: if peer reboots, clock may go
1012 out-of-sync and half-open connections will not be reset.
1013 Actually, the problem would be not existing if all
1014 the implementations followed draft about maintaining clock
1015 via reboots. Linux-2.2 DOES NOT!
1017 However, we can relax time bounds for RST segments to MSL.
1019 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1024 #define TCP_CHECK_TIMER(sk) do { } while (0)
1026 static inline void tcp_mib_init(struct net *net)
1029 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1030 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1031 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1032 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1036 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1038 tp->lost_skb_hint = NULL;
1039 tp->scoreboard_skb_hint = NULL;
1040 tp->retransmit_skb_hint = NULL;
1041 tp->forward_skb_hint = NULL;
1044 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1046 tcp_clear_retrans_hints_partial(tp);
1052 /* - key database */
1053 struct tcp_md5sig_key {
1058 struct tcp4_md5sig_key {
1059 struct tcp_md5sig_key base;
1063 struct tcp6_md5sig_key {
1064 struct tcp_md5sig_key base;
1066 u32 scope_id; /* XXX */
1068 struct in6_addr addr;
1072 struct tcp_md5sig_info {
1073 struct tcp4_md5sig_key *keys4;
1074 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1075 struct tcp6_md5sig_key *keys6;
1083 /* - pseudo header */
1084 struct tcp4_pseudohdr {
1092 struct tcp6_pseudohdr {
1093 struct in6_addr saddr;
1094 struct in6_addr daddr;
1096 __be32 protocol; /* including padding */
1099 union tcp_md5sum_block {
1100 struct tcp4_pseudohdr ip4;
1101 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1102 struct tcp6_pseudohdr ip6;
1106 /* - pool: digest algorithm, hash description and scratch buffer */
1107 struct tcp_md5sig_pool {
1108 struct hash_desc md5_desc;
1109 union tcp_md5sum_block md5_blk;
1112 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1115 extern int tcp_calc_md5_hash(char *md5_hash,
1116 struct tcp_md5sig_key *key,
1119 unsigned int tcplen,
1120 struct tcp_md5sig_pool *hp);
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 unsigned 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 #ifdef CONFIG_TCP_MD5SIG
1141 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1142 &(struct tcp_md5sig_key) { \
1143 .key = (twsk)->tw_md5_key, \
1144 .keylen = (twsk)->tw_md5_keylen, \
1147 #define tcp_twsk_md5_key(twsk) NULL
1150 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1151 extern void tcp_free_md5sig_pool(void);
1153 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1154 extern void __tcp_put_md5sig_pool(void);
1157 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1159 int cpu = get_cpu();
1160 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1166 static inline void tcp_put_md5sig_pool(void)
1168 __tcp_put_md5sig_pool();
1172 /* write queue abstraction */
1173 static inline void tcp_write_queue_purge(struct sock *sk)
1175 struct sk_buff *skb;
1177 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1178 sk_wmem_free_skb(sk, skb);
1182 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1184 struct sk_buff *skb = sk->sk_write_queue.next;
1185 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1190 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1192 struct sk_buff *skb = sk->sk_write_queue.prev;
1193 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1198 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1203 #define tcp_for_write_queue(skb, sk) \
1204 for (skb = (sk)->sk_write_queue.next; \
1205 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1208 #define tcp_for_write_queue_from(skb, sk) \
1209 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\
1212 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1213 for (tmp = skb->next; \
1214 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1215 skb = tmp, tmp = skb->next)
1217 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1219 return sk->sk_send_head;
1222 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1224 sk->sk_send_head = skb->next;
1225 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue)
1226 sk->sk_send_head = NULL;
1229 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1231 if (sk->sk_send_head == skb_unlinked)
1232 sk->sk_send_head = NULL;
1235 static inline void tcp_init_send_head(struct sock *sk)
1237 sk->sk_send_head = NULL;
1240 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1242 __skb_queue_tail(&sk->sk_write_queue, skb);
1245 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1247 __tcp_add_write_queue_tail(sk, skb);
1249 /* Queue it, remembering where we must start sending. */
1250 if (sk->sk_send_head == NULL) {
1251 sk->sk_send_head = skb;
1253 if (tcp_sk(sk)->highest_sack == NULL)
1254 tcp_sk(sk)->highest_sack = skb;
1258 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1260 __skb_queue_head(&sk->sk_write_queue, skb);
1263 /* Insert buff after skb on the write queue of sk. */
1264 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1265 struct sk_buff *buff,
1268 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1271 /* Insert skb between prev and next on the write queue of sk. */
1272 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1273 struct sk_buff *skb,
1276 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue);
1278 if (sk->sk_send_head == skb)
1279 sk->sk_send_head = new;
1282 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1284 __skb_unlink(skb, &sk->sk_write_queue);
1287 static inline int tcp_skb_is_last(const struct sock *sk,
1288 const struct sk_buff *skb)
1290 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1293 static inline int tcp_write_queue_empty(struct sock *sk)
1295 return skb_queue_empty(&sk->sk_write_queue);
1298 /* Start sequence of the highest skb with SACKed bit, valid only if
1299 * sacked > 0 or when the caller has ensured validity by itself.
1301 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1303 if (!tp->sacked_out)
1306 if (tp->highest_sack == NULL)
1309 return TCP_SKB_CB(tp->highest_sack)->seq;
1312 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1314 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1315 tcp_write_queue_next(sk, skb);
1318 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1320 return tcp_sk(sk)->highest_sack;
1323 static inline void tcp_highest_sack_reset(struct sock *sk)
1325 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1328 /* Called when old skb is about to be deleted (to be combined with new skb) */
1329 static inline void tcp_highest_sack_combine(struct sock *sk,
1330 struct sk_buff *old,
1331 struct sk_buff *new)
1333 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1334 tcp_sk(sk)->highest_sack = new;
1338 enum tcp_seq_states {
1339 TCP_SEQ_STATE_LISTENING,
1340 TCP_SEQ_STATE_OPENREQ,
1341 TCP_SEQ_STATE_ESTABLISHED,
1342 TCP_SEQ_STATE_TIME_WAIT,
1345 struct tcp_seq_afinfo {
1348 struct file_operations seq_fops;
1349 struct seq_operations seq_ops;
1352 struct tcp_iter_state {
1353 struct seq_net_private p;
1355 enum tcp_seq_states state;
1356 struct sock *syn_wait_sk;
1357 int bucket, sbucket, num, uid;
1360 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1361 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1363 extern struct request_sock_ops tcp_request_sock_ops;
1364 extern struct request_sock_ops tcp6_request_sock_ops;
1366 extern void tcp_v4_destroy_sock(struct sock *sk);
1368 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1369 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1371 #ifdef CONFIG_PROC_FS
1372 extern int tcp4_proc_init(void);
1373 extern void tcp4_proc_exit(void);
1376 /* TCP af-specific functions */
1377 struct tcp_sock_af_ops {
1378 #ifdef CONFIG_TCP_MD5SIG
1379 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1380 struct sock *addr_sk);
1381 int (*calc_md5_hash) (char *location,
1382 struct tcp_md5sig_key *md5,
1384 struct dst_entry *dst,
1385 struct request_sock *req,
1388 int (*md5_add) (struct sock *sk,
1389 struct sock *addr_sk,
1392 int (*md5_parse) (struct sock *sk,
1393 char __user *optval,
1398 struct tcp_request_sock_ops {
1399 #ifdef CONFIG_TCP_MD5SIG
1400 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1401 struct request_sock *req);
1405 extern void tcp_v4_init(void);
1406 extern void tcp_init(void);