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
142 #define MAX_TCP_ACCEPT_DEFERRED 65535
144 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
146 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
147 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
148 * after this time. It should be equal
149 * (or greater than) TCP_TIMEWAIT_LEN
150 * to provide reliability equal to one
151 * provided by timewait state.
153 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
154 * timestamps. It must be less than
155 * minimal timewait lifetime.
161 #define TCPOPT_NOP 1 /* Padding */
162 #define TCPOPT_EOL 0 /* End of options */
163 #define TCPOPT_MSS 2 /* Segment size negotiating */
164 #define TCPOPT_WINDOW 3 /* Window scaling */
165 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
166 #define TCPOPT_SACK 5 /* SACK Block */
167 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
168 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
174 #define TCPOLEN_MSS 4
175 #define TCPOLEN_WINDOW 3
176 #define TCPOLEN_SACK_PERM 2
177 #define TCPOLEN_TIMESTAMP 10
178 #define TCPOLEN_MD5SIG 18
180 /* But this is what stacks really send out. */
181 #define TCPOLEN_TSTAMP_ALIGNED 12
182 #define TCPOLEN_WSCALE_ALIGNED 4
183 #define TCPOLEN_SACKPERM_ALIGNED 4
184 #define TCPOLEN_SACK_BASE 2
185 #define TCPOLEN_SACK_BASE_ALIGNED 4
186 #define TCPOLEN_SACK_PERBLOCK 8
187 #define TCPOLEN_MD5SIG_ALIGNED 20
189 /* Flags in tp->nonagle */
190 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
191 #define TCP_NAGLE_CORK 2 /* Socket is corked */
192 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
194 extern struct inet_timewait_death_row tcp_death_row;
196 /* sysctl variables for tcp */
197 extern int sysctl_tcp_timestamps;
198 extern int sysctl_tcp_window_scaling;
199 extern int sysctl_tcp_sack;
200 extern int sysctl_tcp_fin_timeout;
201 extern int sysctl_tcp_keepalive_time;
202 extern int sysctl_tcp_keepalive_probes;
203 extern int sysctl_tcp_keepalive_intvl;
204 extern int sysctl_tcp_syn_retries;
205 extern int sysctl_tcp_synack_retries;
206 extern int sysctl_tcp_retries1;
207 extern int sysctl_tcp_retries2;
208 extern int sysctl_tcp_orphan_retries;
209 extern int sysctl_tcp_syncookies;
210 extern int sysctl_tcp_retrans_collapse;
211 extern int sysctl_tcp_stdurg;
212 extern int sysctl_tcp_rfc1337;
213 extern int sysctl_tcp_abort_on_overflow;
214 extern int sysctl_tcp_max_orphans;
215 extern int sysctl_tcp_fack;
216 extern int sysctl_tcp_reordering;
217 extern int sysctl_tcp_ecn;
218 extern int sysctl_tcp_dsack;
219 extern int sysctl_tcp_mem[3];
220 extern int sysctl_tcp_wmem[3];
221 extern int sysctl_tcp_rmem[3];
222 extern int sysctl_tcp_app_win;
223 extern int sysctl_tcp_adv_win_scale;
224 extern int sysctl_tcp_tw_reuse;
225 extern int sysctl_tcp_frto;
226 extern int sysctl_tcp_frto_response;
227 extern int sysctl_tcp_low_latency;
228 extern int sysctl_tcp_dma_copybreak;
229 extern int sysctl_tcp_nometrics_save;
230 extern int sysctl_tcp_moderate_rcvbuf;
231 extern int sysctl_tcp_tso_win_divisor;
232 extern int sysctl_tcp_abc;
233 extern int sysctl_tcp_mtu_probing;
234 extern int sysctl_tcp_base_mss;
235 extern int sysctl_tcp_workaround_signed_windows;
236 extern int sysctl_tcp_slow_start_after_idle;
237 extern int sysctl_tcp_max_ssthresh;
239 extern atomic_t tcp_memory_allocated;
240 extern atomic_t tcp_sockets_allocated;
241 extern int tcp_memory_pressure;
244 * The next routines deal with comparing 32 bit unsigned ints
245 * and worry about wraparound (automatic with unsigned arithmetic).
248 static inline int before(__u32 seq1, __u32 seq2)
250 return (__s32)(seq1-seq2) < 0;
252 #define after(seq2, seq1) before(seq1, seq2)
254 /* is s2<=s1<=s3 ? */
255 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
257 return seq3 - seq2 >= seq1 - seq2;
260 static inline int tcp_too_many_orphans(struct sock *sk, int num)
262 return (num > sysctl_tcp_max_orphans) ||
263 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
264 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
267 extern struct proto tcp_prot;
269 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
270 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
271 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
272 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
273 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
274 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
275 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
277 extern void tcp_v4_err(struct sk_buff *skb, u32);
279 extern void tcp_shutdown (struct sock *sk, int how);
281 extern int tcp_v4_rcv(struct sk_buff *skb);
283 extern int tcp_v4_remember_stamp(struct sock *sk);
285 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
287 extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
288 struct msghdr *msg, size_t size);
289 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
291 extern int tcp_ioctl(struct sock *sk,
295 extern int tcp_rcv_state_process(struct sock *sk,
300 extern int tcp_rcv_established(struct sock *sk,
305 extern void tcp_rcv_space_adjust(struct sock *sk);
307 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
309 extern int tcp_twsk_unique(struct sock *sk,
310 struct sock *sktw, void *twp);
312 extern void tcp_twsk_destructor(struct sock *sk);
314 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
315 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
317 static inline void tcp_dec_quickack_mode(struct sock *sk,
318 const unsigned int pkts)
320 struct inet_connection_sock *icsk = inet_csk(sk);
322 if (icsk->icsk_ack.quick) {
323 if (pkts >= icsk->icsk_ack.quick) {
324 icsk->icsk_ack.quick = 0;
325 /* Leaving quickack mode we deflate ATO. */
326 icsk->icsk_ack.ato = TCP_ATO_MIN;
328 icsk->icsk_ack.quick -= pkts;
332 extern void tcp_enter_quickack_mode(struct sock *sk);
334 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
336 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
340 #define TCP_ECN_QUEUE_CWR 2
341 #define TCP_ECN_DEMAND_CWR 4
343 static __inline__ void
344 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
346 if (sysctl_tcp_ecn && th->ece && th->cwr)
347 inet_rsk(req)->ecn_ok = 1;
359 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
361 const struct tcphdr *th);
363 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
364 struct request_sock *req,
365 struct request_sock **prev);
366 extern int tcp_child_process(struct sock *parent,
368 struct sk_buff *skb);
369 extern int tcp_use_frto(struct sock *sk);
370 extern void tcp_enter_frto(struct sock *sk);
371 extern void tcp_enter_loss(struct sock *sk, int how);
372 extern void tcp_clear_retrans(struct tcp_sock *tp);
373 extern void tcp_update_metrics(struct sock *sk);
375 extern void tcp_close(struct sock *sk,
377 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
379 extern int tcp_getsockopt(struct sock *sk, int level,
383 extern int tcp_setsockopt(struct sock *sk, int level,
384 int optname, char __user *optval,
386 extern int compat_tcp_getsockopt(struct sock *sk,
387 int level, int optname,
388 char __user *optval, int __user *optlen);
389 extern int compat_tcp_setsockopt(struct sock *sk,
390 int level, int optname,
391 char __user *optval, int optlen);
392 extern void tcp_set_keepalive(struct sock *sk, int val);
393 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
395 size_t len, int nonblock,
396 int flags, int *addr_len);
398 extern void tcp_parse_options(struct sk_buff *skb,
399 struct tcp_options_received *opt_rx,
402 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
405 * TCP v4 functions exported for the inet6 API
408 extern void tcp_v4_send_check(struct sock *sk, int len,
409 struct sk_buff *skb);
411 extern int tcp_v4_conn_request(struct sock *sk,
412 struct sk_buff *skb);
414 extern struct sock * tcp_create_openreq_child(struct sock *sk,
415 struct request_sock *req,
416 struct sk_buff *skb);
418 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
420 struct request_sock *req,
421 struct dst_entry *dst);
423 extern int tcp_v4_do_rcv(struct sock *sk,
424 struct sk_buff *skb);
426 extern int tcp_v4_connect(struct sock *sk,
427 struct sockaddr *uaddr,
430 extern int tcp_connect(struct sock *sk);
432 extern struct sk_buff * tcp_make_synack(struct sock *sk,
433 struct dst_entry *dst,
434 struct request_sock *req);
436 extern int tcp_disconnect(struct sock *sk, int flags);
438 extern void tcp_unhash(struct sock *sk);
440 /* From syncookies.c */
441 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
442 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
443 struct ip_options *opt);
444 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
447 extern __u32 cookie_init_timestamp(struct request_sock *req);
448 extern void cookie_check_timestamp(struct tcp_options_received *tcp_opt);
450 /* From net/ipv6/syncookies.c */
451 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
452 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
457 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
459 extern int tcp_may_send_now(struct sock *sk);
460 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
461 extern void tcp_xmit_retransmit_queue(struct sock *);
462 extern void tcp_simple_retransmit(struct sock *);
463 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
464 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
466 extern void tcp_send_probe0(struct sock *);
467 extern void tcp_send_partial(struct sock *);
468 extern int tcp_write_wakeup(struct sock *);
469 extern void tcp_send_fin(struct sock *sk);
470 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
471 extern int tcp_send_synack(struct sock *);
472 extern void tcp_push_one(struct sock *, unsigned int mss_now);
473 extern void tcp_send_ack(struct sock *sk);
474 extern void tcp_send_delayed_ack(struct sock *sk);
477 extern void tcp_cwnd_application_limited(struct sock *sk);
480 extern void tcp_init_xmit_timers(struct sock *);
481 static inline void tcp_clear_xmit_timers(struct sock *sk)
483 inet_csk_clear_xmit_timers(sk);
486 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
487 extern unsigned int tcp_current_mss(struct sock *sk, int large);
490 extern void tcp_get_info(struct sock *, struct tcp_info *);
492 /* Read 'sendfile()'-style from a TCP socket */
493 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
494 unsigned int, size_t);
495 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
496 sk_read_actor_t recv_actor);
498 extern void tcp_initialize_rcv_mss(struct sock *sk);
500 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
501 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
502 extern void tcp_mtup_init(struct sock *sk);
504 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
506 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
507 ntohl(TCP_FLAG_ACK) |
511 static inline void tcp_fast_path_on(struct tcp_sock *tp)
513 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
516 static inline void tcp_fast_path_check(struct sock *sk)
518 struct tcp_sock *tp = tcp_sk(sk);
520 if (skb_queue_empty(&tp->out_of_order_queue) &&
522 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
524 tcp_fast_path_on(tp);
527 /* Compute the actual receive window we are currently advertising.
528 * Rcv_nxt can be after the window if our peer push more data
529 * than the offered window.
531 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
533 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
540 /* Choose a new window, without checks for shrinking, and without
541 * scaling applied to the result. The caller does these things
542 * if necessary. This is a "raw" window selection.
544 extern u32 __tcp_select_window(struct sock *sk);
546 /* TCP timestamps are only 32-bits, this causes a slight
547 * complication on 64-bit systems since we store a snapshot
548 * of jiffies in the buffer control blocks below. We decided
549 * to use only the low 32-bits of jiffies and hide the ugly
550 * casts with the following macro.
552 #define tcp_time_stamp ((__u32)(jiffies))
554 /* This is what the send packet queuing engine uses to pass
555 * TCP per-packet control information to the transmission
556 * code. We also store the host-order sequence numbers in
557 * here too. This is 36 bytes on 32-bit architectures,
558 * 40 bytes on 64-bit machines, if this grows please adjust
559 * skbuff.h:skbuff->cb[xxx] size appropriately.
563 struct inet_skb_parm h4;
564 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
565 struct inet6_skb_parm h6;
567 } header; /* For incoming frames */
568 __u32 seq; /* Starting sequence number */
569 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
570 __u32 when; /* used to compute rtt's */
571 __u8 flags; /* TCP header flags. */
573 /* NOTE: These must match up to the flags byte in a
576 #define TCPCB_FLAG_FIN 0x01
577 #define TCPCB_FLAG_SYN 0x02
578 #define TCPCB_FLAG_RST 0x04
579 #define TCPCB_FLAG_PSH 0x08
580 #define TCPCB_FLAG_ACK 0x10
581 #define TCPCB_FLAG_URG 0x20
582 #define TCPCB_FLAG_ECE 0x40
583 #define TCPCB_FLAG_CWR 0x80
585 __u8 sacked; /* State flags for SACK/FACK. */
586 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
587 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
588 #define TCPCB_LOST 0x04 /* SKB is lost */
589 #define TCPCB_TAGBITS 0x07 /* All tag bits */
591 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
592 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
594 __u16 urg_ptr; /* Valid w/URG flags is set. */
595 __u32 ack_seq; /* Sequence number ACK'd */
598 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
600 /* Due to TSO, an SKB can be composed of multiple actual
601 * packets. To keep these tracked properly, we use this.
603 static inline int tcp_skb_pcount(const struct sk_buff *skb)
605 return skb_shinfo(skb)->gso_segs;
608 /* This is valid iff tcp_skb_pcount() > 1. */
609 static inline int tcp_skb_mss(const struct sk_buff *skb)
611 return skb_shinfo(skb)->gso_size;
614 static inline void tcp_dec_pcount_approx_int(__u32 *count, const int decr)
623 static inline void tcp_dec_pcount_approx(__u32 *count,
624 const struct sk_buff *skb)
626 tcp_dec_pcount_approx_int(count, tcp_skb_pcount(skb));
629 /* Events passed to congestion control interface */
631 CA_EVENT_TX_START, /* first transmit when no packets in flight */
632 CA_EVENT_CWND_RESTART, /* congestion window restart */
633 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
634 CA_EVENT_FRTO, /* fast recovery timeout */
635 CA_EVENT_LOSS, /* loss timeout */
636 CA_EVENT_FAST_ACK, /* in sequence ack */
637 CA_EVENT_SLOW_ACK, /* other ack */
641 * Interface for adding new TCP congestion control handlers
643 #define TCP_CA_NAME_MAX 16
644 #define TCP_CA_MAX 128
645 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
647 #define TCP_CONG_NON_RESTRICTED 0x1
648 #define TCP_CONG_RTT_STAMP 0x2
650 struct tcp_congestion_ops {
651 struct list_head list;
654 /* initialize private data (optional) */
655 void (*init)(struct sock *sk);
656 /* cleanup private data (optional) */
657 void (*release)(struct sock *sk);
659 /* return slow start threshold (required) */
660 u32 (*ssthresh)(struct sock *sk);
661 /* lower bound for congestion window (optional) */
662 u32 (*min_cwnd)(const struct sock *sk);
663 /* do new cwnd calculation (required) */
664 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
665 /* call before changing ca_state (optional) */
666 void (*set_state)(struct sock *sk, u8 new_state);
667 /* call when cwnd event occurs (optional) */
668 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
669 /* new value of cwnd after loss (optional) */
670 u32 (*undo_cwnd)(struct sock *sk);
671 /* hook for packet ack accounting (optional) */
672 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
673 /* get info for inet_diag (optional) */
674 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
676 char name[TCP_CA_NAME_MAX];
677 struct module *owner;
680 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
681 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
683 extern void tcp_init_congestion_control(struct sock *sk);
684 extern void tcp_cleanup_congestion_control(struct sock *sk);
685 extern int tcp_set_default_congestion_control(const char *name);
686 extern void tcp_get_default_congestion_control(char *name);
687 extern void tcp_get_available_congestion_control(char *buf, size_t len);
688 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
689 extern int tcp_set_allowed_congestion_control(char *allowed);
690 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
691 extern void tcp_slow_start(struct tcp_sock *tp);
693 extern struct tcp_congestion_ops tcp_init_congestion_ops;
694 extern u32 tcp_reno_ssthresh(struct sock *sk);
695 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
696 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
697 extern struct tcp_congestion_ops tcp_reno;
699 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
701 struct inet_connection_sock *icsk = inet_csk(sk);
703 if (icsk->icsk_ca_ops->set_state)
704 icsk->icsk_ca_ops->set_state(sk, ca_state);
705 icsk->icsk_ca_state = ca_state;
708 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
710 const struct inet_connection_sock *icsk = inet_csk(sk);
712 if (icsk->icsk_ca_ops->cwnd_event)
713 icsk->icsk_ca_ops->cwnd_event(sk, event);
716 /* These functions determine how the current flow behaves in respect of SACK
717 * handling. SACK is negotiated with the peer, and therefore it can vary
718 * between different flows.
720 * tcp_is_sack - SACK enabled
721 * tcp_is_reno - No SACK
722 * tcp_is_fack - FACK enabled, implies SACK enabled
724 static inline int tcp_is_sack(const struct tcp_sock *tp)
726 return tp->rx_opt.sack_ok;
729 static inline int tcp_is_reno(const struct tcp_sock *tp)
731 return !tcp_is_sack(tp);
734 static inline int tcp_is_fack(const struct tcp_sock *tp)
736 return tp->rx_opt.sack_ok & 2;
739 static inline void tcp_enable_fack(struct tcp_sock *tp)
741 tp->rx_opt.sack_ok |= 2;
744 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
746 return tp->sacked_out + tp->lost_out;
749 /* This determines how many packets are "in the network" to the best
750 * of our knowledge. In many cases it is conservative, but where
751 * detailed information is available from the receiver (via SACK
752 * blocks etc.) we can make more aggressive calculations.
754 * Use this for decisions involving congestion control, use just
755 * tp->packets_out to determine if the send queue is empty or not.
757 * Read this equation as:
759 * "Packets sent once on transmission queue" MINUS
760 * "Packets left network, but not honestly ACKed yet" PLUS
761 * "Packets fast retransmitted"
763 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
765 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
768 extern int tcp_limit_reno_sacked(struct tcp_sock *tp);
770 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
771 * The exception is rate halving phase, when cwnd is decreasing towards
774 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
776 const struct tcp_sock *tp = tcp_sk(sk);
777 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
778 return tp->snd_ssthresh;
780 return max(tp->snd_ssthresh,
781 ((tp->snd_cwnd >> 1) +
782 (tp->snd_cwnd >> 2)));
785 /* Use define here intentionally to get WARN_ON location shown at the caller */
786 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
788 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
789 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
791 /* Slow start with delack produces 3 packets of burst, so that
792 * it is safe "de facto". This will be the default - same as
793 * the default reordering threshold - but if reordering increases,
794 * we must be able to allow cwnd to burst at least this much in order
795 * to not pull it back when holes are filled.
797 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
799 return tp->reordering;
802 /* Returns end sequence number of the receiver's advertised window */
803 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
805 return tp->snd_una + tp->snd_wnd;
807 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
809 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
810 const struct sk_buff *skb)
813 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
816 static inline void tcp_check_probe_timer(struct sock *sk)
818 struct tcp_sock *tp = tcp_sk(sk);
819 const struct inet_connection_sock *icsk = inet_csk(sk);
821 if (!tp->packets_out && !icsk->icsk_pending)
822 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
823 icsk->icsk_rto, TCP_RTO_MAX);
826 static inline void tcp_push_pending_frames(struct sock *sk)
828 struct tcp_sock *tp = tcp_sk(sk);
830 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle);
833 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
838 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
844 * Calculate(/check) TCP checksum
846 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
847 __be32 daddr, __wsum base)
849 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
852 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
854 return __skb_checksum_complete(skb);
857 static inline int tcp_checksum_complete(struct sk_buff *skb)
859 return !skb_csum_unnecessary(skb) &&
860 __tcp_checksum_complete(skb);
863 /* Prequeue for VJ style copy to user, combined with checksumming. */
865 static inline void tcp_prequeue_init(struct tcp_sock *tp)
867 tp->ucopy.task = NULL;
869 tp->ucopy.memory = 0;
870 skb_queue_head_init(&tp->ucopy.prequeue);
871 #ifdef CONFIG_NET_DMA
872 tp->ucopy.dma_chan = NULL;
873 tp->ucopy.wakeup = 0;
874 tp->ucopy.pinned_list = NULL;
875 tp->ucopy.dma_cookie = 0;
879 /* Packet is added to VJ-style prequeue for processing in process
880 * context, if a reader task is waiting. Apparently, this exciting
881 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
882 * failed somewhere. Latency? Burstiness? Well, at least now we will
883 * see, why it failed. 8)8) --ANK
885 * NOTE: is this not too big to inline?
887 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
889 struct tcp_sock *tp = tcp_sk(sk);
891 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
892 __skb_queue_tail(&tp->ucopy.prequeue, skb);
893 tp->ucopy.memory += skb->truesize;
894 if (tp->ucopy.memory > sk->sk_rcvbuf) {
895 struct sk_buff *skb1;
897 BUG_ON(sock_owned_by_user(sk));
899 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
900 sk->sk_backlog_rcv(sk, skb1);
901 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
904 tp->ucopy.memory = 0;
905 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
906 wake_up_interruptible(sk->sk_sleep);
907 if (!inet_csk_ack_scheduled(sk))
908 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
909 (3 * TCP_RTO_MIN) / 4,
921 static const char *statename[]={
922 "Unused","Established","Syn Sent","Syn Recv",
923 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
924 "Close Wait","Last ACK","Listen","Closing"
927 extern void tcp_set_state(struct sock *sk, int state);
929 extern void tcp_done(struct sock *sk);
931 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
934 rx_opt->eff_sacks = 0;
935 rx_opt->num_sacks = 0;
938 /* Determine a window scaling and initial window to offer. */
939 extern void tcp_select_initial_window(int __space, __u32 mss,
940 __u32 *rcv_wnd, __u32 *window_clamp,
941 int wscale_ok, __u8 *rcv_wscale);
943 static inline int tcp_win_from_space(int space)
945 return sysctl_tcp_adv_win_scale<=0 ?
946 (space>>(-sysctl_tcp_adv_win_scale)) :
947 space - (space>>sysctl_tcp_adv_win_scale);
950 /* Note: caller must be prepared to deal with negative returns */
951 static inline int tcp_space(const struct sock *sk)
953 return tcp_win_from_space(sk->sk_rcvbuf -
954 atomic_read(&sk->sk_rmem_alloc));
957 static inline int tcp_full_space(const struct sock *sk)
959 return tcp_win_from_space(sk->sk_rcvbuf);
962 static inline void tcp_openreq_init(struct request_sock *req,
963 struct tcp_options_received *rx_opt,
966 struct inet_request_sock *ireq = inet_rsk(req);
968 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
970 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
971 req->mss = rx_opt->mss_clamp;
972 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
973 ireq->tstamp_ok = rx_opt->tstamp_ok;
974 ireq->sack_ok = rx_opt->sack_ok;
975 ireq->snd_wscale = rx_opt->snd_wscale;
976 ireq->wscale_ok = rx_opt->wscale_ok;
979 ireq->rmt_port = tcp_hdr(skb)->source;
982 extern void tcp_enter_memory_pressure(void);
984 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
986 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
989 static inline int keepalive_time_when(const struct tcp_sock *tp)
991 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
994 static inline int tcp_fin_time(const struct sock *sk)
996 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
997 const int rto = inet_csk(sk)->icsk_rto;
999 if (fin_timeout < (rto << 2) - (rto >> 1))
1000 fin_timeout = (rto << 2) - (rto >> 1);
1005 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1007 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1009 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1012 /* RST segments are not recommended to carry timestamp,
1013 and, if they do, it is recommended to ignore PAWS because
1014 "their cleanup function should take precedence over timestamps."
1015 Certainly, it is mistake. It is necessary to understand the reasons
1016 of this constraint to relax it: if peer reboots, clock may go
1017 out-of-sync and half-open connections will not be reset.
1018 Actually, the problem would be not existing if all
1019 the implementations followed draft about maintaining clock
1020 via reboots. Linux-2.2 DOES NOT!
1022 However, we can relax time bounds for RST segments to MSL.
1024 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1029 #define TCP_CHECK_TIMER(sk) do { } while (0)
1031 static inline void tcp_mib_init(void)
1034 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1035 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1036 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1037 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1041 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1043 tp->lost_skb_hint = NULL;
1044 tp->scoreboard_skb_hint = NULL;
1045 tp->retransmit_skb_hint = NULL;
1046 tp->forward_skb_hint = NULL;
1049 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1051 tcp_clear_retrans_hints_partial(tp);
1057 /* - key database */
1058 struct tcp_md5sig_key {
1063 struct tcp4_md5sig_key {
1064 struct tcp_md5sig_key base;
1068 struct tcp6_md5sig_key {
1069 struct tcp_md5sig_key base;
1071 u32 scope_id; /* XXX */
1073 struct in6_addr addr;
1077 struct tcp_md5sig_info {
1078 struct tcp4_md5sig_key *keys4;
1079 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1080 struct tcp6_md5sig_key *keys6;
1088 /* - pseudo header */
1089 struct tcp4_pseudohdr {
1097 struct tcp6_pseudohdr {
1098 struct in6_addr saddr;
1099 struct in6_addr daddr;
1101 __be32 protocol; /* including padding */
1104 union tcp_md5sum_block {
1105 struct tcp4_pseudohdr ip4;
1106 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1107 struct tcp6_pseudohdr ip6;
1111 /* - pool: digest algorithm, hash description and scratch buffer */
1112 struct tcp_md5sig_pool {
1113 struct hash_desc md5_desc;
1114 union tcp_md5sum_block md5_blk;
1117 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1120 extern int tcp_v4_calc_md5_hash(char *md5_hash,
1121 struct tcp_md5sig_key *key,
1123 struct dst_entry *dst,
1124 struct request_sock *req,
1127 unsigned int tcplen);
1128 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1129 struct sock *addr_sk);
1131 extern int tcp_v4_md5_do_add(struct sock *sk,
1136 extern int tcp_v4_md5_do_del(struct sock *sk,
1139 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1140 extern void tcp_free_md5sig_pool(void);
1142 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1143 extern void __tcp_put_md5sig_pool(void);
1146 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1148 int cpu = get_cpu();
1149 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1155 static inline void tcp_put_md5sig_pool(void)
1157 __tcp_put_md5sig_pool();
1161 /* write queue abstraction */
1162 static inline void tcp_write_queue_purge(struct sock *sk)
1164 struct sk_buff *skb;
1166 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1167 sk_wmem_free_skb(sk, skb);
1171 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1173 struct sk_buff *skb = sk->sk_write_queue.next;
1174 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1179 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1181 struct sk_buff *skb = sk->sk_write_queue.prev;
1182 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1187 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1192 #define tcp_for_write_queue(skb, sk) \
1193 for (skb = (sk)->sk_write_queue.next; \
1194 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1197 #define tcp_for_write_queue_from(skb, sk) \
1198 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\
1201 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1202 for (tmp = skb->next; \
1203 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1204 skb = tmp, tmp = skb->next)
1206 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1208 return sk->sk_send_head;
1211 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1213 sk->sk_send_head = skb->next;
1214 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue)
1215 sk->sk_send_head = NULL;
1218 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1220 if (sk->sk_send_head == skb_unlinked)
1221 sk->sk_send_head = NULL;
1224 static inline void tcp_init_send_head(struct sock *sk)
1226 sk->sk_send_head = NULL;
1229 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1231 __skb_queue_tail(&sk->sk_write_queue, skb);
1234 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1236 __tcp_add_write_queue_tail(sk, skb);
1238 /* Queue it, remembering where we must start sending. */
1239 if (sk->sk_send_head == NULL) {
1240 sk->sk_send_head = skb;
1242 if (tcp_sk(sk)->highest_sack == NULL)
1243 tcp_sk(sk)->highest_sack = skb;
1247 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1249 __skb_queue_head(&sk->sk_write_queue, skb);
1252 /* Insert buff after skb on the write queue of sk. */
1253 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1254 struct sk_buff *buff,
1257 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1260 /* Insert skb between prev and next on the write queue of sk. */
1261 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1262 struct sk_buff *skb,
1265 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue);
1267 if (sk->sk_send_head == skb)
1268 sk->sk_send_head = new;
1271 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1273 __skb_unlink(skb, &sk->sk_write_queue);
1276 static inline int tcp_skb_is_last(const struct sock *sk,
1277 const struct sk_buff *skb)
1279 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1282 static inline int tcp_write_queue_empty(struct sock *sk)
1284 return skb_queue_empty(&sk->sk_write_queue);
1287 /* Start sequence of the highest skb with SACKed bit, valid only if
1288 * sacked > 0 or when the caller has ensured validity by itself.
1290 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1292 if (!tp->sacked_out)
1295 if (tp->highest_sack == NULL)
1298 return TCP_SKB_CB(tp->highest_sack)->seq;
1301 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1303 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1304 tcp_write_queue_next(sk, skb);
1307 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1309 return tcp_sk(sk)->highest_sack;
1312 static inline void tcp_highest_sack_reset(struct sock *sk)
1314 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1317 /* Called when old skb is about to be deleted (to be combined with new skb) */
1318 static inline void tcp_highest_sack_combine(struct sock *sk,
1319 struct sk_buff *old,
1320 struct sk_buff *new)
1322 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1323 tcp_sk(sk)->highest_sack = new;
1327 enum tcp_seq_states {
1328 TCP_SEQ_STATE_LISTENING,
1329 TCP_SEQ_STATE_OPENREQ,
1330 TCP_SEQ_STATE_ESTABLISHED,
1331 TCP_SEQ_STATE_TIME_WAIT,
1334 struct tcp_seq_afinfo {
1337 struct file_operations seq_fops;
1338 struct seq_operations seq_ops;
1341 struct tcp_iter_state {
1342 struct seq_net_private p;
1344 enum tcp_seq_states state;
1345 struct sock *syn_wait_sk;
1346 int bucket, sbucket, num, uid;
1349 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1350 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1352 extern struct request_sock_ops tcp_request_sock_ops;
1353 extern struct request_sock_ops tcp6_request_sock_ops;
1355 extern int tcp_v4_destroy_sock(struct sock *sk);
1357 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1358 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1360 #ifdef CONFIG_PROC_FS
1361 extern int tcp4_proc_init(void);
1362 extern void tcp4_proc_exit(void);
1365 /* TCP af-specific functions */
1366 struct tcp_sock_af_ops {
1367 #ifdef CONFIG_TCP_MD5SIG
1368 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1369 struct sock *addr_sk);
1370 int (*calc_md5_hash) (char *location,
1371 struct tcp_md5sig_key *md5,
1373 struct dst_entry *dst,
1374 struct request_sock *req,
1378 int (*md5_add) (struct sock *sk,
1379 struct sock *addr_sk,
1382 int (*md5_parse) (struct sock *sk,
1383 char __user *optval,
1388 struct tcp_request_sock_ops {
1389 #ifdef CONFIG_TCP_MD5SIG
1390 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1391 struct request_sock *req);
1395 extern void tcp_v4_init(void);
1396 extern void tcp_init(void);