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/config.h>
25 #include <linux/list.h>
26 #include <linux/tcp.h>
27 #include <linux/slab.h>
28 #include <linux/cache.h>
29 #include <linux/percpu.h>
30 #include <linux/skbuff.h>
31 #include <linux/dmaengine.h>
33 #include <net/inet_connection_sock.h>
34 #include <net/inet_timewait_sock.h>
35 #include <net/inet_hashtables.h>
36 #include <net/checksum.h>
37 #include <net/request_sock.h>
41 #include <net/tcp_states.h>
43 #include <linux/seq_file.h>
45 extern struct inet_hashinfo tcp_hashinfo;
47 extern atomic_t tcp_orphan_count;
48 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
50 #define MAX_TCP_HEADER (128 + MAX_HEADER)
53 * Never offer a window over 32767 without using window scaling. Some
54 * poor stacks do signed 16bit maths!
56 #define MAX_TCP_WINDOW 32767U
58 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
59 #define TCP_MIN_MSS 88U
61 /* Minimal RCV_MSS. */
62 #define TCP_MIN_RCVMSS 536U
64 /* The least MTU to use for probing */
65 #define TCP_BASE_MSS 512
67 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
68 #define TCP_FASTRETRANS_THRESH 3
70 /* Maximal reordering. */
71 #define TCP_MAX_REORDERING 127
73 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
74 #define TCP_MAX_QUICKACKS 16U
77 #define TCP_URG_VALID 0x0100
78 #define TCP_URG_NOTYET 0x0200
79 #define TCP_URG_READ 0x0400
81 #define TCP_RETR1 3 /*
82 * This is how many retries it does before it
83 * tries to figure out if the gateway is
84 * down. Minimal RFC value is 3; it corresponds
85 * to ~3sec-8min depending on RTO.
88 #define TCP_RETR2 15 /*
89 * This should take at least
90 * 90 minutes to time out.
91 * RFC1122 says that the limit is 100 sec.
92 * 15 is ~13-30min depending on RTO.
95 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
96 * connection: ~180sec is RFC minimum */
98 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
99 * connection: ~180sec is RFC minimum */
102 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
103 * socket. 7 is ~50sec-16min.
107 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
108 * state, about 60 seconds */
109 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
110 /* BSD style FIN_WAIT2 deadlock breaker.
111 * It used to be 3min, new value is 60sec,
112 * to combine FIN-WAIT-2 timeout with
116 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
118 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
119 #define TCP_ATO_MIN ((unsigned)(HZ/25))
121 #define TCP_DELACK_MIN 4U
122 #define TCP_ATO_MIN 4U
124 #define TCP_RTO_MAX ((unsigned)(120*HZ))
125 #define TCP_RTO_MIN ((unsigned)(HZ/5))
126 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
128 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
129 * for local resources.
132 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
133 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
134 #define TCP_KEEPALIVE_INTVL (75*HZ)
136 #define MAX_TCP_KEEPIDLE 32767
137 #define MAX_TCP_KEEPINTVL 32767
138 #define MAX_TCP_KEEPCNT 127
139 #define MAX_TCP_SYNCNT 127
141 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
142 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
144 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
145 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
146 * after this time. It should be equal
147 * (or greater than) TCP_TIMEWAIT_LEN
148 * to provide reliability equal to one
149 * provided by timewait state.
151 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
152 * timestamps. It must be less than
153 * minimal timewait lifetime.
159 #define TCPOPT_NOP 1 /* Padding */
160 #define TCPOPT_EOL 0 /* End of options */
161 #define TCPOPT_MSS 2 /* Segment size negotiating */
162 #define TCPOPT_WINDOW 3 /* Window scaling */
163 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
164 #define TCPOPT_SACK 5 /* SACK Block */
165 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
171 #define TCPOLEN_MSS 4
172 #define TCPOLEN_WINDOW 3
173 #define TCPOLEN_SACK_PERM 2
174 #define TCPOLEN_TIMESTAMP 10
176 /* But this is what stacks really send out. */
177 #define TCPOLEN_TSTAMP_ALIGNED 12
178 #define TCPOLEN_WSCALE_ALIGNED 4
179 #define TCPOLEN_SACKPERM_ALIGNED 4
180 #define TCPOLEN_SACK_BASE 2
181 #define TCPOLEN_SACK_BASE_ALIGNED 4
182 #define TCPOLEN_SACK_PERBLOCK 8
184 /* Flags in tp->nonagle */
185 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
186 #define TCP_NAGLE_CORK 2 /* Socket is corked */
187 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
189 extern struct inet_timewait_death_row tcp_death_row;
191 /* sysctl variables for tcp */
192 extern int sysctl_tcp_timestamps;
193 extern int sysctl_tcp_window_scaling;
194 extern int sysctl_tcp_sack;
195 extern int sysctl_tcp_fin_timeout;
196 extern int sysctl_tcp_keepalive_time;
197 extern int sysctl_tcp_keepalive_probes;
198 extern int sysctl_tcp_keepalive_intvl;
199 extern int sysctl_tcp_syn_retries;
200 extern int sysctl_tcp_synack_retries;
201 extern int sysctl_tcp_retries1;
202 extern int sysctl_tcp_retries2;
203 extern int sysctl_tcp_orphan_retries;
204 extern int sysctl_tcp_syncookies;
205 extern int sysctl_tcp_retrans_collapse;
206 extern int sysctl_tcp_stdurg;
207 extern int sysctl_tcp_rfc1337;
208 extern int sysctl_tcp_abort_on_overflow;
209 extern int sysctl_tcp_max_orphans;
210 extern int sysctl_tcp_fack;
211 extern int sysctl_tcp_reordering;
212 extern int sysctl_tcp_ecn;
213 extern int sysctl_tcp_dsack;
214 extern int sysctl_tcp_mem[3];
215 extern int sysctl_tcp_wmem[3];
216 extern int sysctl_tcp_rmem[3];
217 extern int sysctl_tcp_app_win;
218 extern int sysctl_tcp_adv_win_scale;
219 extern int sysctl_tcp_tw_reuse;
220 extern int sysctl_tcp_frto;
221 extern int sysctl_tcp_low_latency;
222 extern int sysctl_tcp_dma_copybreak;
223 extern int sysctl_tcp_nometrics_save;
224 extern int sysctl_tcp_moderate_rcvbuf;
225 extern int sysctl_tcp_tso_win_divisor;
226 extern int sysctl_tcp_abc;
227 extern int sysctl_tcp_mtu_probing;
228 extern int sysctl_tcp_base_mss;
229 extern int sysctl_tcp_workaround_signed_windows;
230 extern int sysctl_tcp_slow_start_after_idle;
232 extern atomic_t tcp_memory_allocated;
233 extern atomic_t tcp_sockets_allocated;
234 extern int tcp_memory_pressure;
237 * The next routines deal with comparing 32 bit unsigned ints
238 * and worry about wraparound (automatic with unsigned arithmetic).
241 static inline int before(__u32 seq1, __u32 seq2)
243 return (__s32)(seq1-seq2) < 0;
246 static inline int after(__u32 seq1, __u32 seq2)
248 return (__s32)(seq2-seq1) < 0;
252 /* is s2<=s1<=s3 ? */
253 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
255 return seq3 - seq2 >= seq1 - seq2;
259 extern struct proto tcp_prot;
261 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
262 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
263 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
264 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
265 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
266 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
267 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
269 extern void tcp_v4_err(struct sk_buff *skb, u32);
271 extern void tcp_shutdown (struct sock *sk, int how);
273 extern int tcp_v4_rcv(struct sk_buff *skb);
275 extern int tcp_v4_remember_stamp(struct sock *sk);
277 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
279 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
280 struct msghdr *msg, size_t size);
281 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
283 extern int tcp_ioctl(struct sock *sk,
287 extern int tcp_rcv_state_process(struct sock *sk,
292 extern int tcp_rcv_established(struct sock *sk,
297 extern void tcp_rcv_space_adjust(struct sock *sk);
299 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
301 extern int tcp_twsk_unique(struct sock *sk,
302 struct sock *sktw, void *twp);
304 static inline void tcp_dec_quickack_mode(struct sock *sk,
305 const unsigned int pkts)
307 struct inet_connection_sock *icsk = inet_csk(sk);
309 if (icsk->icsk_ack.quick) {
310 if (pkts >= icsk->icsk_ack.quick) {
311 icsk->icsk_ack.quick = 0;
312 /* Leaving quickack mode we deflate ATO. */
313 icsk->icsk_ack.ato = TCP_ATO_MIN;
315 icsk->icsk_ack.quick -= pkts;
319 extern void tcp_enter_quickack_mode(struct sock *sk);
321 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
323 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
335 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
337 const struct tcphdr *th);
339 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
340 struct request_sock *req,
341 struct request_sock **prev);
342 extern int tcp_child_process(struct sock *parent,
344 struct sk_buff *skb);
345 extern void tcp_enter_frto(struct sock *sk);
346 extern void tcp_enter_loss(struct sock *sk, int how);
347 extern void tcp_clear_retrans(struct tcp_sock *tp);
348 extern void tcp_update_metrics(struct sock *sk);
350 extern void tcp_close(struct sock *sk,
352 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
354 extern int tcp_getsockopt(struct sock *sk, int level,
358 extern int tcp_setsockopt(struct sock *sk, int level,
359 int optname, char __user *optval,
361 extern int compat_tcp_getsockopt(struct sock *sk,
362 int level, int optname,
363 char __user *optval, int __user *optlen);
364 extern int compat_tcp_setsockopt(struct sock *sk,
365 int level, int optname,
366 char __user *optval, int optlen);
367 extern void tcp_set_keepalive(struct sock *sk, int val);
368 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
370 size_t len, int nonblock,
371 int flags, int *addr_len);
373 extern void tcp_parse_options(struct sk_buff *skb,
374 struct tcp_options_received *opt_rx,
378 * TCP v4 functions exported for the inet6 API
381 extern void tcp_v4_send_check(struct sock *sk, int len,
382 struct sk_buff *skb);
384 extern int tcp_v4_conn_request(struct sock *sk,
385 struct sk_buff *skb);
387 extern struct sock * tcp_create_openreq_child(struct sock *sk,
388 struct request_sock *req,
389 struct sk_buff *skb);
391 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
393 struct request_sock *req,
394 struct dst_entry *dst);
396 extern int tcp_v4_do_rcv(struct sock *sk,
397 struct sk_buff *skb);
399 extern int tcp_v4_connect(struct sock *sk,
400 struct sockaddr *uaddr,
403 extern int tcp_connect(struct sock *sk);
405 extern struct sk_buff * tcp_make_synack(struct sock *sk,
406 struct dst_entry *dst,
407 struct request_sock *req);
409 extern int tcp_disconnect(struct sock *sk, int flags);
411 extern void tcp_unhash(struct sock *sk);
413 /* From syncookies.c */
414 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
415 struct ip_options *opt);
416 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
421 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
422 unsigned int cur_mss, int nonagle);
423 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
424 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
425 extern void tcp_xmit_retransmit_queue(struct sock *);
426 extern void tcp_simple_retransmit(struct sock *);
427 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
428 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
430 extern void tcp_send_probe0(struct sock *);
431 extern void tcp_send_partial(struct sock *);
432 extern int tcp_write_wakeup(struct sock *);
433 extern void tcp_send_fin(struct sock *sk);
434 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
435 extern int tcp_send_synack(struct sock *);
436 extern void tcp_push_one(struct sock *, unsigned int mss_now);
437 extern void tcp_send_ack(struct sock *sk);
438 extern void tcp_send_delayed_ack(struct sock *sk);
441 extern void tcp_cwnd_application_limited(struct sock *sk);
444 extern void tcp_init_xmit_timers(struct sock *);
445 static inline void tcp_clear_xmit_timers(struct sock *sk)
447 inet_csk_clear_xmit_timers(sk);
450 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
451 extern unsigned int tcp_current_mss(struct sock *sk, int large);
454 extern void tcp_get_info(struct sock *, struct tcp_info *);
456 /* Read 'sendfile()'-style from a TCP socket */
457 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
458 unsigned int, size_t);
459 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
460 sk_read_actor_t recv_actor);
462 extern void tcp_initialize_rcv_mss(struct sock *sk);
464 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
465 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
466 extern void tcp_mtup_init(struct sock *sk);
468 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
470 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
471 ntohl(TCP_FLAG_ACK) |
475 static inline void tcp_fast_path_on(struct tcp_sock *tp)
477 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
480 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
482 if (skb_queue_empty(&tp->out_of_order_queue) &&
484 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
486 tcp_fast_path_on(tp);
489 /* Compute the actual receive window we are currently advertising.
490 * Rcv_nxt can be after the window if our peer push more data
491 * than the offered window.
493 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
495 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
502 /* Choose a new window, without checks for shrinking, and without
503 * scaling applied to the result. The caller does these things
504 * if necessary. This is a "raw" window selection.
506 extern u32 __tcp_select_window(struct sock *sk);
508 /* TCP timestamps are only 32-bits, this causes a slight
509 * complication on 64-bit systems since we store a snapshot
510 * of jiffies in the buffer control blocks below. We decided
511 * to use only the low 32-bits of jiffies and hide the ugly
512 * casts with the following macro.
514 #define tcp_time_stamp ((__u32)(jiffies))
516 /* This is what the send packet queuing engine uses to pass
517 * TCP per-packet control information to the transmission
518 * code. We also store the host-order sequence numbers in
519 * here too. This is 36 bytes on 32-bit architectures,
520 * 40 bytes on 64-bit machines, if this grows please adjust
521 * skbuff.h:skbuff->cb[xxx] size appropriately.
525 struct inet_skb_parm h4;
526 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
527 struct inet6_skb_parm h6;
529 } header; /* For incoming frames */
530 __u32 seq; /* Starting sequence number */
531 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
532 __u32 when; /* used to compute rtt's */
533 __u8 flags; /* TCP header flags. */
535 /* NOTE: These must match up to the flags byte in a
538 #define TCPCB_FLAG_FIN 0x01
539 #define TCPCB_FLAG_SYN 0x02
540 #define TCPCB_FLAG_RST 0x04
541 #define TCPCB_FLAG_PSH 0x08
542 #define TCPCB_FLAG_ACK 0x10
543 #define TCPCB_FLAG_URG 0x20
544 #define TCPCB_FLAG_ECE 0x40
545 #define TCPCB_FLAG_CWR 0x80
547 __u8 sacked; /* State flags for SACK/FACK. */
548 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
549 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
550 #define TCPCB_LOST 0x04 /* SKB is lost */
551 #define TCPCB_TAGBITS 0x07 /* All tag bits */
553 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
554 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
556 #define TCPCB_URG 0x20 /* Urgent pointer advanced here */
558 #define TCPCB_AT_TAIL (TCPCB_URG)
560 __u16 urg_ptr; /* Valid w/URG flags is set. */
561 __u32 ack_seq; /* Sequence number ACK'd */
564 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
566 #include <net/tcp_ecn.h>
568 /* Due to TSO, an SKB can be composed of multiple actual
569 * packets. To keep these tracked properly, we use this.
571 static inline int tcp_skb_pcount(const struct sk_buff *skb)
573 return skb_shinfo(skb)->tso_segs;
576 /* This is valid iff tcp_skb_pcount() > 1. */
577 static inline int tcp_skb_mss(const struct sk_buff *skb)
579 return skb_shinfo(skb)->tso_size;
582 static inline void tcp_dec_pcount_approx(__u32 *count,
583 const struct sk_buff *skb)
586 *count -= tcp_skb_pcount(skb);
592 static inline void tcp_packets_out_inc(struct sock *sk,
594 const struct sk_buff *skb)
596 int orig = tp->packets_out;
598 tp->packets_out += tcp_skb_pcount(skb);
600 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
601 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
604 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
605 const struct sk_buff *skb)
607 tp->packets_out -= tcp_skb_pcount(skb);
610 /* Events passed to congestion control interface */
612 CA_EVENT_TX_START, /* first transmit when no packets in flight */
613 CA_EVENT_CWND_RESTART, /* congestion window restart */
614 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
615 CA_EVENT_FRTO, /* fast recovery timeout */
616 CA_EVENT_LOSS, /* loss timeout */
617 CA_EVENT_FAST_ACK, /* in sequence ack */
618 CA_EVENT_SLOW_ACK, /* other ack */
622 * Interface for adding new TCP congestion control handlers
624 #define TCP_CA_NAME_MAX 16
625 struct tcp_congestion_ops {
626 struct list_head list;
628 /* initialize private data (optional) */
629 void (*init)(struct sock *sk);
630 /* cleanup private data (optional) */
631 void (*release)(struct sock *sk);
633 /* return slow start threshold (required) */
634 u32 (*ssthresh)(struct sock *sk);
635 /* lower bound for congestion window (optional) */
636 u32 (*min_cwnd)(const struct sock *sk);
637 /* do new cwnd calculation (required) */
638 void (*cong_avoid)(struct sock *sk, u32 ack,
639 u32 rtt, u32 in_flight, int good_ack);
640 /* round trip time sample per acked packet (optional) */
641 void (*rtt_sample)(struct sock *sk, u32 usrtt);
642 /* call before changing ca_state (optional) */
643 void (*set_state)(struct sock *sk, u8 new_state);
644 /* call when cwnd event occurs (optional) */
645 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
646 /* new value of cwnd after loss (optional) */
647 u32 (*undo_cwnd)(struct sock *sk);
648 /* hook for packet ack accounting (optional) */
649 void (*pkts_acked)(struct sock *sk, u32 num_acked);
650 /* get info for inet_diag (optional) */
651 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
653 char name[TCP_CA_NAME_MAX];
654 struct module *owner;
657 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
658 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
660 extern void tcp_init_congestion_control(struct sock *sk);
661 extern void tcp_cleanup_congestion_control(struct sock *sk);
662 extern int tcp_set_default_congestion_control(const char *name);
663 extern void tcp_get_default_congestion_control(char *name);
664 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
665 extern void tcp_slow_start(struct tcp_sock *tp);
667 extern struct tcp_congestion_ops tcp_init_congestion_ops;
668 extern u32 tcp_reno_ssthresh(struct sock *sk);
669 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
670 u32 rtt, u32 in_flight, int flag);
671 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
672 extern struct tcp_congestion_ops tcp_reno;
674 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
676 struct inet_connection_sock *icsk = inet_csk(sk);
678 if (icsk->icsk_ca_ops->set_state)
679 icsk->icsk_ca_ops->set_state(sk, ca_state);
680 icsk->icsk_ca_state = ca_state;
683 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
685 const struct inet_connection_sock *icsk = inet_csk(sk);
687 if (icsk->icsk_ca_ops->cwnd_event)
688 icsk->icsk_ca_ops->cwnd_event(sk, event);
691 /* This determines how many packets are "in the network" to the best
692 * of our knowledge. In many cases it is conservative, but where
693 * detailed information is available from the receiver (via SACK
694 * blocks etc.) we can make more aggressive calculations.
696 * Use this for decisions involving congestion control, use just
697 * tp->packets_out to determine if the send queue is empty or not.
699 * Read this equation as:
701 * "Packets sent once on transmission queue" MINUS
702 * "Packets left network, but not honestly ACKed yet" PLUS
703 * "Packets fast retransmitted"
705 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
707 return (tp->packets_out - tp->left_out + tp->retrans_out);
710 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
711 * The exception is rate halving phase, when cwnd is decreasing towards
714 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
716 const struct tcp_sock *tp = tcp_sk(sk);
717 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
718 return tp->snd_ssthresh;
720 return max(tp->snd_ssthresh,
721 ((tp->snd_cwnd >> 1) +
722 (tp->snd_cwnd >> 2)));
725 static inline void tcp_sync_left_out(struct tcp_sock *tp)
727 if (tp->rx_opt.sack_ok &&
728 (tp->sacked_out >= tp->packets_out - tp->lost_out))
729 tp->sacked_out = tp->packets_out - tp->lost_out;
730 tp->left_out = tp->sacked_out + tp->lost_out;
733 extern void tcp_enter_cwr(struct sock *sk);
734 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
736 /* Slow start with delack produces 3 packets of burst, so that
737 * it is safe "de facto".
739 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
744 /* RFC2861 Check whether we are limited by application or congestion window
745 * This is the inverse of cwnd check in tcp_tso_should_defer
747 static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
749 const struct tcp_sock *tp = tcp_sk(sk);
752 if (in_flight >= tp->snd_cwnd)
755 if (!(sk->sk_route_caps & NETIF_F_TSO))
758 left = tp->snd_cwnd - in_flight;
759 if (sysctl_tcp_tso_win_divisor)
760 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
762 return left <= tcp_max_burst(tp);
765 static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
766 const struct sk_buff *skb)
769 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
772 static inline void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
774 const struct inet_connection_sock *icsk = inet_csk(sk);
775 if (!tp->packets_out && !icsk->icsk_pending)
776 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
777 icsk->icsk_rto, TCP_RTO_MAX);
780 static inline void tcp_push_pending_frames(struct sock *sk,
783 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
786 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
791 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
797 * Calculate(/check) TCP checksum
799 static inline u16 tcp_v4_check(struct tcphdr *th, int len,
800 unsigned long saddr, unsigned long daddr,
803 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
806 static inline int __tcp_checksum_complete(struct sk_buff *skb)
808 return __skb_checksum_complete(skb);
811 static inline int tcp_checksum_complete(struct sk_buff *skb)
813 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
814 __tcp_checksum_complete(skb);
817 /* Prequeue for VJ style copy to user, combined with checksumming. */
819 static inline void tcp_prequeue_init(struct tcp_sock *tp)
821 tp->ucopy.task = NULL;
823 tp->ucopy.memory = 0;
824 skb_queue_head_init(&tp->ucopy.prequeue);
825 #ifdef CONFIG_NET_DMA
826 tp->ucopy.dma_chan = NULL;
827 tp->ucopy.wakeup = 0;
828 tp->ucopy.pinned_list = NULL;
829 tp->ucopy.dma_cookie = 0;
833 /* Packet is added to VJ-style prequeue for processing in process
834 * context, if a reader task is waiting. Apparently, this exciting
835 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
836 * failed somewhere. Latency? Burstiness? Well, at least now we will
837 * see, why it failed. 8)8) --ANK
839 * NOTE: is this not too big to inline?
841 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
843 struct tcp_sock *tp = tcp_sk(sk);
845 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
846 __skb_queue_tail(&tp->ucopy.prequeue, skb);
847 tp->ucopy.memory += skb->truesize;
848 if (tp->ucopy.memory > sk->sk_rcvbuf) {
849 struct sk_buff *skb1;
851 BUG_ON(sock_owned_by_user(sk));
853 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
854 sk->sk_backlog_rcv(sk, skb1);
855 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
858 tp->ucopy.memory = 0;
859 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
860 wake_up_interruptible(sk->sk_sleep);
861 if (!inet_csk_ack_scheduled(sk))
862 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
863 (3 * TCP_RTO_MIN) / 4,
875 static const char *statename[]={
876 "Unused","Established","Syn Sent","Syn Recv",
877 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
878 "Close Wait","Last ACK","Listen","Closing"
882 static inline void tcp_set_state(struct sock *sk, int state)
884 int oldstate = sk->sk_state;
887 case TCP_ESTABLISHED:
888 if (oldstate != TCP_ESTABLISHED)
889 TCP_INC_STATS(TCP_MIB_CURRESTAB);
893 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
894 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
896 sk->sk_prot->unhash(sk);
897 if (inet_csk(sk)->icsk_bind_hash &&
898 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
899 inet_put_port(&tcp_hashinfo, sk);
902 if (oldstate==TCP_ESTABLISHED)
903 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
906 /* Change state AFTER socket is unhashed to avoid closed
907 * socket sitting in hash tables.
909 sk->sk_state = state;
912 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
916 static inline void tcp_done(struct sock *sk)
918 tcp_set_state(sk, TCP_CLOSE);
919 tcp_clear_xmit_timers(sk);
921 sk->sk_shutdown = SHUTDOWN_MASK;
923 if (!sock_flag(sk, SOCK_DEAD))
924 sk->sk_state_change(sk);
926 inet_csk_destroy_sock(sk);
929 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
932 rx_opt->eff_sacks = 0;
933 rx_opt->num_sacks = 0;
936 /* Determine a window scaling and initial window to offer. */
937 extern void tcp_select_initial_window(int __space, __u32 mss,
938 __u32 *rcv_wnd, __u32 *window_clamp,
939 int wscale_ok, __u8 *rcv_wscale);
941 static inline int tcp_win_from_space(int space)
943 return sysctl_tcp_adv_win_scale<=0 ?
944 (space>>(-sysctl_tcp_adv_win_scale)) :
945 space - (space>>sysctl_tcp_adv_win_scale);
948 /* Note: caller must be prepared to deal with negative returns */
949 static inline int tcp_space(const struct sock *sk)
951 return tcp_win_from_space(sk->sk_rcvbuf -
952 atomic_read(&sk->sk_rmem_alloc));
955 static inline int tcp_full_space(const struct sock *sk)
957 return tcp_win_from_space(sk->sk_rcvbuf);
960 static inline void tcp_openreq_init(struct request_sock *req,
961 struct tcp_options_received *rx_opt,
964 struct inet_request_sock *ireq = inet_rsk(req);
966 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
967 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
968 req->mss = rx_opt->mss_clamp;
969 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
970 ireq->tstamp_ok = rx_opt->tstamp_ok;
971 ireq->sack_ok = rx_opt->sack_ok;
972 ireq->snd_wscale = rx_opt->snd_wscale;
973 ireq->wscale_ok = rx_opt->wscale_ok;
976 ireq->rmt_port = skb->h.th->source;
979 extern void tcp_enter_memory_pressure(void);
981 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
983 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
986 static inline int keepalive_time_when(const struct tcp_sock *tp)
988 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
991 static inline int tcp_fin_time(const struct sock *sk)
993 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
994 const int rto = inet_csk(sk)->icsk_rto;
996 if (fin_timeout < (rto << 2) - (rto >> 1))
997 fin_timeout = (rto << 2) - (rto >> 1);
1002 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1004 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1006 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1009 /* RST segments are not recommended to carry timestamp,
1010 and, if they do, it is recommended to ignore PAWS because
1011 "their cleanup function should take precedence over timestamps."
1012 Certainly, it is mistake. It is necessary to understand the reasons
1013 of this constraint to relax it: if peer reboots, clock may go
1014 out-of-sync and half-open connections will not be reset.
1015 Actually, the problem would be not existing if all
1016 the implementations followed draft about maintaining clock
1017 via reboots. Linux-2.2 DOES NOT!
1019 However, we can relax time bounds for RST segments to MSL.
1021 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1026 #define TCP_CHECK_TIMER(sk) do { } while (0)
1028 static inline int tcp_use_frto(const struct sock *sk)
1030 const struct tcp_sock *tp = tcp_sk(sk);
1032 /* F-RTO must be activated in sysctl and there must be some
1033 * unsent new data, and the advertised window should allow
1036 return (sysctl_tcp_frto && sk->sk_send_head &&
1037 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1038 tp->snd_una + tp->snd_wnd));
1041 static inline void tcp_mib_init(void)
1044 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1045 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1046 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1047 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1051 static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1052 tp->lost_skb_hint = NULL;
1053 tp->scoreboard_skb_hint = NULL;
1054 tp->retransmit_skb_hint = NULL;
1055 tp->forward_skb_hint = NULL;
1056 tp->fastpath_skb_hint = NULL;
1060 enum tcp_seq_states {
1061 TCP_SEQ_STATE_LISTENING,
1062 TCP_SEQ_STATE_OPENREQ,
1063 TCP_SEQ_STATE_ESTABLISHED,
1064 TCP_SEQ_STATE_TIME_WAIT,
1067 struct tcp_seq_afinfo {
1068 struct module *owner;
1071 int (*seq_show) (struct seq_file *m, void *v);
1072 struct file_operations *seq_fops;
1075 struct tcp_iter_state {
1077 enum tcp_seq_states state;
1078 struct sock *syn_wait_sk;
1079 int bucket, sbucket, num, uid;
1080 struct seq_operations seq_ops;
1083 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1084 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1086 extern struct request_sock_ops tcp_request_sock_ops;
1088 extern int tcp_v4_destroy_sock(struct sock *sk);
1090 #ifdef CONFIG_PROC_FS
1091 extern int tcp4_proc_init(void);
1092 extern void tcp4_proc_exit(void);
1095 extern void tcp_v4_init(struct net_proto_family *ops);
1096 extern void tcp_init(void);