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>
31 #include <net/inet_connection_sock.h>
32 #include <net/inet_hashtables.h>
33 #include <net/checksum.h>
34 #include <net/request_sock.h>
38 #include <net/tcp_states.h>
40 #include <linux/seq_file.h>
42 extern struct inet_hashinfo tcp_hashinfo;
44 extern atomic_t tcp_orphan_count;
45 extern int tcp_tw_count;
46 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
47 extern void tcp_tw_deschedule(struct inet_timewait_sock *tw);
49 #define MAX_TCP_HEADER (128 + MAX_HEADER)
52 * Never offer a window over 32767 without using window scaling. Some
53 * poor stacks do signed 16bit maths!
55 #define MAX_TCP_WINDOW 32767U
57 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
58 #define TCP_MIN_MSS 88U
60 /* Minimal RCV_MSS. */
61 #define TCP_MIN_RCVMSS 536U
63 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
64 #define TCP_FASTRETRANS_THRESH 3
66 /* Maximal reordering. */
67 #define TCP_MAX_REORDERING 127
69 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
70 #define TCP_MAX_QUICKACKS 16U
73 #define TCP_URG_VALID 0x0100
74 #define TCP_URG_NOTYET 0x0200
75 #define TCP_URG_READ 0x0400
77 #define TCP_RETR1 3 /*
78 * This is how many retries it does before it
79 * tries to figure out if the gateway is
80 * down. Minimal RFC value is 3; it corresponds
81 * to ~3sec-8min depending on RTO.
84 #define TCP_RETR2 15 /*
85 * This should take at least
86 * 90 minutes to time out.
87 * RFC1122 says that the limit is 100 sec.
88 * 15 is ~13-30min depending on RTO.
91 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
92 * connection: ~180sec is RFC minumum */
94 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
95 * connection: ~180sec is RFC minumum */
98 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
99 * socket. 7 is ~50sec-16min.
103 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
104 * state, about 60 seconds */
105 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
106 /* BSD style FIN_WAIT2 deadlock breaker.
107 * It used to be 3min, new value is 60sec,
108 * to combine FIN-WAIT-2 timeout with
112 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
114 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
115 #define TCP_ATO_MIN ((unsigned)(HZ/25))
117 #define TCP_DELACK_MIN 4U
118 #define TCP_ATO_MIN 4U
120 #define TCP_RTO_MAX ((unsigned)(120*HZ))
121 #define TCP_RTO_MIN ((unsigned)(HZ/5))
122 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
124 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
125 * for local resources.
128 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
129 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
130 #define TCP_KEEPALIVE_INTVL (75*HZ)
132 #define MAX_TCP_KEEPIDLE 32767
133 #define MAX_TCP_KEEPINTVL 32767
134 #define MAX_TCP_KEEPCNT 127
135 #define MAX_TCP_SYNCNT 127
137 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
138 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
140 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
141 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
142 * after this time. It should be equal
143 * (or greater than) TCP_TIMEWAIT_LEN
144 * to provide reliability equal to one
145 * provided by timewait state.
147 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
148 * timestamps. It must be less than
149 * minimal timewait lifetime.
152 #define TCP_TW_RECYCLE_SLOTS_LOG 5
153 #define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG)
155 /* If time > 4sec, it is "slow" path, no recycling is required,
156 so that we select tick to get range about 4 seconds.
159 #if HZ <= 16 || HZ > 4096
160 # error Unsupported: HZ <= 16 or HZ > 4096
162 # define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
164 # define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
166 # define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
168 # define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
170 # define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
172 # define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
174 # define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
176 # define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
182 #define TCPOPT_NOP 1 /* Padding */
183 #define TCPOPT_EOL 0 /* End of options */
184 #define TCPOPT_MSS 2 /* Segment size negotiating */
185 #define TCPOPT_WINDOW 3 /* Window scaling */
186 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
187 #define TCPOPT_SACK 5 /* SACK Block */
188 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
194 #define TCPOLEN_MSS 4
195 #define TCPOLEN_WINDOW 3
196 #define TCPOLEN_SACK_PERM 2
197 #define TCPOLEN_TIMESTAMP 10
199 /* But this is what stacks really send out. */
200 #define TCPOLEN_TSTAMP_ALIGNED 12
201 #define TCPOLEN_WSCALE_ALIGNED 4
202 #define TCPOLEN_SACKPERM_ALIGNED 4
203 #define TCPOLEN_SACK_BASE 2
204 #define TCPOLEN_SACK_BASE_ALIGNED 4
205 #define TCPOLEN_SACK_PERBLOCK 8
207 /* Flags in tp->nonagle */
208 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
209 #define TCP_NAGLE_CORK 2 /* Socket is corked */
210 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
212 /* sysctl variables for tcp */
213 extern int sysctl_tcp_timestamps;
214 extern int sysctl_tcp_window_scaling;
215 extern int sysctl_tcp_sack;
216 extern int sysctl_tcp_fin_timeout;
217 extern int sysctl_tcp_tw_recycle;
218 extern int sysctl_tcp_keepalive_time;
219 extern int sysctl_tcp_keepalive_probes;
220 extern int sysctl_tcp_keepalive_intvl;
221 extern int sysctl_tcp_syn_retries;
222 extern int sysctl_tcp_synack_retries;
223 extern int sysctl_tcp_retries1;
224 extern int sysctl_tcp_retries2;
225 extern int sysctl_tcp_orphan_retries;
226 extern int sysctl_tcp_syncookies;
227 extern int sysctl_tcp_retrans_collapse;
228 extern int sysctl_tcp_stdurg;
229 extern int sysctl_tcp_rfc1337;
230 extern int sysctl_tcp_abort_on_overflow;
231 extern int sysctl_tcp_max_orphans;
232 extern int sysctl_tcp_max_tw_buckets;
233 extern int sysctl_tcp_fack;
234 extern int sysctl_tcp_reordering;
235 extern int sysctl_tcp_ecn;
236 extern int sysctl_tcp_dsack;
237 extern int sysctl_tcp_mem[3];
238 extern int sysctl_tcp_wmem[3];
239 extern int sysctl_tcp_rmem[3];
240 extern int sysctl_tcp_app_win;
241 extern int sysctl_tcp_adv_win_scale;
242 extern int sysctl_tcp_tw_reuse;
243 extern int sysctl_tcp_frto;
244 extern int sysctl_tcp_low_latency;
245 extern int sysctl_tcp_nometrics_save;
246 extern int sysctl_tcp_moderate_rcvbuf;
247 extern int sysctl_tcp_tso_win_divisor;
249 extern atomic_t tcp_memory_allocated;
250 extern atomic_t tcp_sockets_allocated;
251 extern int tcp_memory_pressure;
254 * Pointers to address related TCP functions
255 * (i.e. things that depend on the address family)
259 int (*queue_xmit) (struct sk_buff *skb,
262 void (*send_check) (struct sock *sk,
265 struct sk_buff *skb);
267 int (*rebuild_header) (struct sock *sk);
269 int (*conn_request) (struct sock *sk,
270 struct sk_buff *skb);
272 struct sock * (*syn_recv_sock) (struct sock *sk,
274 struct request_sock *req,
275 struct dst_entry *dst);
277 int (*remember_stamp) (struct sock *sk);
279 __u16 net_header_len;
281 int (*setsockopt) (struct sock *sk,
287 int (*getsockopt) (struct sock *sk,
294 void (*addr2sockaddr) (struct sock *sk,
301 * The next routines deal with comparing 32 bit unsigned ints
302 * and worry about wraparound (automatic with unsigned arithmetic).
305 static inline int before(__u32 seq1, __u32 seq2)
307 return (__s32)(seq1-seq2) < 0;
310 static inline int after(__u32 seq1, __u32 seq2)
312 return (__s32)(seq2-seq1) < 0;
316 /* is s2<=s1<=s3 ? */
317 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
319 return seq3 - seq2 >= seq1 - seq2;
323 extern struct proto tcp_prot;
325 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
326 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
327 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
328 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
329 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
330 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
331 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
333 extern void tcp_v4_err(struct sk_buff *skb, u32);
335 extern void tcp_shutdown (struct sock *sk, int how);
337 extern int tcp_v4_rcv(struct sk_buff *skb);
339 extern int tcp_v4_remember_stamp(struct sock *sk);
341 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
343 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
344 struct msghdr *msg, size_t size);
345 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
347 extern int tcp_ioctl(struct sock *sk,
351 extern int tcp_rcv_state_process(struct sock *sk,
356 extern int tcp_rcv_established(struct sock *sk,
361 extern void tcp_rcv_space_adjust(struct sock *sk);
363 static inline void tcp_dec_quickack_mode(struct sock *sk,
364 const unsigned int pkts)
366 struct inet_connection_sock *icsk = inet_csk(sk);
368 if (icsk->icsk_ack.quick) {
369 if (pkts >= icsk->icsk_ack.quick) {
370 icsk->icsk_ack.quick = 0;
371 /* Leaving quickack mode we deflate ATO. */
372 icsk->icsk_ack.ato = TCP_ATO_MIN;
374 icsk->icsk_ack.quick -= pkts;
378 extern void tcp_enter_quickack_mode(struct sock *sk);
380 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
382 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
394 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
396 const struct tcphdr *th);
398 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
399 struct request_sock *req,
400 struct request_sock **prev);
401 extern int tcp_child_process(struct sock *parent,
403 struct sk_buff *skb);
404 extern void tcp_enter_frto(struct sock *sk);
405 extern void tcp_enter_loss(struct sock *sk, int how);
406 extern void tcp_clear_retrans(struct tcp_sock *tp);
407 extern void tcp_update_metrics(struct sock *sk);
409 extern void tcp_close(struct sock *sk,
411 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
413 extern int tcp_getsockopt(struct sock *sk, int level,
417 extern int tcp_setsockopt(struct sock *sk, int level,
418 int optname, char __user *optval,
420 extern void tcp_set_keepalive(struct sock *sk, int val);
421 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
423 size_t len, int nonblock,
424 int flags, int *addr_len);
426 extern void tcp_parse_options(struct sk_buff *skb,
427 struct tcp_options_received *opt_rx,
431 * TCP v4 functions exported for the inet6 API
434 extern void tcp_v4_send_check(struct sock *sk,
435 struct tcphdr *th, int len,
436 struct sk_buff *skb);
438 extern int tcp_v4_conn_request(struct sock *sk,
439 struct sk_buff *skb);
441 extern struct sock * tcp_create_openreq_child(struct sock *sk,
442 struct request_sock *req,
443 struct sk_buff *skb);
445 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
447 struct request_sock *req,
448 struct dst_entry *dst);
450 extern int tcp_v4_do_rcv(struct sock *sk,
451 struct sk_buff *skb);
453 extern int tcp_v4_connect(struct sock *sk,
454 struct sockaddr *uaddr,
457 extern int tcp_connect(struct sock *sk);
459 extern struct sk_buff * tcp_make_synack(struct sock *sk,
460 struct dst_entry *dst,
461 struct request_sock *req);
463 extern int tcp_disconnect(struct sock *sk, int flags);
465 extern void tcp_unhash(struct sock *sk);
467 extern int tcp_v4_hash_connecting(struct sock *sk);
470 /* From syncookies.c */
471 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
472 struct ip_options *opt);
473 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
478 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
479 unsigned int cur_mss, int nonagle);
480 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
481 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
482 extern void tcp_xmit_retransmit_queue(struct sock *);
483 extern void tcp_simple_retransmit(struct sock *);
484 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
486 extern void tcp_send_probe0(struct sock *);
487 extern void tcp_send_partial(struct sock *);
488 extern int tcp_write_wakeup(struct sock *);
489 extern void tcp_send_fin(struct sock *sk);
490 extern void tcp_send_active_reset(struct sock *sk,
491 unsigned int __nocast priority);
492 extern int tcp_send_synack(struct sock *);
493 extern void tcp_push_one(struct sock *, unsigned int mss_now);
494 extern void tcp_send_ack(struct sock *sk);
495 extern void tcp_send_delayed_ack(struct sock *sk);
498 extern void tcp_cwnd_application_limited(struct sock *sk);
501 extern void tcp_init_xmit_timers(struct sock *);
502 static inline void tcp_clear_xmit_timers(struct sock *sk)
504 inet_csk_clear_xmit_timers(sk);
507 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
508 extern unsigned int tcp_current_mss(struct sock *sk, int large);
511 extern void tcp_get_info(struct sock *, struct tcp_info *);
513 /* Read 'sendfile()'-style from a TCP socket */
514 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
515 unsigned int, size_t);
516 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
517 sk_read_actor_t recv_actor);
519 /* Initialize RCV_MSS value.
520 * RCV_MSS is an our guess about MSS used by the peer.
521 * We haven't any direct information about the MSS.
522 * It's better to underestimate the RCV_MSS rather than overestimate.
523 * Overestimations make us ACKing less frequently than needed.
524 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
527 static inline void tcp_initialize_rcv_mss(struct sock *sk)
529 struct tcp_sock *tp = tcp_sk(sk);
530 unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);
532 hint = min(hint, tp->rcv_wnd/2);
533 hint = min(hint, TCP_MIN_RCVMSS);
534 hint = max(hint, TCP_MIN_MSS);
536 inet_csk(sk)->icsk_ack.rcv_mss = hint;
539 static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
541 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
542 ntohl(TCP_FLAG_ACK) |
546 static __inline__ void tcp_fast_path_on(struct tcp_sock *tp)
548 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
551 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
553 if (skb_queue_empty(&tp->out_of_order_queue) &&
555 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
557 tcp_fast_path_on(tp);
560 /* Compute the actual receive window we are currently advertising.
561 * Rcv_nxt can be after the window if our peer push more data
562 * than the offered window.
564 static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp)
566 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
573 /* Choose a new window, without checks for shrinking, and without
574 * scaling applied to the result. The caller does these things
575 * if necessary. This is a "raw" window selection.
577 extern u32 __tcp_select_window(struct sock *sk);
579 /* TCP timestamps are only 32-bits, this causes a slight
580 * complication on 64-bit systems since we store a snapshot
581 * of jiffies in the buffer control blocks below. We decidely
582 * only use of the low 32-bits of jiffies and hide the ugly
583 * casts with the following macro.
585 #define tcp_time_stamp ((__u32)(jiffies))
587 /* This is what the send packet queueing engine uses to pass
588 * TCP per-packet control information to the transmission
589 * code. We also store the host-order sequence numbers in
590 * here too. This is 36 bytes on 32-bit architectures,
591 * 40 bytes on 64-bit machines, if this grows please adjust
592 * skbuff.h:skbuff->cb[xxx] size appropriately.
596 struct inet_skb_parm h4;
597 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
598 struct inet6_skb_parm h6;
600 } header; /* For incoming frames */
601 __u32 seq; /* Starting sequence number */
602 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
603 __u32 when; /* used to compute rtt's */
604 __u8 flags; /* TCP header flags. */
606 /* NOTE: These must match up to the flags byte in a
609 #define TCPCB_FLAG_FIN 0x01
610 #define TCPCB_FLAG_SYN 0x02
611 #define TCPCB_FLAG_RST 0x04
612 #define TCPCB_FLAG_PSH 0x08
613 #define TCPCB_FLAG_ACK 0x10
614 #define TCPCB_FLAG_URG 0x20
615 #define TCPCB_FLAG_ECE 0x40
616 #define TCPCB_FLAG_CWR 0x80
618 __u8 sacked; /* State flags for SACK/FACK. */
619 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
620 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
621 #define TCPCB_LOST 0x04 /* SKB is lost */
622 #define TCPCB_TAGBITS 0x07 /* All tag bits */
624 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
625 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
627 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
629 #define TCPCB_AT_TAIL (TCPCB_URG)
631 __u16 urg_ptr; /* Valid w/URG flags is set. */
632 __u32 ack_seq; /* Sequence number ACK'd */
635 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
637 #include <net/tcp_ecn.h>
639 /* Due to TSO, an SKB can be composed of multiple actual
640 * packets. To keep these tracked properly, we use this.
642 static inline int tcp_skb_pcount(const struct sk_buff *skb)
644 return skb_shinfo(skb)->tso_segs;
647 /* This is valid iff tcp_skb_pcount() > 1. */
648 static inline int tcp_skb_mss(const struct sk_buff *skb)
650 return skb_shinfo(skb)->tso_size;
653 static inline void tcp_dec_pcount_approx(__u32 *count,
654 const struct sk_buff *skb)
657 *count -= tcp_skb_pcount(skb);
663 static inline void tcp_packets_out_inc(struct sock *sk,
665 const struct sk_buff *skb)
667 int orig = tp->packets_out;
669 tp->packets_out += tcp_skb_pcount(skb);
671 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
672 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
675 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
676 const struct sk_buff *skb)
678 tp->packets_out -= tcp_skb_pcount(skb);
681 /* Events passed to congestion control interface */
683 CA_EVENT_TX_START, /* first transmit when no packets in flight */
684 CA_EVENT_CWND_RESTART, /* congestion window restart */
685 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
686 CA_EVENT_FRTO, /* fast recovery timeout */
687 CA_EVENT_LOSS, /* loss timeout */
688 CA_EVENT_FAST_ACK, /* in sequence ack */
689 CA_EVENT_SLOW_ACK, /* other ack */
693 * Interface for adding new TCP congestion control handlers
695 #define TCP_CA_NAME_MAX 16
696 struct tcp_congestion_ops {
697 struct list_head list;
699 /* initialize private data (optional) */
700 void (*init)(struct tcp_sock *tp);
701 /* cleanup private data (optional) */
702 void (*release)(struct tcp_sock *tp);
704 /* return slow start threshold (required) */
705 u32 (*ssthresh)(struct tcp_sock *tp);
706 /* lower bound for congestion window (optional) */
707 u32 (*min_cwnd)(struct tcp_sock *tp);
708 /* do new cwnd calculation (required) */
709 void (*cong_avoid)(struct tcp_sock *tp, u32 ack,
710 u32 rtt, u32 in_flight, int good_ack);
711 /* round trip time sample per acked packet (optional) */
712 void (*rtt_sample)(struct tcp_sock *tp, u32 usrtt);
713 /* call before changing ca_state (optional) */
714 void (*set_state)(struct tcp_sock *tp, u8 new_state);
715 /* call when cwnd event occurs (optional) */
716 void (*cwnd_event)(struct tcp_sock *tp, enum tcp_ca_event ev);
717 /* new value of cwnd after loss (optional) */
718 u32 (*undo_cwnd)(struct tcp_sock *tp);
719 /* hook for packet ack accounting (optional) */
720 void (*pkts_acked)(struct tcp_sock *tp, u32 num_acked);
721 /* get info for tcp_diag (optional) */
722 void (*get_info)(struct tcp_sock *tp, u32 ext, struct sk_buff *skb);
724 char name[TCP_CA_NAME_MAX];
725 struct module *owner;
728 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
729 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
731 extern void tcp_init_congestion_control(struct tcp_sock *tp);
732 extern void tcp_cleanup_congestion_control(struct tcp_sock *tp);
733 extern int tcp_set_default_congestion_control(const char *name);
734 extern void tcp_get_default_congestion_control(char *name);
735 extern int tcp_set_congestion_control(struct tcp_sock *tp, const char *name);
737 extern struct tcp_congestion_ops tcp_init_congestion_ops;
738 extern u32 tcp_reno_ssthresh(struct tcp_sock *tp);
739 extern void tcp_reno_cong_avoid(struct tcp_sock *tp, u32 ack,
740 u32 rtt, u32 in_flight, int flag);
741 extern u32 tcp_reno_min_cwnd(struct tcp_sock *tp);
742 extern struct tcp_congestion_ops tcp_reno;
744 static inline void tcp_set_ca_state(struct tcp_sock *tp, u8 ca_state)
746 if (tp->ca_ops->set_state)
747 tp->ca_ops->set_state(tp, ca_state);
748 tp->ca_state = ca_state;
751 static inline void tcp_ca_event(struct tcp_sock *tp, enum tcp_ca_event event)
753 if (tp->ca_ops->cwnd_event)
754 tp->ca_ops->cwnd_event(tp, event);
757 /* This determines how many packets are "in the network" to the best
758 * of our knowledge. In many cases it is conservative, but where
759 * detailed information is available from the receiver (via SACK
760 * blocks etc.) we can make more aggressive calculations.
762 * Use this for decisions involving congestion control, use just
763 * tp->packets_out to determine if the send queue is empty or not.
765 * Read this equation as:
767 * "Packets sent once on transmission queue" MINUS
768 * "Packets left network, but not honestly ACKed yet" PLUS
769 * "Packets fast retransmitted"
771 static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
773 return (tp->packets_out - tp->left_out + tp->retrans_out);
776 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
777 * The exception is rate halving phase, when cwnd is decreasing towards
780 static inline __u32 tcp_current_ssthresh(struct tcp_sock *tp)
782 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
783 return tp->snd_ssthresh;
785 return max(tp->snd_ssthresh,
786 ((tp->snd_cwnd >> 1) +
787 (tp->snd_cwnd >> 2)));
790 static inline void tcp_sync_left_out(struct tcp_sock *tp)
792 if (tp->rx_opt.sack_ok &&
793 (tp->sacked_out >= tp->packets_out - tp->lost_out))
794 tp->sacked_out = tp->packets_out - tp->lost_out;
795 tp->left_out = tp->sacked_out + tp->lost_out;
798 /* Set slow start threshold and cwnd not falling to slow start */
799 static inline void __tcp_enter_cwr(struct tcp_sock *tp)
802 tp->snd_ssthresh = tp->ca_ops->ssthresh(tp);
803 tp->snd_cwnd = min(tp->snd_cwnd,
804 tcp_packets_in_flight(tp) + 1U);
805 tp->snd_cwnd_cnt = 0;
806 tp->high_seq = tp->snd_nxt;
807 tp->snd_cwnd_stamp = tcp_time_stamp;
808 TCP_ECN_queue_cwr(tp);
811 static inline void tcp_enter_cwr(struct tcp_sock *tp)
813 tp->prior_ssthresh = 0;
814 if (tp->ca_state < TCP_CA_CWR) {
816 tcp_set_ca_state(tp, TCP_CA_CWR);
820 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
822 /* Slow start with delack produces 3 packets of burst, so that
823 * it is safe "de facto".
825 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
830 static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss,
831 const struct sk_buff *skb)
834 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
837 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
839 const struct inet_connection_sock *icsk = inet_csk(sk);
840 if (!tp->packets_out && !icsk->icsk_pending)
841 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
842 icsk->icsk_rto, TCP_RTO_MAX);
845 static __inline__ void tcp_push_pending_frames(struct sock *sk,
848 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
851 static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
856 static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
862 * Calculate(/check) TCP checksum
864 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
865 unsigned long saddr, unsigned long daddr,
868 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
871 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
873 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
876 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
878 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
879 __tcp_checksum_complete(skb);
882 /* Prequeue for VJ style copy to user, combined with checksumming. */
884 static __inline__ void tcp_prequeue_init(struct tcp_sock *tp)
886 tp->ucopy.task = NULL;
888 tp->ucopy.memory = 0;
889 skb_queue_head_init(&tp->ucopy.prequeue);
892 /* Packet is added to VJ-style prequeue for processing in process
893 * context, if a reader task is waiting. Apparently, this exciting
894 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
895 * failed somewhere. Latency? Burstiness? Well, at least now we will
896 * see, why it failed. 8)8) --ANK
898 * NOTE: is this not too big to inline?
900 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
902 struct tcp_sock *tp = tcp_sk(sk);
904 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
905 __skb_queue_tail(&tp->ucopy.prequeue, skb);
906 tp->ucopy.memory += skb->truesize;
907 if (tp->ucopy.memory > sk->sk_rcvbuf) {
908 struct sk_buff *skb1;
910 BUG_ON(sock_owned_by_user(sk));
912 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
913 sk->sk_backlog_rcv(sk, skb1);
914 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
917 tp->ucopy.memory = 0;
918 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
919 wake_up_interruptible(sk->sk_sleep);
920 if (!inet_csk_ack_scheduled(sk))
921 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
922 (3 * TCP_RTO_MIN) / 4,
934 static const char *statename[]={
935 "Unused","Established","Syn Sent","Syn Recv",
936 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
937 "Close Wait","Last ACK","Listen","Closing"
941 static __inline__ void tcp_set_state(struct sock *sk, int state)
943 int oldstate = sk->sk_state;
946 case TCP_ESTABLISHED:
947 if (oldstate != TCP_ESTABLISHED)
948 TCP_INC_STATS(TCP_MIB_CURRESTAB);
952 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
953 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
955 sk->sk_prot->unhash(sk);
956 if (inet_csk(sk)->icsk_bind_hash &&
957 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
958 inet_put_port(&tcp_hashinfo, sk);
961 if (oldstate==TCP_ESTABLISHED)
962 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
965 /* Change state AFTER socket is unhashed to avoid closed
966 * socket sitting in hash tables.
968 sk->sk_state = state;
971 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
975 static __inline__ void tcp_done(struct sock *sk)
977 tcp_set_state(sk, TCP_CLOSE);
978 tcp_clear_xmit_timers(sk);
980 sk->sk_shutdown = SHUTDOWN_MASK;
982 if (!sock_flag(sk, SOCK_DEAD))
983 sk->sk_state_change(sk);
985 inet_csk_destroy_sock(sk);
988 static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt)
991 rx_opt->eff_sacks = 0;
992 rx_opt->num_sacks = 0;
995 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp)
997 if (tp->rx_opt.tstamp_ok) {
998 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1000 (TCPOPT_TIMESTAMP << 8) |
1002 *ptr++ = htonl(tstamp);
1003 *ptr++ = htonl(tp->rx_opt.ts_recent);
1005 if (tp->rx_opt.eff_sacks) {
1006 struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks;
1009 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1010 (TCPOPT_NOP << 16) |
1011 (TCPOPT_SACK << 8) |
1012 (TCPOLEN_SACK_BASE +
1013 (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1014 for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) {
1015 *ptr++ = htonl(sp[this_sack].start_seq);
1016 *ptr++ = htonl(sp[this_sack].end_seq);
1018 if (tp->rx_opt.dsack) {
1019 tp->rx_opt.dsack = 0;
1020 tp->rx_opt.eff_sacks--;
1025 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1026 * If this is every changed make sure to change the definition of
1027 * MAX_SYN_SIZE to match the new maximum number of options that you
1030 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1031 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1033 /* We always get an MSS option.
1034 * The option bytes which will be seen in normal data
1035 * packets should timestamps be used, must be in the MSS
1036 * advertised. But we subtract them from tp->mss_cache so
1037 * that calculations in tcp_sendmsg are simpler etc.
1038 * So account for this fact here if necessary. If we
1039 * don't do this correctly, as a receiver we won't
1040 * recognize data packets as being full sized when we
1041 * should, and thus we won't abide by the delayed ACK
1043 * SACKs don't matter, we never delay an ACK when we
1044 * have any of those going out.
1046 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1049 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1050 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1052 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1053 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1054 *ptr++ = htonl(tstamp); /* TSVAL */
1055 *ptr++ = htonl(ts_recent); /* TSECR */
1057 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1058 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1060 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1063 /* Determine a window scaling and initial window to offer. */
1064 extern void tcp_select_initial_window(int __space, __u32 mss,
1065 __u32 *rcv_wnd, __u32 *window_clamp,
1066 int wscale_ok, __u8 *rcv_wscale);
1068 static inline int tcp_win_from_space(int space)
1070 return sysctl_tcp_adv_win_scale<=0 ?
1071 (space>>(-sysctl_tcp_adv_win_scale)) :
1072 space - (space>>sysctl_tcp_adv_win_scale);
1075 /* Note: caller must be prepared to deal with negative returns */
1076 static inline int tcp_space(const struct sock *sk)
1078 return tcp_win_from_space(sk->sk_rcvbuf -
1079 atomic_read(&sk->sk_rmem_alloc));
1082 static inline int tcp_full_space(const struct sock *sk)
1084 return tcp_win_from_space(sk->sk_rcvbuf);
1087 static __inline__ void tcp_openreq_init(struct request_sock *req,
1088 struct tcp_options_received *rx_opt,
1089 struct sk_buff *skb)
1091 struct inet_request_sock *ireq = inet_rsk(req);
1093 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1094 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1095 req->mss = rx_opt->mss_clamp;
1096 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1097 ireq->tstamp_ok = rx_opt->tstamp_ok;
1098 ireq->sack_ok = rx_opt->sack_ok;
1099 ireq->snd_wscale = rx_opt->snd_wscale;
1100 ireq->wscale_ok = rx_opt->wscale_ok;
1103 ireq->rmt_port = skb->h.th->source;
1106 extern void tcp_enter_memory_pressure(void);
1108 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1110 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1113 static inline int keepalive_time_when(const struct tcp_sock *tp)
1115 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1118 static inline int tcp_fin_time(const struct sock *sk)
1120 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1121 const int rto = inet_csk(sk)->icsk_rto;
1123 if (fin_timeout < (rto << 2) - (rto >> 1))
1124 fin_timeout = (rto << 2) - (rto >> 1);
1129 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1131 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1133 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1136 /* RST segments are not recommended to carry timestamp,
1137 and, if they do, it is recommended to ignore PAWS because
1138 "their cleanup function should take precedence over timestamps."
1139 Certainly, it is mistake. It is necessary to understand the reasons
1140 of this constraint to relax it: if peer reboots, clock may go
1141 out-of-sync and half-open connections will not be reset.
1142 Actually, the problem would be not existing if all
1143 the implementations followed draft about maintaining clock
1144 via reboots. Linux-2.2 DOES NOT!
1146 However, we can relax time bounds for RST segments to MSL.
1148 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1153 #define TCP_CHECK_TIMER(sk) do { } while (0)
1155 static inline int tcp_use_frto(const struct sock *sk)
1157 const struct tcp_sock *tp = tcp_sk(sk);
1159 /* F-RTO must be activated in sysctl and there must be some
1160 * unsent new data, and the advertised window should allow
1163 return (sysctl_tcp_frto && sk->sk_send_head &&
1164 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1165 tp->snd_una + tp->snd_wnd));
1168 static inline void tcp_mib_init(void)
1171 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1172 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1173 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1174 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1178 enum tcp_seq_states {
1179 TCP_SEQ_STATE_LISTENING,
1180 TCP_SEQ_STATE_OPENREQ,
1181 TCP_SEQ_STATE_ESTABLISHED,
1182 TCP_SEQ_STATE_TIME_WAIT,
1185 struct tcp_seq_afinfo {
1186 struct module *owner;
1189 int (*seq_show) (struct seq_file *m, void *v);
1190 struct file_operations *seq_fops;
1193 struct tcp_iter_state {
1195 enum tcp_seq_states state;
1196 struct sock *syn_wait_sk;
1197 int bucket, sbucket, num, uid;
1198 struct seq_operations seq_ops;
1201 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1202 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);