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 /* Cancel timers, when they are not required. */
25 #undef TCP_CLEAR_TIMERS
27 #include <linux/config.h>
28 #include <linux/list.h>
29 #include <linux/tcp.h>
30 #include <linux/slab.h>
31 #include <linux/cache.h>
32 #include <linux/percpu.h>
33 #include <net/inet_hashtables.h>
34 #include <net/checksum.h>
35 #include <net/request_sock.h>
39 #include <net/tcp_states.h>
41 #include <linux/seq_file.h>
43 extern struct inet_hashinfo tcp_hashinfo;
45 extern atomic_t tcp_orphan_count;
46 extern int tcp_tw_count;
47 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
48 extern void tcp_tw_deschedule(struct inet_timewait_sock *tw);
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 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
65 #define TCP_FASTRETRANS_THRESH 3
67 /* Maximal reordering. */
68 #define TCP_MAX_REORDERING 127
70 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
71 #define TCP_MAX_QUICKACKS 16U
74 #define TCP_URG_VALID 0x0100
75 #define TCP_URG_NOTYET 0x0200
76 #define TCP_URG_READ 0x0400
78 #define TCP_RETR1 3 /*
79 * This is how many retries it does before it
80 * tries to figure out if the gateway is
81 * down. Minimal RFC value is 3; it corresponds
82 * to ~3sec-8min depending on RTO.
85 #define TCP_RETR2 15 /*
86 * This should take at least
87 * 90 minutes to time out.
88 * RFC1122 says that the limit is 100 sec.
89 * 15 is ~13-30min depending on RTO.
92 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
93 * connection: ~180sec is RFC minumum */
95 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
96 * connection: ~180sec is RFC minumum */
99 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
100 * socket. 7 is ~50sec-16min.
104 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
105 * state, about 60 seconds */
106 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
107 /* BSD style FIN_WAIT2 deadlock breaker.
108 * It used to be 3min, new value is 60sec,
109 * to combine FIN-WAIT-2 timeout with
113 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
115 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
116 #define TCP_ATO_MIN ((unsigned)(HZ/25))
118 #define TCP_DELACK_MIN 4U
119 #define TCP_ATO_MIN 4U
121 #define TCP_RTO_MAX ((unsigned)(120*HZ))
122 #define TCP_RTO_MIN ((unsigned)(HZ/5))
123 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
125 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
126 * for local resources.
129 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
130 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
131 #define TCP_KEEPALIVE_INTVL (75*HZ)
133 #define MAX_TCP_KEEPIDLE 32767
134 #define MAX_TCP_KEEPINTVL 32767
135 #define MAX_TCP_KEEPCNT 127
136 #define MAX_TCP_SYNCNT 127
138 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
139 #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */
141 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
142 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
143 * after this time. It should be equal
144 * (or greater than) TCP_TIMEWAIT_LEN
145 * to provide reliability equal to one
146 * provided by timewait state.
148 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
149 * timestamps. It must be less than
150 * minimal timewait lifetime.
153 #define TCP_TW_RECYCLE_SLOTS_LOG 5
154 #define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG)
156 /* If time > 4sec, it is "slow" path, no recycling is required,
157 so that we select tick to get range about 4 seconds.
160 #if HZ <= 16 || HZ > 4096
161 # error Unsupported: HZ <= 16 or HZ > 4096
163 # define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG)
165 # define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG)
167 # define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG)
169 # define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG)
171 # define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG)
173 # define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG)
175 # define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG)
177 # define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG)
183 #define TCPOPT_NOP 1 /* Padding */
184 #define TCPOPT_EOL 0 /* End of options */
185 #define TCPOPT_MSS 2 /* Segment size negotiating */
186 #define TCPOPT_WINDOW 3 /* Window scaling */
187 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
188 #define TCPOPT_SACK 5 /* SACK Block */
189 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
195 #define TCPOLEN_MSS 4
196 #define TCPOLEN_WINDOW 3
197 #define TCPOLEN_SACK_PERM 2
198 #define TCPOLEN_TIMESTAMP 10
200 /* But this is what stacks really send out. */
201 #define TCPOLEN_TSTAMP_ALIGNED 12
202 #define TCPOLEN_WSCALE_ALIGNED 4
203 #define TCPOLEN_SACKPERM_ALIGNED 4
204 #define TCPOLEN_SACK_BASE 2
205 #define TCPOLEN_SACK_BASE_ALIGNED 4
206 #define TCPOLEN_SACK_PERBLOCK 8
208 #define TCP_TIME_RETRANS 1 /* Retransmit timer */
209 #define TCP_TIME_DACK 2 /* Delayed ack timer */
210 #define TCP_TIME_PROBE0 3 /* Zero window probe timer */
211 #define TCP_TIME_KEEPOPEN 4 /* Keepalive timer */
213 /* Flags in tp->nonagle */
214 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
215 #define TCP_NAGLE_CORK 2 /* Socket is corked */
216 #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */
218 /* sysctl variables for tcp */
219 extern int sysctl_tcp_timestamps;
220 extern int sysctl_tcp_window_scaling;
221 extern int sysctl_tcp_sack;
222 extern int sysctl_tcp_fin_timeout;
223 extern int sysctl_tcp_tw_recycle;
224 extern int sysctl_tcp_keepalive_time;
225 extern int sysctl_tcp_keepalive_probes;
226 extern int sysctl_tcp_keepalive_intvl;
227 extern int sysctl_tcp_syn_retries;
228 extern int sysctl_tcp_synack_retries;
229 extern int sysctl_tcp_retries1;
230 extern int sysctl_tcp_retries2;
231 extern int sysctl_tcp_orphan_retries;
232 extern int sysctl_tcp_syncookies;
233 extern int sysctl_tcp_retrans_collapse;
234 extern int sysctl_tcp_stdurg;
235 extern int sysctl_tcp_rfc1337;
236 extern int sysctl_tcp_abort_on_overflow;
237 extern int sysctl_tcp_max_orphans;
238 extern int sysctl_tcp_max_tw_buckets;
239 extern int sysctl_tcp_fack;
240 extern int sysctl_tcp_reordering;
241 extern int sysctl_tcp_ecn;
242 extern int sysctl_tcp_dsack;
243 extern int sysctl_tcp_mem[3];
244 extern int sysctl_tcp_wmem[3];
245 extern int sysctl_tcp_rmem[3];
246 extern int sysctl_tcp_app_win;
247 extern int sysctl_tcp_adv_win_scale;
248 extern int sysctl_tcp_tw_reuse;
249 extern int sysctl_tcp_frto;
250 extern int sysctl_tcp_low_latency;
251 extern int sysctl_tcp_nometrics_save;
252 extern int sysctl_tcp_moderate_rcvbuf;
253 extern int sysctl_tcp_tso_win_divisor;
255 extern atomic_t tcp_memory_allocated;
256 extern atomic_t tcp_sockets_allocated;
257 extern int tcp_memory_pressure;
259 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
260 #define TCP_INET_FAMILY(fam) ((fam) == AF_INET)
262 #define TCP_INET_FAMILY(fam) 1
266 * Pointers to address related TCP functions
267 * (i.e. things that depend on the address family)
271 int (*queue_xmit) (struct sk_buff *skb,
274 void (*send_check) (struct sock *sk,
277 struct sk_buff *skb);
279 int (*rebuild_header) (struct sock *sk);
281 int (*conn_request) (struct sock *sk,
282 struct sk_buff *skb);
284 struct sock * (*syn_recv_sock) (struct sock *sk,
286 struct request_sock *req,
287 struct dst_entry *dst);
289 int (*remember_stamp) (struct sock *sk);
291 __u16 net_header_len;
293 int (*setsockopt) (struct sock *sk,
299 int (*getsockopt) (struct sock *sk,
306 void (*addr2sockaddr) (struct sock *sk,
313 * The next routines deal with comparing 32 bit unsigned ints
314 * and worry about wraparound (automatic with unsigned arithmetic).
317 static inline int before(__u32 seq1, __u32 seq2)
319 return (__s32)(seq1-seq2) < 0;
322 static inline int after(__u32 seq1, __u32 seq2)
324 return (__s32)(seq2-seq1) < 0;
328 /* is s2<=s1<=s3 ? */
329 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
331 return seq3 - seq2 >= seq1 - seq2;
335 extern struct proto tcp_prot;
337 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
338 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
339 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
340 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
341 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
342 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
343 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
345 extern void tcp_v4_err(struct sk_buff *skb, u32);
347 extern void tcp_shutdown (struct sock *sk, int how);
349 extern int tcp_v4_rcv(struct sk_buff *skb);
351 extern int tcp_v4_remember_stamp(struct sock *sk);
353 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
355 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
356 struct msghdr *msg, size_t size);
357 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
359 extern int tcp_ioctl(struct sock *sk,
363 extern int tcp_rcv_state_process(struct sock *sk,
368 extern int tcp_rcv_established(struct sock *sk,
373 extern void tcp_rcv_space_adjust(struct sock *sk);
382 static inline void tcp_schedule_ack(struct tcp_sock *tp)
384 tp->ack.pending |= TCP_ACK_SCHED;
387 static inline int tcp_ack_scheduled(struct tcp_sock *tp)
389 return tp->ack.pending&TCP_ACK_SCHED;
392 static __inline__ void tcp_dec_quickack_mode(struct tcp_sock *tp, unsigned int pkts)
395 if (pkts >= tp->ack.quick) {
398 /* Leaving quickack mode we deflate ATO. */
399 tp->ack.ato = TCP_ATO_MIN;
401 tp->ack.quick -= pkts;
405 extern void tcp_enter_quickack_mode(struct tcp_sock *tp);
407 static __inline__ void tcp_delack_init(struct tcp_sock *tp)
409 memset(&tp->ack, 0, sizeof(tp->ack));
412 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
414 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
426 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
428 const struct tcphdr *th);
430 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
431 struct request_sock *req,
432 struct request_sock **prev);
433 extern int tcp_child_process(struct sock *parent,
435 struct sk_buff *skb);
436 extern void tcp_enter_frto(struct sock *sk);
437 extern void tcp_enter_loss(struct sock *sk, int how);
438 extern void tcp_clear_retrans(struct tcp_sock *tp);
439 extern void tcp_update_metrics(struct sock *sk);
441 extern void tcp_close(struct sock *sk,
443 extern struct sock * tcp_accept(struct sock *sk, int flags, int *err);
444 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
446 extern int tcp_getsockopt(struct sock *sk, int level,
450 extern int tcp_setsockopt(struct sock *sk, int level,
451 int optname, char __user *optval,
453 extern void tcp_set_keepalive(struct sock *sk, int val);
454 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
456 size_t len, int nonblock,
457 int flags, int *addr_len);
459 extern int tcp_listen_start(struct sock *sk);
461 extern void tcp_parse_options(struct sk_buff *skb,
462 struct tcp_options_received *opt_rx,
466 * TCP v4 functions exported for the inet6 API
469 extern void tcp_v4_send_check(struct sock *sk,
470 struct tcphdr *th, int len,
471 struct sk_buff *skb);
473 extern int tcp_v4_conn_request(struct sock *sk,
474 struct sk_buff *skb);
476 extern struct sock * tcp_create_openreq_child(struct sock *sk,
477 struct request_sock *req,
478 struct sk_buff *skb);
480 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
482 struct request_sock *req,
483 struct dst_entry *dst);
485 extern int tcp_v4_do_rcv(struct sock *sk,
486 struct sk_buff *skb);
488 extern int tcp_v4_connect(struct sock *sk,
489 struct sockaddr *uaddr,
492 extern int tcp_connect(struct sock *sk);
494 extern struct sk_buff * tcp_make_synack(struct sock *sk,
495 struct dst_entry *dst,
496 struct request_sock *req);
498 extern int tcp_disconnect(struct sock *sk, int flags);
500 extern void tcp_unhash(struct sock *sk);
502 extern int tcp_v4_hash_connecting(struct sock *sk);
505 /* From syncookies.c */
506 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
507 struct ip_options *opt);
508 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
513 extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp,
514 unsigned int cur_mss, int nonagle);
515 extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp);
516 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
517 extern void tcp_xmit_retransmit_queue(struct sock *);
518 extern void tcp_simple_retransmit(struct sock *);
519 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
521 extern void tcp_send_probe0(struct sock *);
522 extern void tcp_send_partial(struct sock *);
523 extern int tcp_write_wakeup(struct sock *);
524 extern void tcp_send_fin(struct sock *sk);
525 extern void tcp_send_active_reset(struct sock *sk,
526 unsigned int __nocast priority);
527 extern int tcp_send_synack(struct sock *);
528 extern void tcp_push_one(struct sock *, unsigned int mss_now);
529 extern void tcp_send_ack(struct sock *sk);
530 extern void tcp_send_delayed_ack(struct sock *sk);
533 extern void tcp_cwnd_application_limited(struct sock *sk);
536 extern void tcp_init_xmit_timers(struct sock *);
537 extern void tcp_clear_xmit_timers(struct sock *);
539 extern void tcp_delete_keepalive_timer(struct sock *);
540 extern void tcp_reset_keepalive_timer(struct sock *, unsigned long);
541 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
542 extern unsigned int tcp_current_mss(struct sock *sk, int large);
545 extern const char tcp_timer_bug_msg[];
549 extern void tcp_get_info(struct sock *, struct tcp_info *);
551 /* Read 'sendfile()'-style from a TCP socket */
552 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
553 unsigned int, size_t);
554 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
555 sk_read_actor_t recv_actor);
557 static inline void tcp_clear_xmit_timer(struct sock *sk, int what)
559 struct tcp_sock *tp = tcp_sk(sk);
562 case TCP_TIME_RETRANS:
563 case TCP_TIME_PROBE0:
566 #ifdef TCP_CLEAR_TIMERS
567 sk_stop_timer(sk, &tp->retransmit_timer);
574 #ifdef TCP_CLEAR_TIMERS
575 sk_stop_timer(sk, &tp->delack_timer);
580 printk(tcp_timer_bug_msg);
588 * Reset the retransmission timer
590 static inline void tcp_reset_xmit_timer(struct sock *sk, int what, unsigned long when)
592 struct tcp_sock *tp = tcp_sk(sk);
594 if (when > TCP_RTO_MAX) {
596 printk(KERN_DEBUG "reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk, what, when, current_text_addr());
602 case TCP_TIME_RETRANS:
603 case TCP_TIME_PROBE0:
605 tp->timeout = jiffies+when;
606 sk_reset_timer(sk, &tp->retransmit_timer, tp->timeout);
610 tp->ack.pending |= TCP_ACK_TIMER;
611 tp->ack.timeout = jiffies+when;
612 sk_reset_timer(sk, &tp->delack_timer, tp->ack.timeout);
617 printk(tcp_timer_bug_msg);
623 /* Initialize RCV_MSS value.
624 * RCV_MSS is an our guess about MSS used by the peer.
625 * We haven't any direct information about the MSS.
626 * It's better to underestimate the RCV_MSS rather than overestimate.
627 * Overestimations make us ACKing less frequently than needed.
628 * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss().
631 static inline void tcp_initialize_rcv_mss(struct sock *sk)
633 struct tcp_sock *tp = tcp_sk(sk);
634 unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache);
636 hint = min(hint, tp->rcv_wnd/2);
637 hint = min(hint, TCP_MIN_RCVMSS);
638 hint = max(hint, TCP_MIN_MSS);
640 tp->ack.rcv_mss = hint;
643 static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
645 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
646 ntohl(TCP_FLAG_ACK) |
650 static __inline__ void tcp_fast_path_on(struct tcp_sock *tp)
652 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
655 static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp)
657 if (skb_queue_empty(&tp->out_of_order_queue) &&
659 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
661 tcp_fast_path_on(tp);
664 /* Compute the actual receive window we are currently advertising.
665 * Rcv_nxt can be after the window if our peer push more data
666 * than the offered window.
668 static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp)
670 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
677 /* Choose a new window, without checks for shrinking, and without
678 * scaling applied to the result. The caller does these things
679 * if necessary. This is a "raw" window selection.
681 extern u32 __tcp_select_window(struct sock *sk);
683 /* TCP timestamps are only 32-bits, this causes a slight
684 * complication on 64-bit systems since we store a snapshot
685 * of jiffies in the buffer control blocks below. We decidely
686 * only use of the low 32-bits of jiffies and hide the ugly
687 * casts with the following macro.
689 #define tcp_time_stamp ((__u32)(jiffies))
691 /* This is what the send packet queueing engine uses to pass
692 * TCP per-packet control information to the transmission
693 * code. We also store the host-order sequence numbers in
694 * here too. This is 36 bytes on 32-bit architectures,
695 * 40 bytes on 64-bit machines, if this grows please adjust
696 * skbuff.h:skbuff->cb[xxx] size appropriately.
700 struct inet_skb_parm h4;
701 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
702 struct inet6_skb_parm h6;
704 } header; /* For incoming frames */
705 __u32 seq; /* Starting sequence number */
706 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
707 __u32 when; /* used to compute rtt's */
708 __u8 flags; /* TCP header flags. */
710 /* NOTE: These must match up to the flags byte in a
713 #define TCPCB_FLAG_FIN 0x01
714 #define TCPCB_FLAG_SYN 0x02
715 #define TCPCB_FLAG_RST 0x04
716 #define TCPCB_FLAG_PSH 0x08
717 #define TCPCB_FLAG_ACK 0x10
718 #define TCPCB_FLAG_URG 0x20
719 #define TCPCB_FLAG_ECE 0x40
720 #define TCPCB_FLAG_CWR 0x80
722 __u8 sacked; /* State flags for SACK/FACK. */
723 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
724 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
725 #define TCPCB_LOST 0x04 /* SKB is lost */
726 #define TCPCB_TAGBITS 0x07 /* All tag bits */
728 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
729 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
731 #define TCPCB_URG 0x20 /* Urgent pointer advenced here */
733 #define TCPCB_AT_TAIL (TCPCB_URG)
735 __u16 urg_ptr; /* Valid w/URG flags is set. */
736 __u32 ack_seq; /* Sequence number ACK'd */
739 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
741 #include <net/tcp_ecn.h>
743 /* Due to TSO, an SKB can be composed of multiple actual
744 * packets. To keep these tracked properly, we use this.
746 static inline int tcp_skb_pcount(const struct sk_buff *skb)
748 return skb_shinfo(skb)->tso_segs;
751 /* This is valid iff tcp_skb_pcount() > 1. */
752 static inline int tcp_skb_mss(const struct sk_buff *skb)
754 return skb_shinfo(skb)->tso_size;
757 static inline void tcp_dec_pcount_approx(__u32 *count,
758 const struct sk_buff *skb)
761 *count -= tcp_skb_pcount(skb);
767 static inline void tcp_packets_out_inc(struct sock *sk,
769 const struct sk_buff *skb)
771 int orig = tp->packets_out;
773 tp->packets_out += tcp_skb_pcount(skb);
775 tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto);
778 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
779 const struct sk_buff *skb)
781 tp->packets_out -= tcp_skb_pcount(skb);
784 /* Events passed to congestion control interface */
786 CA_EVENT_TX_START, /* first transmit when no packets in flight */
787 CA_EVENT_CWND_RESTART, /* congestion window restart */
788 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
789 CA_EVENT_FRTO, /* fast recovery timeout */
790 CA_EVENT_LOSS, /* loss timeout */
791 CA_EVENT_FAST_ACK, /* in sequence ack */
792 CA_EVENT_SLOW_ACK, /* other ack */
796 * Interface for adding new TCP congestion control handlers
798 #define TCP_CA_NAME_MAX 16
799 struct tcp_congestion_ops {
800 struct list_head list;
802 /* initialize private data (optional) */
803 void (*init)(struct tcp_sock *tp);
804 /* cleanup private data (optional) */
805 void (*release)(struct tcp_sock *tp);
807 /* return slow start threshold (required) */
808 u32 (*ssthresh)(struct tcp_sock *tp);
809 /* lower bound for congestion window (optional) */
810 u32 (*min_cwnd)(struct tcp_sock *tp);
811 /* do new cwnd calculation (required) */
812 void (*cong_avoid)(struct tcp_sock *tp, u32 ack,
813 u32 rtt, u32 in_flight, int good_ack);
814 /* round trip time sample per acked packet (optional) */
815 void (*rtt_sample)(struct tcp_sock *tp, u32 usrtt);
816 /* call before changing ca_state (optional) */
817 void (*set_state)(struct tcp_sock *tp, u8 new_state);
818 /* call when cwnd event occurs (optional) */
819 void (*cwnd_event)(struct tcp_sock *tp, enum tcp_ca_event ev);
820 /* new value of cwnd after loss (optional) */
821 u32 (*undo_cwnd)(struct tcp_sock *tp);
822 /* hook for packet ack accounting (optional) */
823 void (*pkts_acked)(struct tcp_sock *tp, u32 num_acked);
824 /* get info for tcp_diag (optional) */
825 void (*get_info)(struct tcp_sock *tp, u32 ext, struct sk_buff *skb);
827 char name[TCP_CA_NAME_MAX];
828 struct module *owner;
831 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
832 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
834 extern void tcp_init_congestion_control(struct tcp_sock *tp);
835 extern void tcp_cleanup_congestion_control(struct tcp_sock *tp);
836 extern int tcp_set_default_congestion_control(const char *name);
837 extern void tcp_get_default_congestion_control(char *name);
838 extern int tcp_set_congestion_control(struct tcp_sock *tp, const char *name);
840 extern struct tcp_congestion_ops tcp_init_congestion_ops;
841 extern u32 tcp_reno_ssthresh(struct tcp_sock *tp);
842 extern void tcp_reno_cong_avoid(struct tcp_sock *tp, u32 ack,
843 u32 rtt, u32 in_flight, int flag);
844 extern u32 tcp_reno_min_cwnd(struct tcp_sock *tp);
845 extern struct tcp_congestion_ops tcp_reno;
847 static inline void tcp_set_ca_state(struct tcp_sock *tp, u8 ca_state)
849 if (tp->ca_ops->set_state)
850 tp->ca_ops->set_state(tp, ca_state);
851 tp->ca_state = ca_state;
854 static inline void tcp_ca_event(struct tcp_sock *tp, enum tcp_ca_event event)
856 if (tp->ca_ops->cwnd_event)
857 tp->ca_ops->cwnd_event(tp, event);
860 /* This determines how many packets are "in the network" to the best
861 * of our knowledge. In many cases it is conservative, but where
862 * detailed information is available from the receiver (via SACK
863 * blocks etc.) we can make more aggressive calculations.
865 * Use this for decisions involving congestion control, use just
866 * tp->packets_out to determine if the send queue is empty or not.
868 * Read this equation as:
870 * "Packets sent once on transmission queue" MINUS
871 * "Packets left network, but not honestly ACKed yet" PLUS
872 * "Packets fast retransmitted"
874 static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
876 return (tp->packets_out - tp->left_out + tp->retrans_out);
879 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
880 * The exception is rate halving phase, when cwnd is decreasing towards
883 static inline __u32 tcp_current_ssthresh(struct tcp_sock *tp)
885 if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery))
886 return tp->snd_ssthresh;
888 return max(tp->snd_ssthresh,
889 ((tp->snd_cwnd >> 1) +
890 (tp->snd_cwnd >> 2)));
893 static inline void tcp_sync_left_out(struct tcp_sock *tp)
895 if (tp->rx_opt.sack_ok &&
896 (tp->sacked_out >= tp->packets_out - tp->lost_out))
897 tp->sacked_out = tp->packets_out - tp->lost_out;
898 tp->left_out = tp->sacked_out + tp->lost_out;
901 /* Set slow start threshold and cwnd not falling to slow start */
902 static inline void __tcp_enter_cwr(struct tcp_sock *tp)
905 tp->snd_ssthresh = tp->ca_ops->ssthresh(tp);
906 tp->snd_cwnd = min(tp->snd_cwnd,
907 tcp_packets_in_flight(tp) + 1U);
908 tp->snd_cwnd_cnt = 0;
909 tp->high_seq = tp->snd_nxt;
910 tp->snd_cwnd_stamp = tcp_time_stamp;
911 TCP_ECN_queue_cwr(tp);
914 static inline void tcp_enter_cwr(struct tcp_sock *tp)
916 tp->prior_ssthresh = 0;
917 if (tp->ca_state < TCP_CA_CWR) {
919 tcp_set_ca_state(tp, TCP_CA_CWR);
923 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
925 /* Slow start with delack produces 3 packets of burst, so that
926 * it is safe "de facto".
928 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
933 static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss,
934 const struct sk_buff *skb)
937 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
940 static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp)
942 if (!tp->packets_out && !tp->pending)
943 tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto);
946 static __inline__ void tcp_push_pending_frames(struct sock *sk,
949 __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle);
952 static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
957 static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
962 extern void tcp_destroy_sock(struct sock *sk);
966 * Calculate(/check) TCP checksum
968 static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len,
969 unsigned long saddr, unsigned long daddr,
972 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
975 static __inline__ int __tcp_checksum_complete(struct sk_buff *skb)
977 return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum));
980 static __inline__ int tcp_checksum_complete(struct sk_buff *skb)
982 return skb->ip_summed != CHECKSUM_UNNECESSARY &&
983 __tcp_checksum_complete(skb);
986 /* Prequeue for VJ style copy to user, combined with checksumming. */
988 static __inline__ void tcp_prequeue_init(struct tcp_sock *tp)
990 tp->ucopy.task = NULL;
992 tp->ucopy.memory = 0;
993 skb_queue_head_init(&tp->ucopy.prequeue);
996 /* Packet is added to VJ-style prequeue for processing in process
997 * context, if a reader task is waiting. Apparently, this exciting
998 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
999 * failed somewhere. Latency? Burstiness? Well, at least now we will
1000 * see, why it failed. 8)8) --ANK
1002 * NOTE: is this not too big to inline?
1004 static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1006 struct tcp_sock *tp = tcp_sk(sk);
1008 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
1009 __skb_queue_tail(&tp->ucopy.prequeue, skb);
1010 tp->ucopy.memory += skb->truesize;
1011 if (tp->ucopy.memory > sk->sk_rcvbuf) {
1012 struct sk_buff *skb1;
1014 BUG_ON(sock_owned_by_user(sk));
1016 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
1017 sk->sk_backlog_rcv(sk, skb1);
1018 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
1021 tp->ucopy.memory = 0;
1022 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
1023 wake_up_interruptible(sk->sk_sleep);
1024 if (!tcp_ack_scheduled(tp))
1025 tcp_reset_xmit_timer(sk, TCP_TIME_DACK, (3*TCP_RTO_MIN)/4);
1036 static const char *statename[]={
1037 "Unused","Established","Syn Sent","Syn Recv",
1038 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1039 "Close Wait","Last ACK","Listen","Closing"
1043 static __inline__ void tcp_set_state(struct sock *sk, int state)
1045 int oldstate = sk->sk_state;
1048 case TCP_ESTABLISHED:
1049 if (oldstate != TCP_ESTABLISHED)
1050 TCP_INC_STATS(TCP_MIB_CURRESTAB);
1054 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
1055 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
1057 sk->sk_prot->unhash(sk);
1058 if (inet_sk(sk)->bind_hash &&
1059 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
1060 inet_put_port(&tcp_hashinfo, sk);
1063 if (oldstate==TCP_ESTABLISHED)
1064 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
1067 /* Change state AFTER socket is unhashed to avoid closed
1068 * socket sitting in hash tables.
1070 sk->sk_state = state;
1073 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
1077 static __inline__ void tcp_done(struct sock *sk)
1079 tcp_set_state(sk, TCP_CLOSE);
1080 tcp_clear_xmit_timers(sk);
1082 sk->sk_shutdown = SHUTDOWN_MASK;
1084 if (!sock_flag(sk, SOCK_DEAD))
1085 sk->sk_state_change(sk);
1087 tcp_destroy_sock(sk);
1090 static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt)
1093 rx_opt->eff_sacks = 0;
1094 rx_opt->num_sacks = 0;
1097 static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp)
1099 if (tp->rx_opt.tstamp_ok) {
1100 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1101 (TCPOPT_NOP << 16) |
1102 (TCPOPT_TIMESTAMP << 8) |
1104 *ptr++ = htonl(tstamp);
1105 *ptr++ = htonl(tp->rx_opt.ts_recent);
1107 if (tp->rx_opt.eff_sacks) {
1108 struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks;
1111 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) |
1112 (TCPOPT_NOP << 16) |
1113 (TCPOPT_SACK << 8) |
1114 (TCPOLEN_SACK_BASE +
1115 (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK)));
1116 for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) {
1117 *ptr++ = htonl(sp[this_sack].start_seq);
1118 *ptr++ = htonl(sp[this_sack].end_seq);
1120 if (tp->rx_opt.dsack) {
1121 tp->rx_opt.dsack = 0;
1122 tp->rx_opt.eff_sacks--;
1127 /* Construct a tcp options header for a SYN or SYN_ACK packet.
1128 * If this is every changed make sure to change the definition of
1129 * MAX_SYN_SIZE to match the new maximum number of options that you
1132 static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack,
1133 int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent)
1135 /* We always get an MSS option.
1136 * The option bytes which will be seen in normal data
1137 * packets should timestamps be used, must be in the MSS
1138 * advertised. But we subtract them from tp->mss_cache so
1139 * that calculations in tcp_sendmsg are simpler etc.
1140 * So account for this fact here if necessary. If we
1141 * don't do this correctly, as a receiver we won't
1142 * recognize data packets as being full sized when we
1143 * should, and thus we won't abide by the delayed ACK
1145 * SACKs don't matter, we never delay an ACK when we
1146 * have any of those going out.
1148 *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss);
1151 *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) |
1152 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1154 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1155 (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP);
1156 *ptr++ = htonl(tstamp); /* TSVAL */
1157 *ptr++ = htonl(ts_recent); /* TSECR */
1159 *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) |
1160 (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM);
1162 *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale));
1165 /* Determine a window scaling and initial window to offer. */
1166 extern void tcp_select_initial_window(int __space, __u32 mss,
1167 __u32 *rcv_wnd, __u32 *window_clamp,
1168 int wscale_ok, __u8 *rcv_wscale);
1170 static inline int tcp_win_from_space(int space)
1172 return sysctl_tcp_adv_win_scale<=0 ?
1173 (space>>(-sysctl_tcp_adv_win_scale)) :
1174 space - (space>>sysctl_tcp_adv_win_scale);
1177 /* Note: caller must be prepared to deal with negative returns */
1178 static inline int tcp_space(const struct sock *sk)
1180 return tcp_win_from_space(sk->sk_rcvbuf -
1181 atomic_read(&sk->sk_rmem_alloc));
1184 static inline int tcp_full_space(const struct sock *sk)
1186 return tcp_win_from_space(sk->sk_rcvbuf);
1189 static inline void tcp_acceptq_queue(struct sock *sk, struct request_sock *req,
1192 reqsk_queue_add(&tcp_sk(sk)->accept_queue, req, sk, child);
1196 tcp_synq_removed(struct sock *sk, struct request_sock *req)
1198 if (reqsk_queue_removed(&tcp_sk(sk)->accept_queue, req) == 0)
1199 tcp_delete_keepalive_timer(sk);
1202 static inline void tcp_synq_added(struct sock *sk)
1204 if (reqsk_queue_added(&tcp_sk(sk)->accept_queue) == 0)
1205 tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT);
1208 static inline int tcp_synq_len(struct sock *sk)
1210 return reqsk_queue_len(&tcp_sk(sk)->accept_queue);
1213 static inline int tcp_synq_young(struct sock *sk)
1215 return reqsk_queue_len_young(&tcp_sk(sk)->accept_queue);
1218 static inline int tcp_synq_is_full(struct sock *sk)
1220 return reqsk_queue_is_full(&tcp_sk(sk)->accept_queue);
1223 static inline void tcp_synq_unlink(struct tcp_sock *tp, struct request_sock *req,
1224 struct request_sock **prev)
1226 reqsk_queue_unlink(&tp->accept_queue, req, prev);
1229 static inline void tcp_synq_drop(struct sock *sk, struct request_sock *req,
1230 struct request_sock **prev)
1232 tcp_synq_unlink(tcp_sk(sk), req, prev);
1233 tcp_synq_removed(sk, req);
1237 static __inline__ void tcp_openreq_init(struct request_sock *req,
1238 struct tcp_options_received *rx_opt,
1239 struct sk_buff *skb)
1241 struct inet_request_sock *ireq = inet_rsk(req);
1243 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1244 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1245 req->mss = rx_opt->mss_clamp;
1246 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1247 ireq->tstamp_ok = rx_opt->tstamp_ok;
1248 ireq->sack_ok = rx_opt->sack_ok;
1249 ireq->snd_wscale = rx_opt->snd_wscale;
1250 ireq->wscale_ok = rx_opt->wscale_ok;
1253 ireq->rmt_port = skb->h.th->source;
1256 extern void tcp_enter_memory_pressure(void);
1258 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1260 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1263 static inline int keepalive_time_when(const struct tcp_sock *tp)
1265 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1268 static inline int tcp_fin_time(const struct tcp_sock *tp)
1270 int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout;
1272 if (fin_timeout < (tp->rto<<2) - (tp->rto>>1))
1273 fin_timeout = (tp->rto<<2) - (tp->rto>>1);
1278 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1280 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1282 if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1285 /* RST segments are not recommended to carry timestamp,
1286 and, if they do, it is recommended to ignore PAWS because
1287 "their cleanup function should take precedence over timestamps."
1288 Certainly, it is mistake. It is necessary to understand the reasons
1289 of this constraint to relax it: if peer reboots, clock may go
1290 out-of-sync and half-open connections will not be reset.
1291 Actually, the problem would be not existing if all
1292 the implementations followed draft about maintaining clock
1293 via reboots. Linux-2.2 DOES NOT!
1295 However, we can relax time bounds for RST segments to MSL.
1297 if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1302 #define TCP_CHECK_TIMER(sk) do { } while (0)
1304 static inline int tcp_use_frto(const struct sock *sk)
1306 const struct tcp_sock *tp = tcp_sk(sk);
1308 /* F-RTO must be activated in sysctl and there must be some
1309 * unsent new data, and the advertised window should allow
1312 return (sysctl_tcp_frto && sk->sk_send_head &&
1313 !after(TCP_SKB_CB(sk->sk_send_head)->end_seq,
1314 tp->snd_una + tp->snd_wnd));
1317 static inline void tcp_mib_init(void)
1320 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1321 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1322 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1323 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
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 {
1335 struct module *owner;
1338 int (*seq_show) (struct seq_file *m, void *v);
1339 struct file_operations *seq_fops;
1342 struct tcp_iter_state {
1344 enum tcp_seq_states state;
1345 struct sock *syn_wait_sk;
1346 int bucket, sbucket, num, uid;
1347 struct seq_operations seq_ops;
1350 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1351 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);