2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the TCP module.
8 * Version: @(#)tcp.h 1.0.5 05/23/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
22 #define FASTRETRANS_DEBUG 1
24 #include <linux/list.h>
25 #include <linux/tcp.h>
26 #include <linux/slab.h>
27 #include <linux/cache.h>
28 #include <linux/percpu.h>
29 #include <linux/skbuff.h>
30 #include <linux/dmaengine.h>
31 #include <linux/crypto.h>
32 #include <linux/cryptohash.h>
34 #include <net/inet_connection_sock.h>
35 #include <net/inet_timewait_sock.h>
36 #include <net/inet_hashtables.h>
37 #include <net/checksum.h>
38 #include <net/request_sock.h>
42 #include <net/tcp_states.h>
43 #include <net/inet_ecn.h>
45 #include <linux/seq_file.h>
47 extern struct inet_hashinfo tcp_hashinfo;
49 extern atomic_t tcp_orphan_count;
50 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
52 #define MAX_TCP_HEADER (128 + MAX_HEADER)
53 #define MAX_TCP_OPTION_SPACE 40
56 * Never offer a window over 32767 without using window scaling. Some
57 * poor stacks do signed 16bit maths!
59 #define MAX_TCP_WINDOW 32767U
61 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
62 #define TCP_MIN_MSS 88U
64 /* Minimal RCV_MSS. */
65 #define TCP_MIN_RCVMSS 536U
67 /* The least MTU to use for probing */
68 #define TCP_BASE_MSS 512
70 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
71 #define TCP_FASTRETRANS_THRESH 3
73 /* Maximal reordering. */
74 #define TCP_MAX_REORDERING 127
76 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
77 #define TCP_MAX_QUICKACKS 16U
80 #define TCP_URG_VALID 0x0100
81 #define TCP_URG_NOTYET 0x0200
82 #define TCP_URG_READ 0x0400
84 #define TCP_RETR1 3 /*
85 * This is how many retries it does before it
86 * tries to figure out if the gateway is
87 * down. Minimal RFC value is 3; it corresponds
88 * to ~3sec-8min depending on RTO.
91 #define TCP_RETR2 15 /*
92 * This should take at least
93 * 90 minutes to time out.
94 * RFC1122 says that the limit is 100 sec.
95 * 15 is ~13-30min depending on RTO.
98 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
99 * connection: ~180sec is RFC minimum */
101 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
102 * connection: ~180sec is RFC minimum */
105 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
106 * socket. 7 is ~50sec-16min.
110 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
111 * state, about 60 seconds */
112 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
113 /* BSD style FIN_WAIT2 deadlock breaker.
114 * It used to be 3min, new value is 60sec,
115 * to combine FIN-WAIT-2 timeout with
119 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
121 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
122 #define TCP_ATO_MIN ((unsigned)(HZ/25))
124 #define TCP_DELACK_MIN 4U
125 #define TCP_ATO_MIN 4U
127 #define TCP_RTO_MAX ((unsigned)(120*HZ))
128 #define TCP_RTO_MIN ((unsigned)(HZ/5))
129 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
131 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
132 * for local resources.
135 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
136 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
137 #define TCP_KEEPALIVE_INTVL (75*HZ)
139 #define MAX_TCP_KEEPIDLE 32767
140 #define MAX_TCP_KEEPINTVL 32767
141 #define MAX_TCP_KEEPCNT 127
142 #define MAX_TCP_SYNCNT 127
144 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
146 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
147 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
148 * after this time. It should be equal
149 * (or greater than) TCP_TIMEWAIT_LEN
150 * to provide reliability equal to one
151 * provided by timewait state.
153 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
154 * timestamps. It must be less than
155 * minimal timewait lifetime.
161 #define TCPOPT_NOP 1 /* Padding */
162 #define TCPOPT_EOL 0 /* End of options */
163 #define TCPOPT_MSS 2 /* Segment size negotiating */
164 #define TCPOPT_WINDOW 3 /* Window scaling */
165 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
166 #define TCPOPT_SACK 5 /* SACK Block */
167 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
168 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
174 #define TCPOLEN_MSS 4
175 #define TCPOLEN_WINDOW 3
176 #define TCPOLEN_SACK_PERM 2
177 #define TCPOLEN_TIMESTAMP 10
178 #define TCPOLEN_MD5SIG 18
180 /* But this is what stacks really send out. */
181 #define TCPOLEN_TSTAMP_ALIGNED 12
182 #define TCPOLEN_WSCALE_ALIGNED 4
183 #define TCPOLEN_SACKPERM_ALIGNED 4
184 #define TCPOLEN_SACK_BASE 2
185 #define TCPOLEN_SACK_BASE_ALIGNED 4
186 #define TCPOLEN_SACK_PERBLOCK 8
187 #define TCPOLEN_MD5SIG_ALIGNED 20
188 #define TCPOLEN_MSS_ALIGNED 4
190 /* Flags in tp->nonagle */
191 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
192 #define TCP_NAGLE_CORK 2 /* Socket is corked */
193 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
195 extern struct inet_timewait_death_row tcp_death_row;
197 /* sysctl variables for tcp */
198 extern int sysctl_tcp_timestamps;
199 extern int sysctl_tcp_window_scaling;
200 extern int sysctl_tcp_sack;
201 extern int sysctl_tcp_fin_timeout;
202 extern int sysctl_tcp_keepalive_time;
203 extern int sysctl_tcp_keepalive_probes;
204 extern int sysctl_tcp_keepalive_intvl;
205 extern int sysctl_tcp_syn_retries;
206 extern int sysctl_tcp_synack_retries;
207 extern int sysctl_tcp_retries1;
208 extern int sysctl_tcp_retries2;
209 extern int sysctl_tcp_orphan_retries;
210 extern int sysctl_tcp_syncookies;
211 extern int sysctl_tcp_retrans_collapse;
212 extern int sysctl_tcp_stdurg;
213 extern int sysctl_tcp_rfc1337;
214 extern int sysctl_tcp_abort_on_overflow;
215 extern int sysctl_tcp_max_orphans;
216 extern int sysctl_tcp_fack;
217 extern int sysctl_tcp_reordering;
218 extern int sysctl_tcp_ecn;
219 extern int sysctl_tcp_dsack;
220 extern int sysctl_tcp_mem[3];
221 extern int sysctl_tcp_wmem[3];
222 extern int sysctl_tcp_rmem[3];
223 extern int sysctl_tcp_app_win;
224 extern int sysctl_tcp_adv_win_scale;
225 extern int sysctl_tcp_tw_reuse;
226 extern int sysctl_tcp_frto;
227 extern int sysctl_tcp_frto_response;
228 extern int sysctl_tcp_low_latency;
229 extern int sysctl_tcp_dma_copybreak;
230 extern int sysctl_tcp_nometrics_save;
231 extern int sysctl_tcp_moderate_rcvbuf;
232 extern int sysctl_tcp_tso_win_divisor;
233 extern int sysctl_tcp_abc;
234 extern int sysctl_tcp_mtu_probing;
235 extern int sysctl_tcp_base_mss;
236 extern int sysctl_tcp_workaround_signed_windows;
237 extern int sysctl_tcp_slow_start_after_idle;
238 extern int sysctl_tcp_max_ssthresh;
240 extern atomic_t tcp_memory_allocated;
241 extern atomic_t tcp_sockets_allocated;
242 extern int tcp_memory_pressure;
245 * The next routines deal with comparing 32 bit unsigned ints
246 * and worry about wraparound (automatic with unsigned arithmetic).
249 static inline int before(__u32 seq1, __u32 seq2)
251 return (__s32)(seq1-seq2) < 0;
253 #define after(seq2, seq1) before(seq1, seq2)
255 /* is s2<=s1<=s3 ? */
256 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
258 return seq3 - seq2 >= seq1 - seq2;
261 static inline int tcp_too_many_orphans(struct sock *sk, int num)
263 return (num > sysctl_tcp_max_orphans) ||
264 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
265 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
268 extern struct proto tcp_prot;
270 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
271 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
272 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
273 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
275 extern void tcp_v4_err(struct sk_buff *skb, u32);
277 extern void tcp_shutdown (struct sock *sk, int how);
279 extern int tcp_v4_rcv(struct sk_buff *skb);
281 extern int tcp_v4_remember_stamp(struct sock *sk);
283 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
285 extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
286 struct msghdr *msg, size_t size);
287 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
289 extern int tcp_ioctl(struct sock *sk,
293 extern int tcp_rcv_state_process(struct sock *sk,
298 extern int tcp_rcv_established(struct sock *sk,
303 extern void tcp_rcv_space_adjust(struct sock *sk);
305 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
307 extern int tcp_twsk_unique(struct sock *sk,
308 struct sock *sktw, void *twp);
310 extern void tcp_twsk_destructor(struct sock *sk);
312 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
313 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
315 static inline void tcp_dec_quickack_mode(struct sock *sk,
316 const unsigned int pkts)
318 struct inet_connection_sock *icsk = inet_csk(sk);
320 if (icsk->icsk_ack.quick) {
321 if (pkts >= icsk->icsk_ack.quick) {
322 icsk->icsk_ack.quick = 0;
323 /* Leaving quickack mode we deflate ATO. */
324 icsk->icsk_ack.ato = TCP_ATO_MIN;
326 icsk->icsk_ack.quick -= pkts;
330 extern void tcp_enter_quickack_mode(struct sock *sk);
332 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
334 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
338 #define TCP_ECN_QUEUE_CWR 2
339 #define TCP_ECN_DEMAND_CWR 4
341 static __inline__ void
342 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
344 if (sysctl_tcp_ecn && th->ece && th->cwr)
345 inet_rsk(req)->ecn_ok = 1;
357 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
359 const struct tcphdr *th);
361 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
362 struct request_sock *req,
363 struct request_sock **prev);
364 extern int tcp_child_process(struct sock *parent,
366 struct sk_buff *skb);
367 extern int tcp_use_frto(struct sock *sk);
368 extern void tcp_enter_frto(struct sock *sk);
369 extern void tcp_enter_loss(struct sock *sk, int how);
370 extern void tcp_clear_retrans(struct tcp_sock *tp);
371 extern void tcp_update_metrics(struct sock *sk);
373 extern void tcp_close(struct sock *sk,
375 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
377 extern int tcp_getsockopt(struct sock *sk, int level,
381 extern int tcp_setsockopt(struct sock *sk, int level,
382 int optname, char __user *optval,
384 extern int compat_tcp_getsockopt(struct sock *sk,
385 int level, int optname,
386 char __user *optval, int __user *optlen);
387 extern int compat_tcp_setsockopt(struct sock *sk,
388 int level, int optname,
389 char __user *optval, int optlen);
390 extern void tcp_set_keepalive(struct sock *sk, int val);
391 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
393 size_t len, int nonblock,
394 int flags, int *addr_len);
396 extern void tcp_parse_options(struct sk_buff *skb,
397 struct tcp_options_received *opt_rx,
400 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
403 * TCP v4 functions exported for the inet6 API
406 extern void tcp_v4_send_check(struct sock *sk, int len,
407 struct sk_buff *skb);
409 extern int tcp_v4_conn_request(struct sock *sk,
410 struct sk_buff *skb);
412 extern struct sock * tcp_create_openreq_child(struct sock *sk,
413 struct request_sock *req,
414 struct sk_buff *skb);
416 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
418 struct request_sock *req,
419 struct dst_entry *dst);
421 extern int tcp_v4_do_rcv(struct sock *sk,
422 struct sk_buff *skb);
424 extern int tcp_v4_connect(struct sock *sk,
425 struct sockaddr *uaddr,
428 extern int tcp_connect(struct sock *sk);
430 extern struct sk_buff * tcp_make_synack(struct sock *sk,
431 struct dst_entry *dst,
432 struct request_sock *req);
434 extern int tcp_disconnect(struct sock *sk, int flags);
437 /* From syncookies.c */
438 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
439 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
440 struct ip_options *opt);
441 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
444 extern __u32 cookie_init_timestamp(struct request_sock *req);
445 extern void cookie_check_timestamp(struct tcp_options_received *tcp_opt);
447 /* From net/ipv6/syncookies.c */
448 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
449 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
454 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
456 extern int tcp_may_send_now(struct sock *sk);
457 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
458 extern void tcp_xmit_retransmit_queue(struct sock *);
459 extern void tcp_simple_retransmit(struct sock *);
460 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
461 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
463 extern void tcp_send_probe0(struct sock *);
464 extern void tcp_send_partial(struct sock *);
465 extern int tcp_write_wakeup(struct sock *);
466 extern void tcp_send_fin(struct sock *sk);
467 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
468 extern int tcp_send_synack(struct sock *);
469 extern void tcp_push_one(struct sock *, unsigned int mss_now);
470 extern void tcp_send_ack(struct sock *sk);
471 extern void tcp_send_delayed_ack(struct sock *sk);
474 extern void tcp_cwnd_application_limited(struct sock *sk);
475 extern void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp,
476 struct sk_buff *skb);
479 extern void tcp_init_xmit_timers(struct sock *);
480 static inline void tcp_clear_xmit_timers(struct sock *sk)
482 inet_csk_clear_xmit_timers(sk);
485 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
486 extern unsigned int tcp_current_mss(struct sock *sk, int large);
489 extern void tcp_get_info(struct sock *, struct tcp_info *);
491 /* Read 'sendfile()'-style from a TCP socket */
492 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
493 unsigned int, size_t);
494 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
495 sk_read_actor_t recv_actor);
497 extern void tcp_initialize_rcv_mss(struct sock *sk);
499 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
500 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
501 extern void tcp_mtup_init(struct sock *sk);
503 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
505 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
506 ntohl(TCP_FLAG_ACK) |
510 static inline void tcp_fast_path_on(struct tcp_sock *tp)
512 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
515 static inline void tcp_fast_path_check(struct sock *sk)
517 struct tcp_sock *tp = tcp_sk(sk);
519 if (skb_queue_empty(&tp->out_of_order_queue) &&
521 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
523 tcp_fast_path_on(tp);
526 /* Compute the actual receive window we are currently advertising.
527 * Rcv_nxt can be after the window if our peer push more data
528 * than the offered window.
530 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
532 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
539 /* Choose a new window, without checks for shrinking, and without
540 * scaling applied to the result. The caller does these things
541 * if necessary. This is a "raw" window selection.
543 extern u32 __tcp_select_window(struct sock *sk);
545 /* TCP timestamps are only 32-bits, this causes a slight
546 * complication on 64-bit systems since we store a snapshot
547 * of jiffies in the buffer control blocks below. We decided
548 * to use only the low 32-bits of jiffies and hide the ugly
549 * casts with the following macro.
551 #define tcp_time_stamp ((__u32)(jiffies))
553 /* This is what the send packet queuing engine uses to pass
554 * TCP per-packet control information to the transmission
555 * code. We also store the host-order sequence numbers in
556 * here too. This is 36 bytes on 32-bit architectures,
557 * 40 bytes on 64-bit machines, if this grows please adjust
558 * skbuff.h:skbuff->cb[xxx] size appropriately.
562 struct inet_skb_parm h4;
563 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
564 struct inet6_skb_parm h6;
566 } header; /* For incoming frames */
567 __u32 seq; /* Starting sequence number */
568 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
569 __u32 when; /* used to compute rtt's */
570 __u8 flags; /* TCP header flags. */
572 /* NOTE: These must match up to the flags byte in a
575 #define TCPCB_FLAG_FIN 0x01
576 #define TCPCB_FLAG_SYN 0x02
577 #define TCPCB_FLAG_RST 0x04
578 #define TCPCB_FLAG_PSH 0x08
579 #define TCPCB_FLAG_ACK 0x10
580 #define TCPCB_FLAG_URG 0x20
581 #define TCPCB_FLAG_ECE 0x40
582 #define TCPCB_FLAG_CWR 0x80
584 __u8 sacked; /* State flags for SACK/FACK. */
585 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
586 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
587 #define TCPCB_LOST 0x04 /* SKB is lost */
588 #define TCPCB_TAGBITS 0x07 /* All tag bits */
590 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
591 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
593 __u32 ack_seq; /* Sequence number ACK'd */
596 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
598 /* Due to TSO, an SKB can be composed of multiple actual
599 * packets. To keep these tracked properly, we use this.
601 static inline int tcp_skb_pcount(const struct sk_buff *skb)
603 return skb_shinfo(skb)->gso_segs;
606 /* This is valid iff tcp_skb_pcount() > 1. */
607 static inline int tcp_skb_mss(const struct sk_buff *skb)
609 return skb_shinfo(skb)->gso_size;
612 static inline void tcp_dec_pcount_approx_int(__u32 *count, const int decr)
621 static inline void tcp_dec_pcount_approx(__u32 *count,
622 const struct sk_buff *skb)
624 tcp_dec_pcount_approx_int(count, tcp_skb_pcount(skb));
627 /* Events passed to congestion control interface */
629 CA_EVENT_TX_START, /* first transmit when no packets in flight */
630 CA_EVENT_CWND_RESTART, /* congestion window restart */
631 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
632 CA_EVENT_FRTO, /* fast recovery timeout */
633 CA_EVENT_LOSS, /* loss timeout */
634 CA_EVENT_FAST_ACK, /* in sequence ack */
635 CA_EVENT_SLOW_ACK, /* other ack */
639 * Interface for adding new TCP congestion control handlers
641 #define TCP_CA_NAME_MAX 16
642 #define TCP_CA_MAX 128
643 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
645 #define TCP_CONG_NON_RESTRICTED 0x1
646 #define TCP_CONG_RTT_STAMP 0x2
648 struct tcp_congestion_ops {
649 struct list_head list;
652 /* initialize private data (optional) */
653 void (*init)(struct sock *sk);
654 /* cleanup private data (optional) */
655 void (*release)(struct sock *sk);
657 /* return slow start threshold (required) */
658 u32 (*ssthresh)(struct sock *sk);
659 /* lower bound for congestion window (optional) */
660 u32 (*min_cwnd)(const struct sock *sk);
661 /* do new cwnd calculation (required) */
662 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
663 /* call before changing ca_state (optional) */
664 void (*set_state)(struct sock *sk, u8 new_state);
665 /* call when cwnd event occurs (optional) */
666 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
667 /* new value of cwnd after loss (optional) */
668 u32 (*undo_cwnd)(struct sock *sk);
669 /* hook for packet ack accounting (optional) */
670 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
671 /* get info for inet_diag (optional) */
672 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
674 char name[TCP_CA_NAME_MAX];
675 struct module *owner;
678 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
679 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
681 extern void tcp_init_congestion_control(struct sock *sk);
682 extern void tcp_cleanup_congestion_control(struct sock *sk);
683 extern int tcp_set_default_congestion_control(const char *name);
684 extern void tcp_get_default_congestion_control(char *name);
685 extern void tcp_get_available_congestion_control(char *buf, size_t len);
686 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
687 extern int tcp_set_allowed_congestion_control(char *allowed);
688 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
689 extern void tcp_slow_start(struct tcp_sock *tp);
691 extern struct tcp_congestion_ops tcp_init_congestion_ops;
692 extern u32 tcp_reno_ssthresh(struct sock *sk);
693 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
694 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
695 extern struct tcp_congestion_ops tcp_reno;
697 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
699 struct inet_connection_sock *icsk = inet_csk(sk);
701 if (icsk->icsk_ca_ops->set_state)
702 icsk->icsk_ca_ops->set_state(sk, ca_state);
703 icsk->icsk_ca_state = ca_state;
706 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
708 const struct inet_connection_sock *icsk = inet_csk(sk);
710 if (icsk->icsk_ca_ops->cwnd_event)
711 icsk->icsk_ca_ops->cwnd_event(sk, event);
714 /* These functions determine how the current flow behaves in respect of SACK
715 * handling. SACK is negotiated with the peer, and therefore it can vary
716 * between different flows.
718 * tcp_is_sack - SACK enabled
719 * tcp_is_reno - No SACK
720 * tcp_is_fack - FACK enabled, implies SACK enabled
722 static inline int tcp_is_sack(const struct tcp_sock *tp)
724 return tp->rx_opt.sack_ok;
727 static inline int tcp_is_reno(const struct tcp_sock *tp)
729 return !tcp_is_sack(tp);
732 static inline int tcp_is_fack(const struct tcp_sock *tp)
734 return tp->rx_opt.sack_ok & 2;
737 static inline void tcp_enable_fack(struct tcp_sock *tp)
739 tp->rx_opt.sack_ok |= 2;
742 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
744 return tp->sacked_out + tp->lost_out;
747 /* This determines how many packets are "in the network" to the best
748 * of our knowledge. In many cases it is conservative, but where
749 * detailed information is available from the receiver (via SACK
750 * blocks etc.) we can make more aggressive calculations.
752 * Use this for decisions involving congestion control, use just
753 * tp->packets_out to determine if the send queue is empty or not.
755 * Read this equation as:
757 * "Packets sent once on transmission queue" MINUS
758 * "Packets left network, but not honestly ACKed yet" PLUS
759 * "Packets fast retransmitted"
761 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
763 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
766 extern int tcp_limit_reno_sacked(struct tcp_sock *tp);
768 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
769 * The exception is rate halving phase, when cwnd is decreasing towards
772 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
774 const struct tcp_sock *tp = tcp_sk(sk);
775 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
776 return tp->snd_ssthresh;
778 return max(tp->snd_ssthresh,
779 ((tp->snd_cwnd >> 1) +
780 (tp->snd_cwnd >> 2)));
783 /* Use define here intentionally to get WARN_ON location shown at the caller */
784 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
786 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
787 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
789 /* Slow start with delack produces 3 packets of burst, so that
790 * it is safe "de facto". This will be the default - same as
791 * the default reordering threshold - but if reordering increases,
792 * we must be able to allow cwnd to burst at least this much in order
793 * to not pull it back when holes are filled.
795 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
797 return tp->reordering;
800 /* Returns end sequence number of the receiver's advertised window */
801 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
803 return tp->snd_una + tp->snd_wnd;
805 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
807 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
808 const struct sk_buff *skb)
811 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
814 static inline void tcp_check_probe_timer(struct sock *sk)
816 struct tcp_sock *tp = tcp_sk(sk);
817 const struct inet_connection_sock *icsk = inet_csk(sk);
819 if (!tp->packets_out && !icsk->icsk_pending)
820 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
821 icsk->icsk_rto, TCP_RTO_MAX);
824 static inline void tcp_push_pending_frames(struct sock *sk)
826 struct tcp_sock *tp = tcp_sk(sk);
828 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle);
831 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
836 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
842 * Calculate(/check) TCP checksum
844 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
845 __be32 daddr, __wsum base)
847 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
850 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
852 return __skb_checksum_complete(skb);
855 static inline int tcp_checksum_complete(struct sk_buff *skb)
857 return !skb_csum_unnecessary(skb) &&
858 __tcp_checksum_complete(skb);
861 /* Prequeue for VJ style copy to user, combined with checksumming. */
863 static inline void tcp_prequeue_init(struct tcp_sock *tp)
865 tp->ucopy.task = NULL;
867 tp->ucopy.memory = 0;
868 skb_queue_head_init(&tp->ucopy.prequeue);
869 #ifdef CONFIG_NET_DMA
870 tp->ucopy.dma_chan = NULL;
871 tp->ucopy.wakeup = 0;
872 tp->ucopy.pinned_list = NULL;
873 tp->ucopy.dma_cookie = 0;
877 /* Packet is added to VJ-style prequeue for processing in process
878 * context, if a reader task is waiting. Apparently, this exciting
879 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
880 * failed somewhere. Latency? Burstiness? Well, at least now we will
881 * see, why it failed. 8)8) --ANK
883 * NOTE: is this not too big to inline?
885 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
887 struct tcp_sock *tp = tcp_sk(sk);
889 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
890 __skb_queue_tail(&tp->ucopy.prequeue, skb);
891 tp->ucopy.memory += skb->truesize;
892 if (tp->ucopy.memory > sk->sk_rcvbuf) {
893 struct sk_buff *skb1;
895 BUG_ON(sock_owned_by_user(sk));
897 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
898 sk_backlog_rcv(sk, skb1);
899 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPPREQUEUEDROPPED);
902 tp->ucopy.memory = 0;
903 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
904 wake_up_interruptible(sk->sk_sleep);
905 if (!inet_csk_ack_scheduled(sk))
906 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
907 (3 * TCP_RTO_MIN) / 4,
919 static const char *statename[]={
920 "Unused","Established","Syn Sent","Syn Recv",
921 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
922 "Close Wait","Last ACK","Listen","Closing"
925 extern void tcp_set_state(struct sock *sk, int state);
927 extern void tcp_done(struct 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! */
968 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
969 req->mss = rx_opt->mss_clamp;
970 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
971 ireq->tstamp_ok = rx_opt->tstamp_ok;
972 ireq->sack_ok = rx_opt->sack_ok;
973 ireq->snd_wscale = rx_opt->snd_wscale;
974 ireq->wscale_ok = rx_opt->wscale_ok;
977 ireq->rmt_port = tcp_hdr(skb)->source;
978 ireq->loc_port = tcp_hdr(skb)->dest;
981 extern void tcp_enter_memory_pressure(struct sock *sk);
983 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
985 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
988 static inline int keepalive_time_when(const struct tcp_sock *tp)
990 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
993 static inline int tcp_fin_time(const struct sock *sk)
995 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
996 const int rto = inet_csk(sk)->icsk_rto;
998 if (fin_timeout < (rto << 2) - (rto >> 1))
999 fin_timeout = (rto << 2) - (rto >> 1);
1004 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1006 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1008 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1011 /* RST segments are not recommended to carry timestamp,
1012 and, if they do, it is recommended to ignore PAWS because
1013 "their cleanup function should take precedence over timestamps."
1014 Certainly, it is mistake. It is necessary to understand the reasons
1015 of this constraint to relax it: if peer reboots, clock may go
1016 out-of-sync and half-open connections will not be reset.
1017 Actually, the problem would be not existing if all
1018 the implementations followed draft about maintaining clock
1019 via reboots. Linux-2.2 DOES NOT!
1021 However, we can relax time bounds for RST segments to MSL.
1023 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1028 #define TCP_CHECK_TIMER(sk) do { } while (0)
1030 static inline void tcp_mib_init(struct net *net)
1033 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1034 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1035 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1036 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1040 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1042 tp->lost_skb_hint = NULL;
1043 tp->scoreboard_skb_hint = NULL;
1046 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1048 tcp_clear_retrans_hints_partial(tp);
1049 tp->retransmit_skb_hint = NULL;
1055 /* - key database */
1056 struct tcp_md5sig_key {
1061 struct tcp4_md5sig_key {
1062 struct tcp_md5sig_key base;
1066 struct tcp6_md5sig_key {
1067 struct tcp_md5sig_key base;
1069 u32 scope_id; /* XXX */
1071 struct in6_addr addr;
1075 struct tcp_md5sig_info {
1076 struct tcp4_md5sig_key *keys4;
1077 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1078 struct tcp6_md5sig_key *keys6;
1086 /* - pseudo header */
1087 struct tcp4_pseudohdr {
1095 struct tcp6_pseudohdr {
1096 struct in6_addr saddr;
1097 struct in6_addr daddr;
1099 __be32 protocol; /* including padding */
1102 union tcp_md5sum_block {
1103 struct tcp4_pseudohdr ip4;
1104 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1105 struct tcp6_pseudohdr ip6;
1109 /* - pool: digest algorithm, hash description and scratch buffer */
1110 struct tcp_md5sig_pool {
1111 struct hash_desc md5_desc;
1112 union tcp_md5sum_block md5_blk;
1115 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1118 extern int tcp_v4_md5_hash_skb(char *md5_hash,
1119 struct tcp_md5sig_key *key,
1121 struct request_sock *req,
1122 struct sk_buff *skb);
1124 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1125 struct sock *addr_sk);
1127 extern int tcp_v4_md5_do_add(struct sock *sk,
1132 extern int tcp_v4_md5_do_del(struct sock *sk,
1135 #ifdef CONFIG_TCP_MD5SIG
1136 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1137 &(struct tcp_md5sig_key) { \
1138 .key = (twsk)->tw_md5_key, \
1139 .keylen = (twsk)->tw_md5_keylen, \
1142 #define tcp_twsk_md5_key(twsk) NULL
1145 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1146 extern void tcp_free_md5sig_pool(void);
1148 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1149 extern void __tcp_put_md5sig_pool(void);
1150 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1151 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1152 unsigned header_len);
1153 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1154 struct tcp_md5sig_key *key);
1157 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1159 int cpu = get_cpu();
1160 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1166 static inline void tcp_put_md5sig_pool(void)
1168 __tcp_put_md5sig_pool();
1172 /* write queue abstraction */
1173 static inline void tcp_write_queue_purge(struct sock *sk)
1175 struct sk_buff *skb;
1177 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1178 sk_wmem_free_skb(sk, skb);
1182 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1184 return skb_peek(&sk->sk_write_queue);
1187 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1189 return skb_peek_tail(&sk->sk_write_queue);
1192 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1194 return skb_queue_next(&sk->sk_write_queue, skb);
1197 #define tcp_for_write_queue(skb, sk) \
1198 skb_queue_walk(&(sk)->sk_write_queue, skb)
1200 #define tcp_for_write_queue_from(skb, sk) \
1201 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1203 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1204 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1206 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1208 return sk->sk_send_head;
1211 static inline bool tcp_skb_is_last(const struct sock *sk,
1212 const struct sk_buff *skb)
1214 return skb_queue_is_last(&sk->sk_write_queue, skb);
1217 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1219 if (tcp_skb_is_last(sk, skb))
1220 sk->sk_send_head = NULL;
1222 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1225 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1227 if (sk->sk_send_head == skb_unlinked)
1228 sk->sk_send_head = NULL;
1231 static inline void tcp_init_send_head(struct sock *sk)
1233 sk->sk_send_head = NULL;
1236 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1238 __skb_queue_tail(&sk->sk_write_queue, skb);
1241 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1243 __tcp_add_write_queue_tail(sk, skb);
1245 /* Queue it, remembering where we must start sending. */
1246 if (sk->sk_send_head == NULL) {
1247 sk->sk_send_head = skb;
1249 if (tcp_sk(sk)->highest_sack == NULL)
1250 tcp_sk(sk)->highest_sack = skb;
1254 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1256 __skb_queue_head(&sk->sk_write_queue, skb);
1259 /* Insert buff after skb on the write queue of sk. */
1260 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1261 struct sk_buff *buff,
1264 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1267 /* Insert new before skb on the write queue of sk. */
1268 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1269 struct sk_buff *skb,
1272 __skb_queue_before(&sk->sk_write_queue, skb, new);
1274 if (sk->sk_send_head == skb)
1275 sk->sk_send_head = new;
1278 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1280 __skb_unlink(skb, &sk->sk_write_queue);
1283 static inline int tcp_write_queue_empty(struct sock *sk)
1285 return skb_queue_empty(&sk->sk_write_queue);
1288 /* Start sequence of the highest skb with SACKed bit, valid only if
1289 * sacked > 0 or when the caller has ensured validity by itself.
1291 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1293 if (!tp->sacked_out)
1296 if (tp->highest_sack == NULL)
1299 return TCP_SKB_CB(tp->highest_sack)->seq;
1302 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1304 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1305 tcp_write_queue_next(sk, skb);
1308 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1310 return tcp_sk(sk)->highest_sack;
1313 static inline void tcp_highest_sack_reset(struct sock *sk)
1315 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1318 /* Called when old skb is about to be deleted (to be combined with new skb) */
1319 static inline void tcp_highest_sack_combine(struct sock *sk,
1320 struct sk_buff *old,
1321 struct sk_buff *new)
1323 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1324 tcp_sk(sk)->highest_sack = new;
1328 enum tcp_seq_states {
1329 TCP_SEQ_STATE_LISTENING,
1330 TCP_SEQ_STATE_OPENREQ,
1331 TCP_SEQ_STATE_ESTABLISHED,
1332 TCP_SEQ_STATE_TIME_WAIT,
1335 struct tcp_seq_afinfo {
1338 struct file_operations seq_fops;
1339 struct seq_operations seq_ops;
1342 struct tcp_iter_state {
1343 struct seq_net_private p;
1345 enum tcp_seq_states state;
1346 struct sock *syn_wait_sk;
1347 int bucket, sbucket, num, uid;
1350 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1351 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1353 extern struct request_sock_ops tcp_request_sock_ops;
1354 extern struct request_sock_ops tcp6_request_sock_ops;
1356 extern void tcp_v4_destroy_sock(struct sock *sk);
1358 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1359 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1361 #ifdef CONFIG_PROC_FS
1362 extern int tcp4_proc_init(void);
1363 extern void tcp4_proc_exit(void);
1366 /* TCP af-specific functions */
1367 struct tcp_sock_af_ops {
1368 #ifdef CONFIG_TCP_MD5SIG
1369 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1370 struct sock *addr_sk);
1371 int (*calc_md5_hash) (char *location,
1372 struct tcp_md5sig_key *md5,
1374 struct request_sock *req,
1375 struct sk_buff *skb);
1376 int (*md5_add) (struct sock *sk,
1377 struct sock *addr_sk,
1380 int (*md5_parse) (struct sock *sk,
1381 char __user *optval,
1386 struct tcp_request_sock_ops {
1387 #ifdef CONFIG_TCP_MD5SIG
1388 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1389 struct request_sock *req);
1393 extern void tcp_v4_init(void);
1394 extern void tcp_init(void);