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>
33 #include <net/inet_connection_sock.h>
34 #include <net/inet_timewait_sock.h>
35 #include <net/inet_hashtables.h>
36 #include <net/checksum.h>
37 #include <net/request_sock.h>
41 #include <net/tcp_states.h>
42 #include <net/inet_ecn.h>
44 #include <linux/seq_file.h>
46 extern struct inet_hashinfo tcp_hashinfo;
48 extern atomic_t tcp_orphan_count;
49 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
51 #define MAX_TCP_HEADER (128 + MAX_HEADER)
54 * Never offer a window over 32767 without using window scaling. Some
55 * poor stacks do signed 16bit maths!
57 #define MAX_TCP_WINDOW 32767U
59 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
60 #define TCP_MIN_MSS 88U
62 /* Minimal RCV_MSS. */
63 #define TCP_MIN_RCVMSS 536U
65 /* The least MTU to use for probing */
66 #define TCP_BASE_MSS 512
68 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
69 #define TCP_FASTRETRANS_THRESH 3
71 /* Maximal reordering. */
72 #define TCP_MAX_REORDERING 127
74 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
75 #define TCP_MAX_QUICKACKS 16U
78 #define TCP_URG_VALID 0x0100
79 #define TCP_URG_NOTYET 0x0200
80 #define TCP_URG_READ 0x0400
82 #define TCP_RETR1 3 /*
83 * This is how many retries it does before it
84 * tries to figure out if the gateway is
85 * down. Minimal RFC value is 3; it corresponds
86 * to ~3sec-8min depending on RTO.
89 #define TCP_RETR2 15 /*
90 * This should take at least
91 * 90 minutes to time out.
92 * RFC1122 says that the limit is 100 sec.
93 * 15 is ~13-30min depending on RTO.
96 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
97 * connection: ~180sec is RFC minimum */
99 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
100 * connection: ~180sec is RFC minimum */
103 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
104 * socket. 7 is ~50sec-16min.
108 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
109 * state, about 60 seconds */
110 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
111 /* BSD style FIN_WAIT2 deadlock breaker.
112 * It used to be 3min, new value is 60sec,
113 * to combine FIN-WAIT-2 timeout with
117 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
119 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
120 #define TCP_ATO_MIN ((unsigned)(HZ/25))
122 #define TCP_DELACK_MIN 4U
123 #define TCP_ATO_MIN 4U
125 #define TCP_RTO_MAX ((unsigned)(120*HZ))
126 #define TCP_RTO_MIN ((unsigned)(HZ/5))
127 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
129 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
130 * for local resources.
133 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
134 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
135 #define TCP_KEEPALIVE_INTVL (75*HZ)
137 #define MAX_TCP_KEEPIDLE 32767
138 #define MAX_TCP_KEEPINTVL 32767
139 #define MAX_TCP_KEEPCNT 127
140 #define MAX_TCP_SYNCNT 127
142 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
144 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
145 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
146 * after this time. It should be equal
147 * (or greater than) TCP_TIMEWAIT_LEN
148 * to provide reliability equal to one
149 * provided by timewait state.
151 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
152 * timestamps. It must be less than
153 * minimal timewait lifetime.
159 #define TCPOPT_NOP 1 /* Padding */
160 #define TCPOPT_EOL 0 /* End of options */
161 #define TCPOPT_MSS 2 /* Segment size negotiating */
162 #define TCPOPT_WINDOW 3 /* Window scaling */
163 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
164 #define TCPOPT_SACK 5 /* SACK Block */
165 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
166 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
172 #define TCPOLEN_MSS 4
173 #define TCPOLEN_WINDOW 3
174 #define TCPOLEN_SACK_PERM 2
175 #define TCPOLEN_TIMESTAMP 10
176 #define TCPOLEN_MD5SIG 18
178 /* But this is what stacks really send out. */
179 #define TCPOLEN_TSTAMP_ALIGNED 12
180 #define TCPOLEN_WSCALE_ALIGNED 4
181 #define TCPOLEN_SACKPERM_ALIGNED 4
182 #define TCPOLEN_SACK_BASE 2
183 #define TCPOLEN_SACK_BASE_ALIGNED 4
184 #define TCPOLEN_SACK_PERBLOCK 8
185 #define TCPOLEN_MD5SIG_ALIGNED 20
187 /* Flags in tp->nonagle */
188 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
189 #define TCP_NAGLE_CORK 2 /* Socket is corked */
190 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
192 extern struct inet_timewait_death_row tcp_death_row;
194 /* sysctl variables for tcp */
195 extern int sysctl_tcp_timestamps;
196 extern int sysctl_tcp_window_scaling;
197 extern int sysctl_tcp_sack;
198 extern int sysctl_tcp_fin_timeout;
199 extern int sysctl_tcp_keepalive_time;
200 extern int sysctl_tcp_keepalive_probes;
201 extern int sysctl_tcp_keepalive_intvl;
202 extern int sysctl_tcp_syn_retries;
203 extern int sysctl_tcp_synack_retries;
204 extern int sysctl_tcp_retries1;
205 extern int sysctl_tcp_retries2;
206 extern int sysctl_tcp_orphan_retries;
207 extern int sysctl_tcp_syncookies;
208 extern int sysctl_tcp_retrans_collapse;
209 extern int sysctl_tcp_stdurg;
210 extern int sysctl_tcp_rfc1337;
211 extern int sysctl_tcp_abort_on_overflow;
212 extern int sysctl_tcp_max_orphans;
213 extern int sysctl_tcp_fack;
214 extern int sysctl_tcp_reordering;
215 extern int sysctl_tcp_ecn;
216 extern int sysctl_tcp_dsack;
217 extern int sysctl_tcp_mem[3];
218 extern int sysctl_tcp_wmem[3];
219 extern int sysctl_tcp_rmem[3];
220 extern int sysctl_tcp_app_win;
221 extern int sysctl_tcp_adv_win_scale;
222 extern int sysctl_tcp_tw_reuse;
223 extern int sysctl_tcp_frto;
224 extern int sysctl_tcp_frto_response;
225 extern int sysctl_tcp_low_latency;
226 extern int sysctl_tcp_dma_copybreak;
227 extern int sysctl_tcp_nometrics_save;
228 extern int sysctl_tcp_moderate_rcvbuf;
229 extern int sysctl_tcp_tso_win_divisor;
230 extern int sysctl_tcp_abc;
231 extern int sysctl_tcp_mtu_probing;
232 extern int sysctl_tcp_base_mss;
233 extern int sysctl_tcp_workaround_signed_windows;
234 extern int sysctl_tcp_slow_start_after_idle;
235 extern int sysctl_tcp_max_ssthresh;
237 extern atomic_t tcp_memory_allocated;
238 extern atomic_t tcp_sockets_allocated;
239 extern int tcp_memory_pressure;
242 * The next routines deal with comparing 32 bit unsigned ints
243 * and worry about wraparound (automatic with unsigned arithmetic).
246 static inline int before(__u32 seq1, __u32 seq2)
248 return (__s32)(seq1-seq2) < 0;
250 #define after(seq2, seq1) before(seq1, seq2)
252 /* is s2<=s1<=s3 ? */
253 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
255 return seq3 - seq2 >= seq1 - seq2;
258 static inline int tcp_too_many_orphans(struct sock *sk, int num)
260 return (num > sysctl_tcp_max_orphans) ||
261 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
262 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
265 extern struct proto tcp_prot;
267 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
268 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
269 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
270 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
271 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
272 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
273 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(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,
401 * TCP v4 functions exported for the inet6 API
404 extern void tcp_v4_send_check(struct sock *sk, int len,
405 struct sk_buff *skb);
407 extern int tcp_v4_conn_request(struct sock *sk,
408 struct sk_buff *skb);
410 extern struct sock * tcp_create_openreq_child(struct sock *sk,
411 struct request_sock *req,
412 struct sk_buff *skb);
414 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
416 struct request_sock *req,
417 struct dst_entry *dst);
419 extern int tcp_v4_do_rcv(struct sock *sk,
420 struct sk_buff *skb);
422 extern int tcp_v4_connect(struct sock *sk,
423 struct sockaddr *uaddr,
426 extern int tcp_connect(struct sock *sk);
428 extern struct sk_buff * tcp_make_synack(struct sock *sk,
429 struct dst_entry *dst,
430 struct request_sock *req);
432 extern int tcp_disconnect(struct sock *sk, int flags);
434 extern void tcp_unhash(struct sock *sk);
436 /* From syncookies.c */
437 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
438 struct ip_options *opt);
439 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
444 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
446 extern int tcp_may_send_now(struct sock *sk);
447 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
448 extern void tcp_xmit_retransmit_queue(struct sock *);
449 extern void tcp_simple_retransmit(struct sock *);
450 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
451 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
453 extern void tcp_send_probe0(struct sock *);
454 extern void tcp_send_partial(struct sock *);
455 extern int tcp_write_wakeup(struct sock *);
456 extern void tcp_send_fin(struct sock *sk);
457 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
458 extern int tcp_send_synack(struct sock *);
459 extern void tcp_push_one(struct sock *, unsigned int mss_now);
460 extern void tcp_send_ack(struct sock *sk);
461 extern void tcp_send_delayed_ack(struct sock *sk);
464 extern void tcp_cwnd_application_limited(struct sock *sk);
467 extern void tcp_init_xmit_timers(struct sock *);
468 static inline void tcp_clear_xmit_timers(struct sock *sk)
470 inet_csk_clear_xmit_timers(sk);
473 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
474 extern unsigned int tcp_current_mss(struct sock *sk, int large);
477 extern void tcp_get_info(struct sock *, struct tcp_info *);
479 /* Read 'sendfile()'-style from a TCP socket */
480 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
481 unsigned int, size_t);
482 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
483 sk_read_actor_t recv_actor);
485 extern void tcp_initialize_rcv_mss(struct sock *sk);
487 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
488 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
489 extern void tcp_mtup_init(struct sock *sk);
491 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
493 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
494 ntohl(TCP_FLAG_ACK) |
498 static inline void tcp_fast_path_on(struct tcp_sock *tp)
500 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
503 static inline void tcp_fast_path_check(struct sock *sk)
505 struct tcp_sock *tp = tcp_sk(sk);
507 if (skb_queue_empty(&tp->out_of_order_queue) &&
509 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
511 tcp_fast_path_on(tp);
514 /* Compute the actual receive window we are currently advertising.
515 * Rcv_nxt can be after the window if our peer push more data
516 * than the offered window.
518 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
520 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
527 /* Choose a new window, without checks for shrinking, and without
528 * scaling applied to the result. The caller does these things
529 * if necessary. This is a "raw" window selection.
531 extern u32 __tcp_select_window(struct sock *sk);
533 /* TCP timestamps are only 32-bits, this causes a slight
534 * complication on 64-bit systems since we store a snapshot
535 * of jiffies in the buffer control blocks below. We decided
536 * to use only the low 32-bits of jiffies and hide the ugly
537 * casts with the following macro.
539 #define tcp_time_stamp ((__u32)(jiffies))
541 /* This is what the send packet queuing engine uses to pass
542 * TCP per-packet control information to the transmission
543 * code. We also store the host-order sequence numbers in
544 * here too. This is 36 bytes on 32-bit architectures,
545 * 40 bytes on 64-bit machines, if this grows please adjust
546 * skbuff.h:skbuff->cb[xxx] size appropriately.
550 struct inet_skb_parm h4;
551 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
552 struct inet6_skb_parm h6;
554 } header; /* For incoming frames */
555 __u32 seq; /* Starting sequence number */
556 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
557 __u32 when; /* used to compute rtt's */
558 __u8 flags; /* TCP header flags. */
560 /* NOTE: These must match up to the flags byte in a
563 #define TCPCB_FLAG_FIN 0x01
564 #define TCPCB_FLAG_SYN 0x02
565 #define TCPCB_FLAG_RST 0x04
566 #define TCPCB_FLAG_PSH 0x08
567 #define TCPCB_FLAG_ACK 0x10
568 #define TCPCB_FLAG_URG 0x20
569 #define TCPCB_FLAG_ECE 0x40
570 #define TCPCB_FLAG_CWR 0x80
572 __u8 sacked; /* State flags for SACK/FACK. */
573 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
574 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
575 #define TCPCB_LOST 0x04 /* SKB is lost */
576 #define TCPCB_TAGBITS 0x07 /* All tag bits */
578 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
579 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
581 __u16 urg_ptr; /* Valid w/URG flags is set. */
582 __u32 ack_seq; /* Sequence number ACK'd */
585 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
587 /* Due to TSO, an SKB can be composed of multiple actual
588 * packets. To keep these tracked properly, we use this.
590 static inline int tcp_skb_pcount(const struct sk_buff *skb)
592 return skb_shinfo(skb)->gso_segs;
595 /* This is valid iff tcp_skb_pcount() > 1. */
596 static inline int tcp_skb_mss(const struct sk_buff *skb)
598 return skb_shinfo(skb)->gso_size;
601 static inline void tcp_dec_pcount_approx_int(__u32 *count, const int decr)
610 static inline void tcp_dec_pcount_approx(__u32 *count,
611 const struct sk_buff *skb)
613 tcp_dec_pcount_approx_int(count, tcp_skb_pcount(skb));
616 /* Events passed to congestion control interface */
618 CA_EVENT_TX_START, /* first transmit when no packets in flight */
619 CA_EVENT_CWND_RESTART, /* congestion window restart */
620 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
621 CA_EVENT_FRTO, /* fast recovery timeout */
622 CA_EVENT_LOSS, /* loss timeout */
623 CA_EVENT_FAST_ACK, /* in sequence ack */
624 CA_EVENT_SLOW_ACK, /* other ack */
628 * Interface for adding new TCP congestion control handlers
630 #define TCP_CA_NAME_MAX 16
631 #define TCP_CA_MAX 128
632 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
634 #define TCP_CONG_NON_RESTRICTED 0x1
635 #define TCP_CONG_RTT_STAMP 0x2
637 struct tcp_congestion_ops {
638 struct list_head list;
641 /* initialize private data (optional) */
642 void (*init)(struct sock *sk);
643 /* cleanup private data (optional) */
644 void (*release)(struct sock *sk);
646 /* return slow start threshold (required) */
647 u32 (*ssthresh)(struct sock *sk);
648 /* lower bound for congestion window (optional) */
649 u32 (*min_cwnd)(const struct sock *sk);
650 /* do new cwnd calculation (required) */
651 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
652 /* call before changing ca_state (optional) */
653 void (*set_state)(struct sock *sk, u8 new_state);
654 /* call when cwnd event occurs (optional) */
655 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
656 /* new value of cwnd after loss (optional) */
657 u32 (*undo_cwnd)(struct sock *sk);
658 /* hook for packet ack accounting (optional) */
659 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
660 /* get info for inet_diag (optional) */
661 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
663 char name[TCP_CA_NAME_MAX];
664 struct module *owner;
667 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
668 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
670 extern void tcp_init_congestion_control(struct sock *sk);
671 extern void tcp_cleanup_congestion_control(struct sock *sk);
672 extern int tcp_set_default_congestion_control(const char *name);
673 extern void tcp_get_default_congestion_control(char *name);
674 extern void tcp_get_available_congestion_control(char *buf, size_t len);
675 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
676 extern int tcp_set_allowed_congestion_control(char *allowed);
677 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
678 extern void tcp_slow_start(struct tcp_sock *tp);
680 extern struct tcp_congestion_ops tcp_init_congestion_ops;
681 extern u32 tcp_reno_ssthresh(struct sock *sk);
682 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
683 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
684 extern struct tcp_congestion_ops tcp_reno;
686 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
688 struct inet_connection_sock *icsk = inet_csk(sk);
690 if (icsk->icsk_ca_ops->set_state)
691 icsk->icsk_ca_ops->set_state(sk, ca_state);
692 icsk->icsk_ca_state = ca_state;
695 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
697 const struct inet_connection_sock *icsk = inet_csk(sk);
699 if (icsk->icsk_ca_ops->cwnd_event)
700 icsk->icsk_ca_ops->cwnd_event(sk, event);
703 /* These functions determine how the current flow behaves in respect of SACK
704 * handling. SACK is negotiated with the peer, and therefore it can vary
705 * between different flows.
707 * tcp_is_sack - SACK enabled
708 * tcp_is_reno - No SACK
709 * tcp_is_fack - FACK enabled, implies SACK enabled
711 static inline int tcp_is_sack(const struct tcp_sock *tp)
713 return tp->rx_opt.sack_ok;
716 static inline int tcp_is_reno(const struct tcp_sock *tp)
718 return !tcp_is_sack(tp);
721 static inline int tcp_is_fack(const struct tcp_sock *tp)
723 return tp->rx_opt.sack_ok & 2;
726 static inline void tcp_enable_fack(struct tcp_sock *tp)
728 tp->rx_opt.sack_ok |= 2;
731 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
733 return tp->sacked_out + tp->lost_out;
736 /* This determines how many packets are "in the network" to the best
737 * of our knowledge. In many cases it is conservative, but where
738 * detailed information is available from the receiver (via SACK
739 * blocks etc.) we can make more aggressive calculations.
741 * Use this for decisions involving congestion control, use just
742 * tp->packets_out to determine if the send queue is empty or not.
744 * Read this equation as:
746 * "Packets sent once on transmission queue" MINUS
747 * "Packets left network, but not honestly ACKed yet" PLUS
748 * "Packets fast retransmitted"
750 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
752 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
755 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
756 * The exception is rate halving phase, when cwnd is decreasing towards
759 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
761 const struct tcp_sock *tp = tcp_sk(sk);
762 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
763 return tp->snd_ssthresh;
765 return max(tp->snd_ssthresh,
766 ((tp->snd_cwnd >> 1) +
767 (tp->snd_cwnd >> 2)));
770 /* Use define here intentionally to get WARN_ON location shown at the caller */
771 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
773 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
774 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
776 /* Slow start with delack produces 3 packets of burst, so that
777 * it is safe "de facto".
779 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
784 /* Returns end sequence number of the receiver's advertised window */
785 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
787 return tp->snd_una + tp->snd_wnd;
789 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
791 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
792 const struct sk_buff *skb)
795 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
798 static inline void tcp_check_probe_timer(struct sock *sk)
800 struct tcp_sock *tp = tcp_sk(sk);
801 const struct inet_connection_sock *icsk = inet_csk(sk);
803 if (!tp->packets_out && !icsk->icsk_pending)
804 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
805 icsk->icsk_rto, TCP_RTO_MAX);
808 static inline void tcp_push_pending_frames(struct sock *sk)
810 struct tcp_sock *tp = tcp_sk(sk);
812 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle);
815 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
820 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
826 * Calculate(/check) TCP checksum
828 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
829 __be32 daddr, __wsum base)
831 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
834 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
836 return __skb_checksum_complete(skb);
839 static inline int tcp_checksum_complete(struct sk_buff *skb)
841 return !skb_csum_unnecessary(skb) &&
842 __tcp_checksum_complete(skb);
845 /* Prequeue for VJ style copy to user, combined with checksumming. */
847 static inline void tcp_prequeue_init(struct tcp_sock *tp)
849 tp->ucopy.task = NULL;
851 tp->ucopy.memory = 0;
852 skb_queue_head_init(&tp->ucopy.prequeue);
853 #ifdef CONFIG_NET_DMA
854 tp->ucopy.dma_chan = NULL;
855 tp->ucopy.wakeup = 0;
856 tp->ucopy.pinned_list = NULL;
857 tp->ucopy.dma_cookie = 0;
861 /* Packet is added to VJ-style prequeue for processing in process
862 * context, if a reader task is waiting. Apparently, this exciting
863 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
864 * failed somewhere. Latency? Burstiness? Well, at least now we will
865 * see, why it failed. 8)8) --ANK
867 * NOTE: is this not too big to inline?
869 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
871 struct tcp_sock *tp = tcp_sk(sk);
873 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
874 __skb_queue_tail(&tp->ucopy.prequeue, skb);
875 tp->ucopy.memory += skb->truesize;
876 if (tp->ucopy.memory > sk->sk_rcvbuf) {
877 struct sk_buff *skb1;
879 BUG_ON(sock_owned_by_user(sk));
881 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
882 sk->sk_backlog_rcv(sk, skb1);
883 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
886 tp->ucopy.memory = 0;
887 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
888 wake_up_interruptible(sk->sk_sleep);
889 if (!inet_csk_ack_scheduled(sk))
890 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
891 (3 * TCP_RTO_MIN) / 4,
903 static const char *statename[]={
904 "Unused","Established","Syn Sent","Syn Recv",
905 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
906 "Close Wait","Last ACK","Listen","Closing"
909 extern void tcp_set_state(struct sock *sk, int state);
911 extern void tcp_done(struct sock *sk);
913 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
916 rx_opt->eff_sacks = 0;
917 rx_opt->num_sacks = 0;
920 /* Determine a window scaling and initial window to offer. */
921 extern void tcp_select_initial_window(int __space, __u32 mss,
922 __u32 *rcv_wnd, __u32 *window_clamp,
923 int wscale_ok, __u8 *rcv_wscale);
925 static inline int tcp_win_from_space(int space)
927 return sysctl_tcp_adv_win_scale<=0 ?
928 (space>>(-sysctl_tcp_adv_win_scale)) :
929 space - (space>>sysctl_tcp_adv_win_scale);
932 /* Note: caller must be prepared to deal with negative returns */
933 static inline int tcp_space(const struct sock *sk)
935 return tcp_win_from_space(sk->sk_rcvbuf -
936 atomic_read(&sk->sk_rmem_alloc));
939 static inline int tcp_full_space(const struct sock *sk)
941 return tcp_win_from_space(sk->sk_rcvbuf);
944 static inline void tcp_openreq_init(struct request_sock *req,
945 struct tcp_options_received *rx_opt,
948 struct inet_request_sock *ireq = inet_rsk(req);
950 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
951 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
952 req->mss = rx_opt->mss_clamp;
953 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
954 ireq->tstamp_ok = rx_opt->tstamp_ok;
955 ireq->sack_ok = rx_opt->sack_ok;
956 ireq->snd_wscale = rx_opt->snd_wscale;
957 ireq->wscale_ok = rx_opt->wscale_ok;
960 ireq->rmt_port = tcp_hdr(skb)->source;
963 extern void tcp_enter_memory_pressure(void);
965 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
967 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
970 static inline int keepalive_time_when(const struct tcp_sock *tp)
972 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
975 static inline int tcp_fin_time(const struct sock *sk)
977 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
978 const int rto = inet_csk(sk)->icsk_rto;
980 if (fin_timeout < (rto << 2) - (rto >> 1))
981 fin_timeout = (rto << 2) - (rto >> 1);
986 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
988 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
990 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
993 /* RST segments are not recommended to carry timestamp,
994 and, if they do, it is recommended to ignore PAWS because
995 "their cleanup function should take precedence over timestamps."
996 Certainly, it is mistake. It is necessary to understand the reasons
997 of this constraint to relax it: if peer reboots, clock may go
998 out-of-sync and half-open connections will not be reset.
999 Actually, the problem would be not existing if all
1000 the implementations followed draft about maintaining clock
1001 via reboots. Linux-2.2 DOES NOT!
1003 However, we can relax time bounds for RST segments to MSL.
1005 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1010 #define TCP_CHECK_TIMER(sk) do { } while (0)
1012 static inline void tcp_mib_init(void)
1015 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1016 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1017 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1018 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1022 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1024 tp->lost_skb_hint = NULL;
1025 tp->scoreboard_skb_hint = NULL;
1026 tp->retransmit_skb_hint = NULL;
1027 tp->forward_skb_hint = NULL;
1030 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1032 tcp_clear_retrans_hints_partial(tp);
1038 /* - key database */
1039 struct tcp_md5sig_key {
1044 struct tcp4_md5sig_key {
1045 struct tcp_md5sig_key base;
1049 struct tcp6_md5sig_key {
1050 struct tcp_md5sig_key base;
1052 u32 scope_id; /* XXX */
1054 struct in6_addr addr;
1058 struct tcp_md5sig_info {
1059 struct tcp4_md5sig_key *keys4;
1060 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1061 struct tcp6_md5sig_key *keys6;
1069 /* - pseudo header */
1070 struct tcp4_pseudohdr {
1078 struct tcp6_pseudohdr {
1079 struct in6_addr saddr;
1080 struct in6_addr daddr;
1082 __be32 protocol; /* including padding */
1085 union tcp_md5sum_block {
1086 struct tcp4_pseudohdr ip4;
1087 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1088 struct tcp6_pseudohdr ip6;
1092 /* - pool: digest algorithm, hash description and scratch buffer */
1093 struct tcp_md5sig_pool {
1094 struct hash_desc md5_desc;
1095 union tcp_md5sum_block md5_blk;
1098 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1101 extern int tcp_v4_calc_md5_hash(char *md5_hash,
1102 struct tcp_md5sig_key *key,
1104 struct dst_entry *dst,
1105 struct request_sock *req,
1108 unsigned int tcplen);
1109 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1110 struct sock *addr_sk);
1112 extern int tcp_v4_md5_do_add(struct sock *sk,
1117 extern int tcp_v4_md5_do_del(struct sock *sk,
1120 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1121 extern void tcp_free_md5sig_pool(void);
1123 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1124 extern void __tcp_put_md5sig_pool(void);
1127 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1129 int cpu = get_cpu();
1130 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1136 static inline void tcp_put_md5sig_pool(void)
1138 __tcp_put_md5sig_pool();
1142 /* write queue abstraction */
1143 static inline void tcp_write_queue_purge(struct sock *sk)
1145 struct sk_buff *skb;
1147 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1148 sk_wmem_free_skb(sk, skb);
1152 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1154 struct sk_buff *skb = sk->sk_write_queue.next;
1155 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1160 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1162 struct sk_buff *skb = sk->sk_write_queue.prev;
1163 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1168 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1173 #define tcp_for_write_queue(skb, sk) \
1174 for (skb = (sk)->sk_write_queue.next; \
1175 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1178 #define tcp_for_write_queue_from(skb, sk) \
1179 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\
1182 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1183 for (tmp = skb->next; \
1184 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1185 skb = tmp, tmp = skb->next)
1187 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1189 return sk->sk_send_head;
1192 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1194 sk->sk_send_head = skb->next;
1195 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue)
1196 sk->sk_send_head = NULL;
1199 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1201 if (sk->sk_send_head == skb_unlinked)
1202 sk->sk_send_head = NULL;
1205 static inline void tcp_init_send_head(struct sock *sk)
1207 sk->sk_send_head = NULL;
1210 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1212 __skb_queue_tail(&sk->sk_write_queue, skb);
1215 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1217 __tcp_add_write_queue_tail(sk, skb);
1219 /* Queue it, remembering where we must start sending. */
1220 if (sk->sk_send_head == NULL) {
1221 sk->sk_send_head = skb;
1223 if (tcp_sk(sk)->highest_sack == NULL)
1224 tcp_sk(sk)->highest_sack = skb;
1228 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1230 __skb_queue_head(&sk->sk_write_queue, skb);
1233 /* Insert buff after skb on the write queue of sk. */
1234 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1235 struct sk_buff *buff,
1238 __skb_append(skb, buff, &sk->sk_write_queue);
1241 /* Insert skb between prev and next on the write queue of sk. */
1242 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1243 struct sk_buff *skb,
1246 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue);
1248 if (sk->sk_send_head == skb)
1249 sk->sk_send_head = new;
1252 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1254 __skb_unlink(skb, &sk->sk_write_queue);
1257 static inline int tcp_skb_is_last(const struct sock *sk,
1258 const struct sk_buff *skb)
1260 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1263 static inline int tcp_write_queue_empty(struct sock *sk)
1265 return skb_queue_empty(&sk->sk_write_queue);
1268 /* Start sequence of the highest skb with SACKed bit, valid only if
1269 * sacked > 0 or when the caller has ensured validity by itself.
1271 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1273 if (!tp->sacked_out)
1276 if (tp->highest_sack == NULL)
1279 return TCP_SKB_CB(tp->highest_sack)->seq;
1282 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1284 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1285 tcp_write_queue_next(sk, skb);
1288 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1290 return tcp_sk(sk)->highest_sack;
1293 static inline void tcp_highest_sack_reset(struct sock *sk)
1295 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1298 /* Called when old skb is about to be deleted (to be combined with new skb) */
1299 static inline void tcp_highest_sack_combine(struct sock *sk,
1300 struct sk_buff *old,
1301 struct sk_buff *new)
1303 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1304 tcp_sk(sk)->highest_sack = new;
1308 enum tcp_seq_states {
1309 TCP_SEQ_STATE_LISTENING,
1310 TCP_SEQ_STATE_OPENREQ,
1311 TCP_SEQ_STATE_ESTABLISHED,
1312 TCP_SEQ_STATE_TIME_WAIT,
1315 struct tcp_seq_afinfo {
1316 struct module *owner;
1319 int (*seq_show) (struct seq_file *m, void *v);
1320 struct file_operations *seq_fops;
1323 struct tcp_iter_state {
1325 enum tcp_seq_states state;
1326 struct sock *syn_wait_sk;
1327 int bucket, sbucket, num, uid;
1328 struct seq_operations seq_ops;
1331 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1332 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1334 extern struct request_sock_ops tcp_request_sock_ops;
1336 extern int tcp_v4_destroy_sock(struct sock *sk);
1338 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1339 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1341 #ifdef CONFIG_PROC_FS
1342 extern int tcp4_proc_init(void);
1343 extern void tcp4_proc_exit(void);
1346 /* TCP af-specific functions */
1347 struct tcp_sock_af_ops {
1348 #ifdef CONFIG_TCP_MD5SIG
1349 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1350 struct sock *addr_sk);
1351 int (*calc_md5_hash) (char *location,
1352 struct tcp_md5sig_key *md5,
1354 struct dst_entry *dst,
1355 struct request_sock *req,
1359 int (*md5_add) (struct sock *sk,
1360 struct sock *addr_sk,
1363 int (*md5_parse) (struct sock *sk,
1364 char __user *optval,
1369 struct tcp_request_sock_ops {
1370 #ifdef CONFIG_TCP_MD5SIG
1371 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1372 struct request_sock *req);
1376 extern void tcp_v4_init(struct net_proto_family *ops);
1377 extern void tcp_init(void);