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 static inline void tcp_dec_quickack_mode(struct sock *sk,
313 const unsigned int pkts)
315 struct inet_connection_sock *icsk = inet_csk(sk);
317 if (icsk->icsk_ack.quick) {
318 if (pkts >= icsk->icsk_ack.quick) {
319 icsk->icsk_ack.quick = 0;
320 /* Leaving quickack mode we deflate ATO. */
321 icsk->icsk_ack.ato = TCP_ATO_MIN;
323 icsk->icsk_ack.quick -= pkts;
327 extern void tcp_enter_quickack_mode(struct sock *sk);
329 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
331 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
335 #define TCP_ECN_QUEUE_CWR 2
336 #define TCP_ECN_DEMAND_CWR 4
338 static __inline__ void
339 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
341 if (sysctl_tcp_ecn && th->ece && th->cwr)
342 inet_rsk(req)->ecn_ok = 1;
354 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
356 const struct tcphdr *th);
358 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
359 struct request_sock *req,
360 struct request_sock **prev);
361 extern int tcp_child_process(struct sock *parent,
363 struct sk_buff *skb);
364 extern int tcp_use_frto(struct sock *sk);
365 extern void tcp_enter_frto(struct sock *sk);
366 extern void tcp_enter_loss(struct sock *sk, int how);
367 extern void tcp_clear_retrans(struct tcp_sock *tp);
368 extern void tcp_update_metrics(struct sock *sk);
370 extern void tcp_close(struct sock *sk,
372 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
374 extern int tcp_getsockopt(struct sock *sk, int level,
378 extern int tcp_setsockopt(struct sock *sk, int level,
379 int optname, char __user *optval,
381 extern int compat_tcp_getsockopt(struct sock *sk,
382 int level, int optname,
383 char __user *optval, int __user *optlen);
384 extern int compat_tcp_setsockopt(struct sock *sk,
385 int level, int optname,
386 char __user *optval, int optlen);
387 extern void tcp_set_keepalive(struct sock *sk, int val);
388 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
390 size_t len, int nonblock,
391 int flags, int *addr_len);
393 extern void tcp_parse_options(struct sk_buff *skb,
394 struct tcp_options_received *opt_rx,
398 * TCP v4 functions exported for the inet6 API
401 extern void tcp_v4_send_check(struct sock *sk, int len,
402 struct sk_buff *skb);
404 extern int tcp_v4_conn_request(struct sock *sk,
405 struct sk_buff *skb);
407 extern struct sock * tcp_create_openreq_child(struct sock *sk,
408 struct request_sock *req,
409 struct sk_buff *skb);
411 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
413 struct request_sock *req,
414 struct dst_entry *dst);
416 extern int tcp_v4_do_rcv(struct sock *sk,
417 struct sk_buff *skb);
419 extern int tcp_v4_connect(struct sock *sk,
420 struct sockaddr *uaddr,
423 extern int tcp_connect(struct sock *sk);
425 extern struct sk_buff * tcp_make_synack(struct sock *sk,
426 struct dst_entry *dst,
427 struct request_sock *req);
429 extern int tcp_disconnect(struct sock *sk, int flags);
431 extern void tcp_unhash(struct sock *sk);
433 /* From syncookies.c */
434 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
435 struct ip_options *opt);
436 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
441 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
443 extern int tcp_may_send_now(struct sock *sk);
444 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
445 extern void tcp_xmit_retransmit_queue(struct sock *);
446 extern void tcp_simple_retransmit(struct sock *);
447 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
448 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
450 extern void tcp_send_probe0(struct sock *);
451 extern void tcp_send_partial(struct sock *);
452 extern int tcp_write_wakeup(struct sock *);
453 extern void tcp_send_fin(struct sock *sk);
454 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
455 extern int tcp_send_synack(struct sock *);
456 extern void tcp_push_one(struct sock *, unsigned int mss_now);
457 extern void tcp_send_ack(struct sock *sk);
458 extern void tcp_send_delayed_ack(struct sock *sk);
461 extern void tcp_cwnd_application_limited(struct sock *sk);
464 extern void tcp_init_xmit_timers(struct sock *);
465 static inline void tcp_clear_xmit_timers(struct sock *sk)
467 inet_csk_clear_xmit_timers(sk);
470 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
471 extern unsigned int tcp_current_mss(struct sock *sk, int large);
474 extern void tcp_get_info(struct sock *, struct tcp_info *);
476 /* Read 'sendfile()'-style from a TCP socket */
477 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
478 unsigned int, size_t);
479 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
480 sk_read_actor_t recv_actor);
482 extern void tcp_initialize_rcv_mss(struct sock *sk);
484 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
485 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
486 extern void tcp_mtup_init(struct sock *sk);
488 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
490 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
491 ntohl(TCP_FLAG_ACK) |
495 static inline void tcp_fast_path_on(struct tcp_sock *tp)
497 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
500 static inline void tcp_fast_path_check(struct sock *sk)
502 struct tcp_sock *tp = tcp_sk(sk);
504 if (skb_queue_empty(&tp->out_of_order_queue) &&
506 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
508 tcp_fast_path_on(tp);
511 /* Compute the actual receive window we are currently advertising.
512 * Rcv_nxt can be after the window if our peer push more data
513 * than the offered window.
515 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
517 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
524 /* Choose a new window, without checks for shrinking, and without
525 * scaling applied to the result. The caller does these things
526 * if necessary. This is a "raw" window selection.
528 extern u32 __tcp_select_window(struct sock *sk);
530 /* TCP timestamps are only 32-bits, this causes a slight
531 * complication on 64-bit systems since we store a snapshot
532 * of jiffies in the buffer control blocks below. We decided
533 * to use only the low 32-bits of jiffies and hide the ugly
534 * casts with the following macro.
536 #define tcp_time_stamp ((__u32)(jiffies))
538 /* This is what the send packet queuing engine uses to pass
539 * TCP per-packet control information to the transmission
540 * code. We also store the host-order sequence numbers in
541 * here too. This is 36 bytes on 32-bit architectures,
542 * 40 bytes on 64-bit machines, if this grows please adjust
543 * skbuff.h:skbuff->cb[xxx] size appropriately.
547 struct inet_skb_parm h4;
548 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
549 struct inet6_skb_parm h6;
551 } header; /* For incoming frames */
552 __u32 seq; /* Starting sequence number */
553 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
554 __u32 when; /* used to compute rtt's */
555 __u8 flags; /* TCP header flags. */
557 /* NOTE: These must match up to the flags byte in a
560 #define TCPCB_FLAG_FIN 0x01
561 #define TCPCB_FLAG_SYN 0x02
562 #define TCPCB_FLAG_RST 0x04
563 #define TCPCB_FLAG_PSH 0x08
564 #define TCPCB_FLAG_ACK 0x10
565 #define TCPCB_FLAG_URG 0x20
566 #define TCPCB_FLAG_ECE 0x40
567 #define TCPCB_FLAG_CWR 0x80
569 __u8 sacked; /* State flags for SACK/FACK. */
570 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
571 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
572 #define TCPCB_LOST 0x04 /* SKB is lost */
573 #define TCPCB_TAGBITS 0x07 /* All tag bits */
575 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
576 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
578 #define TCPCB_URG 0x20 /* Urgent pointer advanced here */
580 #define TCPCB_AT_TAIL (TCPCB_URG)
582 __u16 urg_ptr; /* Valid w/URG flags is set. */
583 __u32 ack_seq; /* Sequence number ACK'd */
586 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
588 /* Due to TSO, an SKB can be composed of multiple actual
589 * packets. To keep these tracked properly, we use this.
591 static inline int tcp_skb_pcount(const struct sk_buff *skb)
593 return skb_shinfo(skb)->gso_segs;
596 /* This is valid iff tcp_skb_pcount() > 1. */
597 static inline int tcp_skb_mss(const struct sk_buff *skb)
599 return skb_shinfo(skb)->gso_size;
602 static inline void tcp_dec_pcount_approx_int(__u32 *count, const int decr)
611 static inline void tcp_dec_pcount_approx(__u32 *count,
612 const struct sk_buff *skb)
614 tcp_dec_pcount_approx_int(count, tcp_skb_pcount(skb));
617 /* Events passed to congestion control interface */
619 CA_EVENT_TX_START, /* first transmit when no packets in flight */
620 CA_EVENT_CWND_RESTART, /* congestion window restart */
621 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
622 CA_EVENT_FRTO, /* fast recovery timeout */
623 CA_EVENT_LOSS, /* loss timeout */
624 CA_EVENT_FAST_ACK, /* in sequence ack */
625 CA_EVENT_SLOW_ACK, /* other ack */
629 * Interface for adding new TCP congestion control handlers
631 #define TCP_CA_NAME_MAX 16
632 #define TCP_CA_MAX 128
633 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
635 #define TCP_CONG_NON_RESTRICTED 0x1
636 #define TCP_CONG_RTT_STAMP 0x2
638 struct tcp_congestion_ops {
639 struct list_head list;
642 /* initialize private data (optional) */
643 void (*init)(struct sock *sk);
644 /* cleanup private data (optional) */
645 void (*release)(struct sock *sk);
647 /* return slow start threshold (required) */
648 u32 (*ssthresh)(struct sock *sk);
649 /* lower bound for congestion window (optional) */
650 u32 (*min_cwnd)(const struct sock *sk);
651 /* do new cwnd calculation (required) */
652 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight, int good_ack);
653 /* call before changing ca_state (optional) */
654 void (*set_state)(struct sock *sk, u8 new_state);
655 /* call when cwnd event occurs (optional) */
656 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
657 /* new value of cwnd after loss (optional) */
658 u32 (*undo_cwnd)(struct sock *sk);
659 /* hook for packet ack accounting (optional) */
660 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
661 /* get info for inet_diag (optional) */
662 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
664 char name[TCP_CA_NAME_MAX];
665 struct module *owner;
668 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
669 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
671 extern void tcp_init_congestion_control(struct sock *sk);
672 extern void tcp_cleanup_congestion_control(struct sock *sk);
673 extern int tcp_set_default_congestion_control(const char *name);
674 extern void tcp_get_default_congestion_control(char *name);
675 extern void tcp_get_available_congestion_control(char *buf, size_t len);
676 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
677 extern int tcp_set_allowed_congestion_control(char *allowed);
678 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
679 extern void tcp_slow_start(struct tcp_sock *tp);
681 extern struct tcp_congestion_ops tcp_init_congestion_ops;
682 extern u32 tcp_reno_ssthresh(struct sock *sk);
683 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight, int flag);
684 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
685 extern struct tcp_congestion_ops tcp_reno;
687 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
689 struct inet_connection_sock *icsk = inet_csk(sk);
691 if (icsk->icsk_ca_ops->set_state)
692 icsk->icsk_ca_ops->set_state(sk, ca_state);
693 icsk->icsk_ca_state = ca_state;
696 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
698 const struct inet_connection_sock *icsk = inet_csk(sk);
700 if (icsk->icsk_ca_ops->cwnd_event)
701 icsk->icsk_ca_ops->cwnd_event(sk, event);
704 /* These functions determine how the current flow behaves in respect of SACK
705 * handling. SACK is negotiated with the peer, and therefore it can vary
706 * between different flows.
708 * tcp_is_sack - SACK enabled
709 * tcp_is_reno - No SACK
710 * tcp_is_fack - FACK enabled, implies SACK enabled
712 static inline int tcp_is_sack(const struct tcp_sock *tp)
714 return tp->rx_opt.sack_ok;
717 static inline int tcp_is_reno(const struct tcp_sock *tp)
719 return !tcp_is_sack(tp);
722 static inline int tcp_is_fack(const struct tcp_sock *tp)
724 return tp->rx_opt.sack_ok & 2;
727 static inline void tcp_enable_fack(struct tcp_sock *tp)
729 tp->rx_opt.sack_ok |= 2;
732 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
734 return tp->sacked_out + tp->lost_out;
737 /* This determines how many packets are "in the network" to the best
738 * of our knowledge. In many cases it is conservative, but where
739 * detailed information is available from the receiver (via SACK
740 * blocks etc.) we can make more aggressive calculations.
742 * Use this for decisions involving congestion control, use just
743 * tp->packets_out to determine if the send queue is empty or not.
745 * Read this equation as:
747 * "Packets sent once on transmission queue" MINUS
748 * "Packets left network, but not honestly ACKed yet" PLUS
749 * "Packets fast retransmitted"
751 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
753 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
756 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
757 * The exception is rate halving phase, when cwnd is decreasing towards
760 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
762 const struct tcp_sock *tp = tcp_sk(sk);
763 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
764 return tp->snd_ssthresh;
766 return max(tp->snd_ssthresh,
767 ((tp->snd_cwnd >> 1) +
768 (tp->snd_cwnd >> 2)));
771 /* Use define here intentionally to get WARN_ON location shown at the caller */
772 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
774 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
775 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
777 /* Slow start with delack produces 3 packets of burst, so that
778 * it is safe "de facto".
780 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
785 /* RFC2861 Check whether we are limited by application or congestion window
786 * This is the inverse of cwnd check in tcp_tso_should_defer
788 static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
790 const struct tcp_sock *tp = tcp_sk(sk);
793 if (in_flight >= tp->snd_cwnd)
799 left = tp->snd_cwnd - in_flight;
800 if (sysctl_tcp_tso_win_divisor)
801 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
803 return left <= tcp_max_burst(tp);
806 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
807 const struct sk_buff *skb)
810 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
813 static inline void tcp_check_probe_timer(struct sock *sk)
815 struct tcp_sock *tp = tcp_sk(sk);
816 const struct inet_connection_sock *icsk = inet_csk(sk);
818 if (!tp->packets_out && !icsk->icsk_pending)
819 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
820 icsk->icsk_rto, TCP_RTO_MAX);
823 static inline void tcp_push_pending_frames(struct sock *sk)
825 struct tcp_sock *tp = tcp_sk(sk);
827 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle);
830 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
835 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
841 * Calculate(/check) TCP checksum
843 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
844 __be32 daddr, __wsum base)
846 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
849 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
851 return __skb_checksum_complete(skb);
854 static inline int tcp_checksum_complete(struct sk_buff *skb)
856 return !skb_csum_unnecessary(skb) &&
857 __tcp_checksum_complete(skb);
860 /* Prequeue for VJ style copy to user, combined with checksumming. */
862 static inline void tcp_prequeue_init(struct tcp_sock *tp)
864 tp->ucopy.task = NULL;
866 tp->ucopy.memory = 0;
867 skb_queue_head_init(&tp->ucopy.prequeue);
868 #ifdef CONFIG_NET_DMA
869 tp->ucopy.dma_chan = NULL;
870 tp->ucopy.wakeup = 0;
871 tp->ucopy.pinned_list = NULL;
872 tp->ucopy.dma_cookie = 0;
876 /* Packet is added to VJ-style prequeue for processing in process
877 * context, if a reader task is waiting. Apparently, this exciting
878 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
879 * failed somewhere. Latency? Burstiness? Well, at least now we will
880 * see, why it failed. 8)8) --ANK
882 * NOTE: is this not too big to inline?
884 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
886 struct tcp_sock *tp = tcp_sk(sk);
888 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
889 __skb_queue_tail(&tp->ucopy.prequeue, skb);
890 tp->ucopy.memory += skb->truesize;
891 if (tp->ucopy.memory > sk->sk_rcvbuf) {
892 struct sk_buff *skb1;
894 BUG_ON(sock_owned_by_user(sk));
896 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
897 sk->sk_backlog_rcv(sk, skb1);
898 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
901 tp->ucopy.memory = 0;
902 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
903 wake_up_interruptible(sk->sk_sleep);
904 if (!inet_csk_ack_scheduled(sk))
905 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
906 (3 * TCP_RTO_MIN) / 4,
918 static const char *statename[]={
919 "Unused","Established","Syn Sent","Syn Recv",
920 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
921 "Close Wait","Last ACK","Listen","Closing"
925 static inline void tcp_set_state(struct sock *sk, int state)
927 int oldstate = sk->sk_state;
930 case TCP_ESTABLISHED:
931 if (oldstate != TCP_ESTABLISHED)
932 TCP_INC_STATS(TCP_MIB_CURRESTAB);
936 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
937 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
939 sk->sk_prot->unhash(sk);
940 if (inet_csk(sk)->icsk_bind_hash &&
941 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
942 inet_put_port(&tcp_hashinfo, sk);
945 if (oldstate==TCP_ESTABLISHED)
946 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
949 /* Change state AFTER socket is unhashed to avoid closed
950 * socket sitting in hash tables.
952 sk->sk_state = state;
955 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
959 extern void tcp_done(struct sock *sk);
961 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
964 rx_opt->eff_sacks = 0;
965 rx_opt->num_sacks = 0;
968 /* Determine a window scaling and initial window to offer. */
969 extern void tcp_select_initial_window(int __space, __u32 mss,
970 __u32 *rcv_wnd, __u32 *window_clamp,
971 int wscale_ok, __u8 *rcv_wscale);
973 static inline int tcp_win_from_space(int space)
975 return sysctl_tcp_adv_win_scale<=0 ?
976 (space>>(-sysctl_tcp_adv_win_scale)) :
977 space - (space>>sysctl_tcp_adv_win_scale);
980 /* Note: caller must be prepared to deal with negative returns */
981 static inline int tcp_space(const struct sock *sk)
983 return tcp_win_from_space(sk->sk_rcvbuf -
984 atomic_read(&sk->sk_rmem_alloc));
987 static inline int tcp_full_space(const struct sock *sk)
989 return tcp_win_from_space(sk->sk_rcvbuf);
992 static inline void tcp_openreq_init(struct request_sock *req,
993 struct tcp_options_received *rx_opt,
996 struct inet_request_sock *ireq = inet_rsk(req);
998 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
999 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1000 req->mss = rx_opt->mss_clamp;
1001 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1002 ireq->tstamp_ok = rx_opt->tstamp_ok;
1003 ireq->sack_ok = rx_opt->sack_ok;
1004 ireq->snd_wscale = rx_opt->snd_wscale;
1005 ireq->wscale_ok = rx_opt->wscale_ok;
1008 ireq->rmt_port = tcp_hdr(skb)->source;
1011 extern void tcp_enter_memory_pressure(void);
1013 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1015 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1018 static inline int keepalive_time_when(const struct tcp_sock *tp)
1020 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1023 static inline int tcp_fin_time(const struct sock *sk)
1025 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1026 const int rto = inet_csk(sk)->icsk_rto;
1028 if (fin_timeout < (rto << 2) - (rto >> 1))
1029 fin_timeout = (rto << 2) - (rto >> 1);
1034 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1036 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1038 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1041 /* RST segments are not recommended to carry timestamp,
1042 and, if they do, it is recommended to ignore PAWS because
1043 "their cleanup function should take precedence over timestamps."
1044 Certainly, it is mistake. It is necessary to understand the reasons
1045 of this constraint to relax it: if peer reboots, clock may go
1046 out-of-sync and half-open connections will not be reset.
1047 Actually, the problem would be not existing if all
1048 the implementations followed draft about maintaining clock
1049 via reboots. Linux-2.2 DOES NOT!
1051 However, we can relax time bounds for RST segments to MSL.
1053 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1058 #define TCP_CHECK_TIMER(sk) do { } while (0)
1060 static inline void tcp_mib_init(void)
1063 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1064 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1065 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1066 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1070 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1072 tp->lost_skb_hint = NULL;
1073 tp->scoreboard_skb_hint = NULL;
1074 tp->retransmit_skb_hint = NULL;
1075 tp->forward_skb_hint = NULL;
1078 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1080 tcp_clear_retrans_hints_partial(tp);
1081 tp->fastpath_skb_hint = NULL;
1087 /* - key database */
1088 struct tcp_md5sig_key {
1093 struct tcp4_md5sig_key {
1094 struct tcp_md5sig_key base;
1098 struct tcp6_md5sig_key {
1099 struct tcp_md5sig_key base;
1101 u32 scope_id; /* XXX */
1103 struct in6_addr addr;
1107 struct tcp_md5sig_info {
1108 struct tcp4_md5sig_key *keys4;
1109 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1110 struct tcp6_md5sig_key *keys6;
1118 /* - pseudo header */
1119 struct tcp4_pseudohdr {
1127 struct tcp6_pseudohdr {
1128 struct in6_addr saddr;
1129 struct in6_addr daddr;
1131 __be32 protocol; /* including padding */
1134 union tcp_md5sum_block {
1135 struct tcp4_pseudohdr ip4;
1136 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1137 struct tcp6_pseudohdr ip6;
1141 /* - pool: digest algorithm, hash description and scratch buffer */
1142 struct tcp_md5sig_pool {
1143 struct hash_desc md5_desc;
1144 union tcp_md5sum_block md5_blk;
1147 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1150 extern int tcp_v4_calc_md5_hash(char *md5_hash,
1151 struct tcp_md5sig_key *key,
1153 struct dst_entry *dst,
1154 struct request_sock *req,
1156 int protocol, int tcplen);
1157 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1158 struct sock *addr_sk);
1160 extern int tcp_v4_md5_do_add(struct sock *sk,
1165 extern int tcp_v4_md5_do_del(struct sock *sk,
1168 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1169 extern void tcp_free_md5sig_pool(void);
1171 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1172 extern void __tcp_put_md5sig_pool(void);
1175 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1177 int cpu = get_cpu();
1178 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1184 static inline void tcp_put_md5sig_pool(void)
1186 __tcp_put_md5sig_pool();
1190 /* write queue abstraction */
1191 static inline void tcp_write_queue_purge(struct sock *sk)
1193 struct sk_buff *skb;
1195 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1196 sk_stream_free_skb(sk, skb);
1197 sk_stream_mem_reclaim(sk);
1200 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1202 struct sk_buff *skb = sk->sk_write_queue.next;
1203 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1208 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1210 struct sk_buff *skb = sk->sk_write_queue.prev;
1211 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1216 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1221 #define tcp_for_write_queue(skb, sk) \
1222 for (skb = (sk)->sk_write_queue.next; \
1223 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1226 #define tcp_for_write_queue_from(skb, sk) \
1227 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\
1230 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1232 return sk->sk_send_head;
1235 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1237 struct tcp_sock *tp = tcp_sk(sk);
1239 sk->sk_send_head = skb->next;
1240 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue)
1241 sk->sk_send_head = NULL;
1242 /* Don't override Nagle indefinately with F-RTO */
1243 if (tp->frto_counter == 2)
1244 tp->frto_counter = 3;
1247 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1249 if (sk->sk_send_head == skb_unlinked)
1250 sk->sk_send_head = NULL;
1253 static inline void tcp_init_send_head(struct sock *sk)
1255 sk->sk_send_head = NULL;
1258 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1260 __skb_queue_tail(&sk->sk_write_queue, skb);
1263 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1265 __tcp_add_write_queue_tail(sk, skb);
1267 /* Queue it, remembering where we must start sending. */
1268 if (sk->sk_send_head == NULL)
1269 sk->sk_send_head = skb;
1272 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1274 __skb_queue_head(&sk->sk_write_queue, skb);
1277 /* Insert buff after skb on the write queue of sk. */
1278 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1279 struct sk_buff *buff,
1282 __skb_append(skb, buff, &sk->sk_write_queue);
1285 /* Insert skb between prev and next on the write queue of sk. */
1286 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1287 struct sk_buff *skb,
1290 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue);
1293 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1295 __skb_unlink(skb, &sk->sk_write_queue);
1298 static inline int tcp_skb_is_last(const struct sock *sk,
1299 const struct sk_buff *skb)
1301 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1304 static inline int tcp_write_queue_empty(struct sock *sk)
1306 return skb_queue_empty(&sk->sk_write_queue);
1310 enum tcp_seq_states {
1311 TCP_SEQ_STATE_LISTENING,
1312 TCP_SEQ_STATE_OPENREQ,
1313 TCP_SEQ_STATE_ESTABLISHED,
1314 TCP_SEQ_STATE_TIME_WAIT,
1317 struct tcp_seq_afinfo {
1318 struct module *owner;
1321 int (*seq_show) (struct seq_file *m, void *v);
1322 struct file_operations *seq_fops;
1325 struct tcp_iter_state {
1327 enum tcp_seq_states state;
1328 struct sock *syn_wait_sk;
1329 int bucket, sbucket, num, uid;
1330 struct seq_operations seq_ops;
1333 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1334 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1336 extern struct request_sock_ops tcp_request_sock_ops;
1338 extern int tcp_v4_destroy_sock(struct sock *sk);
1340 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1341 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1343 #ifdef CONFIG_PROC_FS
1344 extern int tcp4_proc_init(void);
1345 extern void tcp4_proc_exit(void);
1348 /* TCP af-specific functions */
1349 struct tcp_sock_af_ops {
1350 #ifdef CONFIG_TCP_MD5SIG
1351 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1352 struct sock *addr_sk);
1353 int (*calc_md5_hash) (char *location,
1354 struct tcp_md5sig_key *md5,
1356 struct dst_entry *dst,
1357 struct request_sock *req,
1359 int protocol, int len);
1360 int (*md5_add) (struct sock *sk,
1361 struct sock *addr_sk,
1364 int (*md5_parse) (struct sock *sk,
1365 char __user *optval,
1370 struct tcp_request_sock_ops {
1371 #ifdef CONFIG_TCP_MD5SIG
1372 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1373 struct request_sock *req);
1377 extern void tcp_v4_init(struct net_proto_family *ops);
1378 extern void tcp_init(void);