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>
46 #include <linux/seq_file.h>
48 extern struct inet_hashinfo tcp_hashinfo;
50 extern struct percpu_counter tcp_orphan_count;
51 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
53 #define MAX_TCP_HEADER (128 + MAX_HEADER)
54 #define MAX_TCP_OPTION_SPACE 40
57 * Never offer a window over 32767 without using window scaling. Some
58 * poor stacks do signed 16bit maths!
60 #define MAX_TCP_WINDOW 32767U
62 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
63 #define TCP_MIN_MSS 88U
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
186 #define TCPOLEN_MSS_ALIGNED 4
188 /* Flags in tp->nonagle */
189 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
190 #define TCP_NAGLE_CORK 2 /* Socket is corked */
191 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
193 extern struct inet_timewait_death_row tcp_death_row;
195 /* sysctl variables for tcp */
196 extern int sysctl_tcp_timestamps;
197 extern int sysctl_tcp_window_scaling;
198 extern int sysctl_tcp_sack;
199 extern int sysctl_tcp_fin_timeout;
200 extern int sysctl_tcp_keepalive_time;
201 extern int sysctl_tcp_keepalive_probes;
202 extern int sysctl_tcp_keepalive_intvl;
203 extern int sysctl_tcp_syn_retries;
204 extern int sysctl_tcp_synack_retries;
205 extern int sysctl_tcp_retries1;
206 extern int sysctl_tcp_retries2;
207 extern int sysctl_tcp_orphan_retries;
208 extern int sysctl_tcp_syncookies;
209 extern int sysctl_tcp_retrans_collapse;
210 extern int sysctl_tcp_stdurg;
211 extern int sysctl_tcp_rfc1337;
212 extern int sysctl_tcp_abort_on_overflow;
213 extern int sysctl_tcp_max_orphans;
214 extern int sysctl_tcp_fack;
215 extern int sysctl_tcp_reordering;
216 extern int sysctl_tcp_ecn;
217 extern int sysctl_tcp_dsack;
218 extern int sysctl_tcp_mem[3];
219 extern int sysctl_tcp_wmem[3];
220 extern int sysctl_tcp_rmem[3];
221 extern int sysctl_tcp_app_win;
222 extern int sysctl_tcp_adv_win_scale;
223 extern int sysctl_tcp_tw_reuse;
224 extern int sysctl_tcp_frto;
225 extern int sysctl_tcp_frto_response;
226 extern int sysctl_tcp_low_latency;
227 extern int sysctl_tcp_dma_copybreak;
228 extern int sysctl_tcp_nometrics_save;
229 extern int sysctl_tcp_moderate_rcvbuf;
230 extern int sysctl_tcp_tso_win_divisor;
231 extern int sysctl_tcp_abc;
232 extern int sysctl_tcp_mtu_probing;
233 extern int sysctl_tcp_base_mss;
234 extern int sysctl_tcp_workaround_signed_windows;
235 extern int sysctl_tcp_slow_start_after_idle;
236 extern int sysctl_tcp_max_ssthresh;
237 extern int sysctl_tcp_cookie_size;
239 extern atomic_t tcp_memory_allocated;
240 extern struct percpu_counter tcp_sockets_allocated;
241 extern int tcp_memory_pressure;
244 * The next routines deal with comparing 32 bit unsigned ints
245 * and worry about wraparound (automatic with unsigned arithmetic).
248 static inline int before(__u32 seq1, __u32 seq2)
250 return (__s32)(seq1-seq2) < 0;
252 #define after(seq2, seq1) before(seq1, seq2)
254 /* is s2<=s1<=s3 ? */
255 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
257 return seq3 - seq2 >= seq1 - seq2;
260 static inline int tcp_too_many_orphans(struct sock *sk, int num)
262 return (num > sysctl_tcp_max_orphans) ||
263 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
264 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
267 /* syncookies: remember time of last synqueue overflow */
268 static inline void tcp_synq_overflow(struct sock *sk)
270 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
273 /* syncookies: no recent synqueue overflow on this listening socket? */
274 static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
276 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
277 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT);
280 extern struct proto tcp_prot;
282 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
283 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
284 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
285 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
287 extern void tcp_v4_err(struct sk_buff *skb, u32);
289 extern void tcp_shutdown (struct sock *sk, int how);
291 extern int tcp_v4_rcv(struct sk_buff *skb);
293 extern int tcp_v4_remember_stamp(struct sock *sk);
295 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
297 extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
298 struct msghdr *msg, size_t size);
299 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
301 extern int tcp_ioctl(struct sock *sk,
305 extern int tcp_rcv_state_process(struct sock *sk,
310 extern int tcp_rcv_established(struct sock *sk,
315 extern void tcp_rcv_space_adjust(struct sock *sk);
317 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
319 extern int tcp_twsk_unique(struct sock *sk,
320 struct sock *sktw, void *twp);
322 extern void tcp_twsk_destructor(struct sock *sk);
324 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
325 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
327 static inline void tcp_dec_quickack_mode(struct sock *sk,
328 const unsigned int pkts)
330 struct inet_connection_sock *icsk = inet_csk(sk);
332 if (icsk->icsk_ack.quick) {
333 if (pkts >= icsk->icsk_ack.quick) {
334 icsk->icsk_ack.quick = 0;
335 /* Leaving quickack mode we deflate ATO. */
336 icsk->icsk_ack.ato = TCP_ATO_MIN;
338 icsk->icsk_ack.quick -= pkts;
342 extern void tcp_enter_quickack_mode(struct sock *sk);
345 #define TCP_ECN_QUEUE_CWR 2
346 #define TCP_ECN_DEMAND_CWR 4
348 static __inline__ void
349 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
351 if (sysctl_tcp_ecn && th->ece && th->cwr)
352 inet_rsk(req)->ecn_ok = 1;
363 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
365 const struct tcphdr *th);
367 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
368 struct request_sock *req,
369 struct request_sock **prev);
370 extern int tcp_child_process(struct sock *parent,
372 struct sk_buff *skb);
373 extern int tcp_use_frto(struct sock *sk);
374 extern void tcp_enter_frto(struct sock *sk);
375 extern void tcp_enter_loss(struct sock *sk, int how);
376 extern void tcp_clear_retrans(struct tcp_sock *tp);
377 extern void tcp_update_metrics(struct sock *sk);
379 extern void tcp_close(struct sock *sk,
381 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
383 extern int tcp_getsockopt(struct sock *sk, int level,
387 extern int tcp_setsockopt(struct sock *sk, int level,
388 int optname, char __user *optval,
389 unsigned int optlen);
390 extern int compat_tcp_getsockopt(struct sock *sk,
391 int level, int optname,
392 char __user *optval, int __user *optlen);
393 extern int compat_tcp_setsockopt(struct sock *sk,
394 int level, int optname,
395 char __user *optval, unsigned int optlen);
396 extern void tcp_set_keepalive(struct sock *sk, int val);
397 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
399 size_t len, int nonblock,
400 int flags, int *addr_len);
402 extern void tcp_parse_options(struct sk_buff *skb,
403 struct tcp_options_received *opt_rx,
405 struct dst_entry *dst);
407 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
410 * TCP v4 functions exported for the inet6 API
413 extern void tcp_v4_send_check(struct sock *sk, int len,
414 struct sk_buff *skb);
416 extern int tcp_v4_conn_request(struct sock *sk,
417 struct sk_buff *skb);
419 extern struct sock * tcp_create_openreq_child(struct sock *sk,
420 struct request_sock *req,
421 struct sk_buff *skb);
423 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
425 struct request_sock *req,
426 struct dst_entry *dst);
428 extern int tcp_v4_do_rcv(struct sock *sk,
429 struct sk_buff *skb);
431 extern int tcp_v4_connect(struct sock *sk,
432 struct sockaddr *uaddr,
435 extern int tcp_connect(struct sock *sk);
437 extern struct sk_buff * tcp_make_synack(struct sock *sk,
438 struct dst_entry *dst,
439 struct request_sock *req,
440 struct request_values *rvp);
442 extern int tcp_disconnect(struct sock *sk, int flags);
445 /* From syncookies.c */
446 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
447 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
448 struct ip_options *opt);
449 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
452 extern __u32 cookie_init_timestamp(struct request_sock *req);
453 extern void cookie_check_timestamp(struct tcp_options_received *tcp_opt);
455 /* From net/ipv6/syncookies.c */
456 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
457 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
462 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
464 extern int tcp_may_send_now(struct sock *sk);
465 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
466 extern void tcp_retransmit_timer(struct sock *sk);
467 extern void tcp_xmit_retransmit_queue(struct sock *);
468 extern void tcp_simple_retransmit(struct sock *);
469 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
470 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
472 extern void tcp_send_probe0(struct sock *);
473 extern void tcp_send_partial(struct sock *);
474 extern int tcp_write_wakeup(struct sock *);
475 extern void tcp_send_fin(struct sock *sk);
476 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
477 extern int tcp_send_synack(struct sock *);
478 extern void tcp_push_one(struct sock *, unsigned int mss_now);
479 extern void tcp_send_ack(struct sock *sk);
480 extern void tcp_send_delayed_ack(struct sock *sk);
483 extern void tcp_cwnd_application_limited(struct sock *sk);
486 extern void tcp_init_xmit_timers(struct sock *);
487 static inline void tcp_clear_xmit_timers(struct sock *sk)
489 inet_csk_clear_xmit_timers(sk);
492 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
493 extern unsigned int tcp_current_mss(struct sock *sk);
495 /* Bound MSS / TSO packet size with the half of the window */
496 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
498 if (tp->max_window && pktsize > (tp->max_window >> 1))
499 return max(tp->max_window >> 1, 68U - tp->tcp_header_len);
505 extern void tcp_get_info(struct sock *, struct tcp_info *);
507 /* Read 'sendfile()'-style from a TCP socket */
508 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
509 unsigned int, size_t);
510 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
511 sk_read_actor_t recv_actor);
513 extern void tcp_initialize_rcv_mss(struct sock *sk);
515 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
516 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
517 extern void tcp_mtup_init(struct sock *sk);
519 static inline void tcp_bound_rto(const struct sock *sk)
521 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
522 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
525 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
527 return (tp->srtt >> 3) + tp->rttvar;
530 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
532 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
533 ntohl(TCP_FLAG_ACK) |
537 static inline void tcp_fast_path_on(struct tcp_sock *tp)
539 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
542 static inline void tcp_fast_path_check(struct sock *sk)
544 struct tcp_sock *tp = tcp_sk(sk);
546 if (skb_queue_empty(&tp->out_of_order_queue) &&
548 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
550 tcp_fast_path_on(tp);
553 /* Compute the actual rto_min value */
554 static inline u32 tcp_rto_min(struct sock *sk)
556 struct dst_entry *dst = __sk_dst_get(sk);
557 u32 rto_min = TCP_RTO_MIN;
559 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
560 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
564 /* Compute the actual receive window we are currently advertising.
565 * Rcv_nxt can be after the window if our peer push more data
566 * than the offered window.
568 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
570 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
577 /* Choose a new window, without checks for shrinking, and without
578 * scaling applied to the result. The caller does these things
579 * if necessary. This is a "raw" window selection.
581 extern u32 __tcp_select_window(struct sock *sk);
583 /* TCP timestamps are only 32-bits, this causes a slight
584 * complication on 64-bit systems since we store a snapshot
585 * of jiffies in the buffer control blocks below. We decided
586 * to use only the low 32-bits of jiffies and hide the ugly
587 * casts with the following macro.
589 #define tcp_time_stamp ((__u32)(jiffies))
591 /* This is what the send packet queuing engine uses to pass
592 * TCP per-packet control information to the transmission
593 * code. We also store the host-order sequence numbers in
594 * here too. This is 36 bytes on 32-bit architectures,
595 * 40 bytes on 64-bit machines, if this grows please adjust
596 * skbuff.h:skbuff->cb[xxx] size appropriately.
600 struct inet_skb_parm h4;
601 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
602 struct inet6_skb_parm h6;
604 } header; /* For incoming frames */
605 __u32 seq; /* Starting sequence number */
606 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
607 __u32 when; /* used to compute rtt's */
608 __u8 flags; /* TCP header flags. */
610 /* NOTE: These must match up to the flags byte in a
613 #define TCPCB_FLAG_FIN 0x01
614 #define TCPCB_FLAG_SYN 0x02
615 #define TCPCB_FLAG_RST 0x04
616 #define TCPCB_FLAG_PSH 0x08
617 #define TCPCB_FLAG_ACK 0x10
618 #define TCPCB_FLAG_URG 0x20
619 #define TCPCB_FLAG_ECE 0x40
620 #define TCPCB_FLAG_CWR 0x80
622 __u8 sacked; /* State flags for SACK/FACK. */
623 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
624 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
625 #define TCPCB_LOST 0x04 /* SKB is lost */
626 #define TCPCB_TAGBITS 0x07 /* All tag bits */
628 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
629 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
631 __u32 ack_seq; /* Sequence number ACK'd */
634 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
636 /* Due to TSO, an SKB can be composed of multiple actual
637 * packets. To keep these tracked properly, we use this.
639 static inline int tcp_skb_pcount(const struct sk_buff *skb)
641 return skb_shinfo(skb)->gso_segs;
644 /* This is valid iff tcp_skb_pcount() > 1. */
645 static inline int tcp_skb_mss(const struct sk_buff *skb)
647 return skb_shinfo(skb)->gso_size;
650 /* Events passed to congestion control interface */
652 CA_EVENT_TX_START, /* first transmit when no packets in flight */
653 CA_EVENT_CWND_RESTART, /* congestion window restart */
654 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
655 CA_EVENT_FRTO, /* fast recovery timeout */
656 CA_EVENT_LOSS, /* loss timeout */
657 CA_EVENT_FAST_ACK, /* in sequence ack */
658 CA_EVENT_SLOW_ACK, /* other ack */
662 * Interface for adding new TCP congestion control handlers
664 #define TCP_CA_NAME_MAX 16
665 #define TCP_CA_MAX 128
666 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
668 #define TCP_CONG_NON_RESTRICTED 0x1
669 #define TCP_CONG_RTT_STAMP 0x2
671 struct tcp_congestion_ops {
672 struct list_head list;
675 /* initialize private data (optional) */
676 void (*init)(struct sock *sk);
677 /* cleanup private data (optional) */
678 void (*release)(struct sock *sk);
680 /* return slow start threshold (required) */
681 u32 (*ssthresh)(struct sock *sk);
682 /* lower bound for congestion window (optional) */
683 u32 (*min_cwnd)(const struct sock *sk);
684 /* do new cwnd calculation (required) */
685 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
686 /* call before changing ca_state (optional) */
687 void (*set_state)(struct sock *sk, u8 new_state);
688 /* call when cwnd event occurs (optional) */
689 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
690 /* new value of cwnd after loss (optional) */
691 u32 (*undo_cwnd)(struct sock *sk);
692 /* hook for packet ack accounting (optional) */
693 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
694 /* get info for inet_diag (optional) */
695 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
697 char name[TCP_CA_NAME_MAX];
698 struct module *owner;
701 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
702 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
704 extern void tcp_init_congestion_control(struct sock *sk);
705 extern void tcp_cleanup_congestion_control(struct sock *sk);
706 extern int tcp_set_default_congestion_control(const char *name);
707 extern void tcp_get_default_congestion_control(char *name);
708 extern void tcp_get_available_congestion_control(char *buf, size_t len);
709 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
710 extern int tcp_set_allowed_congestion_control(char *allowed);
711 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
712 extern void tcp_slow_start(struct tcp_sock *tp);
713 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
715 extern struct tcp_congestion_ops tcp_init_congestion_ops;
716 extern u32 tcp_reno_ssthresh(struct sock *sk);
717 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
718 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
719 extern struct tcp_congestion_ops tcp_reno;
721 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
723 struct inet_connection_sock *icsk = inet_csk(sk);
725 if (icsk->icsk_ca_ops->set_state)
726 icsk->icsk_ca_ops->set_state(sk, ca_state);
727 icsk->icsk_ca_state = ca_state;
730 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
732 const struct inet_connection_sock *icsk = inet_csk(sk);
734 if (icsk->icsk_ca_ops->cwnd_event)
735 icsk->icsk_ca_ops->cwnd_event(sk, event);
738 /* These functions determine how the current flow behaves in respect of SACK
739 * handling. SACK is negotiated with the peer, and therefore it can vary
740 * between different flows.
742 * tcp_is_sack - SACK enabled
743 * tcp_is_reno - No SACK
744 * tcp_is_fack - FACK enabled, implies SACK enabled
746 static inline int tcp_is_sack(const struct tcp_sock *tp)
748 return tp->rx_opt.sack_ok;
751 static inline int tcp_is_reno(const struct tcp_sock *tp)
753 return !tcp_is_sack(tp);
756 static inline int tcp_is_fack(const struct tcp_sock *tp)
758 return tp->rx_opt.sack_ok & 2;
761 static inline void tcp_enable_fack(struct tcp_sock *tp)
763 tp->rx_opt.sack_ok |= 2;
766 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
768 return tp->sacked_out + tp->lost_out;
771 /* This determines how many packets are "in the network" to the best
772 * of our knowledge. In many cases it is conservative, but where
773 * detailed information is available from the receiver (via SACK
774 * blocks etc.) we can make more aggressive calculations.
776 * Use this for decisions involving congestion control, use just
777 * tp->packets_out to determine if the send queue is empty or not.
779 * Read this equation as:
781 * "Packets sent once on transmission queue" MINUS
782 * "Packets left network, but not honestly ACKed yet" PLUS
783 * "Packets fast retransmitted"
785 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
787 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
790 #define TCP_INFINITE_SSTHRESH 0x7fffffff
792 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
794 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
797 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
798 * The exception is rate halving phase, when cwnd is decreasing towards
801 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
803 const struct tcp_sock *tp = tcp_sk(sk);
804 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
805 return tp->snd_ssthresh;
807 return max(tp->snd_ssthresh,
808 ((tp->snd_cwnd >> 1) +
809 (tp->snd_cwnd >> 2)));
812 /* Use define here intentionally to get WARN_ON location shown at the caller */
813 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
815 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
816 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
818 /* Slow start with delack produces 3 packets of burst, so that
819 * it is safe "de facto". This will be the default - same as
820 * the default reordering threshold - but if reordering increases,
821 * we must be able to allow cwnd to burst at least this much in order
822 * to not pull it back when holes are filled.
824 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
826 return tp->reordering;
829 /* Returns end sequence number of the receiver's advertised window */
830 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
832 return tp->snd_una + tp->snd_wnd;
834 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
836 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
837 const struct sk_buff *skb)
840 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
843 static inline void tcp_check_probe_timer(struct sock *sk)
845 struct tcp_sock *tp = tcp_sk(sk);
846 const struct inet_connection_sock *icsk = inet_csk(sk);
848 if (!tp->packets_out && !icsk->icsk_pending)
849 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
850 icsk->icsk_rto, TCP_RTO_MAX);
853 static inline void tcp_push_pending_frames(struct sock *sk)
855 struct tcp_sock *tp = tcp_sk(sk);
857 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
860 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
865 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
871 * Calculate(/check) TCP checksum
873 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
874 __be32 daddr, __wsum base)
876 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
879 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
881 return __skb_checksum_complete(skb);
884 static inline int tcp_checksum_complete(struct sk_buff *skb)
886 return !skb_csum_unnecessary(skb) &&
887 __tcp_checksum_complete(skb);
890 /* Prequeue for VJ style copy to user, combined with checksumming. */
892 static inline void tcp_prequeue_init(struct tcp_sock *tp)
894 tp->ucopy.task = NULL;
896 tp->ucopy.memory = 0;
897 skb_queue_head_init(&tp->ucopy.prequeue);
898 #ifdef CONFIG_NET_DMA
899 tp->ucopy.dma_chan = NULL;
900 tp->ucopy.wakeup = 0;
901 tp->ucopy.pinned_list = NULL;
902 tp->ucopy.dma_cookie = 0;
906 /* Packet is added to VJ-style prequeue for processing in process
907 * context, if a reader task is waiting. Apparently, this exciting
908 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
909 * failed somewhere. Latency? Burstiness? Well, at least now we will
910 * see, why it failed. 8)8) --ANK
912 * NOTE: is this not too big to inline?
914 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
916 struct tcp_sock *tp = tcp_sk(sk);
918 if (sysctl_tcp_low_latency || !tp->ucopy.task)
921 __skb_queue_tail(&tp->ucopy.prequeue, skb);
922 tp->ucopy.memory += skb->truesize;
923 if (tp->ucopy.memory > sk->sk_rcvbuf) {
924 struct sk_buff *skb1;
926 BUG_ON(sock_owned_by_user(sk));
928 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
929 sk_backlog_rcv(sk, skb1);
930 NET_INC_STATS_BH(sock_net(sk),
931 LINUX_MIB_TCPPREQUEUEDROPPED);
934 tp->ucopy.memory = 0;
935 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
936 wake_up_interruptible_poll(sk->sk_sleep,
937 POLLIN | POLLRDNORM | POLLRDBAND);
938 if (!inet_csk_ack_scheduled(sk))
939 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
940 (3 * tcp_rto_min(sk)) / 4,
950 static const char *statename[]={
951 "Unused","Established","Syn Sent","Syn Recv",
952 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
953 "Close Wait","Last ACK","Listen","Closing"
956 extern void tcp_set_state(struct sock *sk, int state);
958 extern void tcp_done(struct sock *sk);
960 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
963 rx_opt->num_sacks = 0;
966 /* Determine a window scaling and initial window to offer. */
967 extern void tcp_select_initial_window(int __space, __u32 mss,
968 __u32 *rcv_wnd, __u32 *window_clamp,
969 int wscale_ok, __u8 *rcv_wscale);
971 static inline int tcp_win_from_space(int space)
973 return sysctl_tcp_adv_win_scale<=0 ?
974 (space>>(-sysctl_tcp_adv_win_scale)) :
975 space - (space>>sysctl_tcp_adv_win_scale);
978 /* Note: caller must be prepared to deal with negative returns */
979 static inline int tcp_space(const struct sock *sk)
981 return tcp_win_from_space(sk->sk_rcvbuf -
982 atomic_read(&sk->sk_rmem_alloc));
985 static inline int tcp_full_space(const struct sock *sk)
987 return tcp_win_from_space(sk->sk_rcvbuf);
990 static inline void tcp_openreq_init(struct request_sock *req,
991 struct tcp_options_received *rx_opt,
994 struct inet_request_sock *ireq = inet_rsk(req);
996 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
998 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
999 req->mss = rx_opt->mss_clamp;
1000 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1001 ireq->tstamp_ok = rx_opt->tstamp_ok;
1002 ireq->sack_ok = rx_opt->sack_ok;
1003 ireq->snd_wscale = rx_opt->snd_wscale;
1004 ireq->wscale_ok = rx_opt->wscale_ok;
1007 ireq->rmt_port = tcp_hdr(skb)->source;
1008 ireq->loc_port = tcp_hdr(skb)->dest;
1011 extern void tcp_enter_memory_pressure(struct sock *sk);
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 keepalive_probes(const struct tcp_sock *tp)
1025 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1028 static inline int tcp_fin_time(const struct sock *sk)
1030 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1031 const int rto = inet_csk(sk)->icsk_rto;
1033 if (fin_timeout < (rto << 2) - (rto >> 1))
1034 fin_timeout = (rto << 2) - (rto >> 1);
1039 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1042 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1044 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1050 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1053 if (tcp_paws_check(rx_opt, 0))
1056 /* RST segments are not recommended to carry timestamp,
1057 and, if they do, it is recommended to ignore PAWS because
1058 "their cleanup function should take precedence over timestamps."
1059 Certainly, it is mistake. It is necessary to understand the reasons
1060 of this constraint to relax it: if peer reboots, clock may go
1061 out-of-sync and half-open connections will not be reset.
1062 Actually, the problem would be not existing if all
1063 the implementations followed draft about maintaining clock
1064 via reboots. Linux-2.2 DOES NOT!
1066 However, we can relax time bounds for RST segments to MSL.
1068 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1073 #define TCP_CHECK_TIMER(sk) do { } while (0)
1075 static inline void tcp_mib_init(struct net *net)
1078 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1079 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1080 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1081 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1085 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1087 tp->lost_skb_hint = NULL;
1088 tp->scoreboard_skb_hint = NULL;
1091 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1093 tcp_clear_retrans_hints_partial(tp);
1094 tp->retransmit_skb_hint = NULL;
1100 /* - key database */
1101 struct tcp_md5sig_key {
1106 struct tcp4_md5sig_key {
1107 struct tcp_md5sig_key base;
1111 struct tcp6_md5sig_key {
1112 struct tcp_md5sig_key base;
1114 u32 scope_id; /* XXX */
1116 struct in6_addr addr;
1120 struct tcp_md5sig_info {
1121 struct tcp4_md5sig_key *keys4;
1122 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1123 struct tcp6_md5sig_key *keys6;
1131 /* - pseudo header */
1132 struct tcp4_pseudohdr {
1140 struct tcp6_pseudohdr {
1141 struct in6_addr saddr;
1142 struct in6_addr daddr;
1144 __be32 protocol; /* including padding */
1147 union tcp_md5sum_block {
1148 struct tcp4_pseudohdr ip4;
1149 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1150 struct tcp6_pseudohdr ip6;
1154 /* - pool: digest algorithm, hash description and scratch buffer */
1155 struct tcp_md5sig_pool {
1156 struct hash_desc md5_desc;
1157 union tcp_md5sum_block md5_blk;
1160 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1163 extern int tcp_v4_md5_hash_skb(char *md5_hash,
1164 struct tcp_md5sig_key *key,
1166 struct request_sock *req,
1167 struct sk_buff *skb);
1169 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1170 struct sock *addr_sk);
1172 extern int tcp_v4_md5_do_add(struct sock *sk,
1177 extern int tcp_v4_md5_do_del(struct sock *sk,
1180 #ifdef CONFIG_TCP_MD5SIG
1181 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1182 &(struct tcp_md5sig_key) { \
1183 .key = (twsk)->tw_md5_key, \
1184 .keylen = (twsk)->tw_md5_keylen, \
1187 #define tcp_twsk_md5_key(twsk) NULL
1190 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(struct sock *);
1191 extern void tcp_free_md5sig_pool(void);
1193 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1194 extern void __tcp_put_md5sig_pool(void);
1195 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1196 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1197 unsigned header_len);
1198 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1199 struct tcp_md5sig_key *key);
1202 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1204 int cpu = get_cpu();
1205 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1211 static inline void tcp_put_md5sig_pool(void)
1213 __tcp_put_md5sig_pool();
1217 /* write queue abstraction */
1218 static inline void tcp_write_queue_purge(struct sock *sk)
1220 struct sk_buff *skb;
1222 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1223 sk_wmem_free_skb(sk, skb);
1227 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1229 return skb_peek(&sk->sk_write_queue);
1232 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1234 return skb_peek_tail(&sk->sk_write_queue);
1237 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1239 return skb_queue_next(&sk->sk_write_queue, skb);
1242 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb)
1244 return skb_queue_prev(&sk->sk_write_queue, skb);
1247 #define tcp_for_write_queue(skb, sk) \
1248 skb_queue_walk(&(sk)->sk_write_queue, skb)
1250 #define tcp_for_write_queue_from(skb, sk) \
1251 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1253 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1254 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1256 /* This function calculates a "timeout" which is equivalent to the timeout of a
1257 * TCP connection after "boundary" unsucessful, exponentially backed-off
1258 * retransmissions with an initial RTO of TCP_RTO_MIN.
1260 static inline bool retransmits_timed_out(const struct sock *sk,
1261 unsigned int boundary)
1263 unsigned int timeout, linear_backoff_thresh;
1265 if (!inet_csk(sk)->icsk_retransmits)
1268 linear_backoff_thresh = ilog2(TCP_RTO_MAX/TCP_RTO_MIN);
1270 if (boundary <= linear_backoff_thresh)
1271 timeout = ((2 << boundary) - 1) * TCP_RTO_MIN;
1273 timeout = ((2 << linear_backoff_thresh) - 1) * TCP_RTO_MIN +
1274 (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
1276 return (tcp_time_stamp - tcp_sk(sk)->retrans_stamp) >= timeout;
1279 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1281 return sk->sk_send_head;
1284 static inline bool tcp_skb_is_last(const struct sock *sk,
1285 const struct sk_buff *skb)
1287 return skb_queue_is_last(&sk->sk_write_queue, skb);
1290 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1292 if (tcp_skb_is_last(sk, skb))
1293 sk->sk_send_head = NULL;
1295 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1298 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1300 if (sk->sk_send_head == skb_unlinked)
1301 sk->sk_send_head = NULL;
1304 static inline void tcp_init_send_head(struct sock *sk)
1306 sk->sk_send_head = NULL;
1309 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1311 __skb_queue_tail(&sk->sk_write_queue, skb);
1314 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1316 __tcp_add_write_queue_tail(sk, skb);
1318 /* Queue it, remembering where we must start sending. */
1319 if (sk->sk_send_head == NULL) {
1320 sk->sk_send_head = skb;
1322 if (tcp_sk(sk)->highest_sack == NULL)
1323 tcp_sk(sk)->highest_sack = skb;
1327 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1329 __skb_queue_head(&sk->sk_write_queue, skb);
1332 /* Insert buff after skb on the write queue of sk. */
1333 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1334 struct sk_buff *buff,
1337 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1340 /* Insert new before skb on the write queue of sk. */
1341 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1342 struct sk_buff *skb,
1345 __skb_queue_before(&sk->sk_write_queue, skb, new);
1347 if (sk->sk_send_head == skb)
1348 sk->sk_send_head = new;
1351 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1353 __skb_unlink(skb, &sk->sk_write_queue);
1356 static inline int tcp_write_queue_empty(struct sock *sk)
1358 return skb_queue_empty(&sk->sk_write_queue);
1361 /* Start sequence of the highest skb with SACKed bit, valid only if
1362 * sacked > 0 or when the caller has ensured validity by itself.
1364 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1366 if (!tp->sacked_out)
1369 if (tp->highest_sack == NULL)
1372 return TCP_SKB_CB(tp->highest_sack)->seq;
1375 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1377 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1378 tcp_write_queue_next(sk, skb);
1381 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1383 return tcp_sk(sk)->highest_sack;
1386 static inline void tcp_highest_sack_reset(struct sock *sk)
1388 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1391 /* Called when old skb is about to be deleted (to be combined with new skb) */
1392 static inline void tcp_highest_sack_combine(struct sock *sk,
1393 struct sk_buff *old,
1394 struct sk_buff *new)
1396 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1397 tcp_sk(sk)->highest_sack = new;
1401 enum tcp_seq_states {
1402 TCP_SEQ_STATE_LISTENING,
1403 TCP_SEQ_STATE_OPENREQ,
1404 TCP_SEQ_STATE_ESTABLISHED,
1405 TCP_SEQ_STATE_TIME_WAIT,
1408 struct tcp_seq_afinfo {
1411 struct file_operations seq_fops;
1412 struct seq_operations seq_ops;
1415 struct tcp_iter_state {
1416 struct seq_net_private p;
1418 enum tcp_seq_states state;
1419 struct sock *syn_wait_sk;
1420 int bucket, sbucket, num, uid;
1423 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1424 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1426 extern struct request_sock_ops tcp_request_sock_ops;
1427 extern struct request_sock_ops tcp6_request_sock_ops;
1429 extern void tcp_v4_destroy_sock(struct sock *sk);
1431 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1432 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1433 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1434 struct sk_buff *skb);
1435 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1436 struct sk_buff *skb);
1437 extern int tcp_gro_complete(struct sk_buff *skb);
1438 extern int tcp4_gro_complete(struct sk_buff *skb);
1440 #ifdef CONFIG_PROC_FS
1441 extern int tcp4_proc_init(void);
1442 extern void tcp4_proc_exit(void);
1445 /* TCP af-specific functions */
1446 struct tcp_sock_af_ops {
1447 #ifdef CONFIG_TCP_MD5SIG
1448 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1449 struct sock *addr_sk);
1450 int (*calc_md5_hash) (char *location,
1451 struct tcp_md5sig_key *md5,
1453 struct request_sock *req,
1454 struct sk_buff *skb);
1455 int (*md5_add) (struct sock *sk,
1456 struct sock *addr_sk,
1459 int (*md5_parse) (struct sock *sk,
1460 char __user *optval,
1465 struct tcp_request_sock_ops {
1466 #ifdef CONFIG_TCP_MD5SIG
1467 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1468 struct request_sock *req);
1469 int (*calc_md5_hash) (char *location,
1470 struct tcp_md5sig_key *md5,
1472 struct request_sock *req,
1473 struct sk_buff *skb);
1477 /* Using SHA1 for now, define some constants.
1479 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1480 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1481 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1483 extern int tcp_cookie_generator(u32 *bakery);
1485 extern void tcp_v4_init(void);
1486 extern void tcp_init(void);