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>
33 #include <linux/kref.h>
35 #include <net/inet_connection_sock.h>
36 #include <net/inet_timewait_sock.h>
37 #include <net/inet_hashtables.h>
38 #include <net/checksum.h>
39 #include <net/request_sock.h>
43 #include <net/tcp_states.h>
44 #include <net/inet_ecn.h>
47 #include <linux/seq_file.h>
49 extern struct inet_hashinfo tcp_hashinfo;
51 extern struct percpu_counter tcp_orphan_count;
52 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
54 #define MAX_TCP_HEADER (128 + MAX_HEADER)
55 #define MAX_TCP_OPTION_SPACE 40
58 * Never offer a window over 32767 without using window scaling. Some
59 * poor stacks do signed 16bit maths!
61 #define MAX_TCP_WINDOW 32767U
63 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
64 #define TCP_MIN_MSS 88U
66 /* The least MTU to use for probing */
67 #define TCP_BASE_MSS 512
69 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
70 #define TCP_FASTRETRANS_THRESH 3
72 /* Maximal reordering. */
73 #define TCP_MAX_REORDERING 127
75 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
76 #define TCP_MAX_QUICKACKS 16U
79 #define TCP_URG_VALID 0x0100
80 #define TCP_URG_NOTYET 0x0200
81 #define TCP_URG_READ 0x0400
83 #define TCP_RETR1 3 /*
84 * This is how many retries it does before it
85 * tries to figure out if the gateway is
86 * down. Minimal RFC value is 3; it corresponds
87 * to ~3sec-8min depending on RTO.
90 #define TCP_RETR2 15 /*
91 * This should take at least
92 * 90 minutes to time out.
93 * RFC1122 says that the limit is 100 sec.
94 * 15 is ~13-30min depending on RTO.
97 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
98 * connection: ~180sec is RFC minimum */
100 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
101 * connection: ~180sec is RFC minimum */
104 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
105 * socket. 7 is ~50sec-16min.
109 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
110 * state, about 60 seconds */
111 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
112 /* BSD style FIN_WAIT2 deadlock breaker.
113 * It used to be 3min, new value is 60sec,
114 * to combine FIN-WAIT-2 timeout with
118 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
120 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
121 #define TCP_ATO_MIN ((unsigned)(HZ/25))
123 #define TCP_DELACK_MIN 4U
124 #define TCP_ATO_MIN 4U
126 #define TCP_RTO_MAX ((unsigned)(120*HZ))
127 #define TCP_RTO_MIN ((unsigned)(HZ/5))
128 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
130 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
131 * for local resources.
134 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
135 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
136 #define TCP_KEEPALIVE_INTVL (75*HZ)
138 #define MAX_TCP_KEEPIDLE 32767
139 #define MAX_TCP_KEEPINTVL 32767
140 #define MAX_TCP_KEEPCNT 127
141 #define MAX_TCP_SYNCNT 127
143 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
145 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
146 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
147 * after this time. It should be equal
148 * (or greater than) TCP_TIMEWAIT_LEN
149 * to provide reliability equal to one
150 * provided by timewait state.
152 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
153 * timestamps. It must be less than
154 * minimal timewait lifetime.
160 #define TCPOPT_NOP 1 /* Padding */
161 #define TCPOPT_EOL 0 /* End of options */
162 #define TCPOPT_MSS 2 /* Segment size negotiating */
163 #define TCPOPT_WINDOW 3 /* Window scaling */
164 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
165 #define TCPOPT_SACK 5 /* SACK Block */
166 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
167 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
168 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
174 #define TCPOLEN_MSS 4
175 #define TCPOLEN_WINDOW 3
176 #define TCPOLEN_SACK_PERM 2
177 #define TCPOLEN_TIMESTAMP 10
178 #define TCPOLEN_MD5SIG 18
179 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
180 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
181 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
182 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
184 /* But this is what stacks really send out. */
185 #define TCPOLEN_TSTAMP_ALIGNED 12
186 #define TCPOLEN_WSCALE_ALIGNED 4
187 #define TCPOLEN_SACKPERM_ALIGNED 4
188 #define TCPOLEN_SACK_BASE 2
189 #define TCPOLEN_SACK_BASE_ALIGNED 4
190 #define TCPOLEN_SACK_PERBLOCK 8
191 #define TCPOLEN_MD5SIG_ALIGNED 20
192 #define TCPOLEN_MSS_ALIGNED 4
194 /* Flags in tp->nonagle */
195 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
196 #define TCP_NAGLE_CORK 2 /* Socket is corked */
197 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
199 /* TCP thin-stream limits */
200 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
202 extern struct inet_timewait_death_row tcp_death_row;
204 /* sysctl variables for tcp */
205 extern int sysctl_tcp_timestamps;
206 extern int sysctl_tcp_window_scaling;
207 extern int sysctl_tcp_sack;
208 extern int sysctl_tcp_fin_timeout;
209 extern int sysctl_tcp_keepalive_time;
210 extern int sysctl_tcp_keepalive_probes;
211 extern int sysctl_tcp_keepalive_intvl;
212 extern int sysctl_tcp_syn_retries;
213 extern int sysctl_tcp_synack_retries;
214 extern int sysctl_tcp_retries1;
215 extern int sysctl_tcp_retries2;
216 extern int sysctl_tcp_orphan_retries;
217 extern int sysctl_tcp_syncookies;
218 extern int sysctl_tcp_retrans_collapse;
219 extern int sysctl_tcp_stdurg;
220 extern int sysctl_tcp_rfc1337;
221 extern int sysctl_tcp_abort_on_overflow;
222 extern int sysctl_tcp_max_orphans;
223 extern int sysctl_tcp_fack;
224 extern int sysctl_tcp_reordering;
225 extern int sysctl_tcp_ecn;
226 extern int sysctl_tcp_dsack;
227 extern int sysctl_tcp_mem[3];
228 extern int sysctl_tcp_wmem[3];
229 extern int sysctl_tcp_rmem[3];
230 extern int sysctl_tcp_app_win;
231 extern int sysctl_tcp_adv_win_scale;
232 extern int sysctl_tcp_tw_reuse;
233 extern int sysctl_tcp_frto;
234 extern int sysctl_tcp_frto_response;
235 extern int sysctl_tcp_low_latency;
236 extern int sysctl_tcp_dma_copybreak;
237 extern int sysctl_tcp_nometrics_save;
238 extern int sysctl_tcp_moderate_rcvbuf;
239 extern int sysctl_tcp_tso_win_divisor;
240 extern int sysctl_tcp_abc;
241 extern int sysctl_tcp_mtu_probing;
242 extern int sysctl_tcp_base_mss;
243 extern int sysctl_tcp_workaround_signed_windows;
244 extern int sysctl_tcp_slow_start_after_idle;
245 extern int sysctl_tcp_max_ssthresh;
246 extern int sysctl_tcp_cookie_size;
247 extern int sysctl_tcp_thin_linear_timeouts;
248 extern int sysctl_tcp_thin_dupack;
250 extern atomic_t tcp_memory_allocated;
251 extern struct percpu_counter tcp_sockets_allocated;
252 extern int tcp_memory_pressure;
255 * The next routines deal with comparing 32 bit unsigned ints
256 * and worry about wraparound (automatic with unsigned arithmetic).
259 static inline int before(__u32 seq1, __u32 seq2)
261 return (__s32)(seq1-seq2) < 0;
263 #define after(seq2, seq1) before(seq1, seq2)
265 /* is s2<=s1<=s3 ? */
266 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
268 return seq3 - seq2 >= seq1 - seq2;
271 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
273 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
274 int orphans = percpu_counter_read_positive(ocp);
276 if (orphans << shift > sysctl_tcp_max_orphans) {
277 orphans = percpu_counter_sum_positive(ocp);
278 if (orphans << shift > sysctl_tcp_max_orphans)
282 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
283 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2])
288 /* syncookies: remember time of last synqueue overflow */
289 static inline void tcp_synq_overflow(struct sock *sk)
291 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
294 /* syncookies: no recent synqueue overflow on this listening socket? */
295 static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
297 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
298 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT);
301 extern struct proto tcp_prot;
303 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
304 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
305 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
306 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
307 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
309 extern void tcp_v4_err(struct sk_buff *skb, u32);
311 extern void tcp_shutdown (struct sock *sk, int how);
313 extern int tcp_v4_rcv(struct sk_buff *skb);
315 extern int tcp_v4_remember_stamp(struct sock *sk);
317 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
319 extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
320 struct msghdr *msg, size_t size);
321 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
323 extern int tcp_ioctl(struct sock *sk,
327 extern int tcp_rcv_state_process(struct sock *sk,
332 extern int tcp_rcv_established(struct sock *sk,
337 extern void tcp_rcv_space_adjust(struct sock *sk);
339 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
341 extern int tcp_twsk_unique(struct sock *sk,
342 struct sock *sktw, void *twp);
344 extern void tcp_twsk_destructor(struct sock *sk);
346 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
347 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
349 static inline void tcp_dec_quickack_mode(struct sock *sk,
350 const unsigned int pkts)
352 struct inet_connection_sock *icsk = inet_csk(sk);
354 if (icsk->icsk_ack.quick) {
355 if (pkts >= icsk->icsk_ack.quick) {
356 icsk->icsk_ack.quick = 0;
357 /* Leaving quickack mode we deflate ATO. */
358 icsk->icsk_ack.ato = TCP_ATO_MIN;
360 icsk->icsk_ack.quick -= pkts;
364 extern void tcp_enter_quickack_mode(struct sock *sk);
367 #define TCP_ECN_QUEUE_CWR 2
368 #define TCP_ECN_DEMAND_CWR 4
370 static __inline__ void
371 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
373 if (sysctl_tcp_ecn && th->ece && th->cwr)
374 inet_rsk(req)->ecn_ok = 1;
385 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
387 const struct tcphdr *th);
389 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
390 struct request_sock *req,
391 struct request_sock **prev);
392 extern int tcp_child_process(struct sock *parent,
394 struct sk_buff *skb);
395 extern int tcp_use_frto(struct sock *sk);
396 extern void tcp_enter_frto(struct sock *sk);
397 extern void tcp_enter_loss(struct sock *sk, int how);
398 extern void tcp_clear_retrans(struct tcp_sock *tp);
399 extern void tcp_update_metrics(struct sock *sk);
401 extern void tcp_close(struct sock *sk,
403 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
405 extern int tcp_getsockopt(struct sock *sk, int level,
409 extern int tcp_setsockopt(struct sock *sk, int level,
410 int optname, char __user *optval,
411 unsigned int optlen);
412 extern int compat_tcp_getsockopt(struct sock *sk,
413 int level, int optname,
414 char __user *optval, int __user *optlen);
415 extern int compat_tcp_setsockopt(struct sock *sk,
416 int level, int optname,
417 char __user *optval, unsigned int optlen);
418 extern void tcp_set_keepalive(struct sock *sk, int val);
419 extern void tcp_syn_ack_timeout(struct sock *sk,
420 struct request_sock *req);
421 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
423 size_t len, int nonblock,
424 int flags, int *addr_len);
426 extern void tcp_parse_options(struct sk_buff *skb,
427 struct tcp_options_received *opt_rx,
431 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
434 * TCP v4 functions exported for the inet6 API
437 extern void tcp_v4_send_check(struct sock *sk,
438 struct sk_buff *skb);
440 extern int tcp_v4_conn_request(struct sock *sk,
441 struct sk_buff *skb);
443 extern struct sock * tcp_create_openreq_child(struct sock *sk,
444 struct request_sock *req,
445 struct sk_buff *skb);
447 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
449 struct request_sock *req,
450 struct dst_entry *dst);
452 extern int tcp_v4_do_rcv(struct sock *sk,
453 struct sk_buff *skb);
455 extern int tcp_v4_connect(struct sock *sk,
456 struct sockaddr *uaddr,
459 extern int tcp_connect(struct sock *sk);
461 extern struct sk_buff * tcp_make_synack(struct sock *sk,
462 struct dst_entry *dst,
463 struct request_sock *req,
464 struct request_values *rvp);
466 extern int tcp_disconnect(struct sock *sk, int flags);
469 /* From syncookies.c */
470 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
471 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
472 struct ip_options *opt);
473 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
476 extern __u32 cookie_init_timestamp(struct request_sock *req);
477 extern void cookie_check_timestamp(struct tcp_options_received *tcp_opt);
479 /* From net/ipv6/syncookies.c */
480 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
481 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
486 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
488 extern int tcp_may_send_now(struct sock *sk);
489 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
490 extern void tcp_retransmit_timer(struct sock *sk);
491 extern void tcp_xmit_retransmit_queue(struct sock *);
492 extern void tcp_simple_retransmit(struct sock *);
493 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
494 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
496 extern void tcp_send_probe0(struct sock *);
497 extern void tcp_send_partial(struct sock *);
498 extern int tcp_write_wakeup(struct sock *);
499 extern void tcp_send_fin(struct sock *sk);
500 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
501 extern int tcp_send_synack(struct sock *);
502 extern void tcp_push_one(struct sock *, unsigned int mss_now);
503 extern void tcp_send_ack(struct sock *sk);
504 extern void tcp_send_delayed_ack(struct sock *sk);
507 extern void tcp_cwnd_application_limited(struct sock *sk);
510 extern void tcp_init_xmit_timers(struct sock *);
511 static inline void tcp_clear_xmit_timers(struct sock *sk)
513 inet_csk_clear_xmit_timers(sk);
516 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
517 extern unsigned int tcp_current_mss(struct sock *sk);
519 /* Bound MSS / TSO packet size with the half of the window */
520 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
522 if (tp->max_window && pktsize > (tp->max_window >> 1))
523 return max(tp->max_window >> 1, 68U - tp->tcp_header_len);
529 extern void tcp_get_info(struct sock *, struct tcp_info *);
531 /* Read 'sendfile()'-style from a TCP socket */
532 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
533 unsigned int, size_t);
534 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
535 sk_read_actor_t recv_actor);
537 extern void tcp_initialize_rcv_mss(struct sock *sk);
539 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
540 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
541 extern void tcp_mtup_init(struct sock *sk);
543 static inline void tcp_bound_rto(const struct sock *sk)
545 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
546 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
549 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
551 return (tp->srtt >> 3) + tp->rttvar;
554 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
556 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
557 ntohl(TCP_FLAG_ACK) |
561 static inline void tcp_fast_path_on(struct tcp_sock *tp)
563 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
566 static inline void tcp_fast_path_check(struct sock *sk)
568 struct tcp_sock *tp = tcp_sk(sk);
570 if (skb_queue_empty(&tp->out_of_order_queue) &&
572 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
574 tcp_fast_path_on(tp);
577 /* Compute the actual rto_min value */
578 static inline u32 tcp_rto_min(struct sock *sk)
580 struct dst_entry *dst = __sk_dst_get(sk);
581 u32 rto_min = TCP_RTO_MIN;
583 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
584 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
588 /* Compute the actual receive window we are currently advertising.
589 * Rcv_nxt can be after the window if our peer push more data
590 * than the offered window.
592 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
594 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
601 /* Choose a new window, without checks for shrinking, and without
602 * scaling applied to the result. The caller does these things
603 * if necessary. This is a "raw" window selection.
605 extern u32 __tcp_select_window(struct sock *sk);
607 /* TCP timestamps are only 32-bits, this causes a slight
608 * complication on 64-bit systems since we store a snapshot
609 * of jiffies in the buffer control blocks below. We decided
610 * to use only the low 32-bits of jiffies and hide the ugly
611 * casts with the following macro.
613 #define tcp_time_stamp ((__u32)(jiffies))
615 /* This is what the send packet queuing engine uses to pass
616 * TCP per-packet control information to the transmission
617 * code. We also store the host-order sequence numbers in
618 * here too. This is 36 bytes on 32-bit architectures,
619 * 40 bytes on 64-bit machines, if this grows please adjust
620 * skbuff.h:skbuff->cb[xxx] size appropriately.
624 struct inet_skb_parm h4;
625 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
626 struct inet6_skb_parm h6;
628 } header; /* For incoming frames */
629 __u32 seq; /* Starting sequence number */
630 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
631 __u32 when; /* used to compute rtt's */
632 __u8 flags; /* TCP header flags. */
634 /* NOTE: These must match up to the flags byte in a
637 #define TCPCB_FLAG_FIN 0x01
638 #define TCPCB_FLAG_SYN 0x02
639 #define TCPCB_FLAG_RST 0x04
640 #define TCPCB_FLAG_PSH 0x08
641 #define TCPCB_FLAG_ACK 0x10
642 #define TCPCB_FLAG_URG 0x20
643 #define TCPCB_FLAG_ECE 0x40
644 #define TCPCB_FLAG_CWR 0x80
646 __u8 sacked; /* State flags for SACK/FACK. */
647 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
648 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
649 #define TCPCB_LOST 0x04 /* SKB is lost */
650 #define TCPCB_TAGBITS 0x07 /* All tag bits */
652 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
653 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
655 __u32 ack_seq; /* Sequence number ACK'd */
658 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
660 /* Due to TSO, an SKB can be composed of multiple actual
661 * packets. To keep these tracked properly, we use this.
663 static inline int tcp_skb_pcount(const struct sk_buff *skb)
665 return skb_shinfo(skb)->gso_segs;
668 /* This is valid iff tcp_skb_pcount() > 1. */
669 static inline int tcp_skb_mss(const struct sk_buff *skb)
671 return skb_shinfo(skb)->gso_size;
674 /* Events passed to congestion control interface */
676 CA_EVENT_TX_START, /* first transmit when no packets in flight */
677 CA_EVENT_CWND_RESTART, /* congestion window restart */
678 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
679 CA_EVENT_FRTO, /* fast recovery timeout */
680 CA_EVENT_LOSS, /* loss timeout */
681 CA_EVENT_FAST_ACK, /* in sequence ack */
682 CA_EVENT_SLOW_ACK, /* other ack */
686 * Interface for adding new TCP congestion control handlers
688 #define TCP_CA_NAME_MAX 16
689 #define TCP_CA_MAX 128
690 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
692 #define TCP_CONG_NON_RESTRICTED 0x1
693 #define TCP_CONG_RTT_STAMP 0x2
695 struct tcp_congestion_ops {
696 struct list_head list;
699 /* initialize private data (optional) */
700 void (*init)(struct sock *sk);
701 /* cleanup private data (optional) */
702 void (*release)(struct sock *sk);
704 /* return slow start threshold (required) */
705 u32 (*ssthresh)(struct sock *sk);
706 /* lower bound for congestion window (optional) */
707 u32 (*min_cwnd)(const struct sock *sk);
708 /* do new cwnd calculation (required) */
709 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
710 /* call before changing ca_state (optional) */
711 void (*set_state)(struct sock *sk, u8 new_state);
712 /* call when cwnd event occurs (optional) */
713 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
714 /* new value of cwnd after loss (optional) */
715 u32 (*undo_cwnd)(struct sock *sk);
716 /* hook for packet ack accounting (optional) */
717 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
718 /* get info for inet_diag (optional) */
719 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
721 char name[TCP_CA_NAME_MAX];
722 struct module *owner;
725 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
726 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
728 extern void tcp_init_congestion_control(struct sock *sk);
729 extern void tcp_cleanup_congestion_control(struct sock *sk);
730 extern int tcp_set_default_congestion_control(const char *name);
731 extern void tcp_get_default_congestion_control(char *name);
732 extern void tcp_get_available_congestion_control(char *buf, size_t len);
733 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
734 extern int tcp_set_allowed_congestion_control(char *allowed);
735 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
736 extern void tcp_slow_start(struct tcp_sock *tp);
737 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
739 extern struct tcp_congestion_ops tcp_init_congestion_ops;
740 extern u32 tcp_reno_ssthresh(struct sock *sk);
741 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
742 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
743 extern struct tcp_congestion_ops tcp_reno;
745 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
747 struct inet_connection_sock *icsk = inet_csk(sk);
749 if (icsk->icsk_ca_ops->set_state)
750 icsk->icsk_ca_ops->set_state(sk, ca_state);
751 icsk->icsk_ca_state = ca_state;
754 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
756 const struct inet_connection_sock *icsk = inet_csk(sk);
758 if (icsk->icsk_ca_ops->cwnd_event)
759 icsk->icsk_ca_ops->cwnd_event(sk, event);
762 /* These functions determine how the current flow behaves in respect of SACK
763 * handling. SACK is negotiated with the peer, and therefore it can vary
764 * between different flows.
766 * tcp_is_sack - SACK enabled
767 * tcp_is_reno - No SACK
768 * tcp_is_fack - FACK enabled, implies SACK enabled
770 static inline int tcp_is_sack(const struct tcp_sock *tp)
772 return tp->rx_opt.sack_ok;
775 static inline int tcp_is_reno(const struct tcp_sock *tp)
777 return !tcp_is_sack(tp);
780 static inline int tcp_is_fack(const struct tcp_sock *tp)
782 return tp->rx_opt.sack_ok & 2;
785 static inline void tcp_enable_fack(struct tcp_sock *tp)
787 tp->rx_opt.sack_ok |= 2;
790 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
792 return tp->sacked_out + tp->lost_out;
795 /* This determines how many packets are "in the network" to the best
796 * of our knowledge. In many cases it is conservative, but where
797 * detailed information is available from the receiver (via SACK
798 * blocks etc.) we can make more aggressive calculations.
800 * Use this for decisions involving congestion control, use just
801 * tp->packets_out to determine if the send queue is empty or not.
803 * Read this equation as:
805 * "Packets sent once on transmission queue" MINUS
806 * "Packets left network, but not honestly ACKed yet" PLUS
807 * "Packets fast retransmitted"
809 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
811 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
814 #define TCP_INFINITE_SSTHRESH 0x7fffffff
816 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
818 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
821 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
822 * The exception is rate halving phase, when cwnd is decreasing towards
825 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
827 const struct tcp_sock *tp = tcp_sk(sk);
828 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
829 return tp->snd_ssthresh;
831 return max(tp->snd_ssthresh,
832 ((tp->snd_cwnd >> 1) +
833 (tp->snd_cwnd >> 2)));
836 /* Use define here intentionally to get WARN_ON location shown at the caller */
837 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
839 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
840 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
842 /* Slow start with delack produces 3 packets of burst, so that
843 * it is safe "de facto". This will be the default - same as
844 * the default reordering threshold - but if reordering increases,
845 * we must be able to allow cwnd to burst at least this much in order
846 * to not pull it back when holes are filled.
848 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
850 return tp->reordering;
853 /* Returns end sequence number of the receiver's advertised window */
854 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
856 return tp->snd_una + tp->snd_wnd;
858 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
860 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
861 const struct sk_buff *skb)
864 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
867 static inline void tcp_check_probe_timer(struct sock *sk)
869 struct tcp_sock *tp = tcp_sk(sk);
870 const struct inet_connection_sock *icsk = inet_csk(sk);
872 if (!tp->packets_out && !icsk->icsk_pending)
873 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
874 icsk->icsk_rto, TCP_RTO_MAX);
877 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
882 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
888 * Calculate(/check) TCP checksum
890 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
891 __be32 daddr, __wsum base)
893 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
896 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
898 return __skb_checksum_complete(skb);
901 static inline int tcp_checksum_complete(struct sk_buff *skb)
903 return !skb_csum_unnecessary(skb) &&
904 __tcp_checksum_complete(skb);
907 /* Prequeue for VJ style copy to user, combined with checksumming. */
909 static inline void tcp_prequeue_init(struct tcp_sock *tp)
911 tp->ucopy.task = NULL;
913 tp->ucopy.memory = 0;
914 skb_queue_head_init(&tp->ucopy.prequeue);
915 #ifdef CONFIG_NET_DMA
916 tp->ucopy.dma_chan = NULL;
917 tp->ucopy.wakeup = 0;
918 tp->ucopy.pinned_list = NULL;
919 tp->ucopy.dma_cookie = 0;
923 /* Packet is added to VJ-style prequeue for processing in process
924 * context, if a reader task is waiting. Apparently, this exciting
925 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
926 * failed somewhere. Latency? Burstiness? Well, at least now we will
927 * see, why it failed. 8)8) --ANK
929 * NOTE: is this not too big to inline?
931 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
933 struct tcp_sock *tp = tcp_sk(sk);
935 if (sysctl_tcp_low_latency || !tp->ucopy.task)
938 __skb_queue_tail(&tp->ucopy.prequeue, skb);
939 tp->ucopy.memory += skb->truesize;
940 if (tp->ucopy.memory > sk->sk_rcvbuf) {
941 struct sk_buff *skb1;
943 BUG_ON(sock_owned_by_user(sk));
945 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
946 sk_backlog_rcv(sk, skb1);
947 NET_INC_STATS_BH(sock_net(sk),
948 LINUX_MIB_TCPPREQUEUEDROPPED);
951 tp->ucopy.memory = 0;
952 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
953 wake_up_interruptible_sync_poll(sk_sleep(sk),
954 POLLIN | POLLRDNORM | POLLRDBAND);
955 if (!inet_csk_ack_scheduled(sk))
956 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
957 (3 * tcp_rto_min(sk)) / 4,
967 static const char *statename[]={
968 "Unused","Established","Syn Sent","Syn Recv",
969 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
970 "Close Wait","Last ACK","Listen","Closing"
973 extern void tcp_set_state(struct sock *sk, int state);
975 extern void tcp_done(struct sock *sk);
977 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
980 rx_opt->num_sacks = 0;
983 /* Determine a window scaling and initial window to offer. */
984 extern void tcp_select_initial_window(int __space, __u32 mss,
985 __u32 *rcv_wnd, __u32 *window_clamp,
986 int wscale_ok, __u8 *rcv_wscale,
989 static inline int tcp_win_from_space(int space)
991 return sysctl_tcp_adv_win_scale<=0 ?
992 (space>>(-sysctl_tcp_adv_win_scale)) :
993 space - (space>>sysctl_tcp_adv_win_scale);
996 /* Note: caller must be prepared to deal with negative returns */
997 static inline int tcp_space(const struct sock *sk)
999 return tcp_win_from_space(sk->sk_rcvbuf -
1000 atomic_read(&sk->sk_rmem_alloc));
1003 static inline int tcp_full_space(const struct sock *sk)
1005 return tcp_win_from_space(sk->sk_rcvbuf);
1008 static inline void tcp_openreq_init(struct request_sock *req,
1009 struct tcp_options_received *rx_opt,
1010 struct sk_buff *skb)
1012 struct inet_request_sock *ireq = inet_rsk(req);
1014 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1016 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1017 req->mss = rx_opt->mss_clamp;
1018 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1019 ireq->tstamp_ok = rx_opt->tstamp_ok;
1020 ireq->sack_ok = rx_opt->sack_ok;
1021 ireq->snd_wscale = rx_opt->snd_wscale;
1022 ireq->wscale_ok = rx_opt->wscale_ok;
1025 ireq->rmt_port = tcp_hdr(skb)->source;
1026 ireq->loc_port = tcp_hdr(skb)->dest;
1029 extern void tcp_enter_memory_pressure(struct sock *sk);
1031 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1033 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1036 static inline int keepalive_time_when(const struct tcp_sock *tp)
1038 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1041 static inline int keepalive_probes(const struct tcp_sock *tp)
1043 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1046 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1048 const struct inet_connection_sock *icsk = &tp->inet_conn;
1050 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1051 tcp_time_stamp - tp->rcv_tstamp);
1054 static inline int tcp_fin_time(const struct sock *sk)
1056 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1057 const int rto = inet_csk(sk)->icsk_rto;
1059 if (fin_timeout < (rto << 2) - (rto >> 1))
1060 fin_timeout = (rto << 2) - (rto >> 1);
1065 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1068 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1070 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1076 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1079 if (tcp_paws_check(rx_opt, 0))
1082 /* RST segments are not recommended to carry timestamp,
1083 and, if they do, it is recommended to ignore PAWS because
1084 "their cleanup function should take precedence over timestamps."
1085 Certainly, it is mistake. It is necessary to understand the reasons
1086 of this constraint to relax it: if peer reboots, clock may go
1087 out-of-sync and half-open connections will not be reset.
1088 Actually, the problem would be not existing if all
1089 the implementations followed draft about maintaining clock
1090 via reboots. Linux-2.2 DOES NOT!
1092 However, we can relax time bounds for RST segments to MSL.
1094 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1099 #define TCP_CHECK_TIMER(sk) do { } while (0)
1101 static inline void tcp_mib_init(struct net *net)
1104 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1105 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1106 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1107 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1111 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1113 tp->lost_skb_hint = NULL;
1114 tp->scoreboard_skb_hint = NULL;
1117 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1119 tcp_clear_retrans_hints_partial(tp);
1120 tp->retransmit_skb_hint = NULL;
1126 /* - key database */
1127 struct tcp_md5sig_key {
1132 struct tcp4_md5sig_key {
1133 struct tcp_md5sig_key base;
1137 struct tcp6_md5sig_key {
1138 struct tcp_md5sig_key base;
1140 u32 scope_id; /* XXX */
1142 struct in6_addr addr;
1146 struct tcp_md5sig_info {
1147 struct tcp4_md5sig_key *keys4;
1148 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1149 struct tcp6_md5sig_key *keys6;
1157 /* - pseudo header */
1158 struct tcp4_pseudohdr {
1166 struct tcp6_pseudohdr {
1167 struct in6_addr saddr;
1168 struct in6_addr daddr;
1170 __be32 protocol; /* including padding */
1173 union tcp_md5sum_block {
1174 struct tcp4_pseudohdr ip4;
1175 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1176 struct tcp6_pseudohdr ip6;
1180 /* - pool: digest algorithm, hash description and scratch buffer */
1181 struct tcp_md5sig_pool {
1182 struct hash_desc md5_desc;
1183 union tcp_md5sum_block md5_blk;
1186 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1189 extern int tcp_v4_md5_hash_skb(char *md5_hash,
1190 struct tcp_md5sig_key *key,
1192 struct request_sock *req,
1193 struct sk_buff *skb);
1195 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1196 struct sock *addr_sk);
1198 extern int tcp_v4_md5_do_add(struct sock *sk,
1203 extern int tcp_v4_md5_do_del(struct sock *sk,
1206 #ifdef CONFIG_TCP_MD5SIG
1207 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1208 &(struct tcp_md5sig_key) { \
1209 .key = (twsk)->tw_md5_key, \
1210 .keylen = (twsk)->tw_md5_keylen, \
1213 #define tcp_twsk_md5_key(twsk) NULL
1216 extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *);
1217 extern void tcp_free_md5sig_pool(void);
1219 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1220 extern void tcp_put_md5sig_pool(void);
1222 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1223 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1224 unsigned header_len);
1225 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1226 struct tcp_md5sig_key *key);
1228 /* write queue abstraction */
1229 static inline void tcp_write_queue_purge(struct sock *sk)
1231 struct sk_buff *skb;
1233 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1234 sk_wmem_free_skb(sk, skb);
1236 tcp_clear_all_retrans_hints(tcp_sk(sk));
1239 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1241 return skb_peek(&sk->sk_write_queue);
1244 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1246 return skb_peek_tail(&sk->sk_write_queue);
1249 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1251 return skb_queue_next(&sk->sk_write_queue, skb);
1254 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb)
1256 return skb_queue_prev(&sk->sk_write_queue, skb);
1259 #define tcp_for_write_queue(skb, sk) \
1260 skb_queue_walk(&(sk)->sk_write_queue, skb)
1262 #define tcp_for_write_queue_from(skb, sk) \
1263 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1265 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1266 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1268 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1270 return sk->sk_send_head;
1273 static inline bool tcp_skb_is_last(const struct sock *sk,
1274 const struct sk_buff *skb)
1276 return skb_queue_is_last(&sk->sk_write_queue, skb);
1279 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1281 if (tcp_skb_is_last(sk, skb))
1282 sk->sk_send_head = NULL;
1284 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1287 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1289 if (sk->sk_send_head == skb_unlinked)
1290 sk->sk_send_head = NULL;
1293 static inline void tcp_init_send_head(struct sock *sk)
1295 sk->sk_send_head = NULL;
1298 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1300 __skb_queue_tail(&sk->sk_write_queue, skb);
1303 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1305 __tcp_add_write_queue_tail(sk, skb);
1307 /* Queue it, remembering where we must start sending. */
1308 if (sk->sk_send_head == NULL) {
1309 sk->sk_send_head = skb;
1311 if (tcp_sk(sk)->highest_sack == NULL)
1312 tcp_sk(sk)->highest_sack = skb;
1316 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1318 __skb_queue_head(&sk->sk_write_queue, skb);
1321 /* Insert buff after skb on the write queue of sk. */
1322 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1323 struct sk_buff *buff,
1326 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1329 /* Insert new before skb on the write queue of sk. */
1330 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1331 struct sk_buff *skb,
1334 __skb_queue_before(&sk->sk_write_queue, skb, new);
1336 if (sk->sk_send_head == skb)
1337 sk->sk_send_head = new;
1340 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1342 __skb_unlink(skb, &sk->sk_write_queue);
1345 static inline int tcp_write_queue_empty(struct sock *sk)
1347 return skb_queue_empty(&sk->sk_write_queue);
1350 static inline void tcp_push_pending_frames(struct sock *sk)
1352 if (tcp_send_head(sk)) {
1353 struct tcp_sock *tp = tcp_sk(sk);
1355 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1359 /* Start sequence of the highest skb with SACKed bit, valid only if
1360 * sacked > 0 or when the caller has ensured validity by itself.
1362 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1364 if (!tp->sacked_out)
1367 if (tp->highest_sack == NULL)
1370 return TCP_SKB_CB(tp->highest_sack)->seq;
1373 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1375 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1376 tcp_write_queue_next(sk, skb);
1379 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1381 return tcp_sk(sk)->highest_sack;
1384 static inline void tcp_highest_sack_reset(struct sock *sk)
1386 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1389 /* Called when old skb is about to be deleted (to be combined with new skb) */
1390 static inline void tcp_highest_sack_combine(struct sock *sk,
1391 struct sk_buff *old,
1392 struct sk_buff *new)
1394 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1395 tcp_sk(sk)->highest_sack = new;
1398 /* Determines whether this is a thin stream (which may suffer from
1399 * increased latency). Used to trigger latency-reducing mechanisms.
1401 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp)
1403 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1407 enum tcp_seq_states {
1408 TCP_SEQ_STATE_LISTENING,
1409 TCP_SEQ_STATE_OPENREQ,
1410 TCP_SEQ_STATE_ESTABLISHED,
1411 TCP_SEQ_STATE_TIME_WAIT,
1414 struct tcp_seq_afinfo {
1417 struct file_operations seq_fops;
1418 struct seq_operations seq_ops;
1421 struct tcp_iter_state {
1422 struct seq_net_private p;
1424 enum tcp_seq_states state;
1425 struct sock *syn_wait_sk;
1426 int bucket, sbucket, num, uid;
1429 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1430 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1432 extern struct request_sock_ops tcp_request_sock_ops;
1433 extern struct request_sock_ops tcp6_request_sock_ops;
1435 extern void tcp_v4_destroy_sock(struct sock *sk);
1437 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1438 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1439 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1440 struct sk_buff *skb);
1441 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1442 struct sk_buff *skb);
1443 extern int tcp_gro_complete(struct sk_buff *skb);
1444 extern int tcp4_gro_complete(struct sk_buff *skb);
1446 #ifdef CONFIG_PROC_FS
1447 extern int tcp4_proc_init(void);
1448 extern void tcp4_proc_exit(void);
1451 /* TCP af-specific functions */
1452 struct tcp_sock_af_ops {
1453 #ifdef CONFIG_TCP_MD5SIG
1454 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1455 struct sock *addr_sk);
1456 int (*calc_md5_hash) (char *location,
1457 struct tcp_md5sig_key *md5,
1459 struct request_sock *req,
1460 struct sk_buff *skb);
1461 int (*md5_add) (struct sock *sk,
1462 struct sock *addr_sk,
1465 int (*md5_parse) (struct sock *sk,
1466 char __user *optval,
1471 struct tcp_request_sock_ops {
1472 #ifdef CONFIG_TCP_MD5SIG
1473 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1474 struct request_sock *req);
1475 int (*calc_md5_hash) (char *location,
1476 struct tcp_md5sig_key *md5,
1478 struct request_sock *req,
1479 struct sk_buff *skb);
1483 /* Using SHA1 for now, define some constants.
1485 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1486 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1487 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1489 extern int tcp_cookie_generator(u32 *bakery);
1492 * struct tcp_cookie_values - each socket needs extra space for the
1493 * cookies, together with (optional) space for any SYN data.
1495 * A tcp_sock contains a pointer to the current value, and this is
1496 * cloned to the tcp_timewait_sock.
1498 * @cookie_pair: variable data from the option exchange.
1500 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1501 * indicates default (sysctl_tcp_cookie_size).
1502 * After cookie sent, remembers size of cookie.
1503 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1505 * @s_data_desired: user specified tcpct_s_data_desired. When the
1506 * constant payload is specified (@s_data_constant),
1507 * holds its length instead.
1508 * Range 0 to TCP_MSS_DESIRED.
1510 * @s_data_payload: constant data that is to be included in the
1511 * payload of SYN or SYNACK segments when the
1512 * cookie option is present.
1514 struct tcp_cookie_values {
1516 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE];
1517 u8 cookie_pair_size;
1519 u16 s_data_desired:11,
1524 u8 s_data_payload[0];
1527 static inline void tcp_cookie_values_release(struct kref *kref)
1529 kfree(container_of(kref, struct tcp_cookie_values, kref));
1532 /* The length of constant payload data. Note that s_data_desired is
1533 * overloaded, depending on s_data_constant: either the length of constant
1534 * data (returned here) or the limit on variable data.
1536 static inline int tcp_s_data_size(const struct tcp_sock *tp)
1538 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1539 ? tp->cookie_values->s_data_desired
1544 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1546 * As tcp_request_sock has already been extended in other places, the
1547 * only remaining method is to pass stack values along as function
1548 * parameters. These parameters are not needed after sending SYNACK.
1550 * @cookie_bakery: cryptographic secret and message workspace.
1552 * @cookie_plus: bytes in authenticator/cookie option, copied from
1553 * struct tcp_options_received (above).
1555 struct tcp_extend_values {
1556 struct request_values rv;
1557 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS];
1563 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1565 return (struct tcp_extend_values *)rvp;
1568 extern void tcp_v4_init(void);
1569 extern void tcp_init(void);