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 int tcp_too_many_orphans(struct sock *sk, int num)
273 return (num > sysctl_tcp_max_orphans) ||
274 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
275 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
278 /* syncookies: remember time of last synqueue overflow */
279 static inline void tcp_synq_overflow(struct sock *sk)
281 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
284 /* syncookies: no recent synqueue overflow on this listening socket? */
285 static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
287 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
288 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT);
291 extern struct proto tcp_prot;
293 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
294 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
295 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
296 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
297 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
299 extern void tcp_v4_err(struct sk_buff *skb, u32);
301 extern void tcp_shutdown (struct sock *sk, int how);
303 extern int tcp_v4_rcv(struct sk_buff *skb);
305 extern int tcp_v4_remember_stamp(struct sock *sk);
307 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
309 extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
310 struct msghdr *msg, size_t size);
311 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
313 extern int tcp_ioctl(struct sock *sk,
317 extern int tcp_rcv_state_process(struct sock *sk,
322 extern int tcp_rcv_established(struct sock *sk,
327 extern void tcp_rcv_space_adjust(struct sock *sk);
329 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
331 extern int tcp_twsk_unique(struct sock *sk,
332 struct sock *sktw, void *twp);
334 extern void tcp_twsk_destructor(struct sock *sk);
336 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
337 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
339 static inline void tcp_dec_quickack_mode(struct sock *sk,
340 const unsigned int pkts)
342 struct inet_connection_sock *icsk = inet_csk(sk);
344 if (icsk->icsk_ack.quick) {
345 if (pkts >= icsk->icsk_ack.quick) {
346 icsk->icsk_ack.quick = 0;
347 /* Leaving quickack mode we deflate ATO. */
348 icsk->icsk_ack.ato = TCP_ATO_MIN;
350 icsk->icsk_ack.quick -= pkts;
354 extern void tcp_enter_quickack_mode(struct sock *sk);
357 #define TCP_ECN_QUEUE_CWR 2
358 #define TCP_ECN_DEMAND_CWR 4
360 static __inline__ void
361 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
363 if (sysctl_tcp_ecn && th->ece && th->cwr)
364 inet_rsk(req)->ecn_ok = 1;
375 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
377 const struct tcphdr *th);
379 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
380 struct request_sock *req,
381 struct request_sock **prev);
382 extern int tcp_child_process(struct sock *parent,
384 struct sk_buff *skb);
385 extern int tcp_use_frto(struct sock *sk);
386 extern void tcp_enter_frto(struct sock *sk);
387 extern void tcp_enter_loss(struct sock *sk, int how);
388 extern void tcp_clear_retrans(struct tcp_sock *tp);
389 extern void tcp_update_metrics(struct sock *sk);
391 extern void tcp_close(struct sock *sk,
393 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
395 extern int tcp_getsockopt(struct sock *sk, int level,
399 extern int tcp_setsockopt(struct sock *sk, int level,
400 int optname, char __user *optval,
401 unsigned int optlen);
402 extern int compat_tcp_getsockopt(struct sock *sk,
403 int level, int optname,
404 char __user *optval, int __user *optlen);
405 extern int compat_tcp_setsockopt(struct sock *sk,
406 int level, int optname,
407 char __user *optval, unsigned int optlen);
408 extern void tcp_set_keepalive(struct sock *sk, int val);
409 extern void tcp_syn_ack_timeout(struct sock *sk,
410 struct request_sock *req);
411 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
413 size_t len, int nonblock,
414 int flags, int *addr_len);
416 extern void tcp_parse_options(struct sk_buff *skb,
417 struct tcp_options_received *opt_rx,
421 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
424 * TCP v4 functions exported for the inet6 API
427 extern void tcp_v4_send_check(struct sock *sk,
428 struct sk_buff *skb);
430 extern int tcp_v4_conn_request(struct sock *sk,
431 struct sk_buff *skb);
433 extern struct sock * tcp_create_openreq_child(struct sock *sk,
434 struct request_sock *req,
435 struct sk_buff *skb);
437 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
439 struct request_sock *req,
440 struct dst_entry *dst);
442 extern int tcp_v4_do_rcv(struct sock *sk,
443 struct sk_buff *skb);
445 extern int tcp_v4_connect(struct sock *sk,
446 struct sockaddr *uaddr,
449 extern int tcp_connect(struct sock *sk);
451 extern struct sk_buff * tcp_make_synack(struct sock *sk,
452 struct dst_entry *dst,
453 struct request_sock *req,
454 struct request_values *rvp);
456 extern int tcp_disconnect(struct sock *sk, int flags);
459 /* From syncookies.c */
460 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
461 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
462 struct ip_options *opt);
463 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
466 extern __u32 cookie_init_timestamp(struct request_sock *req);
467 extern bool cookie_check_timestamp(struct tcp_options_received *tcp_opt);
469 /* From net/ipv6/syncookies.c */
470 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
471 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
476 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
478 extern int tcp_may_send_now(struct sock *sk);
479 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
480 extern void tcp_retransmit_timer(struct sock *sk);
481 extern void tcp_xmit_retransmit_queue(struct sock *);
482 extern void tcp_simple_retransmit(struct sock *);
483 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
484 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
486 extern void tcp_send_probe0(struct sock *);
487 extern void tcp_send_partial(struct sock *);
488 extern int tcp_write_wakeup(struct sock *);
489 extern void tcp_send_fin(struct sock *sk);
490 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
491 extern int tcp_send_synack(struct sock *);
492 extern void tcp_push_one(struct sock *, unsigned int mss_now);
493 extern void tcp_send_ack(struct sock *sk);
494 extern void tcp_send_delayed_ack(struct sock *sk);
497 extern void tcp_cwnd_application_limited(struct sock *sk);
500 extern void tcp_init_xmit_timers(struct sock *);
501 static inline void tcp_clear_xmit_timers(struct sock *sk)
503 inet_csk_clear_xmit_timers(sk);
506 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
507 extern unsigned int tcp_current_mss(struct sock *sk);
509 /* Bound MSS / TSO packet size with the half of the window */
510 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
512 if (tp->max_window && pktsize > (tp->max_window >> 1))
513 return max(tp->max_window >> 1, 68U - tp->tcp_header_len);
519 extern void tcp_get_info(struct sock *, struct tcp_info *);
521 /* Read 'sendfile()'-style from a TCP socket */
522 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
523 unsigned int, size_t);
524 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
525 sk_read_actor_t recv_actor);
527 extern void tcp_initialize_rcv_mss(struct sock *sk);
529 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
530 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
531 extern void tcp_mtup_init(struct sock *sk);
533 static inline void tcp_bound_rto(const struct sock *sk)
535 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
536 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
539 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
541 return (tp->srtt >> 3) + tp->rttvar;
544 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
546 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
547 ntohl(TCP_FLAG_ACK) |
551 static inline void tcp_fast_path_on(struct tcp_sock *tp)
553 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
556 static inline void tcp_fast_path_check(struct sock *sk)
558 struct tcp_sock *tp = tcp_sk(sk);
560 if (skb_queue_empty(&tp->out_of_order_queue) &&
562 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
564 tcp_fast_path_on(tp);
567 /* Compute the actual rto_min value */
568 static inline u32 tcp_rto_min(struct sock *sk)
570 struct dst_entry *dst = __sk_dst_get(sk);
571 u32 rto_min = TCP_RTO_MIN;
573 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
574 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
578 /* Compute the actual receive window we are currently advertising.
579 * Rcv_nxt can be after the window if our peer push more data
580 * than the offered window.
582 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
584 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
591 /* Choose a new window, without checks for shrinking, and without
592 * scaling applied to the result. The caller does these things
593 * if necessary. This is a "raw" window selection.
595 extern u32 __tcp_select_window(struct sock *sk);
597 /* TCP timestamps are only 32-bits, this causes a slight
598 * complication on 64-bit systems since we store a snapshot
599 * of jiffies in the buffer control blocks below. We decided
600 * to use only the low 32-bits of jiffies and hide the ugly
601 * casts with the following macro.
603 #define tcp_time_stamp ((__u32)(jiffies))
605 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
607 #define TCPHDR_FIN 0x01
608 #define TCPHDR_SYN 0x02
609 #define TCPHDR_RST 0x04
610 #define TCPHDR_PSH 0x08
611 #define TCPHDR_ACK 0x10
612 #define TCPHDR_URG 0x20
613 #define TCPHDR_ECE 0x40
614 #define TCPHDR_CWR 0x80
616 /* This is what the send packet queuing engine uses to pass
617 * TCP per-packet control information to the transmission
618 * code. We also store the host-order sequence numbers in
619 * here too. This is 36 bytes on 32-bit architectures,
620 * 40 bytes on 64-bit machines, if this grows please adjust
621 * skbuff.h:skbuff->cb[xxx] size appropriately.
625 struct inet_skb_parm h4;
626 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
627 struct inet6_skb_parm h6;
629 } header; /* For incoming frames */
630 __u32 seq; /* Starting sequence number */
631 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
632 __u32 when; /* used to compute rtt's */
633 __u8 flags; /* TCP header flags. */
634 __u8 sacked; /* State flags for SACK/FACK. */
635 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
636 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
637 #define TCPCB_LOST 0x04 /* SKB is lost */
638 #define TCPCB_TAGBITS 0x07 /* All tag bits */
640 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
641 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
643 __u32 ack_seq; /* Sequence number ACK'd */
646 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
648 /* Due to TSO, an SKB can be composed of multiple actual
649 * packets. To keep these tracked properly, we use this.
651 static inline int tcp_skb_pcount(const struct sk_buff *skb)
653 return skb_shinfo(skb)->gso_segs;
656 /* This is valid iff tcp_skb_pcount() > 1. */
657 static inline int tcp_skb_mss(const struct sk_buff *skb)
659 return skb_shinfo(skb)->gso_size;
662 /* Events passed to congestion control interface */
664 CA_EVENT_TX_START, /* first transmit when no packets in flight */
665 CA_EVENT_CWND_RESTART, /* congestion window restart */
666 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
667 CA_EVENT_FRTO, /* fast recovery timeout */
668 CA_EVENT_LOSS, /* loss timeout */
669 CA_EVENT_FAST_ACK, /* in sequence ack */
670 CA_EVENT_SLOW_ACK, /* other ack */
674 * Interface for adding new TCP congestion control handlers
676 #define TCP_CA_NAME_MAX 16
677 #define TCP_CA_MAX 128
678 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
680 #define TCP_CONG_NON_RESTRICTED 0x1
681 #define TCP_CONG_RTT_STAMP 0x2
683 struct tcp_congestion_ops {
684 struct list_head list;
687 /* initialize private data (optional) */
688 void (*init)(struct sock *sk);
689 /* cleanup private data (optional) */
690 void (*release)(struct sock *sk);
692 /* return slow start threshold (required) */
693 u32 (*ssthresh)(struct sock *sk);
694 /* lower bound for congestion window (optional) */
695 u32 (*min_cwnd)(const struct sock *sk);
696 /* do new cwnd calculation (required) */
697 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
698 /* call before changing ca_state (optional) */
699 void (*set_state)(struct sock *sk, u8 new_state);
700 /* call when cwnd event occurs (optional) */
701 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
702 /* new value of cwnd after loss (optional) */
703 u32 (*undo_cwnd)(struct sock *sk);
704 /* hook for packet ack accounting (optional) */
705 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
706 /* get info for inet_diag (optional) */
707 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
709 char name[TCP_CA_NAME_MAX];
710 struct module *owner;
713 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
714 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
716 extern void tcp_init_congestion_control(struct sock *sk);
717 extern void tcp_cleanup_congestion_control(struct sock *sk);
718 extern int tcp_set_default_congestion_control(const char *name);
719 extern void tcp_get_default_congestion_control(char *name);
720 extern void tcp_get_available_congestion_control(char *buf, size_t len);
721 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
722 extern int tcp_set_allowed_congestion_control(char *allowed);
723 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
724 extern void tcp_slow_start(struct tcp_sock *tp);
725 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
727 extern struct tcp_congestion_ops tcp_init_congestion_ops;
728 extern u32 tcp_reno_ssthresh(struct sock *sk);
729 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
730 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
731 extern struct tcp_congestion_ops tcp_reno;
733 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
735 struct inet_connection_sock *icsk = inet_csk(sk);
737 if (icsk->icsk_ca_ops->set_state)
738 icsk->icsk_ca_ops->set_state(sk, ca_state);
739 icsk->icsk_ca_state = ca_state;
742 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
744 const struct inet_connection_sock *icsk = inet_csk(sk);
746 if (icsk->icsk_ca_ops->cwnd_event)
747 icsk->icsk_ca_ops->cwnd_event(sk, event);
750 /* These functions determine how the current flow behaves in respect of SACK
751 * handling. SACK is negotiated with the peer, and therefore it can vary
752 * between different flows.
754 * tcp_is_sack - SACK enabled
755 * tcp_is_reno - No SACK
756 * tcp_is_fack - FACK enabled, implies SACK enabled
758 static inline int tcp_is_sack(const struct tcp_sock *tp)
760 return tp->rx_opt.sack_ok;
763 static inline int tcp_is_reno(const struct tcp_sock *tp)
765 return !tcp_is_sack(tp);
768 static inline int tcp_is_fack(const struct tcp_sock *tp)
770 return tp->rx_opt.sack_ok & 2;
773 static inline void tcp_enable_fack(struct tcp_sock *tp)
775 tp->rx_opt.sack_ok |= 2;
778 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
780 return tp->sacked_out + tp->lost_out;
783 /* This determines how many packets are "in the network" to the best
784 * of our knowledge. In many cases it is conservative, but where
785 * detailed information is available from the receiver (via SACK
786 * blocks etc.) we can make more aggressive calculations.
788 * Use this for decisions involving congestion control, use just
789 * tp->packets_out to determine if the send queue is empty or not.
791 * Read this equation as:
793 * "Packets sent once on transmission queue" MINUS
794 * "Packets left network, but not honestly ACKed yet" PLUS
795 * "Packets fast retransmitted"
797 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
799 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
802 #define TCP_INFINITE_SSTHRESH 0x7fffffff
804 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
806 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
809 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
810 * The exception is rate halving phase, when cwnd is decreasing towards
813 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
815 const struct tcp_sock *tp = tcp_sk(sk);
816 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
817 return tp->snd_ssthresh;
819 return max(tp->snd_ssthresh,
820 ((tp->snd_cwnd >> 1) +
821 (tp->snd_cwnd >> 2)));
824 /* Use define here intentionally to get WARN_ON location shown at the caller */
825 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
827 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
828 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
830 /* Slow start with delack produces 3 packets of burst, so that
831 * it is safe "de facto". This will be the default - same as
832 * the default reordering threshold - but if reordering increases,
833 * we must be able to allow cwnd to burst at least this much in order
834 * to not pull it back when holes are filled.
836 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
838 return tp->reordering;
841 /* Returns end sequence number of the receiver's advertised window */
842 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
844 return tp->snd_una + tp->snd_wnd;
846 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
848 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
849 const struct sk_buff *skb)
852 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
855 static inline void tcp_check_probe_timer(struct sock *sk)
857 struct tcp_sock *tp = tcp_sk(sk);
858 const struct inet_connection_sock *icsk = inet_csk(sk);
860 if (!tp->packets_out && !icsk->icsk_pending)
861 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
862 icsk->icsk_rto, TCP_RTO_MAX);
865 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
870 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
876 * Calculate(/check) TCP checksum
878 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
879 __be32 daddr, __wsum base)
881 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
884 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
886 return __skb_checksum_complete(skb);
889 static inline int tcp_checksum_complete(struct sk_buff *skb)
891 return !skb_csum_unnecessary(skb) &&
892 __tcp_checksum_complete(skb);
895 /* Prequeue for VJ style copy to user, combined with checksumming. */
897 static inline void tcp_prequeue_init(struct tcp_sock *tp)
899 tp->ucopy.task = NULL;
901 tp->ucopy.memory = 0;
902 skb_queue_head_init(&tp->ucopy.prequeue);
903 #ifdef CONFIG_NET_DMA
904 tp->ucopy.dma_chan = NULL;
905 tp->ucopy.wakeup = 0;
906 tp->ucopy.pinned_list = NULL;
907 tp->ucopy.dma_cookie = 0;
911 /* Packet is added to VJ-style prequeue for processing in process
912 * context, if a reader task is waiting. Apparently, this exciting
913 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
914 * failed somewhere. Latency? Burstiness? Well, at least now we will
915 * see, why it failed. 8)8) --ANK
917 * NOTE: is this not too big to inline?
919 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
921 struct tcp_sock *tp = tcp_sk(sk);
923 if (sysctl_tcp_low_latency || !tp->ucopy.task)
926 __skb_queue_tail(&tp->ucopy.prequeue, skb);
927 tp->ucopy.memory += skb->truesize;
928 if (tp->ucopy.memory > sk->sk_rcvbuf) {
929 struct sk_buff *skb1;
931 BUG_ON(sock_owned_by_user(sk));
933 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
934 sk_backlog_rcv(sk, skb1);
935 NET_INC_STATS_BH(sock_net(sk),
936 LINUX_MIB_TCPPREQUEUEDROPPED);
939 tp->ucopy.memory = 0;
940 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
941 wake_up_interruptible_sync_poll(sk_sleep(sk),
942 POLLIN | POLLRDNORM | POLLRDBAND);
943 if (!inet_csk_ack_scheduled(sk))
944 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
945 (3 * tcp_rto_min(sk)) / 4,
955 static const char *statename[]={
956 "Unused","Established","Syn Sent","Syn Recv",
957 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
958 "Close Wait","Last ACK","Listen","Closing"
961 extern void tcp_set_state(struct sock *sk, int state);
963 extern void tcp_done(struct sock *sk);
965 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
968 rx_opt->num_sacks = 0;
971 /* Determine a window scaling and initial window to offer. */
972 extern void tcp_select_initial_window(int __space, __u32 mss,
973 __u32 *rcv_wnd, __u32 *window_clamp,
974 int wscale_ok, __u8 *rcv_wscale,
977 static inline int tcp_win_from_space(int space)
979 return sysctl_tcp_adv_win_scale<=0 ?
980 (space>>(-sysctl_tcp_adv_win_scale)) :
981 space - (space>>sysctl_tcp_adv_win_scale);
984 /* Note: caller must be prepared to deal with negative returns */
985 static inline int tcp_space(const struct sock *sk)
987 return tcp_win_from_space(sk->sk_rcvbuf -
988 atomic_read(&sk->sk_rmem_alloc));
991 static inline int tcp_full_space(const struct sock *sk)
993 return tcp_win_from_space(sk->sk_rcvbuf);
996 static inline void tcp_openreq_init(struct request_sock *req,
997 struct tcp_options_received *rx_opt,
1000 struct inet_request_sock *ireq = inet_rsk(req);
1002 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1004 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1005 req->mss = rx_opt->mss_clamp;
1006 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1007 ireq->tstamp_ok = rx_opt->tstamp_ok;
1008 ireq->sack_ok = rx_opt->sack_ok;
1009 ireq->snd_wscale = rx_opt->snd_wscale;
1010 ireq->wscale_ok = rx_opt->wscale_ok;
1013 ireq->rmt_port = tcp_hdr(skb)->source;
1014 ireq->loc_port = tcp_hdr(skb)->dest;
1017 extern void tcp_enter_memory_pressure(struct sock *sk);
1019 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1021 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1024 static inline int keepalive_time_when(const struct tcp_sock *tp)
1026 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1029 static inline int keepalive_probes(const struct tcp_sock *tp)
1031 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1034 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1036 const struct inet_connection_sock *icsk = &tp->inet_conn;
1038 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1039 tcp_time_stamp - tp->rcv_tstamp);
1042 static inline int tcp_fin_time(const struct sock *sk)
1044 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1045 const int rto = inet_csk(sk)->icsk_rto;
1047 if (fin_timeout < (rto << 2) - (rto >> 1))
1048 fin_timeout = (rto << 2) - (rto >> 1);
1053 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1056 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1058 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1064 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1067 if (tcp_paws_check(rx_opt, 0))
1070 /* RST segments are not recommended to carry timestamp,
1071 and, if they do, it is recommended to ignore PAWS because
1072 "their cleanup function should take precedence over timestamps."
1073 Certainly, it is mistake. It is necessary to understand the reasons
1074 of this constraint to relax it: if peer reboots, clock may go
1075 out-of-sync and half-open connections will not be reset.
1076 Actually, the problem would be not existing if all
1077 the implementations followed draft about maintaining clock
1078 via reboots. Linux-2.2 DOES NOT!
1080 However, we can relax time bounds for RST segments to MSL.
1082 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1087 #define TCP_CHECK_TIMER(sk) do { } while (0)
1089 static inline void tcp_mib_init(struct net *net)
1092 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1093 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1094 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1095 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1099 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1101 tp->lost_skb_hint = NULL;
1102 tp->scoreboard_skb_hint = NULL;
1105 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1107 tcp_clear_retrans_hints_partial(tp);
1108 tp->retransmit_skb_hint = NULL;
1114 /* - key database */
1115 struct tcp_md5sig_key {
1120 struct tcp4_md5sig_key {
1121 struct tcp_md5sig_key base;
1125 struct tcp6_md5sig_key {
1126 struct tcp_md5sig_key base;
1128 u32 scope_id; /* XXX */
1130 struct in6_addr addr;
1134 struct tcp_md5sig_info {
1135 struct tcp4_md5sig_key *keys4;
1136 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1137 struct tcp6_md5sig_key *keys6;
1145 /* - pseudo header */
1146 struct tcp4_pseudohdr {
1154 struct tcp6_pseudohdr {
1155 struct in6_addr saddr;
1156 struct in6_addr daddr;
1158 __be32 protocol; /* including padding */
1161 union tcp_md5sum_block {
1162 struct tcp4_pseudohdr ip4;
1163 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1164 struct tcp6_pseudohdr ip6;
1168 /* - pool: digest algorithm, hash description and scratch buffer */
1169 struct tcp_md5sig_pool {
1170 struct hash_desc md5_desc;
1171 union tcp_md5sum_block md5_blk;
1174 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1177 extern int tcp_v4_md5_hash_skb(char *md5_hash,
1178 struct tcp_md5sig_key *key,
1180 struct request_sock *req,
1181 struct sk_buff *skb);
1183 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1184 struct sock *addr_sk);
1186 extern int tcp_v4_md5_do_add(struct sock *sk,
1191 extern int tcp_v4_md5_do_del(struct sock *sk,
1194 #ifdef CONFIG_TCP_MD5SIG
1195 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1196 &(struct tcp_md5sig_key) { \
1197 .key = (twsk)->tw_md5_key, \
1198 .keylen = (twsk)->tw_md5_keylen, \
1201 #define tcp_twsk_md5_key(twsk) NULL
1204 extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *);
1205 extern void tcp_free_md5sig_pool(void);
1207 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1208 extern void tcp_put_md5sig_pool(void);
1210 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1211 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1212 unsigned header_len);
1213 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1214 struct tcp_md5sig_key *key);
1216 /* write queue abstraction */
1217 static inline void tcp_write_queue_purge(struct sock *sk)
1219 struct sk_buff *skb;
1221 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1222 sk_wmem_free_skb(sk, skb);
1224 tcp_clear_all_retrans_hints(tcp_sk(sk));
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 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1258 return sk->sk_send_head;
1261 static inline bool tcp_skb_is_last(const struct sock *sk,
1262 const struct sk_buff *skb)
1264 return skb_queue_is_last(&sk->sk_write_queue, skb);
1267 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1269 if (tcp_skb_is_last(sk, skb))
1270 sk->sk_send_head = NULL;
1272 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1275 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1277 if (sk->sk_send_head == skb_unlinked)
1278 sk->sk_send_head = NULL;
1281 static inline void tcp_init_send_head(struct sock *sk)
1283 sk->sk_send_head = NULL;
1286 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1288 __skb_queue_tail(&sk->sk_write_queue, skb);
1291 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1293 __tcp_add_write_queue_tail(sk, skb);
1295 /* Queue it, remembering where we must start sending. */
1296 if (sk->sk_send_head == NULL) {
1297 sk->sk_send_head = skb;
1299 if (tcp_sk(sk)->highest_sack == NULL)
1300 tcp_sk(sk)->highest_sack = skb;
1304 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1306 __skb_queue_head(&sk->sk_write_queue, skb);
1309 /* Insert buff after skb on the write queue of sk. */
1310 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1311 struct sk_buff *buff,
1314 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1317 /* Insert new before skb on the write queue of sk. */
1318 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1319 struct sk_buff *skb,
1322 __skb_queue_before(&sk->sk_write_queue, skb, new);
1324 if (sk->sk_send_head == skb)
1325 sk->sk_send_head = new;
1328 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1330 __skb_unlink(skb, &sk->sk_write_queue);
1333 static inline int tcp_write_queue_empty(struct sock *sk)
1335 return skb_queue_empty(&sk->sk_write_queue);
1338 static inline void tcp_push_pending_frames(struct sock *sk)
1340 if (tcp_send_head(sk)) {
1341 struct tcp_sock *tp = tcp_sk(sk);
1343 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1347 /* Start sequence of the highest skb with SACKed bit, valid only if
1348 * sacked > 0 or when the caller has ensured validity by itself.
1350 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1352 if (!tp->sacked_out)
1355 if (tp->highest_sack == NULL)
1358 return TCP_SKB_CB(tp->highest_sack)->seq;
1361 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1363 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1364 tcp_write_queue_next(sk, skb);
1367 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1369 return tcp_sk(sk)->highest_sack;
1372 static inline void tcp_highest_sack_reset(struct sock *sk)
1374 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1377 /* Called when old skb is about to be deleted (to be combined with new skb) */
1378 static inline void tcp_highest_sack_combine(struct sock *sk,
1379 struct sk_buff *old,
1380 struct sk_buff *new)
1382 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1383 tcp_sk(sk)->highest_sack = new;
1386 /* Determines whether this is a thin stream (which may suffer from
1387 * increased latency). Used to trigger latency-reducing mechanisms.
1389 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp)
1391 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1395 enum tcp_seq_states {
1396 TCP_SEQ_STATE_LISTENING,
1397 TCP_SEQ_STATE_OPENREQ,
1398 TCP_SEQ_STATE_ESTABLISHED,
1399 TCP_SEQ_STATE_TIME_WAIT,
1402 struct tcp_seq_afinfo {
1405 struct file_operations seq_fops;
1406 struct seq_operations seq_ops;
1409 struct tcp_iter_state {
1410 struct seq_net_private p;
1412 enum tcp_seq_states state;
1413 struct sock *syn_wait_sk;
1414 int bucket, offset, sbucket, num, uid;
1418 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1419 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1421 extern struct request_sock_ops tcp_request_sock_ops;
1422 extern struct request_sock_ops tcp6_request_sock_ops;
1424 extern void tcp_v4_destroy_sock(struct sock *sk);
1426 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1427 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1428 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1429 struct sk_buff *skb);
1430 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1431 struct sk_buff *skb);
1432 extern int tcp_gro_complete(struct sk_buff *skb);
1433 extern int tcp4_gro_complete(struct sk_buff *skb);
1435 #ifdef CONFIG_PROC_FS
1436 extern int tcp4_proc_init(void);
1437 extern void tcp4_proc_exit(void);
1440 /* TCP af-specific functions */
1441 struct tcp_sock_af_ops {
1442 #ifdef CONFIG_TCP_MD5SIG
1443 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1444 struct sock *addr_sk);
1445 int (*calc_md5_hash) (char *location,
1446 struct tcp_md5sig_key *md5,
1448 struct request_sock *req,
1449 struct sk_buff *skb);
1450 int (*md5_add) (struct sock *sk,
1451 struct sock *addr_sk,
1454 int (*md5_parse) (struct sock *sk,
1455 char __user *optval,
1460 struct tcp_request_sock_ops {
1461 #ifdef CONFIG_TCP_MD5SIG
1462 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1463 struct request_sock *req);
1464 int (*calc_md5_hash) (char *location,
1465 struct tcp_md5sig_key *md5,
1467 struct request_sock *req,
1468 struct sk_buff *skb);
1472 /* Using SHA1 for now, define some constants.
1474 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1475 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1476 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1478 extern int tcp_cookie_generator(u32 *bakery);
1481 * struct tcp_cookie_values - each socket needs extra space for the
1482 * cookies, together with (optional) space for any SYN data.
1484 * A tcp_sock contains a pointer to the current value, and this is
1485 * cloned to the tcp_timewait_sock.
1487 * @cookie_pair: variable data from the option exchange.
1489 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1490 * indicates default (sysctl_tcp_cookie_size).
1491 * After cookie sent, remembers size of cookie.
1492 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1494 * @s_data_desired: user specified tcpct_s_data_desired. When the
1495 * constant payload is specified (@s_data_constant),
1496 * holds its length instead.
1497 * Range 0 to TCP_MSS_DESIRED.
1499 * @s_data_payload: constant data that is to be included in the
1500 * payload of SYN or SYNACK segments when the
1501 * cookie option is present.
1503 struct tcp_cookie_values {
1505 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE];
1506 u8 cookie_pair_size;
1508 u16 s_data_desired:11,
1513 u8 s_data_payload[0];
1516 static inline void tcp_cookie_values_release(struct kref *kref)
1518 kfree(container_of(kref, struct tcp_cookie_values, kref));
1521 /* The length of constant payload data. Note that s_data_desired is
1522 * overloaded, depending on s_data_constant: either the length of constant
1523 * data (returned here) or the limit on variable data.
1525 static inline int tcp_s_data_size(const struct tcp_sock *tp)
1527 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1528 ? tp->cookie_values->s_data_desired
1533 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1535 * As tcp_request_sock has already been extended in other places, the
1536 * only remaining method is to pass stack values along as function
1537 * parameters. These parameters are not needed after sending SYNACK.
1539 * @cookie_bakery: cryptographic secret and message workspace.
1541 * @cookie_plus: bytes in authenticator/cookie option, copied from
1542 * struct tcp_options_received (above).
1544 struct tcp_extend_values {
1545 struct request_values rv;
1546 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS];
1552 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1554 return (struct tcp_extend_values *)rvp;
1557 extern void tcp_v4_init(void);
1558 extern void tcp_init(void);