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 */
103 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
104 * state, about 60 seconds */
105 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
106 /* BSD style FIN_WAIT2 deadlock breaker.
107 * It used to be 3min, new value is 60sec,
108 * to combine FIN-WAIT-2 timeout with
112 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
114 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
115 #define TCP_ATO_MIN ((unsigned)(HZ/25))
117 #define TCP_DELACK_MIN 4U
118 #define TCP_ATO_MIN 4U
120 #define TCP_RTO_MAX ((unsigned)(120*HZ))
121 #define TCP_RTO_MIN ((unsigned)(HZ/5))
122 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
124 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
125 * for local resources.
128 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
129 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
130 #define TCP_KEEPALIVE_INTVL (75*HZ)
132 #define MAX_TCP_KEEPIDLE 32767
133 #define MAX_TCP_KEEPINTVL 32767
134 #define MAX_TCP_KEEPCNT 127
135 #define MAX_TCP_SYNCNT 127
137 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
139 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
140 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
141 * after this time. It should be equal
142 * (or greater than) TCP_TIMEWAIT_LEN
143 * to provide reliability equal to one
144 * provided by timewait state.
146 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
147 * timestamps. It must be less than
148 * minimal timewait lifetime.
154 #define TCPOPT_NOP 1 /* Padding */
155 #define TCPOPT_EOL 0 /* End of options */
156 #define TCPOPT_MSS 2 /* Segment size negotiating */
157 #define TCPOPT_WINDOW 3 /* Window scaling */
158 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
159 #define TCPOPT_SACK 5 /* SACK Block */
160 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
161 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
162 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
168 #define TCPOLEN_MSS 4
169 #define TCPOLEN_WINDOW 3
170 #define TCPOLEN_SACK_PERM 2
171 #define TCPOLEN_TIMESTAMP 10
172 #define TCPOLEN_MD5SIG 18
173 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
174 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
175 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
176 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
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 /* TCP thin-stream limits */
194 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
196 extern struct inet_timewait_death_row tcp_death_row;
198 /* sysctl variables for tcp */
199 extern int sysctl_tcp_timestamps;
200 extern int sysctl_tcp_window_scaling;
201 extern int sysctl_tcp_sack;
202 extern int sysctl_tcp_fin_timeout;
203 extern int sysctl_tcp_keepalive_time;
204 extern int sysctl_tcp_keepalive_probes;
205 extern int sysctl_tcp_keepalive_intvl;
206 extern int sysctl_tcp_syn_retries;
207 extern int sysctl_tcp_synack_retries;
208 extern int sysctl_tcp_retries1;
209 extern int sysctl_tcp_retries2;
210 extern int sysctl_tcp_orphan_retries;
211 extern int sysctl_tcp_syncookies;
212 extern int sysctl_tcp_retrans_collapse;
213 extern int sysctl_tcp_stdurg;
214 extern int sysctl_tcp_rfc1337;
215 extern int sysctl_tcp_abort_on_overflow;
216 extern int sysctl_tcp_max_orphans;
217 extern int sysctl_tcp_fack;
218 extern int sysctl_tcp_reordering;
219 extern int sysctl_tcp_ecn;
220 extern int sysctl_tcp_dsack;
221 extern long sysctl_tcp_mem[3];
222 extern int sysctl_tcp_wmem[3];
223 extern int sysctl_tcp_rmem[3];
224 extern int sysctl_tcp_app_win;
225 extern int sysctl_tcp_adv_win_scale;
226 extern int sysctl_tcp_tw_reuse;
227 extern int sysctl_tcp_frto;
228 extern int sysctl_tcp_frto_response;
229 extern int sysctl_tcp_low_latency;
230 extern int sysctl_tcp_dma_copybreak;
231 extern int sysctl_tcp_nometrics_save;
232 extern int sysctl_tcp_moderate_rcvbuf;
233 extern int sysctl_tcp_tso_win_divisor;
234 extern int sysctl_tcp_abc;
235 extern int sysctl_tcp_mtu_probing;
236 extern int sysctl_tcp_base_mss;
237 extern int sysctl_tcp_workaround_signed_windows;
238 extern int sysctl_tcp_slow_start_after_idle;
239 extern int sysctl_tcp_max_ssthresh;
240 extern int sysctl_tcp_cookie_size;
241 extern int sysctl_tcp_thin_linear_timeouts;
242 extern int sysctl_tcp_thin_dupack;
244 extern atomic_long_t tcp_memory_allocated;
245 extern struct percpu_counter tcp_sockets_allocated;
246 extern int tcp_memory_pressure;
249 * The next routines deal with comparing 32 bit unsigned ints
250 * and worry about wraparound (automatic with unsigned arithmetic).
253 static inline int before(__u32 seq1, __u32 seq2)
255 return (__s32)(seq1-seq2) < 0;
257 #define after(seq2, seq1) before(seq1, seq2)
259 /* is s2<=s1<=s3 ? */
260 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
262 return seq3 - seq2 >= seq1 - seq2;
265 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
267 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
268 int orphans = percpu_counter_read_positive(ocp);
270 if (orphans << shift > sysctl_tcp_max_orphans) {
271 orphans = percpu_counter_sum_positive(ocp);
272 if (orphans << shift > sysctl_tcp_max_orphans)
276 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
277 atomic_long_read(&tcp_memory_allocated) > sysctl_tcp_mem[2])
282 /* syncookies: remember time of last synqueue overflow */
283 static inline void tcp_synq_overflow(struct sock *sk)
285 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
288 /* syncookies: no recent synqueue overflow on this listening socket? */
289 static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
291 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
292 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT);
295 extern struct proto tcp_prot;
297 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
298 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
299 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
300 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
301 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
303 extern void tcp_v4_err(struct sk_buff *skb, u32);
305 extern void tcp_shutdown (struct sock *sk, int how);
307 extern int tcp_v4_rcv(struct sk_buff *skb);
309 extern struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it);
310 extern void *tcp_v4_tw_get_peer(struct sock *sk);
311 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
312 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
314 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
315 size_t size, int flags);
316 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
317 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
318 struct tcphdr *th, unsigned len);
319 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
320 struct tcphdr *th, unsigned len);
321 extern void tcp_rcv_space_adjust(struct sock *sk);
322 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
323 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
324 extern void tcp_twsk_destructor(struct sock *sk);
325 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
326 struct pipe_inode_info *pipe, size_t len,
329 static inline void tcp_dec_quickack_mode(struct sock *sk,
330 const unsigned int pkts)
332 struct inet_connection_sock *icsk = inet_csk(sk);
334 if (icsk->icsk_ack.quick) {
335 if (pkts >= icsk->icsk_ack.quick) {
336 icsk->icsk_ack.quick = 0;
337 /* Leaving quickack mode we deflate ATO. */
338 icsk->icsk_ack.ato = TCP_ATO_MIN;
340 icsk->icsk_ack.quick -= pkts;
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);
366 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
367 struct request_sock *req,
368 struct request_sock **prev);
369 extern int tcp_child_process(struct sock *parent, struct sock *child,
370 struct sk_buff *skb);
371 extern int tcp_use_frto(struct sock *sk);
372 extern void tcp_enter_frto(struct sock *sk);
373 extern void tcp_enter_loss(struct sock *sk, int how);
374 extern void tcp_clear_retrans(struct tcp_sock *tp);
375 extern void tcp_update_metrics(struct sock *sk);
376 extern void tcp_close(struct sock *sk, long timeout);
377 extern unsigned int tcp_poll(struct file * file, struct socket *sock,
378 struct poll_table_struct *wait);
379 extern int tcp_getsockopt(struct sock *sk, int level, int optname,
380 char __user *optval, int __user *optlen);
381 extern int tcp_setsockopt(struct sock *sk, int level, int optname,
382 char __user *optval, unsigned int optlen);
383 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
384 char __user *optval, int __user *optlen);
385 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
386 char __user *optval, unsigned int optlen);
387 extern void tcp_set_keepalive(struct sock *sk, int val);
388 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
389 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
390 size_t len, int nonblock, int flags, int *addr_len);
391 extern void tcp_parse_options(struct sk_buff *skb,
392 struct tcp_options_received *opt_rx, u8 **hvpp,
394 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
397 * TCP v4 functions exported for the inet6 API
400 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
401 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
402 extern struct sock * tcp_create_openreq_child(struct sock *sk,
403 struct request_sock *req,
404 struct sk_buff *skb);
405 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
406 struct request_sock *req,
407 struct dst_entry *dst);
408 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
409 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
411 extern int tcp_connect(struct sock *sk);
412 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
413 struct request_sock *req,
414 struct request_values *rvp);
415 extern int tcp_disconnect(struct sock *sk, int flags);
418 /* From syncookies.c */
419 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
420 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
421 struct ip_options *opt);
422 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
425 extern __u32 cookie_init_timestamp(struct request_sock *req);
426 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *);
428 /* From net/ipv6/syncookies.c */
429 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
430 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
435 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
437 extern int tcp_may_send_now(struct sock *sk);
438 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
439 extern void tcp_retransmit_timer(struct sock *sk);
440 extern void tcp_xmit_retransmit_queue(struct sock *);
441 extern void tcp_simple_retransmit(struct sock *);
442 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
443 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
445 extern void tcp_send_probe0(struct sock *);
446 extern void tcp_send_partial(struct sock *);
447 extern int tcp_write_wakeup(struct sock *);
448 extern void tcp_send_fin(struct sock *sk);
449 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
450 extern int tcp_send_synack(struct sock *);
451 extern void tcp_push_one(struct sock *, unsigned int mss_now);
452 extern void tcp_send_ack(struct sock *sk);
453 extern void tcp_send_delayed_ack(struct sock *sk);
456 extern void tcp_cwnd_application_limited(struct sock *sk);
459 extern void tcp_init_xmit_timers(struct sock *);
460 static inline void tcp_clear_xmit_timers(struct sock *sk)
462 inet_csk_clear_xmit_timers(sk);
465 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
466 extern unsigned int tcp_current_mss(struct sock *sk);
468 /* Bound MSS / TSO packet size with the half of the window */
469 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
473 /* When peer uses tiny windows, there is no use in packetizing
474 * to sub-MSS pieces for the sake of SWS or making sure there
475 * are enough packets in the pipe for fast recovery.
477 * On the other hand, for extremely large MSS devices, handling
478 * smaller than MSS windows in this way does make sense.
480 if (tp->max_window >= 512)
481 cutoff = (tp->max_window >> 1);
483 cutoff = tp->max_window;
485 if (cutoff && pktsize > cutoff)
486 return max_t(int, cutoff, 68U - tp->tcp_header_len);
492 extern void tcp_get_info(struct sock *, struct tcp_info *);
494 /* Read 'sendfile()'-style from a TCP socket */
495 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
496 unsigned int, size_t);
497 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
498 sk_read_actor_t recv_actor);
500 extern void tcp_initialize_rcv_mss(struct sock *sk);
502 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
503 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
504 extern void tcp_mtup_init(struct sock *sk);
506 static inline void tcp_bound_rto(const struct sock *sk)
508 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
509 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
512 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
514 return (tp->srtt >> 3) + tp->rttvar;
517 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
519 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
520 ntohl(TCP_FLAG_ACK) |
524 static inline void tcp_fast_path_on(struct tcp_sock *tp)
526 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
529 static inline void tcp_fast_path_check(struct sock *sk)
531 struct tcp_sock *tp = tcp_sk(sk);
533 if (skb_queue_empty(&tp->out_of_order_queue) &&
535 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
537 tcp_fast_path_on(tp);
540 /* Compute the actual rto_min value */
541 static inline u32 tcp_rto_min(struct sock *sk)
543 struct dst_entry *dst = __sk_dst_get(sk);
544 u32 rto_min = TCP_RTO_MIN;
546 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
547 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
551 /* Compute the actual receive window we are currently advertising.
552 * Rcv_nxt can be after the window if our peer push more data
553 * than the offered window.
555 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
557 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
564 /* Choose a new window, without checks for shrinking, and without
565 * scaling applied to the result. The caller does these things
566 * if necessary. This is a "raw" window selection.
568 extern u32 __tcp_select_window(struct sock *sk);
570 /* TCP timestamps are only 32-bits, this causes a slight
571 * complication on 64-bit systems since we store a snapshot
572 * of jiffies in the buffer control blocks below. We decided
573 * to use only the low 32-bits of jiffies and hide the ugly
574 * casts with the following macro.
576 #define tcp_time_stamp ((__u32)(jiffies))
578 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
580 #define TCPHDR_FIN 0x01
581 #define TCPHDR_SYN 0x02
582 #define TCPHDR_RST 0x04
583 #define TCPHDR_PSH 0x08
584 #define TCPHDR_ACK 0x10
585 #define TCPHDR_URG 0x20
586 #define TCPHDR_ECE 0x40
587 #define TCPHDR_CWR 0x80
589 /* This is what the send packet queuing engine uses to pass
590 * TCP per-packet control information to the transmission code.
591 * We also store the host-order sequence numbers in here too.
592 * This is 44 bytes if IPV6 is enabled.
593 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
597 struct inet_skb_parm h4;
598 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
599 struct inet6_skb_parm h6;
601 } header; /* For incoming frames */
602 __u32 seq; /* Starting sequence number */
603 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
604 __u32 when; /* used to compute rtt's */
605 __u8 flags; /* TCP header flags. */
606 __u8 sacked; /* State flags for SACK/FACK. */
607 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
608 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
609 #define TCPCB_LOST 0x04 /* SKB is lost */
610 #define TCPCB_TAGBITS 0x07 /* All tag bits */
612 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
613 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
615 __u32 ack_seq; /* Sequence number ACK'd */
618 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
620 /* Due to TSO, an SKB can be composed of multiple actual
621 * packets. To keep these tracked properly, we use this.
623 static inline int tcp_skb_pcount(const struct sk_buff *skb)
625 return skb_shinfo(skb)->gso_segs;
628 /* This is valid iff tcp_skb_pcount() > 1. */
629 static inline int tcp_skb_mss(const struct sk_buff *skb)
631 return skb_shinfo(skb)->gso_size;
634 /* Events passed to congestion control interface */
636 CA_EVENT_TX_START, /* first transmit when no packets in flight */
637 CA_EVENT_CWND_RESTART, /* congestion window restart */
638 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
639 CA_EVENT_FRTO, /* fast recovery timeout */
640 CA_EVENT_LOSS, /* loss timeout */
641 CA_EVENT_FAST_ACK, /* in sequence ack */
642 CA_EVENT_SLOW_ACK, /* other ack */
646 * Interface for adding new TCP congestion control handlers
648 #define TCP_CA_NAME_MAX 16
649 #define TCP_CA_MAX 128
650 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
652 #define TCP_CONG_NON_RESTRICTED 0x1
653 #define TCP_CONG_RTT_STAMP 0x2
655 struct tcp_congestion_ops {
656 struct list_head list;
659 /* initialize private data (optional) */
660 void (*init)(struct sock *sk);
661 /* cleanup private data (optional) */
662 void (*release)(struct sock *sk);
664 /* return slow start threshold (required) */
665 u32 (*ssthresh)(struct sock *sk);
666 /* lower bound for congestion window (optional) */
667 u32 (*min_cwnd)(const struct sock *sk);
668 /* do new cwnd calculation (required) */
669 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
670 /* call before changing ca_state (optional) */
671 void (*set_state)(struct sock *sk, u8 new_state);
672 /* call when cwnd event occurs (optional) */
673 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
674 /* new value of cwnd after loss (optional) */
675 u32 (*undo_cwnd)(struct sock *sk);
676 /* hook for packet ack accounting (optional) */
677 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
678 /* get info for inet_diag (optional) */
679 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
681 char name[TCP_CA_NAME_MAX];
682 struct module *owner;
685 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
686 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
688 extern void tcp_init_congestion_control(struct sock *sk);
689 extern void tcp_cleanup_congestion_control(struct sock *sk);
690 extern int tcp_set_default_congestion_control(const char *name);
691 extern void tcp_get_default_congestion_control(char *name);
692 extern void tcp_get_available_congestion_control(char *buf, size_t len);
693 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
694 extern int tcp_set_allowed_congestion_control(char *allowed);
695 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
696 extern void tcp_slow_start(struct tcp_sock *tp);
697 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
699 extern struct tcp_congestion_ops tcp_init_congestion_ops;
700 extern u32 tcp_reno_ssthresh(struct sock *sk);
701 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
702 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
703 extern struct tcp_congestion_ops tcp_reno;
705 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
707 struct inet_connection_sock *icsk = inet_csk(sk);
709 if (icsk->icsk_ca_ops->set_state)
710 icsk->icsk_ca_ops->set_state(sk, ca_state);
711 icsk->icsk_ca_state = ca_state;
714 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
716 const struct inet_connection_sock *icsk = inet_csk(sk);
718 if (icsk->icsk_ca_ops->cwnd_event)
719 icsk->icsk_ca_ops->cwnd_event(sk, event);
722 /* These functions determine how the current flow behaves in respect of SACK
723 * handling. SACK is negotiated with the peer, and therefore it can vary
724 * between different flows.
726 * tcp_is_sack - SACK enabled
727 * tcp_is_reno - No SACK
728 * tcp_is_fack - FACK enabled, implies SACK enabled
730 static inline int tcp_is_sack(const struct tcp_sock *tp)
732 return tp->rx_opt.sack_ok;
735 static inline int tcp_is_reno(const struct tcp_sock *tp)
737 return !tcp_is_sack(tp);
740 static inline int tcp_is_fack(const struct tcp_sock *tp)
742 return tp->rx_opt.sack_ok & 2;
745 static inline void tcp_enable_fack(struct tcp_sock *tp)
747 tp->rx_opt.sack_ok |= 2;
750 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
752 return tp->sacked_out + tp->lost_out;
755 /* This determines how many packets are "in the network" to the best
756 * of our knowledge. In many cases it is conservative, but where
757 * detailed information is available from the receiver (via SACK
758 * blocks etc.) we can make more aggressive calculations.
760 * Use this for decisions involving congestion control, use just
761 * tp->packets_out to determine if the send queue is empty or not.
763 * Read this equation as:
765 * "Packets sent once on transmission queue" MINUS
766 * "Packets left network, but not honestly ACKed yet" PLUS
767 * "Packets fast retransmitted"
769 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
771 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
774 #define TCP_INFINITE_SSTHRESH 0x7fffffff
776 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
778 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
781 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
782 * The exception is rate halving phase, when cwnd is decreasing towards
785 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
787 const struct tcp_sock *tp = tcp_sk(sk);
788 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
789 return tp->snd_ssthresh;
791 return max(tp->snd_ssthresh,
792 ((tp->snd_cwnd >> 1) +
793 (tp->snd_cwnd >> 2)));
796 /* Use define here intentionally to get WARN_ON location shown at the caller */
797 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
800 * Convert RFC 3390 larger initial window into an equivalent number of packets.
801 * This is based on the numbers specified in RFC 5681, 3.1.
803 static inline u32 rfc3390_bytes_to_packets(const u32 smss)
805 return smss <= 1095 ? 4 : (smss > 2190 ? 2 : 3);
808 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
809 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
811 /* Slow start with delack produces 3 packets of burst, so that
812 * it is safe "de facto". This will be the default - same as
813 * the default reordering threshold - but if reordering increases,
814 * we must be able to allow cwnd to burst at least this much in order
815 * to not pull it back when holes are filled.
817 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
819 return tp->reordering;
822 /* Returns end sequence number of the receiver's advertised window */
823 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
825 return tp->snd_una + tp->snd_wnd;
827 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
829 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
830 const struct sk_buff *skb)
833 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
836 static inline void tcp_check_probe_timer(struct sock *sk)
838 struct tcp_sock *tp = tcp_sk(sk);
839 const struct inet_connection_sock *icsk = inet_csk(sk);
841 if (!tp->packets_out && !icsk->icsk_pending)
842 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
843 icsk->icsk_rto, TCP_RTO_MAX);
846 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
851 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
857 * Calculate(/check) TCP checksum
859 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
860 __be32 daddr, __wsum base)
862 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
865 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
867 return __skb_checksum_complete(skb);
870 static inline int tcp_checksum_complete(struct sk_buff *skb)
872 return !skb_csum_unnecessary(skb) &&
873 __tcp_checksum_complete(skb);
876 /* Prequeue for VJ style copy to user, combined with checksumming. */
878 static inline void tcp_prequeue_init(struct tcp_sock *tp)
880 tp->ucopy.task = NULL;
882 tp->ucopy.memory = 0;
883 skb_queue_head_init(&tp->ucopy.prequeue);
884 #ifdef CONFIG_NET_DMA
885 tp->ucopy.dma_chan = NULL;
886 tp->ucopy.wakeup = 0;
887 tp->ucopy.pinned_list = NULL;
888 tp->ucopy.dma_cookie = 0;
892 /* Packet is added to VJ-style prequeue for processing in process
893 * context, if a reader task is waiting. Apparently, this exciting
894 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
895 * failed somewhere. Latency? Burstiness? Well, at least now we will
896 * see, why it failed. 8)8) --ANK
898 * NOTE: is this not too big to inline?
900 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
902 struct tcp_sock *tp = tcp_sk(sk);
904 if (sysctl_tcp_low_latency || !tp->ucopy.task)
907 __skb_queue_tail(&tp->ucopy.prequeue, skb);
908 tp->ucopy.memory += skb->truesize;
909 if (tp->ucopy.memory > sk->sk_rcvbuf) {
910 struct sk_buff *skb1;
912 BUG_ON(sock_owned_by_user(sk));
914 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
915 sk_backlog_rcv(sk, skb1);
916 NET_INC_STATS_BH(sock_net(sk),
917 LINUX_MIB_TCPPREQUEUEDROPPED);
920 tp->ucopy.memory = 0;
921 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
922 wake_up_interruptible_sync_poll(sk_sleep(sk),
923 POLLIN | POLLRDNORM | POLLRDBAND);
924 if (!inet_csk_ack_scheduled(sk))
925 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
926 (3 * tcp_rto_min(sk)) / 4,
936 static const char *statename[]={
937 "Unused","Established","Syn Sent","Syn Recv",
938 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
939 "Close Wait","Last ACK","Listen","Closing"
942 extern void tcp_set_state(struct sock *sk, int state);
944 extern void tcp_done(struct sock *sk);
946 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
949 rx_opt->num_sacks = 0;
952 /* Determine a window scaling and initial window to offer. */
953 extern void tcp_select_initial_window(int __space, __u32 mss,
954 __u32 *rcv_wnd, __u32 *window_clamp,
955 int wscale_ok, __u8 *rcv_wscale,
958 static inline int tcp_win_from_space(int space)
960 return sysctl_tcp_adv_win_scale<=0 ?
961 (space>>(-sysctl_tcp_adv_win_scale)) :
962 space - (space>>sysctl_tcp_adv_win_scale);
965 /* Note: caller must be prepared to deal with negative returns */
966 static inline int tcp_space(const struct sock *sk)
968 return tcp_win_from_space(sk->sk_rcvbuf -
969 atomic_read(&sk->sk_rmem_alloc));
972 static inline int tcp_full_space(const struct sock *sk)
974 return tcp_win_from_space(sk->sk_rcvbuf);
977 static inline void tcp_openreq_init(struct request_sock *req,
978 struct tcp_options_received *rx_opt,
981 struct inet_request_sock *ireq = inet_rsk(req);
983 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
985 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
986 req->mss = rx_opt->mss_clamp;
987 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
988 ireq->tstamp_ok = rx_opt->tstamp_ok;
989 ireq->sack_ok = rx_opt->sack_ok;
990 ireq->snd_wscale = rx_opt->snd_wscale;
991 ireq->wscale_ok = rx_opt->wscale_ok;
994 ireq->rmt_port = tcp_hdr(skb)->source;
995 ireq->loc_port = tcp_hdr(skb)->dest;
998 extern void tcp_enter_memory_pressure(struct sock *sk);
1000 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1002 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1005 static inline int keepalive_time_when(const struct tcp_sock *tp)
1007 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1010 static inline int keepalive_probes(const struct tcp_sock *tp)
1012 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1015 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1017 const struct inet_connection_sock *icsk = &tp->inet_conn;
1019 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1020 tcp_time_stamp - tp->rcv_tstamp);
1023 static inline int tcp_fin_time(const struct sock *sk)
1025 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1026 const int rto = inet_csk(sk)->icsk_rto;
1028 if (fin_timeout < (rto << 2) - (rto >> 1))
1029 fin_timeout = (rto << 2) - (rto >> 1);
1034 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1037 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1039 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1042 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1043 * then following tcp messages have valid values. Ignore 0 value,
1044 * or else 'negative' tsval might forbid us to accept their packets.
1046 if (!rx_opt->ts_recent)
1051 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1054 if (tcp_paws_check(rx_opt, 0))
1057 /* RST segments are not recommended to carry timestamp,
1058 and, if they do, it is recommended to ignore PAWS because
1059 "their cleanup function should take precedence over timestamps."
1060 Certainly, it is mistake. It is necessary to understand the reasons
1061 of this constraint to relax it: if peer reboots, clock may go
1062 out-of-sync and half-open connections will not be reset.
1063 Actually, the problem would be not existing if all
1064 the implementations followed draft about maintaining clock
1065 via reboots. Linux-2.2 DOES NOT!
1067 However, we can relax time bounds for RST segments to MSL.
1069 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1074 #define TCP_CHECK_TIMER(sk) do { } while (0)
1076 static inline void tcp_mib_init(struct net *net)
1079 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1080 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1081 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1082 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1086 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1088 tp->lost_skb_hint = NULL;
1089 tp->scoreboard_skb_hint = NULL;
1092 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1094 tcp_clear_retrans_hints_partial(tp);
1095 tp->retransmit_skb_hint = NULL;
1101 /* - key database */
1102 struct tcp_md5sig_key {
1107 struct tcp4_md5sig_key {
1108 struct tcp_md5sig_key base;
1112 struct tcp6_md5sig_key {
1113 struct tcp_md5sig_key base;
1115 u32 scope_id; /* XXX */
1117 struct in6_addr addr;
1121 struct tcp_md5sig_info {
1122 struct tcp4_md5sig_key *keys4;
1123 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1124 struct tcp6_md5sig_key *keys6;
1132 /* - pseudo header */
1133 struct tcp4_pseudohdr {
1141 struct tcp6_pseudohdr {
1142 struct in6_addr saddr;
1143 struct in6_addr daddr;
1145 __be32 protocol; /* including padding */
1148 union tcp_md5sum_block {
1149 struct tcp4_pseudohdr ip4;
1150 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1151 struct tcp6_pseudohdr ip6;
1155 /* - pool: digest algorithm, hash description and scratch buffer */
1156 struct tcp_md5sig_pool {
1157 struct hash_desc md5_desc;
1158 union tcp_md5sum_block md5_blk;
1162 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1163 struct sock *sk, struct request_sock *req,
1164 struct sk_buff *skb);
1165 extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk,
1166 struct sock *addr_sk);
1167 extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey,
1169 extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr);
1171 #ifdef CONFIG_TCP_MD5SIG
1172 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1173 &(struct tcp_md5sig_key) { \
1174 .key = (twsk)->tw_md5_key, \
1175 .keylen = (twsk)->tw_md5_keylen, \
1178 #define tcp_twsk_md5_key(twsk) NULL
1181 extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *);
1182 extern void tcp_free_md5sig_pool(void);
1184 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1185 extern void tcp_put_md5sig_pool(void);
1187 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1188 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1189 unsigned header_len);
1190 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1191 struct tcp_md5sig_key *key);
1193 /* write queue abstraction */
1194 static inline void tcp_write_queue_purge(struct sock *sk)
1196 struct sk_buff *skb;
1198 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1199 sk_wmem_free_skb(sk, skb);
1201 tcp_clear_all_retrans_hints(tcp_sk(sk));
1204 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1206 return skb_peek(&sk->sk_write_queue);
1209 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1211 return skb_peek_tail(&sk->sk_write_queue);
1214 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1216 return skb_queue_next(&sk->sk_write_queue, skb);
1219 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb)
1221 return skb_queue_prev(&sk->sk_write_queue, skb);
1224 #define tcp_for_write_queue(skb, sk) \
1225 skb_queue_walk(&(sk)->sk_write_queue, skb)
1227 #define tcp_for_write_queue_from(skb, sk) \
1228 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1230 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1231 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1233 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1235 return sk->sk_send_head;
1238 static inline bool tcp_skb_is_last(const struct sock *sk,
1239 const struct sk_buff *skb)
1241 return skb_queue_is_last(&sk->sk_write_queue, skb);
1244 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1246 if (tcp_skb_is_last(sk, skb))
1247 sk->sk_send_head = NULL;
1249 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1252 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1254 if (sk->sk_send_head == skb_unlinked)
1255 sk->sk_send_head = NULL;
1258 static inline void tcp_init_send_head(struct sock *sk)
1260 sk->sk_send_head = NULL;
1263 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1265 __skb_queue_tail(&sk->sk_write_queue, skb);
1268 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1270 __tcp_add_write_queue_tail(sk, skb);
1272 /* Queue it, remembering where we must start sending. */
1273 if (sk->sk_send_head == NULL) {
1274 sk->sk_send_head = skb;
1276 if (tcp_sk(sk)->highest_sack == NULL)
1277 tcp_sk(sk)->highest_sack = skb;
1281 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1283 __skb_queue_head(&sk->sk_write_queue, skb);
1286 /* Insert buff after skb on the write queue of sk. */
1287 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1288 struct sk_buff *buff,
1291 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1294 /* Insert new before skb on the write queue of sk. */
1295 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1296 struct sk_buff *skb,
1299 __skb_queue_before(&sk->sk_write_queue, skb, new);
1301 if (sk->sk_send_head == skb)
1302 sk->sk_send_head = new;
1305 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1307 __skb_unlink(skb, &sk->sk_write_queue);
1310 static inline int tcp_write_queue_empty(struct sock *sk)
1312 return skb_queue_empty(&sk->sk_write_queue);
1315 static inline void tcp_push_pending_frames(struct sock *sk)
1317 if (tcp_send_head(sk)) {
1318 struct tcp_sock *tp = tcp_sk(sk);
1320 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1324 /* Start sequence of the highest skb with SACKed bit, valid only if
1325 * sacked > 0 or when the caller has ensured validity by itself.
1327 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1329 if (!tp->sacked_out)
1332 if (tp->highest_sack == NULL)
1335 return TCP_SKB_CB(tp->highest_sack)->seq;
1338 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1340 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1341 tcp_write_queue_next(sk, skb);
1344 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1346 return tcp_sk(sk)->highest_sack;
1349 static inline void tcp_highest_sack_reset(struct sock *sk)
1351 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1354 /* Called when old skb is about to be deleted (to be combined with new skb) */
1355 static inline void tcp_highest_sack_combine(struct sock *sk,
1356 struct sk_buff *old,
1357 struct sk_buff *new)
1359 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1360 tcp_sk(sk)->highest_sack = new;
1363 /* Determines whether this is a thin stream (which may suffer from
1364 * increased latency). Used to trigger latency-reducing mechanisms.
1366 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp)
1368 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1372 enum tcp_seq_states {
1373 TCP_SEQ_STATE_LISTENING,
1374 TCP_SEQ_STATE_OPENREQ,
1375 TCP_SEQ_STATE_ESTABLISHED,
1376 TCP_SEQ_STATE_TIME_WAIT,
1379 struct tcp_seq_afinfo {
1382 struct file_operations seq_fops;
1383 struct seq_operations seq_ops;
1386 struct tcp_iter_state {
1387 struct seq_net_private p;
1389 enum tcp_seq_states state;
1390 struct sock *syn_wait_sk;
1391 int bucket, offset, sbucket, num, uid;
1395 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1396 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1398 extern struct request_sock_ops tcp_request_sock_ops;
1399 extern struct request_sock_ops tcp6_request_sock_ops;
1401 extern void tcp_v4_destroy_sock(struct sock *sk);
1403 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1404 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1405 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1406 struct sk_buff *skb);
1407 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1408 struct sk_buff *skb);
1409 extern int tcp_gro_complete(struct sk_buff *skb);
1410 extern int tcp4_gro_complete(struct sk_buff *skb);
1412 #ifdef CONFIG_PROC_FS
1413 extern int tcp4_proc_init(void);
1414 extern void tcp4_proc_exit(void);
1417 /* TCP af-specific functions */
1418 struct tcp_sock_af_ops {
1419 #ifdef CONFIG_TCP_MD5SIG
1420 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1421 struct sock *addr_sk);
1422 int (*calc_md5_hash) (char *location,
1423 struct tcp_md5sig_key *md5,
1425 struct request_sock *req,
1426 struct sk_buff *skb);
1427 int (*md5_add) (struct sock *sk,
1428 struct sock *addr_sk,
1431 int (*md5_parse) (struct sock *sk,
1432 char __user *optval,
1437 struct tcp_request_sock_ops {
1438 #ifdef CONFIG_TCP_MD5SIG
1439 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1440 struct request_sock *req);
1441 int (*calc_md5_hash) (char *location,
1442 struct tcp_md5sig_key *md5,
1444 struct request_sock *req,
1445 struct sk_buff *skb);
1449 /* Using SHA1 for now, define some constants.
1451 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1452 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1453 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1455 extern int tcp_cookie_generator(u32 *bakery);
1458 * struct tcp_cookie_values - each socket needs extra space for the
1459 * cookies, together with (optional) space for any SYN data.
1461 * A tcp_sock contains a pointer to the current value, and this is
1462 * cloned to the tcp_timewait_sock.
1464 * @cookie_pair: variable data from the option exchange.
1466 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1467 * indicates default (sysctl_tcp_cookie_size).
1468 * After cookie sent, remembers size of cookie.
1469 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1471 * @s_data_desired: user specified tcpct_s_data_desired. When the
1472 * constant payload is specified (@s_data_constant),
1473 * holds its length instead.
1474 * Range 0 to TCP_MSS_DESIRED.
1476 * @s_data_payload: constant data that is to be included in the
1477 * payload of SYN or SYNACK segments when the
1478 * cookie option is present.
1480 struct tcp_cookie_values {
1482 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE];
1483 u8 cookie_pair_size;
1485 u16 s_data_desired:11,
1490 u8 s_data_payload[0];
1493 static inline void tcp_cookie_values_release(struct kref *kref)
1495 kfree(container_of(kref, struct tcp_cookie_values, kref));
1498 /* The length of constant payload data. Note that s_data_desired is
1499 * overloaded, depending on s_data_constant: either the length of constant
1500 * data (returned here) or the limit on variable data.
1502 static inline int tcp_s_data_size(const struct tcp_sock *tp)
1504 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1505 ? tp->cookie_values->s_data_desired
1510 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1512 * As tcp_request_sock has already been extended in other places, the
1513 * only remaining method is to pass stack values along as function
1514 * parameters. These parameters are not needed after sending SYNACK.
1516 * @cookie_bakery: cryptographic secret and message workspace.
1518 * @cookie_plus: bytes in authenticator/cookie option, copied from
1519 * struct tcp_options_received (above).
1521 struct tcp_extend_values {
1522 struct request_values rv;
1523 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS];
1529 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1531 return (struct tcp_extend_values *)rvp;
1534 extern void tcp_v4_init(void);
1535 extern void tcp_init(void);