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.
21 #define FASTRETRANS_DEBUG 1
23 #include <linux/list.h>
24 #include <linux/tcp.h>
25 #include <linux/slab.h>
26 #include <linux/cache.h>
27 #include <linux/percpu.h>
28 #include <linux/skbuff.h>
29 #include <linux/dmaengine.h>
30 #include <linux/crypto.h>
31 #include <linux/cryptohash.h>
32 #include <linux/kref.h>
34 #include <net/inet_connection_sock.h>
35 #include <net/inet_timewait_sock.h>
36 #include <net/inet_hashtables.h>
37 #include <net/checksum.h>
38 #include <net/request_sock.h>
42 #include <net/tcp_states.h>
43 #include <net/inet_ecn.h>
46 #include <linux/seq_file.h>
48 extern struct inet_hashinfo tcp_hashinfo;
50 extern struct percpu_counter tcp_orphan_count;
51 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
53 #define MAX_TCP_HEADER (128 + MAX_HEADER)
54 #define MAX_TCP_OPTION_SPACE 40
55 #define TCP_MIN_SND_MSS 48
56 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
62 #define MAX_TCP_WINDOW 32767U
64 /* Offer an initial receive window of 10 mss. */
65 #define TCP_DEFAULT_INIT_RCVWND 10
67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68 #define TCP_MIN_MSS 88U
70 /* The least MTU to use for probing */
71 #define TCP_BASE_MSS 512
73 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
74 #define TCP_FASTRETRANS_THRESH 3
76 /* Maximal reordering. */
77 #define TCP_MAX_REORDERING 127
79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
80 #define TCP_MAX_QUICKACKS 16U
83 #define TCP_URG_VALID 0x0100
84 #define TCP_URG_NOTYET 0x0200
85 #define TCP_URG_READ 0x0400
87 #define TCP_RETR1 3 /*
88 * This is how many retries it does before it
89 * tries to figure out if the gateway is
90 * down. Minimal RFC value is 3; it corresponds
91 * to ~3sec-8min depending on RTO.
94 #define TCP_RETR2 15 /*
95 * This should take at least
96 * 90 minutes to time out.
97 * RFC1122 says that the limit is 100 sec.
98 * 15 is ~13-30min depending on RTO.
101 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
102 * connection: ~180sec is RFC minimum */
104 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
105 * connection: ~180sec is RFC minimum */
107 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
108 * state, about 60 seconds */
109 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
110 /* BSD style FIN_WAIT2 deadlock breaker.
111 * It used to be 3min, new value is 60sec,
112 * to combine FIN-WAIT-2 timeout with
116 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
118 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
119 #define TCP_ATO_MIN ((unsigned)(HZ/25))
121 #define TCP_DELACK_MIN 4U
122 #define TCP_ATO_MIN 4U
124 #define TCP_RTO_MAX ((unsigned)(120*HZ))
125 #define TCP_RTO_MIN ((unsigned)(HZ/5))
126 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC2988bis initial RTO value */
127 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
128 * used as a fallback RTO for the
129 * initial data transmission if no
130 * valid RTT sample has been acquired,
131 * most likely due to retrans in 3WHS.
134 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
135 * for local resources.
138 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
139 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
140 #define TCP_KEEPALIVE_INTVL (75*HZ)
142 #define MAX_TCP_KEEPIDLE 32767
143 #define MAX_TCP_KEEPINTVL 32767
144 #define MAX_TCP_KEEPCNT 127
145 #define MAX_TCP_SYNCNT 127
147 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
149 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
150 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
151 * after this time. It should be equal
152 * (or greater than) TCP_TIMEWAIT_LEN
153 * to provide reliability equal to one
154 * provided by timewait state.
156 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
157 * timestamps. It must be less than
158 * minimal timewait lifetime.
164 #define TCPOPT_NOP 1 /* Padding */
165 #define TCPOPT_EOL 0 /* End of options */
166 #define TCPOPT_MSS 2 /* Segment size negotiating */
167 #define TCPOPT_WINDOW 3 /* Window scaling */
168 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
169 #define TCPOPT_SACK 5 /* SACK Block */
170 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
171 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
172 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
178 #define TCPOLEN_MSS 4
179 #define TCPOLEN_WINDOW 3
180 #define TCPOLEN_SACK_PERM 2
181 #define TCPOLEN_TIMESTAMP 10
182 #define TCPOLEN_MD5SIG 18
183 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
184 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
185 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
186 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
188 /* But this is what stacks really send out. */
189 #define TCPOLEN_TSTAMP_ALIGNED 12
190 #define TCPOLEN_WSCALE_ALIGNED 4
191 #define TCPOLEN_SACKPERM_ALIGNED 4
192 #define TCPOLEN_SACK_BASE 2
193 #define TCPOLEN_SACK_BASE_ALIGNED 4
194 #define TCPOLEN_SACK_PERBLOCK 8
195 #define TCPOLEN_MD5SIG_ALIGNED 20
196 #define TCPOLEN_MSS_ALIGNED 4
198 /* Flags in tp->nonagle */
199 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
200 #define TCP_NAGLE_CORK 2 /* Socket is corked */
201 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
203 /* TCP thin-stream limits */
204 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
206 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
207 #define TCP_INIT_CWND 10
209 extern struct inet_timewait_death_row tcp_death_row;
211 /* sysctl variables for tcp */
212 extern int sysctl_tcp_timestamps;
213 extern int sysctl_tcp_window_scaling;
214 extern int sysctl_tcp_sack;
215 extern int sysctl_tcp_fin_timeout;
216 extern int sysctl_tcp_keepalive_time;
217 extern int sysctl_tcp_keepalive_probes;
218 extern int sysctl_tcp_keepalive_intvl;
219 extern int sysctl_tcp_syn_retries;
220 extern int sysctl_tcp_synack_retries;
221 extern int sysctl_tcp_retries1;
222 extern int sysctl_tcp_retries2;
223 extern int sysctl_tcp_orphan_retries;
224 extern int sysctl_tcp_syncookies;
225 extern int sysctl_tcp_retrans_collapse;
226 extern int sysctl_tcp_stdurg;
227 extern int sysctl_tcp_rfc1337;
228 extern int sysctl_tcp_abort_on_overflow;
229 extern int sysctl_tcp_max_orphans;
230 extern int sysctl_tcp_fack;
231 extern int sysctl_tcp_reordering;
232 extern int sysctl_tcp_ecn;
233 extern int sysctl_tcp_dsack;
234 extern long sysctl_tcp_mem[3];
235 extern int sysctl_tcp_wmem[3];
236 extern int sysctl_tcp_rmem[3];
237 extern int sysctl_tcp_app_win;
238 extern int sysctl_tcp_adv_win_scale;
239 extern int sysctl_tcp_tw_reuse;
240 extern int sysctl_tcp_frto;
241 extern int sysctl_tcp_frto_response;
242 extern int sysctl_tcp_low_latency;
243 extern int sysctl_tcp_dma_copybreak;
244 extern int sysctl_tcp_nometrics_save;
245 extern int sysctl_tcp_moderate_rcvbuf;
246 extern int sysctl_tcp_tso_win_divisor;
247 extern int sysctl_tcp_abc;
248 extern int sysctl_tcp_mtu_probing;
249 extern int sysctl_tcp_base_mss;
250 extern int sysctl_tcp_workaround_signed_windows;
251 extern int sysctl_tcp_slow_start_after_idle;
252 extern int sysctl_tcp_max_ssthresh;
253 extern int sysctl_tcp_cookie_size;
254 extern int sysctl_tcp_thin_linear_timeouts;
255 extern int sysctl_tcp_thin_dupack;
256 extern int sysctl_tcp_challenge_ack_limit;
258 extern atomic_long_t tcp_memory_allocated;
259 extern struct percpu_counter tcp_sockets_allocated;
260 extern int tcp_memory_pressure;
263 * The next routines deal with comparing 32 bit unsigned ints
264 * and worry about wraparound (automatic with unsigned arithmetic).
267 static inline int before(__u32 seq1, __u32 seq2)
269 return (__s32)(seq1-seq2) < 0;
271 #define after(seq2, seq1) before(seq1, seq2)
273 /* is s2<=s1<=s3 ? */
274 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
276 return seq3 - seq2 >= seq1 - seq2;
279 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
281 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
282 int orphans = percpu_counter_read_positive(ocp);
284 if (orphans << shift > sysctl_tcp_max_orphans) {
285 orphans = percpu_counter_sum_positive(ocp);
286 if (orphans << shift > sysctl_tcp_max_orphans)
290 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
291 atomic_long_read(&tcp_memory_allocated) > sysctl_tcp_mem[2])
296 /* syncookies: remember time of last synqueue overflow */
297 static inline void tcp_synq_overflow(struct sock *sk)
299 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
302 /* syncookies: no recent synqueue overflow on this listening socket? */
303 static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
305 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
306 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
309 extern struct proto tcp_prot;
311 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
312 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
313 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
314 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
315 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
317 extern void tcp_v4_err(struct sk_buff *skb, u32);
319 extern void tcp_shutdown (struct sock *sk, int how);
321 extern int tcp_v4_rcv(struct sk_buff *skb);
323 extern struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it);
324 extern void *tcp_v4_tw_get_peer(struct sock *sk);
325 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
326 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
328 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
329 size_t size, int flags);
330 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
331 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
332 const struct tcphdr *th, unsigned int len);
333 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
334 const struct tcphdr *th, unsigned int len);
335 extern void tcp_rcv_space_adjust(struct sock *sk);
336 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
337 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
338 extern void tcp_twsk_destructor(struct sock *sk);
339 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
340 struct pipe_inode_info *pipe, size_t len,
343 static inline void tcp_dec_quickack_mode(struct sock *sk,
344 const unsigned int pkts)
346 struct inet_connection_sock *icsk = inet_csk(sk);
348 if (icsk->icsk_ack.quick) {
349 if (pkts >= icsk->icsk_ack.quick) {
350 icsk->icsk_ack.quick = 0;
351 /* Leaving quickack mode we deflate ATO. */
352 icsk->icsk_ack.ato = TCP_ATO_MIN;
354 icsk->icsk_ack.quick -= pkts;
359 #define TCP_ECN_QUEUE_CWR 2
360 #define TCP_ECN_DEMAND_CWR 4
361 #define TCP_ECN_SEEN 8
371 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
373 const struct tcphdr *th);
374 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
375 struct request_sock *req,
376 struct request_sock **prev);
377 extern int tcp_child_process(struct sock *parent, struct sock *child,
378 struct sk_buff *skb);
379 extern int tcp_use_frto(struct sock *sk);
380 extern void tcp_enter_frto(struct sock *sk);
381 extern void tcp_enter_loss(struct sock *sk, int how);
382 extern void tcp_clear_retrans(struct tcp_sock *tp);
383 extern void tcp_update_metrics(struct sock *sk);
384 extern void tcp_close(struct sock *sk, long timeout);
385 extern unsigned int tcp_poll(struct file * file, struct socket *sock,
386 struct poll_table_struct *wait);
387 extern int tcp_getsockopt(struct sock *sk, int level, int optname,
388 char __user *optval, int __user *optlen);
389 extern int tcp_setsockopt(struct sock *sk, int level, int optname,
390 char __user *optval, unsigned int optlen);
391 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
392 char __user *optval, int __user *optlen);
393 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
394 char __user *optval, unsigned int optlen);
395 extern void tcp_set_keepalive(struct sock *sk, int val);
396 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
397 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
398 size_t len, int nonblock, int flags, int *addr_len);
399 extern void tcp_parse_options(const struct sk_buff *skb,
400 struct tcp_options_received *opt_rx, const u8 **hvpp,
402 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
405 * TCP v4 functions exported for the inet6 API
408 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
409 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
410 extern struct sock * tcp_create_openreq_child(struct sock *sk,
411 struct request_sock *req,
412 struct sk_buff *skb);
413 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
414 struct request_sock *req,
415 struct dst_entry *dst);
416 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
417 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
419 extern int tcp_connect(struct sock *sk);
420 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
421 struct request_sock *req,
422 struct request_values *rvp);
423 extern int tcp_disconnect(struct sock *sk, int flags);
426 /* From syncookies.c */
427 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
428 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
429 struct ip_options *opt);
430 #ifdef CONFIG_SYN_COOKIES
431 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
434 static inline __u32 cookie_v4_init_sequence(struct sock *sk,
442 extern __u32 cookie_init_timestamp(struct request_sock *req);
443 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *);
445 /* From net/ipv6/syncookies.c */
446 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
447 #ifdef CONFIG_SYN_COOKIES
448 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
451 static inline __u32 cookie_v6_init_sequence(struct sock *sk,
460 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
462 extern int tcp_may_send_now(struct sock *sk);
463 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
464 extern void tcp_retransmit_timer(struct sock *sk);
465 extern void tcp_xmit_retransmit_queue(struct sock *);
466 extern void tcp_simple_retransmit(struct sock *);
467 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
468 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
470 extern void tcp_send_probe0(struct sock *);
471 extern void tcp_send_partial(struct sock *);
472 extern int tcp_write_wakeup(struct sock *);
473 extern void tcp_send_fin(struct sock *sk);
474 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
475 extern int tcp_send_synack(struct sock *);
476 extern int tcp_syn_flood_action(struct sock *sk,
477 const struct sk_buff *skb,
479 extern void tcp_push_one(struct sock *, unsigned int mss_now);
480 extern void tcp_send_ack(struct sock *sk);
481 extern void tcp_send_delayed_ack(struct sock *sk);
484 extern void tcp_cwnd_application_limited(struct sock *sk);
487 extern void tcp_init_xmit_timers(struct sock *);
488 static inline void tcp_clear_xmit_timers(struct sock *sk)
490 inet_csk_clear_xmit_timers(sk);
493 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
494 extern unsigned int tcp_current_mss(struct sock *sk);
496 /* Bound MSS / TSO packet size with the half of the window */
497 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
501 /* When peer uses tiny windows, there is no use in packetizing
502 * to sub-MSS pieces for the sake of SWS or making sure there
503 * are enough packets in the pipe for fast recovery.
505 * On the other hand, for extremely large MSS devices, handling
506 * smaller than MSS windows in this way does make sense.
508 if (tp->max_window >= 512)
509 cutoff = (tp->max_window >> 1);
511 cutoff = tp->max_window;
513 if (cutoff && pktsize > cutoff)
514 return max_t(int, cutoff, 68U - tp->tcp_header_len);
520 extern void tcp_get_info(const struct sock *, struct tcp_info *);
522 /* Read 'sendfile()'-style from a TCP socket */
523 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
524 unsigned int, size_t);
525 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
526 sk_read_actor_t recv_actor);
528 extern void tcp_initialize_rcv_mss(struct sock *sk);
530 extern int tcp_mtu_to_mss(const struct sock *sk, int pmtu);
531 extern int tcp_mss_to_mtu(const struct sock *sk, int mss);
532 extern void tcp_mtup_init(struct sock *sk);
533 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt);
535 static inline void tcp_bound_rto(const struct sock *sk)
537 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
538 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
541 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
543 return (tp->srtt >> 3) + tp->rttvar;
546 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
548 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
549 ntohl(TCP_FLAG_ACK) |
553 static inline void tcp_fast_path_on(struct tcp_sock *tp)
555 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
558 static inline void tcp_fast_path_check(struct sock *sk)
560 struct tcp_sock *tp = tcp_sk(sk);
562 if (skb_queue_empty(&tp->out_of_order_queue) &&
564 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
566 tcp_fast_path_on(tp);
569 /* Compute the actual rto_min value */
570 static inline u32 tcp_rto_min(struct sock *sk)
572 const struct dst_entry *dst = __sk_dst_get(sk);
573 u32 rto_min = TCP_RTO_MIN;
575 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
576 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
580 /* Compute the actual receive window we are currently advertising.
581 * Rcv_nxt can be after the window if our peer push more data
582 * than the offered window.
584 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
586 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
593 /* Choose a new window, without checks for shrinking, and without
594 * scaling applied to the result. The caller does these things
595 * if necessary. This is a "raw" window selection.
597 extern u32 __tcp_select_window(struct sock *sk);
599 /* TCP timestamps are only 32-bits, this causes a slight
600 * complication on 64-bit systems since we store a snapshot
601 * of jiffies in the buffer control blocks below. We decided
602 * to use only the low 32-bits of jiffies and hide the ugly
603 * casts with the following macro.
605 #define tcp_time_stamp ((__u32)(jiffies))
607 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
609 #define TCPHDR_FIN 0x01
610 #define TCPHDR_SYN 0x02
611 #define TCPHDR_RST 0x04
612 #define TCPHDR_PSH 0x08
613 #define TCPHDR_ACK 0x10
614 #define TCPHDR_URG 0x20
615 #define TCPHDR_ECE 0x40
616 #define TCPHDR_CWR 0x80
618 /* This is what the send packet queuing engine uses to pass
619 * TCP per-packet control information to the transmission code.
620 * We also store the host-order sequence numbers in here too.
621 * This is 44 bytes if IPV6 is enabled.
622 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
626 struct inet_skb_parm h4;
627 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
628 struct inet6_skb_parm h6;
630 } header; /* For incoming frames */
631 __u32 seq; /* Starting sequence number */
632 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
633 __u32 when; /* used to compute rtt's */
634 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
635 __u8 sacked; /* State flags for SACK/FACK. */
636 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
637 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
638 #define TCPCB_LOST 0x04 /* SKB is lost */
639 #define TCPCB_TAGBITS 0x07 /* All tag bits */
640 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
642 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
643 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
645 __u32 ack_seq; /* Sequence number ACK'd */
648 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
650 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
652 * If we receive a SYN packet with these bits set, it means a network is
653 * playing bad games with TOS bits. In order to avoid possible false congestion
654 * notifications, we disable TCP ECN negociation.
657 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb)
659 const struct tcphdr *th = tcp_hdr(skb);
661 if (sysctl_tcp_ecn && th->ece && th->cwr &&
662 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield))
663 inet_rsk(req)->ecn_ok = 1;
666 /* Due to TSO, an SKB can be composed of multiple actual
667 * packets. To keep these tracked properly, we use this.
669 static inline int tcp_skb_pcount(const struct sk_buff *skb)
671 return skb_shinfo(skb)->gso_segs;
674 /* This is valid iff tcp_skb_pcount() > 1. */
675 static inline int tcp_skb_mss(const struct sk_buff *skb)
677 return skb_shinfo(skb)->gso_size;
680 /* Events passed to congestion control interface */
682 CA_EVENT_TX_START, /* first transmit when no packets in flight */
683 CA_EVENT_CWND_RESTART, /* congestion window restart */
684 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
685 CA_EVENT_FRTO, /* fast recovery timeout */
686 CA_EVENT_LOSS, /* loss timeout */
687 CA_EVENT_FAST_ACK, /* in sequence ack */
688 CA_EVENT_SLOW_ACK, /* other ack */
692 * Interface for adding new TCP congestion control handlers
694 #define TCP_CA_NAME_MAX 16
695 #define TCP_CA_MAX 128
696 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
698 #define TCP_CONG_NON_RESTRICTED 0x1
699 #define TCP_CONG_RTT_STAMP 0x2
701 struct tcp_congestion_ops {
702 struct list_head list;
705 /* initialize private data (optional) */
706 void (*init)(struct sock *sk);
707 /* cleanup private data (optional) */
708 void (*release)(struct sock *sk);
710 /* return slow start threshold (required) */
711 u32 (*ssthresh)(struct sock *sk);
712 /* lower bound for congestion window (optional) */
713 u32 (*min_cwnd)(const struct sock *sk);
714 /* do new cwnd calculation (required) */
715 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
716 /* call before changing ca_state (optional) */
717 void (*set_state)(struct sock *sk, u8 new_state);
718 /* call when cwnd event occurs (optional) */
719 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
720 /* new value of cwnd after loss (optional) */
721 u32 (*undo_cwnd)(struct sock *sk);
722 /* hook for packet ack accounting (optional) */
723 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
724 /* get info for inet_diag (optional) */
725 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
727 char name[TCP_CA_NAME_MAX];
728 struct module *owner;
731 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
732 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
734 extern void tcp_init_congestion_control(struct sock *sk);
735 extern void tcp_cleanup_congestion_control(struct sock *sk);
736 extern int tcp_set_default_congestion_control(const char *name);
737 extern void tcp_get_default_congestion_control(char *name);
738 extern void tcp_get_available_congestion_control(char *buf, size_t len);
739 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
740 extern int tcp_set_allowed_congestion_control(char *allowed);
741 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
742 extern void tcp_slow_start(struct tcp_sock *tp);
743 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
745 extern struct tcp_congestion_ops tcp_init_congestion_ops;
746 extern u32 tcp_reno_ssthresh(struct sock *sk);
747 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
748 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
749 extern struct tcp_congestion_ops tcp_reno;
751 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
753 struct inet_connection_sock *icsk = inet_csk(sk);
755 if (icsk->icsk_ca_ops->set_state)
756 icsk->icsk_ca_ops->set_state(sk, ca_state);
757 icsk->icsk_ca_state = ca_state;
760 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
762 const struct inet_connection_sock *icsk = inet_csk(sk);
764 if (icsk->icsk_ca_ops->cwnd_event)
765 icsk->icsk_ca_ops->cwnd_event(sk, event);
768 /* These functions determine how the current flow behaves in respect of SACK
769 * handling. SACK is negotiated with the peer, and therefore it can vary
770 * between different flows.
772 * tcp_is_sack - SACK enabled
773 * tcp_is_reno - No SACK
774 * tcp_is_fack - FACK enabled, implies SACK enabled
776 static inline int tcp_is_sack(const struct tcp_sock *tp)
778 return tp->rx_opt.sack_ok;
781 static inline int tcp_is_reno(const struct tcp_sock *tp)
783 return !tcp_is_sack(tp);
786 static inline int tcp_is_fack(const struct tcp_sock *tp)
788 return tp->rx_opt.sack_ok & 2;
791 static inline void tcp_enable_fack(struct tcp_sock *tp)
793 tp->rx_opt.sack_ok |= 2;
796 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
798 return tp->sacked_out + tp->lost_out;
801 /* This determines how many packets are "in the network" to the best
802 * of our knowledge. In many cases it is conservative, but where
803 * detailed information is available from the receiver (via SACK
804 * blocks etc.) we can make more aggressive calculations.
806 * Use this for decisions involving congestion control, use just
807 * tp->packets_out to determine if the send queue is empty or not.
809 * Read this equation as:
811 * "Packets sent once on transmission queue" MINUS
812 * "Packets left network, but not honestly ACKed yet" PLUS
813 * "Packets fast retransmitted"
815 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
817 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
820 #define TCP_INFINITE_SSTHRESH 0x7fffffff
822 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
824 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
827 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
828 * The exception is rate halving phase, when cwnd is decreasing towards
831 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
833 const struct tcp_sock *tp = tcp_sk(sk);
835 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
836 return tp->snd_ssthresh;
838 return max(tp->snd_ssthresh,
839 ((tp->snd_cwnd >> 1) +
840 (tp->snd_cwnd >> 2)));
843 /* Use define here intentionally to get WARN_ON location shown at the caller */
844 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
846 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
847 extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
849 /* Slow start with delack produces 3 packets of burst, so that
850 * it is safe "de facto". This will be the default - same as
851 * the default reordering threshold - but if reordering increases,
852 * we must be able to allow cwnd to burst at least this much in order
853 * to not pull it back when holes are filled.
855 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
857 return tp->reordering;
860 /* Returns end sequence number of the receiver's advertised window */
861 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
863 return tp->snd_una + tp->snd_wnd;
865 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
867 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
868 const struct sk_buff *skb)
871 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
874 static inline void tcp_check_probe_timer(struct sock *sk)
876 const struct tcp_sock *tp = tcp_sk(sk);
877 const struct inet_connection_sock *icsk = inet_csk(sk);
879 if (!tp->packets_out && !icsk->icsk_pending)
880 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
881 icsk->icsk_rto, TCP_RTO_MAX);
884 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
889 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
895 * Calculate(/check) TCP checksum
897 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
898 __be32 daddr, __wsum base)
900 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
903 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
905 return __skb_checksum_complete(skb);
908 static inline int tcp_checksum_complete(struct sk_buff *skb)
910 return !skb_csum_unnecessary(skb) &&
911 __tcp_checksum_complete(skb);
914 /* Prequeue for VJ style copy to user, combined with checksumming. */
916 static inline void tcp_prequeue_init(struct tcp_sock *tp)
918 tp->ucopy.task = NULL;
920 tp->ucopy.memory = 0;
921 skb_queue_head_init(&tp->ucopy.prequeue);
922 #ifdef CONFIG_NET_DMA
923 tp->ucopy.dma_chan = NULL;
924 tp->ucopy.wakeup = 0;
925 tp->ucopy.pinned_list = NULL;
926 tp->ucopy.dma_cookie = 0;
930 /* Packet is added to VJ-style prequeue for processing in process
931 * context, if a reader task is waiting. Apparently, this exciting
932 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
933 * failed somewhere. Latency? Burstiness? Well, at least now we will
934 * see, why it failed. 8)8) --ANK
936 * NOTE: is this not too big to inline?
938 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
940 struct tcp_sock *tp = tcp_sk(sk);
942 if (sysctl_tcp_low_latency || !tp->ucopy.task)
946 __skb_queue_tail(&tp->ucopy.prequeue, skb);
947 tp->ucopy.memory += skb->truesize;
948 if (tp->ucopy.memory > sk->sk_rcvbuf) {
949 struct sk_buff *skb1;
951 BUG_ON(sock_owned_by_user(sk));
953 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
954 sk_backlog_rcv(sk, skb1);
955 NET_INC_STATS_BH(sock_net(sk),
956 LINUX_MIB_TCPPREQUEUEDROPPED);
959 tp->ucopy.memory = 0;
960 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
961 wake_up_interruptible_sync_poll(sk_sleep(sk),
962 POLLIN | POLLRDNORM | POLLRDBAND);
963 if (!inet_csk_ack_scheduled(sk))
964 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
965 (3 * tcp_rto_min(sk)) / 4,
971 int tcp_filter(struct sock *sk, struct sk_buff *skb);
976 static const char *statename[]={
977 "Unused","Established","Syn Sent","Syn Recv",
978 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
979 "Close Wait","Last ACK","Listen","Closing"
982 extern void tcp_set_state(struct sock *sk, int state);
984 extern void tcp_done(struct sock *sk);
986 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
989 rx_opt->num_sacks = 0;
992 /* Determine a window scaling and initial window to offer. */
993 extern void tcp_select_initial_window(int __space, __u32 mss,
994 __u32 *rcv_wnd, __u32 *window_clamp,
995 int wscale_ok, __u8 *rcv_wscale,
998 static inline int tcp_win_from_space(int space)
1000 return sysctl_tcp_adv_win_scale<=0 ?
1001 (space>>(-sysctl_tcp_adv_win_scale)) :
1002 space - (space>>sysctl_tcp_adv_win_scale);
1005 /* Note: caller must be prepared to deal with negative returns */
1006 static inline int tcp_space(const struct sock *sk)
1008 return tcp_win_from_space(sk->sk_rcvbuf -
1009 atomic_read(&sk->sk_rmem_alloc));
1012 static inline int tcp_full_space(const struct sock *sk)
1014 return tcp_win_from_space(sk->sk_rcvbuf);
1017 static inline void tcp_openreq_init(struct request_sock *req,
1018 struct tcp_options_received *rx_opt,
1019 struct sk_buff *skb)
1021 struct inet_request_sock *ireq = inet_rsk(req);
1023 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1025 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1026 req->mss = rx_opt->mss_clamp;
1027 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1028 ireq->tstamp_ok = rx_opt->tstamp_ok;
1029 ireq->sack_ok = rx_opt->sack_ok;
1030 ireq->snd_wscale = rx_opt->snd_wscale;
1031 ireq->wscale_ok = rx_opt->wscale_ok;
1034 ireq->rmt_port = tcp_hdr(skb)->source;
1035 ireq->loc_port = tcp_hdr(skb)->dest;
1038 extern void tcp_enter_memory_pressure(struct sock *sk);
1040 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1042 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1045 static inline int keepalive_time_when(const struct tcp_sock *tp)
1047 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1050 static inline int keepalive_probes(const struct tcp_sock *tp)
1052 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1055 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1057 const struct inet_connection_sock *icsk = &tp->inet_conn;
1059 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1060 tcp_time_stamp - tp->rcv_tstamp);
1063 static inline int tcp_fin_time(const struct sock *sk)
1065 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1066 const int rto = inet_csk(sk)->icsk_rto;
1068 if (fin_timeout < (rto << 2) - (rto >> 1))
1069 fin_timeout = (rto << 2) - (rto >> 1);
1074 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1077 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1079 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1082 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1083 * then following tcp messages have valid values. Ignore 0 value,
1084 * or else 'negative' tsval might forbid us to accept their packets.
1086 if (!rx_opt->ts_recent)
1091 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1094 if (tcp_paws_check(rx_opt, 0))
1097 /* RST segments are not recommended to carry timestamp,
1098 and, if they do, it is recommended to ignore PAWS because
1099 "their cleanup function should take precedence over timestamps."
1100 Certainly, it is mistake. It is necessary to understand the reasons
1101 of this constraint to relax it: if peer reboots, clock may go
1102 out-of-sync and half-open connections will not be reset.
1103 Actually, the problem would be not existing if all
1104 the implementations followed draft about maintaining clock
1105 via reboots. Linux-2.2 DOES NOT!
1107 However, we can relax time bounds for RST segments to MSL.
1109 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1114 static inline void tcp_mib_init(struct net *net)
1117 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1118 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1119 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1120 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1124 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1126 tp->lost_skb_hint = NULL;
1127 tp->scoreboard_skb_hint = NULL;
1130 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1132 tcp_clear_retrans_hints_partial(tp);
1133 tp->retransmit_skb_hint = NULL;
1139 /* - key database */
1140 struct tcp_md5sig_key {
1145 struct tcp4_md5sig_key {
1146 struct tcp_md5sig_key base;
1150 struct tcp6_md5sig_key {
1151 struct tcp_md5sig_key base;
1153 u32 scope_id; /* XXX */
1155 struct in6_addr addr;
1159 struct tcp_md5sig_info {
1160 struct tcp4_md5sig_key *keys4;
1161 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1162 struct tcp6_md5sig_key *keys6;
1170 /* - pseudo header */
1171 struct tcp4_pseudohdr {
1179 struct tcp6_pseudohdr {
1180 struct in6_addr saddr;
1181 struct in6_addr daddr;
1183 __be32 protocol; /* including padding */
1186 union tcp_md5sum_block {
1187 struct tcp4_pseudohdr ip4;
1188 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1189 struct tcp6_pseudohdr ip6;
1193 /* - pool: digest algorithm, hash description and scratch buffer */
1194 struct tcp_md5sig_pool {
1195 struct hash_desc md5_desc;
1196 union tcp_md5sum_block md5_blk;
1200 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1201 const struct sock *sk,
1202 const struct request_sock *req,
1203 const struct sk_buff *skb);
1204 extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk,
1205 struct sock *addr_sk);
1206 extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey,
1208 extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr);
1210 #ifdef CONFIG_TCP_MD5SIG
1211 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1212 &(struct tcp_md5sig_key) { \
1213 .key = (twsk)->tw_md5_key, \
1214 .keylen = (twsk)->tw_md5_keylen, \
1217 #define tcp_twsk_md5_key(twsk) NULL
1220 extern bool tcp_alloc_md5sig_pool(void);
1222 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1223 static inline void tcp_put_md5sig_pool(void)
1228 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1229 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1230 unsigned header_len);
1231 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1232 const struct tcp_md5sig_key *key);
1234 /* write queue abstraction */
1235 static inline void tcp_write_queue_purge(struct sock *sk)
1237 struct sk_buff *skb;
1239 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1240 sk_wmem_free_skb(sk, skb);
1242 tcp_clear_all_retrans_hints(tcp_sk(sk));
1245 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1247 return skb_peek(&sk->sk_write_queue);
1250 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1252 return skb_peek_tail(&sk->sk_write_queue);
1255 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1256 const struct sk_buff *skb)
1258 return skb_queue_next(&sk->sk_write_queue, skb);
1261 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1262 const struct sk_buff *skb)
1264 return skb_queue_prev(&sk->sk_write_queue, skb);
1267 #define tcp_for_write_queue(skb, sk) \
1268 skb_queue_walk(&(sk)->sk_write_queue, skb)
1270 #define tcp_for_write_queue_from(skb, sk) \
1271 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1273 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1274 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1276 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1278 return sk->sk_send_head;
1281 static inline bool tcp_skb_is_last(const struct sock *sk,
1282 const struct sk_buff *skb)
1284 return skb_queue_is_last(&sk->sk_write_queue, skb);
1287 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1289 if (tcp_skb_is_last(sk, skb))
1290 sk->sk_send_head = NULL;
1292 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1295 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1297 if (sk->sk_send_head == skb_unlinked)
1298 sk->sk_send_head = NULL;
1299 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1300 tcp_sk(sk)->highest_sack = NULL;
1303 static inline void tcp_init_send_head(struct sock *sk)
1305 sk->sk_send_head = NULL;
1308 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1310 __skb_queue_tail(&sk->sk_write_queue, skb);
1313 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1315 __tcp_add_write_queue_tail(sk, skb);
1317 /* Queue it, remembering where we must start sending. */
1318 if (sk->sk_send_head == NULL) {
1319 sk->sk_send_head = skb;
1321 if (tcp_sk(sk)->highest_sack == NULL)
1322 tcp_sk(sk)->highest_sack = skb;
1326 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1328 __skb_queue_head(&sk->sk_write_queue, skb);
1331 /* Insert buff after skb on the write queue of sk. */
1332 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1333 struct sk_buff *buff,
1336 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1339 /* Insert new before skb on the write queue of sk. */
1340 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1341 struct sk_buff *skb,
1344 __skb_queue_before(&sk->sk_write_queue, skb, new);
1346 if (sk->sk_send_head == skb)
1347 sk->sk_send_head = new;
1350 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1352 __skb_unlink(skb, &sk->sk_write_queue);
1355 static inline int tcp_write_queue_empty(struct sock *sk)
1357 return skb_queue_empty(&sk->sk_write_queue);
1360 static inline void tcp_push_pending_frames(struct sock *sk)
1362 if (tcp_send_head(sk)) {
1363 struct tcp_sock *tp = tcp_sk(sk);
1365 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1369 /* Start sequence of the highest skb with SACKed bit, valid only if
1370 * sacked > 0 or when the caller has ensured validity by itself.
1372 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1374 if (!tp->sacked_out)
1377 if (tp->highest_sack == NULL)
1380 return TCP_SKB_CB(tp->highest_sack)->seq;
1383 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1385 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1386 tcp_write_queue_next(sk, skb);
1389 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1391 return tcp_sk(sk)->highest_sack;
1394 static inline void tcp_highest_sack_reset(struct sock *sk)
1396 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1399 /* Called when old skb is about to be deleted and replaced by new skb */
1400 static inline void tcp_highest_sack_replace(struct sock *sk,
1401 struct sk_buff *old,
1402 struct sk_buff *new)
1404 if (old == tcp_highest_sack(sk))
1405 tcp_sk(sk)->highest_sack = new;
1408 /* Determines whether this is a thin stream (which may suffer from
1409 * increased latency). Used to trigger latency-reducing mechanisms.
1411 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp)
1413 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1417 enum tcp_seq_states {
1418 TCP_SEQ_STATE_LISTENING,
1419 TCP_SEQ_STATE_OPENREQ,
1420 TCP_SEQ_STATE_ESTABLISHED,
1421 TCP_SEQ_STATE_TIME_WAIT,
1424 int tcp_seq_open(struct inode *inode, struct file *file);
1426 struct tcp_seq_afinfo {
1429 const struct file_operations *seq_fops;
1430 struct seq_operations seq_ops;
1433 struct tcp_iter_state {
1434 struct seq_net_private p;
1436 enum tcp_seq_states state;
1437 struct sock *syn_wait_sk;
1438 int bucket, offset, sbucket, num, uid;
1442 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1443 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1445 extern struct request_sock_ops tcp_request_sock_ops;
1446 extern struct request_sock_ops tcp6_request_sock_ops;
1448 extern void tcp_v4_destroy_sock(struct sock *sk);
1450 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1451 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, u32 features);
1452 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1453 struct sk_buff *skb);
1454 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1455 struct sk_buff *skb);
1456 extern int tcp_gro_complete(struct sk_buff *skb);
1457 extern int tcp4_gro_complete(struct sk_buff *skb);
1459 #ifdef CONFIG_PROC_FS
1460 extern int tcp4_proc_init(void);
1461 extern void tcp4_proc_exit(void);
1464 /* TCP af-specific functions */
1465 struct tcp_sock_af_ops {
1466 #ifdef CONFIG_TCP_MD5SIG
1467 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1468 struct sock *addr_sk);
1469 int (*calc_md5_hash) (char *location,
1470 struct tcp_md5sig_key *md5,
1471 const struct sock *sk,
1472 const struct request_sock *req,
1473 const struct sk_buff *skb);
1474 int (*md5_add) (struct sock *sk,
1475 struct sock *addr_sk,
1478 int (*md5_parse) (struct sock *sk,
1479 char __user *optval,
1484 struct tcp_request_sock_ops {
1485 #ifdef CONFIG_TCP_MD5SIG
1486 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1487 struct request_sock *req);
1488 int (*calc_md5_hash) (char *location,
1489 struct tcp_md5sig_key *md5,
1490 const struct sock *sk,
1491 const struct request_sock *req,
1492 const struct sk_buff *skb);
1496 /* Using SHA1 for now, define some constants.
1498 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1499 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1500 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1502 extern int tcp_cookie_generator(u32 *bakery);
1505 * struct tcp_cookie_values - each socket needs extra space for the
1506 * cookies, together with (optional) space for any SYN data.
1508 * A tcp_sock contains a pointer to the current value, and this is
1509 * cloned to the tcp_timewait_sock.
1511 * @cookie_pair: variable data from the option exchange.
1513 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1514 * indicates default (sysctl_tcp_cookie_size).
1515 * After cookie sent, remembers size of cookie.
1516 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1518 * @s_data_desired: user specified tcpct_s_data_desired. When the
1519 * constant payload is specified (@s_data_constant),
1520 * holds its length instead.
1521 * Range 0 to TCP_MSS_DESIRED.
1523 * @s_data_payload: constant data that is to be included in the
1524 * payload of SYN or SYNACK segments when the
1525 * cookie option is present.
1527 struct tcp_cookie_values {
1529 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE];
1530 u8 cookie_pair_size;
1532 u16 s_data_desired:11,
1537 u8 s_data_payload[0];
1540 static inline void tcp_cookie_values_release(struct kref *kref)
1542 kfree(container_of(kref, struct tcp_cookie_values, kref));
1545 /* The length of constant payload data. Note that s_data_desired is
1546 * overloaded, depending on s_data_constant: either the length of constant
1547 * data (returned here) or the limit on variable data.
1549 static inline int tcp_s_data_size(const struct tcp_sock *tp)
1551 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1552 ? tp->cookie_values->s_data_desired
1557 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1559 * As tcp_request_sock has already been extended in other places, the
1560 * only remaining method is to pass stack values along as function
1561 * parameters. These parameters are not needed after sending SYNACK.
1563 * @cookie_bakery: cryptographic secret and message workspace.
1565 * @cookie_plus: bytes in authenticator/cookie option, copied from
1566 * struct tcp_options_received (above).
1568 struct tcp_extend_values {
1569 struct request_values rv;
1570 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS];
1576 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1578 return (struct tcp_extend_values *)rvp;
1581 extern void tcp_v4_init(void);
1582 extern void tcp_init(void);