2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
6 * Definitions for the TCP module.
8 * Version: @(#)tcp.h 1.0.5 05/23/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
22 #define FASTRETRANS_DEBUG 1
24 #include <linux/list.h>
25 #include <linux/tcp.h>
26 #include <linux/slab.h>
27 #include <linux/cache.h>
28 #include <linux/percpu.h>
29 #include <linux/skbuff.h>
30 #include <linux/dmaengine.h>
31 #include <linux/crypto.h>
32 #include <linux/cryptohash.h>
33 #include <linux/kref.h>
35 #include <net/inet_connection_sock.h>
36 #include <net/inet_timewait_sock.h>
37 #include <net/inet_hashtables.h>
38 #include <net/checksum.h>
39 #include <net/request_sock.h>
43 #include <net/tcp_states.h>
44 #include <net/inet_ecn.h>
47 #include <linux/seq_file.h>
49 extern struct inet_hashinfo tcp_hashinfo;
51 extern struct percpu_counter tcp_orphan_count;
52 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
54 #define MAX_TCP_HEADER (128 + MAX_HEADER)
55 #define MAX_TCP_OPTION_SPACE 40
58 * Never offer a window over 32767 without using window scaling. Some
59 * poor stacks do signed 16bit maths!
61 #define MAX_TCP_WINDOW 32767U
63 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
64 #define TCP_MIN_MSS 88U
66 /* The least MTU to use for probing */
67 #define TCP_BASE_MSS 512
69 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
70 #define TCP_FASTRETRANS_THRESH 3
72 /* Maximal reordering. */
73 #define TCP_MAX_REORDERING 127
75 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
76 #define TCP_MAX_QUICKACKS 16U
79 #define TCP_URG_VALID 0x0100
80 #define TCP_URG_NOTYET 0x0200
81 #define TCP_URG_READ 0x0400
83 #define TCP_RETR1 3 /*
84 * This is how many retries it does before it
85 * tries to figure out if the gateway is
86 * down. Minimal RFC value is 3; it corresponds
87 * to ~3sec-8min depending on RTO.
90 #define TCP_RETR2 15 /*
91 * This should take at least
92 * 90 minutes to time out.
93 * RFC1122 says that the limit is 100 sec.
94 * 15 is ~13-30min depending on RTO.
97 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
98 * connection: ~180sec is RFC minimum */
100 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
101 * connection: ~180sec is RFC minimum */
104 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
105 * socket. 7 is ~50sec-16min.
109 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
110 * state, about 60 seconds */
111 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
112 /* BSD style FIN_WAIT2 deadlock breaker.
113 * It used to be 3min, new value is 60sec,
114 * to combine FIN-WAIT-2 timeout with
118 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
120 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
121 #define TCP_ATO_MIN ((unsigned)(HZ/25))
123 #define TCP_DELACK_MIN 4U
124 #define TCP_ATO_MIN 4U
126 #define TCP_RTO_MAX ((unsigned)(120*HZ))
127 #define TCP_RTO_MIN ((unsigned)(HZ/5))
128 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
130 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
131 * for local resources.
134 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
135 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
136 #define TCP_KEEPALIVE_INTVL (75*HZ)
138 #define MAX_TCP_KEEPIDLE 32767
139 #define MAX_TCP_KEEPINTVL 32767
140 #define MAX_TCP_KEEPCNT 127
141 #define MAX_TCP_SYNCNT 127
143 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
145 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
146 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
147 * after this time. It should be equal
148 * (or greater than) TCP_TIMEWAIT_LEN
149 * to provide reliability equal to one
150 * provided by timewait state.
152 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
153 * timestamps. It must be less than
154 * minimal timewait lifetime.
160 #define TCPOPT_NOP 1 /* Padding */
161 #define TCPOPT_EOL 0 /* End of options */
162 #define TCPOPT_MSS 2 /* Segment size negotiating */
163 #define TCPOPT_WINDOW 3 /* Window scaling */
164 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
165 #define TCPOPT_SACK 5 /* SACK Block */
166 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
167 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
168 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
174 #define TCPOLEN_MSS 4
175 #define TCPOLEN_WINDOW 3
176 #define TCPOLEN_SACK_PERM 2
177 #define TCPOLEN_TIMESTAMP 10
178 #define TCPOLEN_MD5SIG 18
179 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
180 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
181 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
182 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
184 /* But this is what stacks really send out. */
185 #define TCPOLEN_TSTAMP_ALIGNED 12
186 #define TCPOLEN_WSCALE_ALIGNED 4
187 #define TCPOLEN_SACKPERM_ALIGNED 4
188 #define TCPOLEN_SACK_BASE 2
189 #define TCPOLEN_SACK_BASE_ALIGNED 4
190 #define TCPOLEN_SACK_PERBLOCK 8
191 #define TCPOLEN_MD5SIG_ALIGNED 20
192 #define TCPOLEN_MSS_ALIGNED 4
194 /* Flags in tp->nonagle */
195 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
196 #define TCP_NAGLE_CORK 2 /* Socket is corked */
197 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
199 /* TCP thin-stream limits */
200 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
202 extern struct inet_timewait_death_row tcp_death_row;
204 /* sysctl variables for tcp */
205 extern int sysctl_tcp_timestamps;
206 extern int sysctl_tcp_window_scaling;
207 extern int sysctl_tcp_sack;
208 extern int sysctl_tcp_fin_timeout;
209 extern int sysctl_tcp_keepalive_time;
210 extern int sysctl_tcp_keepalive_probes;
211 extern int sysctl_tcp_keepalive_intvl;
212 extern int sysctl_tcp_syn_retries;
213 extern int sysctl_tcp_synack_retries;
214 extern int sysctl_tcp_retries1;
215 extern int sysctl_tcp_retries2;
216 extern int sysctl_tcp_orphan_retries;
217 extern int sysctl_tcp_syncookies;
218 extern int sysctl_tcp_retrans_collapse;
219 extern int sysctl_tcp_stdurg;
220 extern int sysctl_tcp_rfc1337;
221 extern int sysctl_tcp_abort_on_overflow;
222 extern int sysctl_tcp_max_orphans;
223 extern int sysctl_tcp_fack;
224 extern int sysctl_tcp_reordering;
225 extern int sysctl_tcp_ecn;
226 extern int sysctl_tcp_dsack;
227 extern int sysctl_tcp_mem[3];
228 extern int sysctl_tcp_wmem[3];
229 extern int sysctl_tcp_rmem[3];
230 extern int sysctl_tcp_app_win;
231 extern int sysctl_tcp_adv_win_scale;
232 extern int sysctl_tcp_tw_reuse;
233 extern int sysctl_tcp_frto;
234 extern int sysctl_tcp_frto_response;
235 extern int sysctl_tcp_low_latency;
236 extern int sysctl_tcp_dma_copybreak;
237 extern int sysctl_tcp_nometrics_save;
238 extern int sysctl_tcp_moderate_rcvbuf;
239 extern int sysctl_tcp_tso_win_divisor;
240 extern int sysctl_tcp_abc;
241 extern int sysctl_tcp_mtu_probing;
242 extern int sysctl_tcp_base_mss;
243 extern int sysctl_tcp_workaround_signed_windows;
244 extern int sysctl_tcp_slow_start_after_idle;
245 extern int sysctl_tcp_max_ssthresh;
246 extern int sysctl_tcp_cookie_size;
247 extern int sysctl_tcp_thin_linear_timeouts;
248 extern int sysctl_tcp_thin_dupack;
250 extern atomic_t tcp_memory_allocated;
251 extern struct percpu_counter tcp_sockets_allocated;
252 extern int tcp_memory_pressure;
255 * The next routines deal with comparing 32 bit unsigned ints
256 * and worry about wraparound (automatic with unsigned arithmetic).
259 static inline int before(__u32 seq1, __u32 seq2)
261 return (__s32)(seq1-seq2) < 0;
263 #define after(seq2, seq1) before(seq1, seq2)
265 /* is s2<=s1<=s3 ? */
266 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
268 return seq3 - seq2 >= seq1 - seq2;
271 static inline int tcp_too_many_orphans(struct sock *sk, int num)
273 return (num > sysctl_tcp_max_orphans) ||
274 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
275 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
278 /* syncookies: remember time of last synqueue overflow */
279 static inline void tcp_synq_overflow(struct sock *sk)
281 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
284 /* syncookies: no recent synqueue overflow on this listening socket? */
285 static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
287 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
288 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT);
291 extern struct proto tcp_prot;
293 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
294 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
295 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
296 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
297 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
299 extern void tcp_v4_err(struct sk_buff *skb, u32);
301 extern void tcp_shutdown (struct sock *sk, int how);
303 extern int tcp_v4_rcv(struct sk_buff *skb);
305 extern int tcp_v4_remember_stamp(struct sock *sk);
306 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
307 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
309 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
310 size_t size, int flags);
311 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
312 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
313 struct tcphdr *th, unsigned len);
314 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
315 struct tcphdr *th, unsigned len);
316 extern void tcp_rcv_space_adjust(struct sock *sk);
317 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
318 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
319 extern void tcp_twsk_destructor(struct sock *sk);
320 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
321 struct pipe_inode_info *pipe, size_t len,
324 static inline void tcp_dec_quickack_mode(struct sock *sk,
325 const unsigned int pkts)
327 struct inet_connection_sock *icsk = inet_csk(sk);
329 if (icsk->icsk_ack.quick) {
330 if (pkts >= icsk->icsk_ack.quick) {
331 icsk->icsk_ack.quick = 0;
332 /* Leaving quickack mode we deflate ATO. */
333 icsk->icsk_ack.ato = TCP_ATO_MIN;
335 icsk->icsk_ack.quick -= pkts;
339 extern void tcp_enter_quickack_mode(struct sock *sk);
342 #define TCP_ECN_QUEUE_CWR 2
343 #define TCP_ECN_DEMAND_CWR 4
345 static __inline__ void
346 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
348 if (sysctl_tcp_ecn && th->ece && th->cwr)
349 inet_rsk(req)->ecn_ok = 1;
360 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
362 const struct tcphdr *th);
363 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
364 struct request_sock *req,
365 struct request_sock **prev);
366 extern int tcp_child_process(struct sock *parent, struct sock *child,
367 struct sk_buff *skb);
368 extern int tcp_use_frto(struct sock *sk);
369 extern void tcp_enter_frto(struct sock *sk);
370 extern void tcp_enter_loss(struct sock *sk, int how);
371 extern void tcp_clear_retrans(struct tcp_sock *tp);
372 extern void tcp_update_metrics(struct sock *sk);
373 extern void tcp_close(struct sock *sk, long timeout);
374 extern unsigned int tcp_poll(struct file * file, struct socket *sock,
375 struct poll_table_struct *wait);
376 extern int tcp_getsockopt(struct sock *sk, int level, int optname,
377 char __user *optval, int __user *optlen);
378 extern int tcp_setsockopt(struct sock *sk, int level, int optname,
379 char __user *optval, unsigned int optlen);
380 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
381 char __user *optval, int __user *optlen);
382 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
383 char __user *optval, unsigned int optlen);
384 extern void tcp_set_keepalive(struct sock *sk, int val);
385 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
386 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
387 size_t len, int nonblock, int flags, int *addr_len);
388 extern void tcp_parse_options(struct sk_buff *skb,
389 struct tcp_options_received *opt_rx, u8 **hvpp,
391 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
394 * TCP v4 functions exported for the inet6 API
397 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
398 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
399 extern struct sock * tcp_create_openreq_child(struct sock *sk,
400 struct request_sock *req,
401 struct sk_buff *skb);
402 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
403 struct request_sock *req,
404 struct dst_entry *dst);
405 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
406 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
408 extern int tcp_connect(struct sock *sk);
409 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
410 struct request_sock *req,
411 struct request_values *rvp);
412 extern int tcp_disconnect(struct sock *sk, int flags);
415 /* From syncookies.c */
416 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
417 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
418 struct ip_options *opt);
419 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
422 extern __u32 cookie_init_timestamp(struct request_sock *req);
423 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *);
425 /* From net/ipv6/syncookies.c */
426 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
427 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
432 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
434 extern int tcp_may_send_now(struct sock *sk);
435 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
436 extern void tcp_retransmit_timer(struct sock *sk);
437 extern void tcp_xmit_retransmit_queue(struct sock *);
438 extern void tcp_simple_retransmit(struct sock *);
439 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
440 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
442 extern void tcp_send_probe0(struct sock *);
443 extern void tcp_send_partial(struct sock *);
444 extern int tcp_write_wakeup(struct sock *);
445 extern void tcp_send_fin(struct sock *sk);
446 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
447 extern int tcp_send_synack(struct sock *);
448 extern void tcp_push_one(struct sock *, unsigned int mss_now);
449 extern void tcp_send_ack(struct sock *sk);
450 extern void tcp_send_delayed_ack(struct sock *sk);
453 extern void tcp_cwnd_application_limited(struct sock *sk);
456 extern void tcp_init_xmit_timers(struct sock *);
457 static inline void tcp_clear_xmit_timers(struct sock *sk)
459 inet_csk_clear_xmit_timers(sk);
462 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
463 extern unsigned int tcp_current_mss(struct sock *sk);
465 /* Bound MSS / TSO packet size with the half of the window */
466 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
468 if (tp->max_window && pktsize > (tp->max_window >> 1))
469 return max(tp->max_window >> 1, 68U - tp->tcp_header_len);
475 extern void tcp_get_info(struct sock *, struct tcp_info *);
477 /* Read 'sendfile()'-style from a TCP socket */
478 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
479 unsigned int, size_t);
480 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
481 sk_read_actor_t recv_actor);
483 extern void tcp_initialize_rcv_mss(struct sock *sk);
485 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
486 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
487 extern void tcp_mtup_init(struct sock *sk);
489 static inline void tcp_bound_rto(const struct sock *sk)
491 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
492 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
495 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
497 return (tp->srtt >> 3) + tp->rttvar;
500 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
502 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
503 ntohl(TCP_FLAG_ACK) |
507 static inline void tcp_fast_path_on(struct tcp_sock *tp)
509 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
512 static inline void tcp_fast_path_check(struct sock *sk)
514 struct tcp_sock *tp = tcp_sk(sk);
516 if (skb_queue_empty(&tp->out_of_order_queue) &&
518 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
520 tcp_fast_path_on(tp);
523 /* Compute the actual rto_min value */
524 static inline u32 tcp_rto_min(struct sock *sk)
526 struct dst_entry *dst = __sk_dst_get(sk);
527 u32 rto_min = TCP_RTO_MIN;
529 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
530 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
534 /* Compute the actual receive window we are currently advertising.
535 * Rcv_nxt can be after the window if our peer push more data
536 * than the offered window.
538 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
540 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
547 /* Choose a new window, without checks for shrinking, and without
548 * scaling applied to the result. The caller does these things
549 * if necessary. This is a "raw" window selection.
551 extern u32 __tcp_select_window(struct sock *sk);
553 /* TCP timestamps are only 32-bits, this causes a slight
554 * complication on 64-bit systems since we store a snapshot
555 * of jiffies in the buffer control blocks below. We decided
556 * to use only the low 32-bits of jiffies and hide the ugly
557 * casts with the following macro.
559 #define tcp_time_stamp ((__u32)(jiffies))
561 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
563 #define TCPHDR_FIN 0x01
564 #define TCPHDR_SYN 0x02
565 #define TCPHDR_RST 0x04
566 #define TCPHDR_PSH 0x08
567 #define TCPHDR_ACK 0x10
568 #define TCPHDR_URG 0x20
569 #define TCPHDR_ECE 0x40
570 #define TCPHDR_CWR 0x80
572 /* This is what the send packet queuing engine uses to pass
573 * TCP per-packet control information to the transmission code.
574 * We also store the host-order sequence numbers in here too.
575 * This is 44 bytes if IPV6 is enabled.
576 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
580 struct inet_skb_parm h4;
581 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
582 struct inet6_skb_parm h6;
584 } header; /* For incoming frames */
585 __u32 seq; /* Starting sequence number */
586 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
587 __u32 when; /* used to compute rtt's */
588 __u8 flags; /* TCP header flags. */
589 __u8 sacked; /* State flags for SACK/FACK. */
590 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
591 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
592 #define TCPCB_LOST 0x04 /* SKB is lost */
593 #define TCPCB_TAGBITS 0x07 /* All tag bits */
595 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
596 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
598 __u32 ack_seq; /* Sequence number ACK'd */
601 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
603 /* Due to TSO, an SKB can be composed of multiple actual
604 * packets. To keep these tracked properly, we use this.
606 static inline int tcp_skb_pcount(const struct sk_buff *skb)
608 return skb_shinfo(skb)->gso_segs;
611 /* This is valid iff tcp_skb_pcount() > 1. */
612 static inline int tcp_skb_mss(const struct sk_buff *skb)
614 return skb_shinfo(skb)->gso_size;
617 /* Events passed to congestion control interface */
619 CA_EVENT_TX_START, /* first transmit when no packets in flight */
620 CA_EVENT_CWND_RESTART, /* congestion window restart */
621 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
622 CA_EVENT_FRTO, /* fast recovery timeout */
623 CA_EVENT_LOSS, /* loss timeout */
624 CA_EVENT_FAST_ACK, /* in sequence ack */
625 CA_EVENT_SLOW_ACK, /* other ack */
629 * Interface for adding new TCP congestion control handlers
631 #define TCP_CA_NAME_MAX 16
632 #define TCP_CA_MAX 128
633 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
635 #define TCP_CONG_NON_RESTRICTED 0x1
636 #define TCP_CONG_RTT_STAMP 0x2
638 struct tcp_congestion_ops {
639 struct list_head list;
642 /* initialize private data (optional) */
643 void (*init)(struct sock *sk);
644 /* cleanup private data (optional) */
645 void (*release)(struct sock *sk);
647 /* return slow start threshold (required) */
648 u32 (*ssthresh)(struct sock *sk);
649 /* lower bound for congestion window (optional) */
650 u32 (*min_cwnd)(const struct sock *sk);
651 /* do new cwnd calculation (required) */
652 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
653 /* call before changing ca_state (optional) */
654 void (*set_state)(struct sock *sk, u8 new_state);
655 /* call when cwnd event occurs (optional) */
656 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
657 /* new value of cwnd after loss (optional) */
658 u32 (*undo_cwnd)(struct sock *sk);
659 /* hook for packet ack accounting (optional) */
660 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
661 /* get info for inet_diag (optional) */
662 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
664 char name[TCP_CA_NAME_MAX];
665 struct module *owner;
668 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
669 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
671 extern void tcp_init_congestion_control(struct sock *sk);
672 extern void tcp_cleanup_congestion_control(struct sock *sk);
673 extern int tcp_set_default_congestion_control(const char *name);
674 extern void tcp_get_default_congestion_control(char *name);
675 extern void tcp_get_available_congestion_control(char *buf, size_t len);
676 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
677 extern int tcp_set_allowed_congestion_control(char *allowed);
678 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
679 extern void tcp_slow_start(struct tcp_sock *tp);
680 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
682 extern struct tcp_congestion_ops tcp_init_congestion_ops;
683 extern u32 tcp_reno_ssthresh(struct sock *sk);
684 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
685 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
686 extern struct tcp_congestion_ops tcp_reno;
688 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
690 struct inet_connection_sock *icsk = inet_csk(sk);
692 if (icsk->icsk_ca_ops->set_state)
693 icsk->icsk_ca_ops->set_state(sk, ca_state);
694 icsk->icsk_ca_state = ca_state;
697 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
699 const struct inet_connection_sock *icsk = inet_csk(sk);
701 if (icsk->icsk_ca_ops->cwnd_event)
702 icsk->icsk_ca_ops->cwnd_event(sk, event);
705 /* These functions determine how the current flow behaves in respect of SACK
706 * handling. SACK is negotiated with the peer, and therefore it can vary
707 * between different flows.
709 * tcp_is_sack - SACK enabled
710 * tcp_is_reno - No SACK
711 * tcp_is_fack - FACK enabled, implies SACK enabled
713 static inline int tcp_is_sack(const struct tcp_sock *tp)
715 return tp->rx_opt.sack_ok;
718 static inline int tcp_is_reno(const struct tcp_sock *tp)
720 return !tcp_is_sack(tp);
723 static inline int tcp_is_fack(const struct tcp_sock *tp)
725 return tp->rx_opt.sack_ok & 2;
728 static inline void tcp_enable_fack(struct tcp_sock *tp)
730 tp->rx_opt.sack_ok |= 2;
733 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
735 return tp->sacked_out + tp->lost_out;
738 /* This determines how many packets are "in the network" to the best
739 * of our knowledge. In many cases it is conservative, but where
740 * detailed information is available from the receiver (via SACK
741 * blocks etc.) we can make more aggressive calculations.
743 * Use this for decisions involving congestion control, use just
744 * tp->packets_out to determine if the send queue is empty or not.
746 * Read this equation as:
748 * "Packets sent once on transmission queue" MINUS
749 * "Packets left network, but not honestly ACKed yet" PLUS
750 * "Packets fast retransmitted"
752 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
754 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
757 #define TCP_INFINITE_SSTHRESH 0x7fffffff
759 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
761 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
764 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
765 * The exception is rate halving phase, when cwnd is decreasing towards
768 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
770 const struct tcp_sock *tp = tcp_sk(sk);
771 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
772 return tp->snd_ssthresh;
774 return max(tp->snd_ssthresh,
775 ((tp->snd_cwnd >> 1) +
776 (tp->snd_cwnd >> 2)));
779 /* Use define here intentionally to get WARN_ON location shown at the caller */
780 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
783 * Convert RFC 3390 larger initial window into an equivalent number of packets.
785 * John Heffner states:
787 * The RFC specifies a window of no more than 4380 bytes
788 * unless 2*MSS > 4380. Reading the pseudocode in the RFC
789 * is a bit misleading because they use a clamp at 4380 bytes
790 * rather than a multiplier in the relevant range.
792 static inline u32 rfc3390_bytes_to_packets(const u32 smss)
794 return smss <= 1095 ? 4 : (smss > 1460 ? 2 : 3);
797 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
798 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
800 /* Slow start with delack produces 3 packets of burst, so that
801 * it is safe "de facto". This will be the default - same as
802 * the default reordering threshold - but if reordering increases,
803 * we must be able to allow cwnd to burst at least this much in order
804 * to not pull it back when holes are filled.
806 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
808 return tp->reordering;
811 /* Returns end sequence number of the receiver's advertised window */
812 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
814 return tp->snd_una + tp->snd_wnd;
816 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
818 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
819 const struct sk_buff *skb)
822 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
825 static inline void tcp_check_probe_timer(struct sock *sk)
827 struct tcp_sock *tp = tcp_sk(sk);
828 const struct inet_connection_sock *icsk = inet_csk(sk);
830 if (!tp->packets_out && !icsk->icsk_pending)
831 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
832 icsk->icsk_rto, TCP_RTO_MAX);
835 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
840 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
846 * Calculate(/check) TCP checksum
848 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
849 __be32 daddr, __wsum base)
851 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
854 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
856 return __skb_checksum_complete(skb);
859 static inline int tcp_checksum_complete(struct sk_buff *skb)
861 return !skb_csum_unnecessary(skb) &&
862 __tcp_checksum_complete(skb);
865 /* Prequeue for VJ style copy to user, combined with checksumming. */
867 static inline void tcp_prequeue_init(struct tcp_sock *tp)
869 tp->ucopy.task = NULL;
871 tp->ucopy.memory = 0;
872 skb_queue_head_init(&tp->ucopy.prequeue);
873 #ifdef CONFIG_NET_DMA
874 tp->ucopy.dma_chan = NULL;
875 tp->ucopy.wakeup = 0;
876 tp->ucopy.pinned_list = NULL;
877 tp->ucopy.dma_cookie = 0;
881 /* Packet is added to VJ-style prequeue for processing in process
882 * context, if a reader task is waiting. Apparently, this exciting
883 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
884 * failed somewhere. Latency? Burstiness? Well, at least now we will
885 * see, why it failed. 8)8) --ANK
887 * NOTE: is this not too big to inline?
889 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
891 struct tcp_sock *tp = tcp_sk(sk);
893 if (sysctl_tcp_low_latency || !tp->ucopy.task)
896 __skb_queue_tail(&tp->ucopy.prequeue, skb);
897 tp->ucopy.memory += skb->truesize;
898 if (tp->ucopy.memory > sk->sk_rcvbuf) {
899 struct sk_buff *skb1;
901 BUG_ON(sock_owned_by_user(sk));
903 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
904 sk_backlog_rcv(sk, skb1);
905 NET_INC_STATS_BH(sock_net(sk),
906 LINUX_MIB_TCPPREQUEUEDROPPED);
909 tp->ucopy.memory = 0;
910 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
911 wake_up_interruptible_sync_poll(sk_sleep(sk),
912 POLLIN | POLLRDNORM | POLLRDBAND);
913 if (!inet_csk_ack_scheduled(sk))
914 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
915 (3 * tcp_rto_min(sk)) / 4,
925 static const char *statename[]={
926 "Unused","Established","Syn Sent","Syn Recv",
927 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
928 "Close Wait","Last ACK","Listen","Closing"
931 extern void tcp_set_state(struct sock *sk, int state);
933 extern void tcp_done(struct sock *sk);
935 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
938 rx_opt->num_sacks = 0;
941 /* Determine a window scaling and initial window to offer. */
942 extern void tcp_select_initial_window(int __space, __u32 mss,
943 __u32 *rcv_wnd, __u32 *window_clamp,
944 int wscale_ok, __u8 *rcv_wscale,
947 static inline int tcp_win_from_space(int space)
949 return sysctl_tcp_adv_win_scale<=0 ?
950 (space>>(-sysctl_tcp_adv_win_scale)) :
951 space - (space>>sysctl_tcp_adv_win_scale);
954 /* Note: caller must be prepared to deal with negative returns */
955 static inline int tcp_space(const struct sock *sk)
957 return tcp_win_from_space(sk->sk_rcvbuf -
958 atomic_read(&sk->sk_rmem_alloc));
961 static inline int tcp_full_space(const struct sock *sk)
963 return tcp_win_from_space(sk->sk_rcvbuf);
966 static inline void tcp_openreq_init(struct request_sock *req,
967 struct tcp_options_received *rx_opt,
970 struct inet_request_sock *ireq = inet_rsk(req);
972 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
974 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
975 req->mss = rx_opt->mss_clamp;
976 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
977 ireq->tstamp_ok = rx_opt->tstamp_ok;
978 ireq->sack_ok = rx_opt->sack_ok;
979 ireq->snd_wscale = rx_opt->snd_wscale;
980 ireq->wscale_ok = rx_opt->wscale_ok;
983 ireq->rmt_port = tcp_hdr(skb)->source;
984 ireq->loc_port = tcp_hdr(skb)->dest;
987 extern void tcp_enter_memory_pressure(struct sock *sk);
989 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
991 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
994 static inline int keepalive_time_when(const struct tcp_sock *tp)
996 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
999 static inline int keepalive_probes(const struct tcp_sock *tp)
1001 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1004 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1006 const struct inet_connection_sock *icsk = &tp->inet_conn;
1008 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1009 tcp_time_stamp - tp->rcv_tstamp);
1012 static inline int tcp_fin_time(const struct sock *sk)
1014 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1015 const int rto = inet_csk(sk)->icsk_rto;
1017 if (fin_timeout < (rto << 2) - (rto >> 1))
1018 fin_timeout = (rto << 2) - (rto >> 1);
1023 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1026 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1028 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1034 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1037 if (tcp_paws_check(rx_opt, 0))
1040 /* RST segments are not recommended to carry timestamp,
1041 and, if they do, it is recommended to ignore PAWS because
1042 "their cleanup function should take precedence over timestamps."
1043 Certainly, it is mistake. It is necessary to understand the reasons
1044 of this constraint to relax it: if peer reboots, clock may go
1045 out-of-sync and half-open connections will not be reset.
1046 Actually, the problem would be not existing if all
1047 the implementations followed draft about maintaining clock
1048 via reboots. Linux-2.2 DOES NOT!
1050 However, we can relax time bounds for RST segments to MSL.
1052 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1057 #define TCP_CHECK_TIMER(sk) do { } while (0)
1059 static inline void tcp_mib_init(struct net *net)
1062 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1063 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1064 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1065 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1069 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1071 tp->lost_skb_hint = NULL;
1072 tp->scoreboard_skb_hint = NULL;
1075 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1077 tcp_clear_retrans_hints_partial(tp);
1078 tp->retransmit_skb_hint = NULL;
1084 /* - key database */
1085 struct tcp_md5sig_key {
1090 struct tcp4_md5sig_key {
1091 struct tcp_md5sig_key base;
1095 struct tcp6_md5sig_key {
1096 struct tcp_md5sig_key base;
1098 u32 scope_id; /* XXX */
1100 struct in6_addr addr;
1104 struct tcp_md5sig_info {
1105 struct tcp4_md5sig_key *keys4;
1106 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1107 struct tcp6_md5sig_key *keys6;
1115 /* - pseudo header */
1116 struct tcp4_pseudohdr {
1124 struct tcp6_pseudohdr {
1125 struct in6_addr saddr;
1126 struct in6_addr daddr;
1128 __be32 protocol; /* including padding */
1131 union tcp_md5sum_block {
1132 struct tcp4_pseudohdr ip4;
1133 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1134 struct tcp6_pseudohdr ip6;
1138 /* - pool: digest algorithm, hash description and scratch buffer */
1139 struct tcp_md5sig_pool {
1140 struct hash_desc md5_desc;
1141 union tcp_md5sum_block md5_blk;
1144 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1147 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1148 struct sock *sk, struct request_sock *req,
1149 struct sk_buff *skb);
1150 extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk,
1151 struct sock *addr_sk);
1152 extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey,
1154 extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr);
1156 #ifdef CONFIG_TCP_MD5SIG
1157 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1158 &(struct tcp_md5sig_key) { \
1159 .key = (twsk)->tw_md5_key, \
1160 .keylen = (twsk)->tw_md5_keylen, \
1163 #define tcp_twsk_md5_key(twsk) NULL
1166 extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *);
1167 extern void tcp_free_md5sig_pool(void);
1169 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1170 extern void tcp_put_md5sig_pool(void);
1172 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1173 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1174 unsigned header_len);
1175 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1176 struct tcp_md5sig_key *key);
1178 /* write queue abstraction */
1179 static inline void tcp_write_queue_purge(struct sock *sk)
1181 struct sk_buff *skb;
1183 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1184 sk_wmem_free_skb(sk, skb);
1186 tcp_clear_all_retrans_hints(tcp_sk(sk));
1189 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1191 return skb_peek(&sk->sk_write_queue);
1194 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1196 return skb_peek_tail(&sk->sk_write_queue);
1199 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1201 return skb_queue_next(&sk->sk_write_queue, skb);
1204 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb)
1206 return skb_queue_prev(&sk->sk_write_queue, skb);
1209 #define tcp_for_write_queue(skb, sk) \
1210 skb_queue_walk(&(sk)->sk_write_queue, skb)
1212 #define tcp_for_write_queue_from(skb, sk) \
1213 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1215 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1216 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1218 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1220 return sk->sk_send_head;
1223 static inline bool tcp_skb_is_last(const struct sock *sk,
1224 const struct sk_buff *skb)
1226 return skb_queue_is_last(&sk->sk_write_queue, skb);
1229 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1231 if (tcp_skb_is_last(sk, skb))
1232 sk->sk_send_head = NULL;
1234 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1237 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1239 if (sk->sk_send_head == skb_unlinked)
1240 sk->sk_send_head = NULL;
1243 static inline void tcp_init_send_head(struct sock *sk)
1245 sk->sk_send_head = NULL;
1248 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1250 __skb_queue_tail(&sk->sk_write_queue, skb);
1253 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1255 __tcp_add_write_queue_tail(sk, skb);
1257 /* Queue it, remembering where we must start sending. */
1258 if (sk->sk_send_head == NULL) {
1259 sk->sk_send_head = skb;
1261 if (tcp_sk(sk)->highest_sack == NULL)
1262 tcp_sk(sk)->highest_sack = skb;
1266 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1268 __skb_queue_head(&sk->sk_write_queue, skb);
1271 /* Insert buff after skb on the write queue of sk. */
1272 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1273 struct sk_buff *buff,
1276 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1279 /* Insert new before skb on the write queue of sk. */
1280 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1281 struct sk_buff *skb,
1284 __skb_queue_before(&sk->sk_write_queue, skb, new);
1286 if (sk->sk_send_head == skb)
1287 sk->sk_send_head = new;
1290 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1292 __skb_unlink(skb, &sk->sk_write_queue);
1295 static inline int tcp_write_queue_empty(struct sock *sk)
1297 return skb_queue_empty(&sk->sk_write_queue);
1300 static inline void tcp_push_pending_frames(struct sock *sk)
1302 if (tcp_send_head(sk)) {
1303 struct tcp_sock *tp = tcp_sk(sk);
1305 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1309 /* Start sequence of the highest skb with SACKed bit, valid only if
1310 * sacked > 0 or when the caller has ensured validity by itself.
1312 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1314 if (!tp->sacked_out)
1317 if (tp->highest_sack == NULL)
1320 return TCP_SKB_CB(tp->highest_sack)->seq;
1323 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1325 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1326 tcp_write_queue_next(sk, skb);
1329 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1331 return tcp_sk(sk)->highest_sack;
1334 static inline void tcp_highest_sack_reset(struct sock *sk)
1336 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1339 /* Called when old skb is about to be deleted (to be combined with new skb) */
1340 static inline void tcp_highest_sack_combine(struct sock *sk,
1341 struct sk_buff *old,
1342 struct sk_buff *new)
1344 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1345 tcp_sk(sk)->highest_sack = new;
1348 /* Determines whether this is a thin stream (which may suffer from
1349 * increased latency). Used to trigger latency-reducing mechanisms.
1351 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp)
1353 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1357 enum tcp_seq_states {
1358 TCP_SEQ_STATE_LISTENING,
1359 TCP_SEQ_STATE_OPENREQ,
1360 TCP_SEQ_STATE_ESTABLISHED,
1361 TCP_SEQ_STATE_TIME_WAIT,
1364 struct tcp_seq_afinfo {
1367 struct file_operations seq_fops;
1368 struct seq_operations seq_ops;
1371 struct tcp_iter_state {
1372 struct seq_net_private p;
1374 enum tcp_seq_states state;
1375 struct sock *syn_wait_sk;
1376 int bucket, offset, sbucket, num, uid;
1380 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1381 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1383 extern struct request_sock_ops tcp_request_sock_ops;
1384 extern struct request_sock_ops tcp6_request_sock_ops;
1386 extern void tcp_v4_destroy_sock(struct sock *sk);
1388 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1389 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1390 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1391 struct sk_buff *skb);
1392 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1393 struct sk_buff *skb);
1394 extern int tcp_gro_complete(struct sk_buff *skb);
1395 extern int tcp4_gro_complete(struct sk_buff *skb);
1397 #ifdef CONFIG_PROC_FS
1398 extern int tcp4_proc_init(void);
1399 extern void tcp4_proc_exit(void);
1402 /* TCP af-specific functions */
1403 struct tcp_sock_af_ops {
1404 #ifdef CONFIG_TCP_MD5SIG
1405 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1406 struct sock *addr_sk);
1407 int (*calc_md5_hash) (char *location,
1408 struct tcp_md5sig_key *md5,
1410 struct request_sock *req,
1411 struct sk_buff *skb);
1412 int (*md5_add) (struct sock *sk,
1413 struct sock *addr_sk,
1416 int (*md5_parse) (struct sock *sk,
1417 char __user *optval,
1422 struct tcp_request_sock_ops {
1423 #ifdef CONFIG_TCP_MD5SIG
1424 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1425 struct request_sock *req);
1426 int (*calc_md5_hash) (char *location,
1427 struct tcp_md5sig_key *md5,
1429 struct request_sock *req,
1430 struct sk_buff *skb);
1434 /* Using SHA1 for now, define some constants.
1436 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1437 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1438 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1440 extern int tcp_cookie_generator(u32 *bakery);
1443 * struct tcp_cookie_values - each socket needs extra space for the
1444 * cookies, together with (optional) space for any SYN data.
1446 * A tcp_sock contains a pointer to the current value, and this is
1447 * cloned to the tcp_timewait_sock.
1449 * @cookie_pair: variable data from the option exchange.
1451 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1452 * indicates default (sysctl_tcp_cookie_size).
1453 * After cookie sent, remembers size of cookie.
1454 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1456 * @s_data_desired: user specified tcpct_s_data_desired. When the
1457 * constant payload is specified (@s_data_constant),
1458 * holds its length instead.
1459 * Range 0 to TCP_MSS_DESIRED.
1461 * @s_data_payload: constant data that is to be included in the
1462 * payload of SYN or SYNACK segments when the
1463 * cookie option is present.
1465 struct tcp_cookie_values {
1467 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE];
1468 u8 cookie_pair_size;
1470 u16 s_data_desired:11,
1475 u8 s_data_payload[0];
1478 static inline void tcp_cookie_values_release(struct kref *kref)
1480 kfree(container_of(kref, struct tcp_cookie_values, kref));
1483 /* The length of constant payload data. Note that s_data_desired is
1484 * overloaded, depending on s_data_constant: either the length of constant
1485 * data (returned here) or the limit on variable data.
1487 static inline int tcp_s_data_size(const struct tcp_sock *tp)
1489 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1490 ? tp->cookie_values->s_data_desired
1495 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1497 * As tcp_request_sock has already been extended in other places, the
1498 * only remaining method is to pass stack values along as function
1499 * parameters. These parameters are not needed after sending SYNACK.
1501 * @cookie_bakery: cryptographic secret and message workspace.
1503 * @cookie_plus: bytes in authenticator/cookie option, copied from
1504 * struct tcp_options_received (above).
1506 struct tcp_extend_values {
1507 struct request_values rv;
1508 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS];
1514 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1516 return (struct tcp_extend_values *)rvp;
1519 extern void tcp_v4_init(void);
1520 extern void tcp_init(void);