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
57 * Never offer a window over 32767 without using window scaling. Some
58 * poor stacks do signed 16bit maths!
60 #define MAX_TCP_WINDOW 32767U
62 /* Offer an initial receive window of 10 mss. */
63 #define TCP_DEFAULT_INIT_RCVWND 10
65 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
66 #define TCP_MIN_MSS 88U
68 /* The least MTU to use for probing */
69 #define TCP_BASE_MSS 512
71 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
72 #define TCP_FASTRETRANS_THRESH 3
74 /* Maximal reordering. */
75 #define TCP_MAX_REORDERING 127
77 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
78 #define TCP_MAX_QUICKACKS 16U
81 #define TCP_URG_VALID 0x0100
82 #define TCP_URG_NOTYET 0x0200
83 #define TCP_URG_READ 0x0400
85 #define TCP_RETR1 3 /*
86 * This is how many retries it does before it
87 * tries to figure out if the gateway is
88 * down. Minimal RFC value is 3; it corresponds
89 * to ~3sec-8min depending on RTO.
92 #define TCP_RETR2 15 /*
93 * This should take at least
94 * 90 minutes to time out.
95 * RFC1122 says that the limit is 100 sec.
96 * 15 is ~13-30min depending on RTO.
99 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
100 * connection: ~180sec is RFC minimum */
102 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
103 * connection: ~180sec is RFC minimum */
105 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
106 * state, about 60 seconds */
107 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
108 /* BSD style FIN_WAIT2 deadlock breaker.
109 * It used to be 3min, new value is 60sec,
110 * to combine FIN-WAIT-2 timeout with
114 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
116 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
117 #define TCP_ATO_MIN ((unsigned)(HZ/25))
119 #define TCP_DELACK_MIN 4U
120 #define TCP_ATO_MIN 4U
122 #define TCP_RTO_MAX ((unsigned)(120*HZ))
123 #define TCP_RTO_MIN ((unsigned)(HZ/5))
124 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC2988bis initial RTO value */
125 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
126 * used as a fallback RTO for the
127 * initial data transmission if no
128 * valid RTT sample has been acquired,
129 * most likely due to retrans in 3WHS.
132 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
133 * for local resources.
136 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
137 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
138 #define TCP_KEEPALIVE_INTVL (75*HZ)
140 #define MAX_TCP_KEEPIDLE 32767
141 #define MAX_TCP_KEEPINTVL 32767
142 #define MAX_TCP_KEEPCNT 127
143 #define MAX_TCP_SYNCNT 127
145 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
147 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
148 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
149 * after this time. It should be equal
150 * (or greater than) TCP_TIMEWAIT_LEN
151 * to provide reliability equal to one
152 * provided by timewait state.
154 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
155 * timestamps. It must be less than
156 * minimal timewait lifetime.
162 #define TCPOPT_NOP 1 /* Padding */
163 #define TCPOPT_EOL 0 /* End of options */
164 #define TCPOPT_MSS 2 /* Segment size negotiating */
165 #define TCPOPT_WINDOW 3 /* Window scaling */
166 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
167 #define TCPOPT_SACK 5 /* SACK Block */
168 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
169 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
170 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
176 #define TCPOLEN_MSS 4
177 #define TCPOLEN_WINDOW 3
178 #define TCPOLEN_SACK_PERM 2
179 #define TCPOLEN_TIMESTAMP 10
180 #define TCPOLEN_MD5SIG 18
181 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
182 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
183 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
184 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
186 /* But this is what stacks really send out. */
187 #define TCPOLEN_TSTAMP_ALIGNED 12
188 #define TCPOLEN_WSCALE_ALIGNED 4
189 #define TCPOLEN_SACKPERM_ALIGNED 4
190 #define TCPOLEN_SACK_BASE 2
191 #define TCPOLEN_SACK_BASE_ALIGNED 4
192 #define TCPOLEN_SACK_PERBLOCK 8
193 #define TCPOLEN_MD5SIG_ALIGNED 20
194 #define TCPOLEN_MSS_ALIGNED 4
196 /* Flags in tp->nonagle */
197 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
198 #define TCP_NAGLE_CORK 2 /* Socket is corked */
199 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
201 /* TCP thin-stream limits */
202 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
204 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
205 #define TCP_INIT_CWND 10
207 extern struct inet_timewait_death_row tcp_death_row;
209 /* sysctl variables for tcp */
210 extern int sysctl_tcp_timestamps;
211 extern int sysctl_tcp_window_scaling;
212 extern int sysctl_tcp_sack;
213 extern int sysctl_tcp_fin_timeout;
214 extern int sysctl_tcp_keepalive_time;
215 extern int sysctl_tcp_keepalive_probes;
216 extern int sysctl_tcp_keepalive_intvl;
217 extern int sysctl_tcp_syn_retries;
218 extern int sysctl_tcp_synack_retries;
219 extern int sysctl_tcp_retries1;
220 extern int sysctl_tcp_retries2;
221 extern int sysctl_tcp_orphan_retries;
222 extern int sysctl_tcp_syncookies;
223 extern int sysctl_tcp_retrans_collapse;
224 extern int sysctl_tcp_stdurg;
225 extern int sysctl_tcp_rfc1337;
226 extern int sysctl_tcp_abort_on_overflow;
227 extern int sysctl_tcp_max_orphans;
228 extern int sysctl_tcp_fack;
229 extern int sysctl_tcp_reordering;
230 extern int sysctl_tcp_ecn;
231 extern int sysctl_tcp_dsack;
232 extern long sysctl_tcp_mem[3];
233 extern int sysctl_tcp_wmem[3];
234 extern int sysctl_tcp_rmem[3];
235 extern int sysctl_tcp_app_win;
236 extern int sysctl_tcp_adv_win_scale;
237 extern int sysctl_tcp_tw_reuse;
238 extern int sysctl_tcp_frto;
239 extern int sysctl_tcp_frto_response;
240 extern int sysctl_tcp_low_latency;
241 extern int sysctl_tcp_dma_copybreak;
242 extern int sysctl_tcp_nometrics_save;
243 extern int sysctl_tcp_moderate_rcvbuf;
244 extern int sysctl_tcp_tso_win_divisor;
245 extern int sysctl_tcp_abc;
246 extern int sysctl_tcp_mtu_probing;
247 extern int sysctl_tcp_base_mss;
248 extern int sysctl_tcp_workaround_signed_windows;
249 extern int sysctl_tcp_slow_start_after_idle;
250 extern int sysctl_tcp_max_ssthresh;
251 extern int sysctl_tcp_cookie_size;
252 extern int sysctl_tcp_thin_linear_timeouts;
253 extern int sysctl_tcp_thin_dupack;
254 extern int sysctl_tcp_challenge_ack_limit;
256 extern atomic_long_t tcp_memory_allocated;
257 extern struct percpu_counter tcp_sockets_allocated;
258 extern int tcp_memory_pressure;
261 * The next routines deal with comparing 32 bit unsigned ints
262 * and worry about wraparound (automatic with unsigned arithmetic).
265 static inline int before(__u32 seq1, __u32 seq2)
267 return (__s32)(seq1-seq2) < 0;
269 #define after(seq2, seq1) before(seq1, seq2)
271 /* is s2<=s1<=s3 ? */
272 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
274 return seq3 - seq2 >= seq1 - seq2;
277 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
279 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
280 int orphans = percpu_counter_read_positive(ocp);
282 if (orphans << shift > sysctl_tcp_max_orphans) {
283 orphans = percpu_counter_sum_positive(ocp);
284 if (orphans << shift > sysctl_tcp_max_orphans)
288 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
289 atomic_long_read(&tcp_memory_allocated) > sysctl_tcp_mem[2])
294 /* syncookies: remember time of last synqueue overflow */
295 static inline void tcp_synq_overflow(struct sock *sk)
297 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
300 /* syncookies: no recent synqueue overflow on this listening socket? */
301 static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
303 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
304 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
307 extern struct proto tcp_prot;
309 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
310 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
311 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
312 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
313 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
315 extern void tcp_v4_err(struct sk_buff *skb, u32);
317 extern void tcp_shutdown (struct sock *sk, int how);
319 extern int tcp_v4_rcv(struct sk_buff *skb);
321 extern struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it);
322 extern void *tcp_v4_tw_get_peer(struct sock *sk);
323 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
324 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
326 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
327 size_t size, int flags);
328 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
329 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
330 const struct tcphdr *th, unsigned int len);
331 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
332 const struct tcphdr *th, unsigned int len);
333 extern void tcp_rcv_space_adjust(struct sock *sk);
334 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
335 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
336 extern void tcp_twsk_destructor(struct sock *sk);
337 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
338 struct pipe_inode_info *pipe, size_t len,
341 static inline void tcp_dec_quickack_mode(struct sock *sk,
342 const unsigned int pkts)
344 struct inet_connection_sock *icsk = inet_csk(sk);
346 if (icsk->icsk_ack.quick) {
347 if (pkts >= icsk->icsk_ack.quick) {
348 icsk->icsk_ack.quick = 0;
349 /* Leaving quickack mode we deflate ATO. */
350 icsk->icsk_ack.ato = TCP_ATO_MIN;
352 icsk->icsk_ack.quick -= pkts;
357 #define TCP_ECN_QUEUE_CWR 2
358 #define TCP_ECN_DEMAND_CWR 4
359 #define TCP_ECN_SEEN 8
369 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
371 const struct tcphdr *th);
372 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
373 struct request_sock *req,
374 struct request_sock **prev);
375 extern int tcp_child_process(struct sock *parent, struct sock *child,
376 struct sk_buff *skb);
377 extern int tcp_use_frto(struct sock *sk);
378 extern void tcp_enter_frto(struct sock *sk);
379 extern void tcp_enter_loss(struct sock *sk, int how);
380 extern void tcp_clear_retrans(struct tcp_sock *tp);
381 extern void tcp_update_metrics(struct sock *sk);
382 extern void tcp_close(struct sock *sk, long timeout);
383 extern unsigned int tcp_poll(struct file * file, struct socket *sock,
384 struct poll_table_struct *wait);
385 extern int tcp_getsockopt(struct sock *sk, int level, int optname,
386 char __user *optval, int __user *optlen);
387 extern int tcp_setsockopt(struct sock *sk, int level, int optname,
388 char __user *optval, unsigned int optlen);
389 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
390 char __user *optval, int __user *optlen);
391 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
392 char __user *optval, unsigned int optlen);
393 extern void tcp_set_keepalive(struct sock *sk, int val);
394 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
395 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
396 size_t len, int nonblock, int flags, int *addr_len);
397 extern void tcp_parse_options(const struct sk_buff *skb,
398 struct tcp_options_received *opt_rx, const u8 **hvpp,
400 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
403 * TCP v4 functions exported for the inet6 API
406 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
407 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
408 extern struct sock * tcp_create_openreq_child(struct sock *sk,
409 struct request_sock *req,
410 struct sk_buff *skb);
411 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
412 struct request_sock *req,
413 struct dst_entry *dst);
414 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
415 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
417 extern int tcp_connect(struct sock *sk);
418 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
419 struct request_sock *req,
420 struct request_values *rvp);
421 extern int tcp_disconnect(struct sock *sk, int flags);
424 /* From syncookies.c */
425 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
426 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
427 struct ip_options *opt);
428 #ifdef CONFIG_SYN_COOKIES
429 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
432 static inline __u32 cookie_v4_init_sequence(struct sock *sk,
440 extern __u32 cookie_init_timestamp(struct request_sock *req);
441 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *);
443 /* From net/ipv6/syncookies.c */
444 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
445 #ifdef CONFIG_SYN_COOKIES
446 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
449 static inline __u32 cookie_v6_init_sequence(struct sock *sk,
458 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
460 extern int tcp_may_send_now(struct sock *sk);
461 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
462 extern void tcp_retransmit_timer(struct sock *sk);
463 extern void tcp_xmit_retransmit_queue(struct sock *);
464 extern void tcp_simple_retransmit(struct sock *);
465 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
466 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
468 extern void tcp_send_probe0(struct sock *);
469 extern void tcp_send_partial(struct sock *);
470 extern int tcp_write_wakeup(struct sock *);
471 extern void tcp_send_fin(struct sock *sk);
472 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
473 extern int tcp_send_synack(struct sock *);
474 extern int tcp_syn_flood_action(struct sock *sk,
475 const struct sk_buff *skb,
477 extern void tcp_push_one(struct sock *, unsigned int mss_now);
478 extern void tcp_send_ack(struct sock *sk);
479 extern void tcp_send_delayed_ack(struct sock *sk);
482 extern void tcp_cwnd_application_limited(struct sock *sk);
485 extern void tcp_init_xmit_timers(struct sock *);
486 static inline void tcp_clear_xmit_timers(struct sock *sk)
488 inet_csk_clear_xmit_timers(sk);
491 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
492 extern unsigned int tcp_current_mss(struct sock *sk);
494 /* Bound MSS / TSO packet size with the half of the window */
495 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
499 /* When peer uses tiny windows, there is no use in packetizing
500 * to sub-MSS pieces for the sake of SWS or making sure there
501 * are enough packets in the pipe for fast recovery.
503 * On the other hand, for extremely large MSS devices, handling
504 * smaller than MSS windows in this way does make sense.
506 if (tp->max_window >= 512)
507 cutoff = (tp->max_window >> 1);
509 cutoff = tp->max_window;
511 if (cutoff && pktsize > cutoff)
512 return max_t(int, cutoff, 68U - tp->tcp_header_len);
518 extern void tcp_get_info(const struct sock *, struct tcp_info *);
520 /* Read 'sendfile()'-style from a TCP socket */
521 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
522 unsigned int, size_t);
523 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
524 sk_read_actor_t recv_actor);
526 extern void tcp_initialize_rcv_mss(struct sock *sk);
528 extern int tcp_mtu_to_mss(const struct sock *sk, int pmtu);
529 extern int tcp_mss_to_mtu(const struct sock *sk, int mss);
530 extern void tcp_mtup_init(struct sock *sk);
531 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt);
533 static inline void tcp_bound_rto(const struct sock *sk)
535 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
536 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
539 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
541 return (tp->srtt >> 3) + tp->rttvar;
544 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
546 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
547 ntohl(TCP_FLAG_ACK) |
551 static inline void tcp_fast_path_on(struct tcp_sock *tp)
553 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
556 static inline void tcp_fast_path_check(struct sock *sk)
558 struct tcp_sock *tp = tcp_sk(sk);
560 if (skb_queue_empty(&tp->out_of_order_queue) &&
562 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
564 tcp_fast_path_on(tp);
567 /* Compute the actual rto_min value */
568 static inline u32 tcp_rto_min(struct sock *sk)
570 const struct dst_entry *dst = __sk_dst_get(sk);
571 u32 rto_min = TCP_RTO_MIN;
573 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
574 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
578 /* Compute the actual receive window we are currently advertising.
579 * Rcv_nxt can be after the window if our peer push more data
580 * than the offered window.
582 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
584 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
591 /* Choose a new window, without checks for shrinking, and without
592 * scaling applied to the result. The caller does these things
593 * if necessary. This is a "raw" window selection.
595 extern u32 __tcp_select_window(struct sock *sk);
597 /* TCP timestamps are only 32-bits, this causes a slight
598 * complication on 64-bit systems since we store a snapshot
599 * of jiffies in the buffer control blocks below. We decided
600 * to use only the low 32-bits of jiffies and hide the ugly
601 * casts with the following macro.
603 #define tcp_time_stamp ((__u32)(jiffies))
605 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
607 #define TCPHDR_FIN 0x01
608 #define TCPHDR_SYN 0x02
609 #define TCPHDR_RST 0x04
610 #define TCPHDR_PSH 0x08
611 #define TCPHDR_ACK 0x10
612 #define TCPHDR_URG 0x20
613 #define TCPHDR_ECE 0x40
614 #define TCPHDR_CWR 0x80
616 /* This is what the send packet queuing engine uses to pass
617 * TCP per-packet control information to the transmission code.
618 * We also store the host-order sequence numbers in here too.
619 * This is 44 bytes if IPV6 is enabled.
620 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
624 struct inet_skb_parm h4;
625 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
626 struct inet6_skb_parm h6;
628 } header; /* For incoming frames */
629 __u32 seq; /* Starting sequence number */
630 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
631 __u32 when; /* used to compute rtt's */
632 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
633 __u8 sacked; /* State flags for SACK/FACK. */
634 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
635 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
636 #define TCPCB_LOST 0x04 /* SKB is lost */
637 #define TCPCB_TAGBITS 0x07 /* All tag bits */
638 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
640 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
641 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
643 __u32 ack_seq; /* Sequence number ACK'd */
646 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
648 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
650 * If we receive a SYN packet with these bits set, it means a network is
651 * playing bad games with TOS bits. In order to avoid possible false congestion
652 * notifications, we disable TCP ECN negociation.
655 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb)
657 const struct tcphdr *th = tcp_hdr(skb);
659 if (sysctl_tcp_ecn && th->ece && th->cwr &&
660 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield))
661 inet_rsk(req)->ecn_ok = 1;
664 /* Due to TSO, an SKB can be composed of multiple actual
665 * packets. To keep these tracked properly, we use this.
667 static inline int tcp_skb_pcount(const struct sk_buff *skb)
669 return skb_shinfo(skb)->gso_segs;
672 /* This is valid iff tcp_skb_pcount() > 1. */
673 static inline int tcp_skb_mss(const struct sk_buff *skb)
675 return skb_shinfo(skb)->gso_size;
678 /* Events passed to congestion control interface */
680 CA_EVENT_TX_START, /* first transmit when no packets in flight */
681 CA_EVENT_CWND_RESTART, /* congestion window restart */
682 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
683 CA_EVENT_FRTO, /* fast recovery timeout */
684 CA_EVENT_LOSS, /* loss timeout */
685 CA_EVENT_FAST_ACK, /* in sequence ack */
686 CA_EVENT_SLOW_ACK, /* other ack */
690 * Interface for adding new TCP congestion control handlers
692 #define TCP_CA_NAME_MAX 16
693 #define TCP_CA_MAX 128
694 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
696 #define TCP_CONG_NON_RESTRICTED 0x1
697 #define TCP_CONG_RTT_STAMP 0x2
699 struct tcp_congestion_ops {
700 struct list_head list;
703 /* initialize private data (optional) */
704 void (*init)(struct sock *sk);
705 /* cleanup private data (optional) */
706 void (*release)(struct sock *sk);
708 /* return slow start threshold (required) */
709 u32 (*ssthresh)(struct sock *sk);
710 /* lower bound for congestion window (optional) */
711 u32 (*min_cwnd)(const struct sock *sk);
712 /* do new cwnd calculation (required) */
713 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
714 /* call before changing ca_state (optional) */
715 void (*set_state)(struct sock *sk, u8 new_state);
716 /* call when cwnd event occurs (optional) */
717 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
718 /* new value of cwnd after loss (optional) */
719 u32 (*undo_cwnd)(struct sock *sk);
720 /* hook for packet ack accounting (optional) */
721 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
722 /* get info for inet_diag (optional) */
723 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
725 char name[TCP_CA_NAME_MAX];
726 struct module *owner;
729 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
730 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
732 extern void tcp_init_congestion_control(struct sock *sk);
733 extern void tcp_cleanup_congestion_control(struct sock *sk);
734 extern int tcp_set_default_congestion_control(const char *name);
735 extern void tcp_get_default_congestion_control(char *name);
736 extern void tcp_get_available_congestion_control(char *buf, size_t len);
737 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
738 extern int tcp_set_allowed_congestion_control(char *allowed);
739 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
740 extern void tcp_slow_start(struct tcp_sock *tp);
741 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
743 extern struct tcp_congestion_ops tcp_init_congestion_ops;
744 extern u32 tcp_reno_ssthresh(struct sock *sk);
745 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
746 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
747 extern struct tcp_congestion_ops tcp_reno;
749 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
751 struct inet_connection_sock *icsk = inet_csk(sk);
753 if (icsk->icsk_ca_ops->set_state)
754 icsk->icsk_ca_ops->set_state(sk, ca_state);
755 icsk->icsk_ca_state = ca_state;
758 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
760 const struct inet_connection_sock *icsk = inet_csk(sk);
762 if (icsk->icsk_ca_ops->cwnd_event)
763 icsk->icsk_ca_ops->cwnd_event(sk, event);
766 /* These functions determine how the current flow behaves in respect of SACK
767 * handling. SACK is negotiated with the peer, and therefore it can vary
768 * between different flows.
770 * tcp_is_sack - SACK enabled
771 * tcp_is_reno - No SACK
772 * tcp_is_fack - FACK enabled, implies SACK enabled
774 static inline int tcp_is_sack(const struct tcp_sock *tp)
776 return tp->rx_opt.sack_ok;
779 static inline int tcp_is_reno(const struct tcp_sock *tp)
781 return !tcp_is_sack(tp);
784 static inline int tcp_is_fack(const struct tcp_sock *tp)
786 return tp->rx_opt.sack_ok & 2;
789 static inline void tcp_enable_fack(struct tcp_sock *tp)
791 tp->rx_opt.sack_ok |= 2;
794 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
796 return tp->sacked_out + tp->lost_out;
799 /* This determines how many packets are "in the network" to the best
800 * of our knowledge. In many cases it is conservative, but where
801 * detailed information is available from the receiver (via SACK
802 * blocks etc.) we can make more aggressive calculations.
804 * Use this for decisions involving congestion control, use just
805 * tp->packets_out to determine if the send queue is empty or not.
807 * Read this equation as:
809 * "Packets sent once on transmission queue" MINUS
810 * "Packets left network, but not honestly ACKed yet" PLUS
811 * "Packets fast retransmitted"
813 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
815 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
818 #define TCP_INFINITE_SSTHRESH 0x7fffffff
820 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
822 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
825 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
826 * The exception is rate halving phase, when cwnd is decreasing towards
829 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
831 const struct tcp_sock *tp = tcp_sk(sk);
833 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
834 return tp->snd_ssthresh;
836 return max(tp->snd_ssthresh,
837 ((tp->snd_cwnd >> 1) +
838 (tp->snd_cwnd >> 2)));
841 /* Use define here intentionally to get WARN_ON location shown at the caller */
842 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
844 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
845 extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
847 /* Slow start with delack produces 3 packets of burst, so that
848 * it is safe "de facto". This will be the default - same as
849 * the default reordering threshold - but if reordering increases,
850 * we must be able to allow cwnd to burst at least this much in order
851 * to not pull it back when holes are filled.
853 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
855 return tp->reordering;
858 /* Returns end sequence number of the receiver's advertised window */
859 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
861 return tp->snd_una + tp->snd_wnd;
863 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
865 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
866 const struct sk_buff *skb)
869 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
872 static inline void tcp_check_probe_timer(struct sock *sk)
874 const struct tcp_sock *tp = tcp_sk(sk);
875 const struct inet_connection_sock *icsk = inet_csk(sk);
877 if (!tp->packets_out && !icsk->icsk_pending)
878 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
879 icsk->icsk_rto, TCP_RTO_MAX);
882 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
887 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
893 * Calculate(/check) TCP checksum
895 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
896 __be32 daddr, __wsum base)
898 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
901 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
903 return __skb_checksum_complete(skb);
906 static inline int tcp_checksum_complete(struct sk_buff *skb)
908 return !skb_csum_unnecessary(skb) &&
909 __tcp_checksum_complete(skb);
912 /* Prequeue for VJ style copy to user, combined with checksumming. */
914 static inline void tcp_prequeue_init(struct tcp_sock *tp)
916 tp->ucopy.task = NULL;
918 tp->ucopy.memory = 0;
919 skb_queue_head_init(&tp->ucopy.prequeue);
920 #ifdef CONFIG_NET_DMA
921 tp->ucopy.dma_chan = NULL;
922 tp->ucopy.wakeup = 0;
923 tp->ucopy.pinned_list = NULL;
924 tp->ucopy.dma_cookie = 0;
928 /* Packet is added to VJ-style prequeue for processing in process
929 * context, if a reader task is waiting. Apparently, this exciting
930 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
931 * failed somewhere. Latency? Burstiness? Well, at least now we will
932 * see, why it failed. 8)8) --ANK
934 * NOTE: is this not too big to inline?
936 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
938 struct tcp_sock *tp = tcp_sk(sk);
940 if (sysctl_tcp_low_latency || !tp->ucopy.task)
944 __skb_queue_tail(&tp->ucopy.prequeue, skb);
945 tp->ucopy.memory += skb->truesize;
946 if (tp->ucopy.memory > sk->sk_rcvbuf) {
947 struct sk_buff *skb1;
949 BUG_ON(sock_owned_by_user(sk));
951 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
952 sk_backlog_rcv(sk, skb1);
953 NET_INC_STATS_BH(sock_net(sk),
954 LINUX_MIB_TCPPREQUEUEDROPPED);
957 tp->ucopy.memory = 0;
958 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
959 wake_up_interruptible_sync_poll(sk_sleep(sk),
960 POLLIN | POLLRDNORM | POLLRDBAND);
961 if (!inet_csk_ack_scheduled(sk))
962 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
963 (3 * tcp_rto_min(sk)) / 4,
973 static const char *statename[]={
974 "Unused","Established","Syn Sent","Syn Recv",
975 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
976 "Close Wait","Last ACK","Listen","Closing"
979 extern void tcp_set_state(struct sock *sk, int state);
981 extern void tcp_done(struct sock *sk);
983 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
986 rx_opt->num_sacks = 0;
989 /* Determine a window scaling and initial window to offer. */
990 extern void tcp_select_initial_window(int __space, __u32 mss,
991 __u32 *rcv_wnd, __u32 *window_clamp,
992 int wscale_ok, __u8 *rcv_wscale,
995 static inline int tcp_win_from_space(int space)
997 return sysctl_tcp_adv_win_scale<=0 ?
998 (space>>(-sysctl_tcp_adv_win_scale)) :
999 space - (space>>sysctl_tcp_adv_win_scale);
1002 /* Note: caller must be prepared to deal with negative returns */
1003 static inline int tcp_space(const struct sock *sk)
1005 return tcp_win_from_space(sk->sk_rcvbuf -
1006 atomic_read(&sk->sk_rmem_alloc));
1009 static inline int tcp_full_space(const struct sock *sk)
1011 return tcp_win_from_space(sk->sk_rcvbuf);
1014 static inline void tcp_openreq_init(struct request_sock *req,
1015 struct tcp_options_received *rx_opt,
1016 struct sk_buff *skb)
1018 struct inet_request_sock *ireq = inet_rsk(req);
1020 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1022 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1023 req->mss = rx_opt->mss_clamp;
1024 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1025 ireq->tstamp_ok = rx_opt->tstamp_ok;
1026 ireq->sack_ok = rx_opt->sack_ok;
1027 ireq->snd_wscale = rx_opt->snd_wscale;
1028 ireq->wscale_ok = rx_opt->wscale_ok;
1031 ireq->rmt_port = tcp_hdr(skb)->source;
1032 ireq->loc_port = tcp_hdr(skb)->dest;
1035 extern void tcp_enter_memory_pressure(struct sock *sk);
1037 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1039 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1042 static inline int keepalive_time_when(const struct tcp_sock *tp)
1044 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1047 static inline int keepalive_probes(const struct tcp_sock *tp)
1049 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1052 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1054 const struct inet_connection_sock *icsk = &tp->inet_conn;
1056 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1057 tcp_time_stamp - tp->rcv_tstamp);
1060 static inline int tcp_fin_time(const struct sock *sk)
1062 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1063 const int rto = inet_csk(sk)->icsk_rto;
1065 if (fin_timeout < (rto << 2) - (rto >> 1))
1066 fin_timeout = (rto << 2) - (rto >> 1);
1071 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1074 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1076 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1079 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1080 * then following tcp messages have valid values. Ignore 0 value,
1081 * or else 'negative' tsval might forbid us to accept their packets.
1083 if (!rx_opt->ts_recent)
1088 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1091 if (tcp_paws_check(rx_opt, 0))
1094 /* RST segments are not recommended to carry timestamp,
1095 and, if they do, it is recommended to ignore PAWS because
1096 "their cleanup function should take precedence over timestamps."
1097 Certainly, it is mistake. It is necessary to understand the reasons
1098 of this constraint to relax it: if peer reboots, clock may go
1099 out-of-sync and half-open connections will not be reset.
1100 Actually, the problem would be not existing if all
1101 the implementations followed draft about maintaining clock
1102 via reboots. Linux-2.2 DOES NOT!
1104 However, we can relax time bounds for RST segments to MSL.
1106 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1111 static inline void tcp_mib_init(struct net *net)
1114 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1115 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1116 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1117 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1121 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1123 tp->lost_skb_hint = NULL;
1124 tp->scoreboard_skb_hint = NULL;
1127 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1129 tcp_clear_retrans_hints_partial(tp);
1130 tp->retransmit_skb_hint = NULL;
1136 /* - key database */
1137 struct tcp_md5sig_key {
1142 struct tcp4_md5sig_key {
1143 struct tcp_md5sig_key base;
1147 struct tcp6_md5sig_key {
1148 struct tcp_md5sig_key base;
1150 u32 scope_id; /* XXX */
1152 struct in6_addr addr;
1156 struct tcp_md5sig_info {
1157 struct tcp4_md5sig_key *keys4;
1158 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1159 struct tcp6_md5sig_key *keys6;
1167 /* - pseudo header */
1168 struct tcp4_pseudohdr {
1176 struct tcp6_pseudohdr {
1177 struct in6_addr saddr;
1178 struct in6_addr daddr;
1180 __be32 protocol; /* including padding */
1183 union tcp_md5sum_block {
1184 struct tcp4_pseudohdr ip4;
1185 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1186 struct tcp6_pseudohdr ip6;
1190 /* - pool: digest algorithm, hash description and scratch buffer */
1191 struct tcp_md5sig_pool {
1192 struct hash_desc md5_desc;
1193 union tcp_md5sum_block md5_blk;
1197 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1198 const struct sock *sk,
1199 const struct request_sock *req,
1200 const struct sk_buff *skb);
1201 extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk,
1202 struct sock *addr_sk);
1203 extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey,
1205 extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr);
1207 #ifdef CONFIG_TCP_MD5SIG
1208 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1209 &(struct tcp_md5sig_key) { \
1210 .key = (twsk)->tw_md5_key, \
1211 .keylen = (twsk)->tw_md5_keylen, \
1214 #define tcp_twsk_md5_key(twsk) NULL
1217 extern bool tcp_alloc_md5sig_pool(void);
1219 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1220 static inline void tcp_put_md5sig_pool(void)
1225 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1226 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1227 unsigned header_len);
1228 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1229 const struct tcp_md5sig_key *key);
1231 /* write queue abstraction */
1232 static inline void tcp_write_queue_purge(struct sock *sk)
1234 struct sk_buff *skb;
1236 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1237 sk_wmem_free_skb(sk, skb);
1239 tcp_clear_all_retrans_hints(tcp_sk(sk));
1242 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1244 return skb_peek(&sk->sk_write_queue);
1247 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1249 return skb_peek_tail(&sk->sk_write_queue);
1252 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1253 const struct sk_buff *skb)
1255 return skb_queue_next(&sk->sk_write_queue, skb);
1258 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1259 const struct sk_buff *skb)
1261 return skb_queue_prev(&sk->sk_write_queue, skb);
1264 #define tcp_for_write_queue(skb, sk) \
1265 skb_queue_walk(&(sk)->sk_write_queue, skb)
1267 #define tcp_for_write_queue_from(skb, sk) \
1268 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1270 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1271 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1273 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1275 return sk->sk_send_head;
1278 static inline bool tcp_skb_is_last(const struct sock *sk,
1279 const struct sk_buff *skb)
1281 return skb_queue_is_last(&sk->sk_write_queue, skb);
1284 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1286 if (tcp_skb_is_last(sk, skb))
1287 sk->sk_send_head = NULL;
1289 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1292 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1294 if (sk->sk_send_head == skb_unlinked)
1295 sk->sk_send_head = NULL;
1298 static inline void tcp_init_send_head(struct sock *sk)
1300 sk->sk_send_head = NULL;
1303 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1305 __skb_queue_tail(&sk->sk_write_queue, skb);
1308 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1310 __tcp_add_write_queue_tail(sk, skb);
1312 /* Queue it, remembering where we must start sending. */
1313 if (sk->sk_send_head == NULL) {
1314 sk->sk_send_head = skb;
1316 if (tcp_sk(sk)->highest_sack == NULL)
1317 tcp_sk(sk)->highest_sack = skb;
1321 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1323 __skb_queue_head(&sk->sk_write_queue, skb);
1326 /* Insert buff after skb on the write queue of sk. */
1327 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1328 struct sk_buff *buff,
1331 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1334 /* Insert new before skb on the write queue of sk. */
1335 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1336 struct sk_buff *skb,
1339 __skb_queue_before(&sk->sk_write_queue, skb, new);
1341 if (sk->sk_send_head == skb)
1342 sk->sk_send_head = new;
1345 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1347 __skb_unlink(skb, &sk->sk_write_queue);
1350 static inline int tcp_write_queue_empty(struct sock *sk)
1352 return skb_queue_empty(&sk->sk_write_queue);
1355 static inline void tcp_push_pending_frames(struct sock *sk)
1357 if (tcp_send_head(sk)) {
1358 struct tcp_sock *tp = tcp_sk(sk);
1360 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1364 /* Start sequence of the highest skb with SACKed bit, valid only if
1365 * sacked > 0 or when the caller has ensured validity by itself.
1367 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1369 if (!tp->sacked_out)
1372 if (tp->highest_sack == NULL)
1375 return TCP_SKB_CB(tp->highest_sack)->seq;
1378 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1380 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1381 tcp_write_queue_next(sk, skb);
1384 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1386 return tcp_sk(sk)->highest_sack;
1389 static inline void tcp_highest_sack_reset(struct sock *sk)
1391 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1394 /* Called when old skb is about to be deleted (to be combined with new skb) */
1395 static inline void tcp_highest_sack_combine(struct sock *sk,
1396 struct sk_buff *old,
1397 struct sk_buff *new)
1399 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1400 tcp_sk(sk)->highest_sack = new;
1403 /* Determines whether this is a thin stream (which may suffer from
1404 * increased latency). Used to trigger latency-reducing mechanisms.
1406 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp)
1408 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1412 enum tcp_seq_states {
1413 TCP_SEQ_STATE_LISTENING,
1414 TCP_SEQ_STATE_OPENREQ,
1415 TCP_SEQ_STATE_ESTABLISHED,
1416 TCP_SEQ_STATE_TIME_WAIT,
1419 int tcp_seq_open(struct inode *inode, struct file *file);
1421 struct tcp_seq_afinfo {
1424 const struct file_operations *seq_fops;
1425 struct seq_operations seq_ops;
1428 struct tcp_iter_state {
1429 struct seq_net_private p;
1431 enum tcp_seq_states state;
1432 struct sock *syn_wait_sk;
1433 int bucket, offset, sbucket, num, uid;
1437 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1438 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1440 extern struct request_sock_ops tcp_request_sock_ops;
1441 extern struct request_sock_ops tcp6_request_sock_ops;
1443 extern void tcp_v4_destroy_sock(struct sock *sk);
1445 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1446 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, u32 features);
1447 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1448 struct sk_buff *skb);
1449 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1450 struct sk_buff *skb);
1451 extern int tcp_gro_complete(struct sk_buff *skb);
1452 extern int tcp4_gro_complete(struct sk_buff *skb);
1454 #ifdef CONFIG_PROC_FS
1455 extern int tcp4_proc_init(void);
1456 extern void tcp4_proc_exit(void);
1459 /* TCP af-specific functions */
1460 struct tcp_sock_af_ops {
1461 #ifdef CONFIG_TCP_MD5SIG
1462 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1463 struct sock *addr_sk);
1464 int (*calc_md5_hash) (char *location,
1465 struct tcp_md5sig_key *md5,
1466 const struct sock *sk,
1467 const struct request_sock *req,
1468 const struct sk_buff *skb);
1469 int (*md5_add) (struct sock *sk,
1470 struct sock *addr_sk,
1473 int (*md5_parse) (struct sock *sk,
1474 char __user *optval,
1479 struct tcp_request_sock_ops {
1480 #ifdef CONFIG_TCP_MD5SIG
1481 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1482 struct request_sock *req);
1483 int (*calc_md5_hash) (char *location,
1484 struct tcp_md5sig_key *md5,
1485 const struct sock *sk,
1486 const struct request_sock *req,
1487 const struct sk_buff *skb);
1491 /* Using SHA1 for now, define some constants.
1493 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1494 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1495 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1497 extern int tcp_cookie_generator(u32 *bakery);
1500 * struct tcp_cookie_values - each socket needs extra space for the
1501 * cookies, together with (optional) space for any SYN data.
1503 * A tcp_sock contains a pointer to the current value, and this is
1504 * cloned to the tcp_timewait_sock.
1506 * @cookie_pair: variable data from the option exchange.
1508 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1509 * indicates default (sysctl_tcp_cookie_size).
1510 * After cookie sent, remembers size of cookie.
1511 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1513 * @s_data_desired: user specified tcpct_s_data_desired. When the
1514 * constant payload is specified (@s_data_constant),
1515 * holds its length instead.
1516 * Range 0 to TCP_MSS_DESIRED.
1518 * @s_data_payload: constant data that is to be included in the
1519 * payload of SYN or SYNACK segments when the
1520 * cookie option is present.
1522 struct tcp_cookie_values {
1524 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE];
1525 u8 cookie_pair_size;
1527 u16 s_data_desired:11,
1532 u8 s_data_payload[0];
1535 static inline void tcp_cookie_values_release(struct kref *kref)
1537 kfree(container_of(kref, struct tcp_cookie_values, kref));
1540 /* The length of constant payload data. Note that s_data_desired is
1541 * overloaded, depending on s_data_constant: either the length of constant
1542 * data (returned here) or the limit on variable data.
1544 static inline int tcp_s_data_size(const struct tcp_sock *tp)
1546 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1547 ? tp->cookie_values->s_data_desired
1552 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1554 * As tcp_request_sock has already been extended in other places, the
1555 * only remaining method is to pass stack values along as function
1556 * parameters. These parameters are not needed after sending SYNACK.
1558 * @cookie_bakery: cryptographic secret and message workspace.
1560 * @cookie_plus: bytes in authenticator/cookie option, copied from
1561 * struct tcp_options_received (above).
1563 struct tcp_extend_values {
1564 struct request_values rv;
1565 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS];
1571 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1573 return (struct tcp_extend_values *)rvp;
1576 extern void tcp_v4_init(void);
1577 extern void tcp_init(void);