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>
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 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
63 #define TCP_MIN_MSS 88U
65 /* Minimal RCV_MSS. */
66 #define TCP_MIN_RCVMSS 536U
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 */
106 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
107 * socket. 7 is ~50sec-16min.
111 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
112 * state, about 60 seconds */
113 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
114 /* BSD style FIN_WAIT2 deadlock breaker.
115 * It used to be 3min, new value is 60sec,
116 * to combine FIN-WAIT-2 timeout with
120 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
122 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
123 #define TCP_ATO_MIN ((unsigned)(HZ/25))
125 #define TCP_DELACK_MIN 4U
126 #define TCP_ATO_MIN 4U
128 #define TCP_RTO_MAX ((unsigned)(120*HZ))
129 #define TCP_RTO_MIN ((unsigned)(HZ/5))
130 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
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) */
175 #define TCPOLEN_MSS 4
176 #define TCPOLEN_WINDOW 3
177 #define TCPOLEN_SACK_PERM 2
178 #define TCPOLEN_TIMESTAMP 10
179 #define TCPOLEN_MD5SIG 18
181 /* But this is what stacks really send out. */
182 #define TCPOLEN_TSTAMP_ALIGNED 12
183 #define TCPOLEN_WSCALE_ALIGNED 4
184 #define TCPOLEN_SACKPERM_ALIGNED 4
185 #define TCPOLEN_SACK_BASE 2
186 #define TCPOLEN_SACK_BASE_ALIGNED 4
187 #define TCPOLEN_SACK_PERBLOCK 8
188 #define TCPOLEN_MD5SIG_ALIGNED 20
189 #define TCPOLEN_MSS_ALIGNED 4
191 /* Flags in tp->nonagle */
192 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
193 #define TCP_NAGLE_CORK 2 /* Socket is corked */
194 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
196 extern struct inet_timewait_death_row tcp_death_row;
198 /* sysctl variables for tcp */
199 extern int sysctl_tcp_timestamps;
200 extern int sysctl_tcp_window_scaling;
201 extern int sysctl_tcp_sack;
202 extern int sysctl_tcp_fin_timeout;
203 extern int sysctl_tcp_keepalive_time;
204 extern int sysctl_tcp_keepalive_probes;
205 extern int sysctl_tcp_keepalive_intvl;
206 extern int sysctl_tcp_syn_retries;
207 extern int sysctl_tcp_synack_retries;
208 extern int sysctl_tcp_retries1;
209 extern int sysctl_tcp_retries2;
210 extern int sysctl_tcp_orphan_retries;
211 extern int sysctl_tcp_syncookies;
212 extern int sysctl_tcp_retrans_collapse;
213 extern int sysctl_tcp_stdurg;
214 extern int sysctl_tcp_rfc1337;
215 extern int sysctl_tcp_abort_on_overflow;
216 extern int sysctl_tcp_max_orphans;
217 extern int sysctl_tcp_fack;
218 extern int sysctl_tcp_reordering;
219 extern int sysctl_tcp_ecn;
220 extern int sysctl_tcp_dsack;
221 extern int sysctl_tcp_mem[3];
222 extern int sysctl_tcp_wmem[3];
223 extern int sysctl_tcp_rmem[3];
224 extern int sysctl_tcp_app_win;
225 extern int sysctl_tcp_adv_win_scale;
226 extern int sysctl_tcp_tw_reuse;
227 extern int sysctl_tcp_frto;
228 extern int sysctl_tcp_frto_response;
229 extern int sysctl_tcp_low_latency;
230 extern int sysctl_tcp_dma_copybreak;
231 extern int sysctl_tcp_nometrics_save;
232 extern int sysctl_tcp_moderate_rcvbuf;
233 extern int sysctl_tcp_tso_win_divisor;
234 extern int sysctl_tcp_abc;
235 extern int sysctl_tcp_mtu_probing;
236 extern int sysctl_tcp_base_mss;
237 extern int sysctl_tcp_workaround_signed_windows;
238 extern int sysctl_tcp_slow_start_after_idle;
239 extern int sysctl_tcp_max_ssthresh;
241 extern atomic_t tcp_memory_allocated;
242 extern struct percpu_counter tcp_sockets_allocated;
243 extern int tcp_memory_pressure;
246 * The next routines deal with comparing 32 bit unsigned ints
247 * and worry about wraparound (automatic with unsigned arithmetic).
250 static inline int before(__u32 seq1, __u32 seq2)
252 return (__s32)(seq1-seq2) < 0;
254 #define after(seq2, seq1) before(seq1, seq2)
256 /* is s2<=s1<=s3 ? */
257 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
259 return seq3 - seq2 >= seq1 - seq2;
262 static inline int tcp_too_many_orphans(struct sock *sk, int num)
264 return (num > sysctl_tcp_max_orphans) ||
265 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
266 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
269 /* syncookies: remember time of last synqueue overflow */
270 static inline void tcp_synq_overflow(struct sock *sk)
272 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
275 /* syncookies: no recent synqueue overflow on this listening socket? */
276 static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
278 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
279 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT);
282 extern struct proto tcp_prot;
284 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
285 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
286 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
287 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
289 extern void tcp_v4_err(struct sk_buff *skb, u32);
291 extern void tcp_shutdown (struct sock *sk, int how);
293 extern int tcp_v4_rcv(struct sk_buff *skb);
295 extern int tcp_v4_remember_stamp(struct sock *sk);
297 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
299 extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
300 struct msghdr *msg, size_t size);
301 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
303 extern int tcp_ioctl(struct sock *sk,
307 extern int tcp_rcv_state_process(struct sock *sk,
312 extern int tcp_rcv_established(struct sock *sk,
317 extern void tcp_rcv_space_adjust(struct sock *sk);
319 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
321 extern int tcp_twsk_unique(struct sock *sk,
322 struct sock *sktw, void *twp);
324 extern void tcp_twsk_destructor(struct sock *sk);
326 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
327 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
329 static inline void tcp_dec_quickack_mode(struct sock *sk,
330 const unsigned int pkts)
332 struct inet_connection_sock *icsk = inet_csk(sk);
334 if (icsk->icsk_ack.quick) {
335 if (pkts >= icsk->icsk_ack.quick) {
336 icsk->icsk_ack.quick = 0;
337 /* Leaving quickack mode we deflate ATO. */
338 icsk->icsk_ack.ato = TCP_ATO_MIN;
340 icsk->icsk_ack.quick -= pkts;
344 extern void tcp_enter_quickack_mode(struct sock *sk);
346 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
348 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
352 #define TCP_ECN_QUEUE_CWR 2
353 #define TCP_ECN_DEMAND_CWR 4
355 static __inline__ void
356 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
358 if (sysctl_tcp_ecn && th->ece && th->cwr)
359 inet_rsk(req)->ecn_ok = 1;
371 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
373 const struct tcphdr *th);
375 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
376 struct request_sock *req,
377 struct request_sock **prev);
378 extern int tcp_child_process(struct sock *parent,
380 struct sk_buff *skb);
381 extern int tcp_use_frto(struct sock *sk);
382 extern void tcp_enter_frto(struct sock *sk);
383 extern void tcp_enter_loss(struct sock *sk, int how);
384 extern void tcp_clear_retrans(struct tcp_sock *tp);
385 extern void tcp_update_metrics(struct sock *sk);
387 extern void tcp_close(struct sock *sk,
389 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
391 extern int tcp_getsockopt(struct sock *sk, int level,
395 extern int tcp_setsockopt(struct sock *sk, int level,
396 int optname, char __user *optval,
398 extern int compat_tcp_getsockopt(struct sock *sk,
399 int level, int optname,
400 char __user *optval, int __user *optlen);
401 extern int compat_tcp_setsockopt(struct sock *sk,
402 int level, int optname,
403 char __user *optval, int optlen);
404 extern void tcp_set_keepalive(struct sock *sk, int val);
405 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
407 size_t len, int nonblock,
408 int flags, int *addr_len);
410 extern void tcp_parse_options(struct sk_buff *skb,
411 struct tcp_options_received *opt_rx,
414 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
417 * TCP v4 functions exported for the inet6 API
420 extern void tcp_v4_send_check(struct sock *sk, int len,
421 struct sk_buff *skb);
423 extern int tcp_v4_conn_request(struct sock *sk,
424 struct sk_buff *skb);
426 extern struct sock * tcp_create_openreq_child(struct sock *sk,
427 struct request_sock *req,
428 struct sk_buff *skb);
430 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
432 struct request_sock *req,
433 struct dst_entry *dst);
435 extern int tcp_v4_do_rcv(struct sock *sk,
436 struct sk_buff *skb);
438 extern int tcp_v4_connect(struct sock *sk,
439 struct sockaddr *uaddr,
442 extern int tcp_connect(struct sock *sk);
444 extern struct sk_buff * tcp_make_synack(struct sock *sk,
445 struct dst_entry *dst,
446 struct request_sock *req);
448 extern int tcp_disconnect(struct sock *sk, int flags);
451 /* From syncookies.c */
452 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
453 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
454 struct ip_options *opt);
455 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
458 extern __u32 cookie_init_timestamp(struct request_sock *req);
459 extern void cookie_check_timestamp(struct tcp_options_received *tcp_opt);
461 /* From net/ipv6/syncookies.c */
462 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
463 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
468 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
470 extern int tcp_may_send_now(struct sock *sk);
471 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
472 extern void tcp_xmit_retransmit_queue(struct sock *);
473 extern void tcp_simple_retransmit(struct sock *);
474 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
475 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
477 extern void tcp_send_probe0(struct sock *);
478 extern void tcp_send_partial(struct sock *);
479 extern int tcp_write_wakeup(struct sock *);
480 extern void tcp_send_fin(struct sock *sk);
481 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
482 extern int tcp_send_synack(struct sock *);
483 extern void tcp_push_one(struct sock *, unsigned int mss_now);
484 extern void tcp_send_ack(struct sock *sk);
485 extern void tcp_send_delayed_ack(struct sock *sk);
488 extern void tcp_cwnd_application_limited(struct sock *sk);
491 extern void tcp_init_xmit_timers(struct sock *);
492 static inline void tcp_clear_xmit_timers(struct sock *sk)
494 inet_csk_clear_xmit_timers(sk);
497 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
498 extern unsigned int tcp_current_mss(struct sock *sk);
500 /* Bound MSS / TSO packet size with the half of the window */
501 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
503 if (tp->max_window && pktsize > (tp->max_window >> 1))
504 return max(tp->max_window >> 1, 68U - tp->tcp_header_len);
510 extern void tcp_get_info(struct sock *, struct tcp_info *);
512 /* Read 'sendfile()'-style from a TCP socket */
513 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
514 unsigned int, size_t);
515 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
516 sk_read_actor_t recv_actor);
518 extern void tcp_initialize_rcv_mss(struct sock *sk);
520 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
521 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
522 extern void tcp_mtup_init(struct sock *sk);
524 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
526 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
527 ntohl(TCP_FLAG_ACK) |
531 static inline void tcp_fast_path_on(struct tcp_sock *tp)
533 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
536 static inline void tcp_fast_path_check(struct sock *sk)
538 struct tcp_sock *tp = tcp_sk(sk);
540 if (skb_queue_empty(&tp->out_of_order_queue) &&
542 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
544 tcp_fast_path_on(tp);
547 /* Compute the actual rto_min value */
548 static inline u32 tcp_rto_min(struct sock *sk)
550 struct dst_entry *dst = __sk_dst_get(sk);
551 u32 rto_min = TCP_RTO_MIN;
553 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
554 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
558 /* Compute the actual receive window we are currently advertising.
559 * Rcv_nxt can be after the window if our peer push more data
560 * than the offered window.
562 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
564 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
571 /* Choose a new window, without checks for shrinking, and without
572 * scaling applied to the result. The caller does these things
573 * if necessary. This is a "raw" window selection.
575 extern u32 __tcp_select_window(struct sock *sk);
577 /* TCP timestamps are only 32-bits, this causes a slight
578 * complication on 64-bit systems since we store a snapshot
579 * of jiffies in the buffer control blocks below. We decided
580 * to use only the low 32-bits of jiffies and hide the ugly
581 * casts with the following macro.
583 #define tcp_time_stamp ((__u32)(jiffies))
585 /* This is what the send packet queuing engine uses to pass
586 * TCP per-packet control information to the transmission
587 * code. We also store the host-order sequence numbers in
588 * here too. This is 36 bytes on 32-bit architectures,
589 * 40 bytes on 64-bit machines, if this grows please adjust
590 * skbuff.h:skbuff->cb[xxx] size appropriately.
594 struct inet_skb_parm h4;
595 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
596 struct inet6_skb_parm h6;
598 } header; /* For incoming frames */
599 __u32 seq; /* Starting sequence number */
600 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
601 __u32 when; /* used to compute rtt's */
602 __u8 flags; /* TCP header flags. */
604 /* NOTE: These must match up to the flags byte in a
607 #define TCPCB_FLAG_FIN 0x01
608 #define TCPCB_FLAG_SYN 0x02
609 #define TCPCB_FLAG_RST 0x04
610 #define TCPCB_FLAG_PSH 0x08
611 #define TCPCB_FLAG_ACK 0x10
612 #define TCPCB_FLAG_URG 0x20
613 #define TCPCB_FLAG_ECE 0x40
614 #define TCPCB_FLAG_CWR 0x80
616 __u8 sacked; /* State flags for SACK/FACK. */
617 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
618 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
619 #define TCPCB_LOST 0x04 /* SKB is lost */
620 #define TCPCB_TAGBITS 0x07 /* All tag bits */
622 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
623 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
625 __u32 ack_seq; /* Sequence number ACK'd */
628 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
630 /* Due to TSO, an SKB can be composed of multiple actual
631 * packets. To keep these tracked properly, we use this.
633 static inline int tcp_skb_pcount(const struct sk_buff *skb)
635 return skb_shinfo(skb)->gso_segs;
638 /* This is valid iff tcp_skb_pcount() > 1. */
639 static inline int tcp_skb_mss(const struct sk_buff *skb)
641 return skb_shinfo(skb)->gso_size;
644 /* Events passed to congestion control interface */
646 CA_EVENT_TX_START, /* first transmit when no packets in flight */
647 CA_EVENT_CWND_RESTART, /* congestion window restart */
648 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
649 CA_EVENT_FRTO, /* fast recovery timeout */
650 CA_EVENT_LOSS, /* loss timeout */
651 CA_EVENT_FAST_ACK, /* in sequence ack */
652 CA_EVENT_SLOW_ACK, /* other ack */
656 * Interface for adding new TCP congestion control handlers
658 #define TCP_CA_NAME_MAX 16
659 #define TCP_CA_MAX 128
660 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
662 #define TCP_CONG_NON_RESTRICTED 0x1
663 #define TCP_CONG_RTT_STAMP 0x2
665 struct tcp_congestion_ops {
666 struct list_head list;
669 /* initialize private data (optional) */
670 void (*init)(struct sock *sk);
671 /* cleanup private data (optional) */
672 void (*release)(struct sock *sk);
674 /* return slow start threshold (required) */
675 u32 (*ssthresh)(struct sock *sk);
676 /* lower bound for congestion window (optional) */
677 u32 (*min_cwnd)(const struct sock *sk);
678 /* do new cwnd calculation (required) */
679 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
680 /* call before changing ca_state (optional) */
681 void (*set_state)(struct sock *sk, u8 new_state);
682 /* call when cwnd event occurs (optional) */
683 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
684 /* new value of cwnd after loss (optional) */
685 u32 (*undo_cwnd)(struct sock *sk);
686 /* hook for packet ack accounting (optional) */
687 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
688 /* get info for inet_diag (optional) */
689 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
691 char name[TCP_CA_NAME_MAX];
692 struct module *owner;
695 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
696 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
698 extern void tcp_init_congestion_control(struct sock *sk);
699 extern void tcp_cleanup_congestion_control(struct sock *sk);
700 extern int tcp_set_default_congestion_control(const char *name);
701 extern void tcp_get_default_congestion_control(char *name);
702 extern void tcp_get_available_congestion_control(char *buf, size_t len);
703 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
704 extern int tcp_set_allowed_congestion_control(char *allowed);
705 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
706 extern void tcp_slow_start(struct tcp_sock *tp);
707 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
709 extern struct tcp_congestion_ops tcp_init_congestion_ops;
710 extern u32 tcp_reno_ssthresh(struct sock *sk);
711 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
712 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
713 extern struct tcp_congestion_ops tcp_reno;
715 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
717 struct inet_connection_sock *icsk = inet_csk(sk);
719 if (icsk->icsk_ca_ops->set_state)
720 icsk->icsk_ca_ops->set_state(sk, ca_state);
721 icsk->icsk_ca_state = ca_state;
724 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
726 const struct inet_connection_sock *icsk = inet_csk(sk);
728 if (icsk->icsk_ca_ops->cwnd_event)
729 icsk->icsk_ca_ops->cwnd_event(sk, event);
732 /* These functions determine how the current flow behaves in respect of SACK
733 * handling. SACK is negotiated with the peer, and therefore it can vary
734 * between different flows.
736 * tcp_is_sack - SACK enabled
737 * tcp_is_reno - No SACK
738 * tcp_is_fack - FACK enabled, implies SACK enabled
740 static inline int tcp_is_sack(const struct tcp_sock *tp)
742 return tp->rx_opt.sack_ok;
745 static inline int tcp_is_reno(const struct tcp_sock *tp)
747 return !tcp_is_sack(tp);
750 static inline int tcp_is_fack(const struct tcp_sock *tp)
752 return tp->rx_opt.sack_ok & 2;
755 static inline void tcp_enable_fack(struct tcp_sock *tp)
757 tp->rx_opt.sack_ok |= 2;
760 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
762 return tp->sacked_out + tp->lost_out;
765 /* This determines how many packets are "in the network" to the best
766 * of our knowledge. In many cases it is conservative, but where
767 * detailed information is available from the receiver (via SACK
768 * blocks etc.) we can make more aggressive calculations.
770 * Use this for decisions involving congestion control, use just
771 * tp->packets_out to determine if the send queue is empty or not.
773 * Read this equation as:
775 * "Packets sent once on transmission queue" MINUS
776 * "Packets left network, but not honestly ACKed yet" PLUS
777 * "Packets fast retransmitted"
779 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
781 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
784 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
785 * The exception is rate halving phase, when cwnd is decreasing towards
788 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
790 const struct tcp_sock *tp = tcp_sk(sk);
791 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
792 return tp->snd_ssthresh;
794 return max(tp->snd_ssthresh,
795 ((tp->snd_cwnd >> 1) +
796 (tp->snd_cwnd >> 2)));
799 /* Use define here intentionally to get WARN_ON location shown at the caller */
800 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
802 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
803 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
805 /* Slow start with delack produces 3 packets of burst, so that
806 * it is safe "de facto". This will be the default - same as
807 * the default reordering threshold - but if reordering increases,
808 * we must be able to allow cwnd to burst at least this much in order
809 * to not pull it back when holes are filled.
811 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
813 return tp->reordering;
816 /* Returns end sequence number of the receiver's advertised window */
817 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
819 return tp->snd_una + tp->snd_wnd;
821 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
823 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
824 const struct sk_buff *skb)
827 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
830 static inline void tcp_check_probe_timer(struct sock *sk)
832 struct tcp_sock *tp = tcp_sk(sk);
833 const struct inet_connection_sock *icsk = inet_csk(sk);
835 if (!tp->packets_out && !icsk->icsk_pending)
836 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
837 icsk->icsk_rto, TCP_RTO_MAX);
840 static inline void tcp_push_pending_frames(struct sock *sk)
842 struct tcp_sock *tp = tcp_sk(sk);
844 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
847 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
852 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
858 * Calculate(/check) TCP checksum
860 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
861 __be32 daddr, __wsum base)
863 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
866 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
868 return __skb_checksum_complete(skb);
871 static inline int tcp_checksum_complete(struct sk_buff *skb)
873 return !skb_csum_unnecessary(skb) &&
874 __tcp_checksum_complete(skb);
877 /* Prequeue for VJ style copy to user, combined with checksumming. */
879 static inline void tcp_prequeue_init(struct tcp_sock *tp)
881 tp->ucopy.task = NULL;
883 tp->ucopy.memory = 0;
884 skb_queue_head_init(&tp->ucopy.prequeue);
885 #ifdef CONFIG_NET_DMA
886 tp->ucopy.dma_chan = NULL;
887 tp->ucopy.wakeup = 0;
888 tp->ucopy.pinned_list = NULL;
889 tp->ucopy.dma_cookie = 0;
893 /* Packet is added to VJ-style prequeue for processing in process
894 * context, if a reader task is waiting. Apparently, this exciting
895 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
896 * failed somewhere. Latency? Burstiness? Well, at least now we will
897 * see, why it failed. 8)8) --ANK
899 * NOTE: is this not too big to inline?
901 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
903 struct tcp_sock *tp = tcp_sk(sk);
905 if (sysctl_tcp_low_latency || !tp->ucopy.task)
908 __skb_queue_tail(&tp->ucopy.prequeue, skb);
909 tp->ucopy.memory += skb->truesize;
910 if (tp->ucopy.memory > sk->sk_rcvbuf) {
911 struct sk_buff *skb1;
913 BUG_ON(sock_owned_by_user(sk));
915 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
916 sk_backlog_rcv(sk, skb1);
917 NET_INC_STATS_BH(sock_net(sk),
918 LINUX_MIB_TCPPREQUEUEDROPPED);
921 tp->ucopy.memory = 0;
922 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
923 wake_up_interruptible_poll(sk->sk_sleep,
924 POLLIN | POLLRDNORM | POLLRDBAND);
925 if (!inet_csk_ack_scheduled(sk))
926 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
927 (3 * tcp_rto_min(sk)) / 4,
937 static const char *statename[]={
938 "Unused","Established","Syn Sent","Syn Recv",
939 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
940 "Close Wait","Last ACK","Listen","Closing"
943 extern void tcp_set_state(struct sock *sk, int state);
945 extern void tcp_done(struct sock *sk);
947 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
950 rx_opt->num_sacks = 0;
953 /* Determine a window scaling and initial window to offer. */
954 extern void tcp_select_initial_window(int __space, __u32 mss,
955 __u32 *rcv_wnd, __u32 *window_clamp,
956 int wscale_ok, __u8 *rcv_wscale);
958 static inline int tcp_win_from_space(int space)
960 return sysctl_tcp_adv_win_scale<=0 ?
961 (space>>(-sysctl_tcp_adv_win_scale)) :
962 space - (space>>sysctl_tcp_adv_win_scale);
965 /* Note: caller must be prepared to deal with negative returns */
966 static inline int tcp_space(const struct sock *sk)
968 return tcp_win_from_space(sk->sk_rcvbuf -
969 atomic_read(&sk->sk_rmem_alloc));
972 static inline int tcp_full_space(const struct sock *sk)
974 return tcp_win_from_space(sk->sk_rcvbuf);
977 static inline void tcp_openreq_init(struct request_sock *req,
978 struct tcp_options_received *rx_opt,
981 struct inet_request_sock *ireq = inet_rsk(req);
983 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
985 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
986 req->mss = rx_opt->mss_clamp;
987 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
988 ireq->tstamp_ok = rx_opt->tstamp_ok;
989 ireq->sack_ok = rx_opt->sack_ok;
990 ireq->snd_wscale = rx_opt->snd_wscale;
991 ireq->wscale_ok = rx_opt->wscale_ok;
994 ireq->rmt_port = tcp_hdr(skb)->source;
995 ireq->loc_port = tcp_hdr(skb)->dest;
998 extern void tcp_enter_memory_pressure(struct sock *sk);
1000 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1002 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1005 static inline int keepalive_time_when(const struct tcp_sock *tp)
1007 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1010 static inline int tcp_fin_time(const struct sock *sk)
1012 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1013 const int rto = inet_csk(sk)->icsk_rto;
1015 if (fin_timeout < (rto << 2) - (rto >> 1))
1016 fin_timeout = (rto << 2) - (rto >> 1);
1021 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1024 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1026 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1032 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1035 if (tcp_paws_check(rx_opt, 0))
1038 /* RST segments are not recommended to carry timestamp,
1039 and, if they do, it is recommended to ignore PAWS because
1040 "their cleanup function should take precedence over timestamps."
1041 Certainly, it is mistake. It is necessary to understand the reasons
1042 of this constraint to relax it: if peer reboots, clock may go
1043 out-of-sync and half-open connections will not be reset.
1044 Actually, the problem would be not existing if all
1045 the implementations followed draft about maintaining clock
1046 via reboots. Linux-2.2 DOES NOT!
1048 However, we can relax time bounds for RST segments to MSL.
1050 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1055 #define TCP_CHECK_TIMER(sk) do { } while (0)
1057 static inline void tcp_mib_init(struct net *net)
1060 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1061 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1062 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1063 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1067 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1069 tp->lost_skb_hint = NULL;
1070 tp->scoreboard_skb_hint = NULL;
1073 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1075 tcp_clear_retrans_hints_partial(tp);
1076 tp->retransmit_skb_hint = NULL;
1082 /* - key database */
1083 struct tcp_md5sig_key {
1088 struct tcp4_md5sig_key {
1089 struct tcp_md5sig_key base;
1093 struct tcp6_md5sig_key {
1094 struct tcp_md5sig_key base;
1096 u32 scope_id; /* XXX */
1098 struct in6_addr addr;
1102 struct tcp_md5sig_info {
1103 struct tcp4_md5sig_key *keys4;
1104 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1105 struct tcp6_md5sig_key *keys6;
1113 /* - pseudo header */
1114 struct tcp4_pseudohdr {
1122 struct tcp6_pseudohdr {
1123 struct in6_addr saddr;
1124 struct in6_addr daddr;
1126 __be32 protocol; /* including padding */
1129 union tcp_md5sum_block {
1130 struct tcp4_pseudohdr ip4;
1131 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1132 struct tcp6_pseudohdr ip6;
1136 /* - pool: digest algorithm, hash description and scratch buffer */
1137 struct tcp_md5sig_pool {
1138 struct hash_desc md5_desc;
1139 union tcp_md5sum_block md5_blk;
1142 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1145 extern int tcp_v4_md5_hash_skb(char *md5_hash,
1146 struct tcp_md5sig_key *key,
1148 struct request_sock *req,
1149 struct sk_buff *skb);
1151 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1152 struct sock *addr_sk);
1154 extern int tcp_v4_md5_do_add(struct sock *sk,
1159 extern int tcp_v4_md5_do_del(struct sock *sk,
1162 #ifdef CONFIG_TCP_MD5SIG
1163 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1164 &(struct tcp_md5sig_key) { \
1165 .key = (twsk)->tw_md5_key, \
1166 .keylen = (twsk)->tw_md5_keylen, \
1169 #define tcp_twsk_md5_key(twsk) NULL
1172 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1173 extern void tcp_free_md5sig_pool(void);
1175 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1176 extern void __tcp_put_md5sig_pool(void);
1177 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1178 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1179 unsigned header_len);
1180 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1181 struct tcp_md5sig_key *key);
1184 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1186 int cpu = get_cpu();
1187 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1193 static inline void tcp_put_md5sig_pool(void)
1195 __tcp_put_md5sig_pool();
1199 /* write queue abstraction */
1200 static inline void tcp_write_queue_purge(struct sock *sk)
1202 struct sk_buff *skb;
1204 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1205 sk_wmem_free_skb(sk, skb);
1209 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1211 return skb_peek(&sk->sk_write_queue);
1214 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1216 return skb_peek_tail(&sk->sk_write_queue);
1219 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1221 return skb_queue_next(&sk->sk_write_queue, skb);
1224 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb)
1226 return skb_queue_prev(&sk->sk_write_queue, skb);
1229 #define tcp_for_write_queue(skb, sk) \
1230 skb_queue_walk(&(sk)->sk_write_queue, skb)
1232 #define tcp_for_write_queue_from(skb, sk) \
1233 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1235 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1236 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1238 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1240 return sk->sk_send_head;
1243 static inline bool tcp_skb_is_last(const struct sock *sk,
1244 const struct sk_buff *skb)
1246 return skb_queue_is_last(&sk->sk_write_queue, skb);
1249 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1251 if (tcp_skb_is_last(sk, skb))
1252 sk->sk_send_head = NULL;
1254 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1257 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1259 if (sk->sk_send_head == skb_unlinked)
1260 sk->sk_send_head = NULL;
1263 static inline void tcp_init_send_head(struct sock *sk)
1265 sk->sk_send_head = NULL;
1268 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1270 __skb_queue_tail(&sk->sk_write_queue, skb);
1273 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1275 __tcp_add_write_queue_tail(sk, skb);
1277 /* Queue it, remembering where we must start sending. */
1278 if (sk->sk_send_head == NULL) {
1279 sk->sk_send_head = skb;
1281 if (tcp_sk(sk)->highest_sack == NULL)
1282 tcp_sk(sk)->highest_sack = skb;
1286 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1288 __skb_queue_head(&sk->sk_write_queue, skb);
1291 /* Insert buff after skb on the write queue of sk. */
1292 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1293 struct sk_buff *buff,
1296 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1299 /* Insert new before skb on the write queue of sk. */
1300 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1301 struct sk_buff *skb,
1304 __skb_queue_before(&sk->sk_write_queue, skb, new);
1306 if (sk->sk_send_head == skb)
1307 sk->sk_send_head = new;
1310 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1312 __skb_unlink(skb, &sk->sk_write_queue);
1315 static inline int tcp_write_queue_empty(struct sock *sk)
1317 return skb_queue_empty(&sk->sk_write_queue);
1320 /* Start sequence of the highest skb with SACKed bit, valid only if
1321 * sacked > 0 or when the caller has ensured validity by itself.
1323 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1325 if (!tp->sacked_out)
1328 if (tp->highest_sack == NULL)
1331 return TCP_SKB_CB(tp->highest_sack)->seq;
1334 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1336 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1337 tcp_write_queue_next(sk, skb);
1340 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1342 return tcp_sk(sk)->highest_sack;
1345 static inline void tcp_highest_sack_reset(struct sock *sk)
1347 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1350 /* Called when old skb is about to be deleted (to be combined with new skb) */
1351 static inline void tcp_highest_sack_combine(struct sock *sk,
1352 struct sk_buff *old,
1353 struct sk_buff *new)
1355 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1356 tcp_sk(sk)->highest_sack = new;
1360 enum tcp_seq_states {
1361 TCP_SEQ_STATE_LISTENING,
1362 TCP_SEQ_STATE_OPENREQ,
1363 TCP_SEQ_STATE_ESTABLISHED,
1364 TCP_SEQ_STATE_TIME_WAIT,
1367 struct tcp_seq_afinfo {
1370 struct file_operations seq_fops;
1371 struct seq_operations seq_ops;
1374 struct tcp_iter_state {
1375 struct seq_net_private p;
1377 enum tcp_seq_states state;
1378 struct sock *syn_wait_sk;
1379 int bucket, sbucket, num, uid;
1382 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1383 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1385 extern struct request_sock_ops tcp_request_sock_ops;
1386 extern struct request_sock_ops tcp6_request_sock_ops;
1388 extern void tcp_v4_destroy_sock(struct sock *sk);
1390 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1391 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1392 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1393 struct sk_buff *skb);
1394 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1395 struct sk_buff *skb);
1396 extern int tcp_gro_complete(struct sk_buff *skb);
1397 extern int tcp4_gro_complete(struct sk_buff *skb);
1399 #ifdef CONFIG_PROC_FS
1400 extern int tcp4_proc_init(void);
1401 extern void tcp4_proc_exit(void);
1404 /* TCP af-specific functions */
1405 struct tcp_sock_af_ops {
1406 #ifdef CONFIG_TCP_MD5SIG
1407 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1408 struct sock *addr_sk);
1409 int (*calc_md5_hash) (char *location,
1410 struct tcp_md5sig_key *md5,
1412 struct request_sock *req,
1413 struct sk_buff *skb);
1414 int (*md5_add) (struct sock *sk,
1415 struct sock *addr_sk,
1418 int (*md5_parse) (struct sock *sk,
1419 char __user *optval,
1424 struct tcp_request_sock_ops {
1425 #ifdef CONFIG_TCP_MD5SIG
1426 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1427 struct request_sock *req);
1431 extern void tcp_v4_init(void);
1432 extern void tcp_init(void);