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 /* The least MTU to use for probing */
66 #define TCP_BASE_MSS 512
68 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
69 #define TCP_FASTRETRANS_THRESH 3
71 /* Maximal reordering. */
72 #define TCP_MAX_REORDERING 127
74 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
75 #define TCP_MAX_QUICKACKS 16U
78 #define TCP_URG_VALID 0x0100
79 #define TCP_URG_NOTYET 0x0200
80 #define TCP_URG_READ 0x0400
82 #define TCP_RETR1 3 /*
83 * This is how many retries it does before it
84 * tries to figure out if the gateway is
85 * down. Minimal RFC value is 3; it corresponds
86 * to ~3sec-8min depending on RTO.
89 #define TCP_RETR2 15 /*
90 * This should take at least
91 * 90 minutes to time out.
92 * RFC1122 says that the limit is 100 sec.
93 * 15 is ~13-30min depending on RTO.
96 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
97 * connection: ~180sec is RFC minimum */
99 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
100 * connection: ~180sec is RFC minimum */
103 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
104 * socket. 7 is ~50sec-16min.
108 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
109 * state, about 60 seconds */
110 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
111 /* BSD style FIN_WAIT2 deadlock breaker.
112 * It used to be 3min, new value is 60sec,
113 * to combine FIN-WAIT-2 timeout with
117 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
119 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
120 #define TCP_ATO_MIN ((unsigned)(HZ/25))
122 #define TCP_DELACK_MIN 4U
123 #define TCP_ATO_MIN 4U
125 #define TCP_RTO_MAX ((unsigned)(120*HZ))
126 #define TCP_RTO_MIN ((unsigned)(HZ/5))
127 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
129 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
130 * for local resources.
133 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
134 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
135 #define TCP_KEEPALIVE_INTVL (75*HZ)
137 #define MAX_TCP_KEEPIDLE 32767
138 #define MAX_TCP_KEEPINTVL 32767
139 #define MAX_TCP_KEEPCNT 127
140 #define MAX_TCP_SYNCNT 127
142 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
144 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
145 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
146 * after this time. It should be equal
147 * (or greater than) TCP_TIMEWAIT_LEN
148 * to provide reliability equal to one
149 * provided by timewait state.
151 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
152 * timestamps. It must be less than
153 * minimal timewait lifetime.
159 #define TCPOPT_NOP 1 /* Padding */
160 #define TCPOPT_EOL 0 /* End of options */
161 #define TCPOPT_MSS 2 /* Segment size negotiating */
162 #define TCPOPT_WINDOW 3 /* Window scaling */
163 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
164 #define TCPOPT_SACK 5 /* SACK Block */
165 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
166 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
172 #define TCPOLEN_MSS 4
173 #define TCPOLEN_WINDOW 3
174 #define TCPOLEN_SACK_PERM 2
175 #define TCPOLEN_TIMESTAMP 10
176 #define TCPOLEN_MD5SIG 18
178 /* But this is what stacks really send out. */
179 #define TCPOLEN_TSTAMP_ALIGNED 12
180 #define TCPOLEN_WSCALE_ALIGNED 4
181 #define TCPOLEN_SACKPERM_ALIGNED 4
182 #define TCPOLEN_SACK_BASE 2
183 #define TCPOLEN_SACK_BASE_ALIGNED 4
184 #define TCPOLEN_SACK_PERBLOCK 8
185 #define TCPOLEN_MD5SIG_ALIGNED 20
186 #define TCPOLEN_MSS_ALIGNED 4
188 /* Flags in tp->nonagle */
189 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
190 #define TCP_NAGLE_CORK 2 /* Socket is corked */
191 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
193 extern struct inet_timewait_death_row tcp_death_row;
195 /* sysctl variables for tcp */
196 extern int sysctl_tcp_timestamps;
197 extern int sysctl_tcp_window_scaling;
198 extern int sysctl_tcp_sack;
199 extern int sysctl_tcp_fin_timeout;
200 extern int sysctl_tcp_keepalive_time;
201 extern int sysctl_tcp_keepalive_probes;
202 extern int sysctl_tcp_keepalive_intvl;
203 extern int sysctl_tcp_syn_retries;
204 extern int sysctl_tcp_synack_retries;
205 extern int sysctl_tcp_retries1;
206 extern int sysctl_tcp_retries2;
207 extern int sysctl_tcp_orphan_retries;
208 extern int sysctl_tcp_syncookies;
209 extern int sysctl_tcp_retrans_collapse;
210 extern int sysctl_tcp_stdurg;
211 extern int sysctl_tcp_rfc1337;
212 extern int sysctl_tcp_abort_on_overflow;
213 extern int sysctl_tcp_max_orphans;
214 extern int sysctl_tcp_fack;
215 extern int sysctl_tcp_reordering;
216 extern int sysctl_tcp_ecn;
217 extern int sysctl_tcp_dsack;
218 extern int sysctl_tcp_mem[3];
219 extern int sysctl_tcp_wmem[3];
220 extern int sysctl_tcp_rmem[3];
221 extern int sysctl_tcp_app_win;
222 extern int sysctl_tcp_adv_win_scale;
223 extern int sysctl_tcp_tw_reuse;
224 extern int sysctl_tcp_frto;
225 extern int sysctl_tcp_frto_response;
226 extern int sysctl_tcp_low_latency;
227 extern int sysctl_tcp_dma_copybreak;
228 extern int sysctl_tcp_nometrics_save;
229 extern int sysctl_tcp_moderate_rcvbuf;
230 extern int sysctl_tcp_tso_win_divisor;
231 extern int sysctl_tcp_abc;
232 extern int sysctl_tcp_mtu_probing;
233 extern int sysctl_tcp_base_mss;
234 extern int sysctl_tcp_workaround_signed_windows;
235 extern int sysctl_tcp_slow_start_after_idle;
236 extern int sysctl_tcp_max_ssthresh;
238 extern atomic_t tcp_memory_allocated;
239 extern struct percpu_counter tcp_sockets_allocated;
240 extern int tcp_memory_pressure;
243 * The next routines deal with comparing 32 bit unsigned ints
244 * and worry about wraparound (automatic with unsigned arithmetic).
247 static inline int before(__u32 seq1, __u32 seq2)
249 return (__s32)(seq1-seq2) < 0;
251 #define after(seq2, seq1) before(seq1, seq2)
253 /* is s2<=s1<=s3 ? */
254 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
256 return seq3 - seq2 >= seq1 - seq2;
259 static inline int tcp_too_many_orphans(struct sock *sk, int num)
261 return (num > sysctl_tcp_max_orphans) ||
262 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
263 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
266 /* syncookies: remember time of last synqueue overflow */
267 static inline void tcp_synq_overflow(struct sock *sk)
269 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
272 /* syncookies: no recent synqueue overflow on this listening socket? */
273 static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
275 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
276 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT);
279 extern struct proto tcp_prot;
281 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
282 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
283 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
284 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
286 extern void tcp_v4_err(struct sk_buff *skb, u32);
288 extern void tcp_shutdown (struct sock *sk, int how);
290 extern int tcp_v4_rcv(struct sk_buff *skb);
292 extern int tcp_v4_remember_stamp(struct sock *sk);
294 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
296 extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
297 struct msghdr *msg, size_t size);
298 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
300 extern int tcp_ioctl(struct sock *sk,
304 extern int tcp_rcv_state_process(struct sock *sk,
309 extern int tcp_rcv_established(struct sock *sk,
314 extern void tcp_rcv_space_adjust(struct sock *sk);
316 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
318 extern int tcp_twsk_unique(struct sock *sk,
319 struct sock *sktw, void *twp);
321 extern void tcp_twsk_destructor(struct sock *sk);
323 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
324 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
326 static inline void tcp_dec_quickack_mode(struct sock *sk,
327 const unsigned int pkts)
329 struct inet_connection_sock *icsk = inet_csk(sk);
331 if (icsk->icsk_ack.quick) {
332 if (pkts >= icsk->icsk_ack.quick) {
333 icsk->icsk_ack.quick = 0;
334 /* Leaving quickack mode we deflate ATO. */
335 icsk->icsk_ack.ato = TCP_ATO_MIN;
337 icsk->icsk_ack.quick -= pkts;
341 extern void tcp_enter_quickack_mode(struct sock *sk);
343 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
345 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
349 #define TCP_ECN_QUEUE_CWR 2
350 #define TCP_ECN_DEMAND_CWR 4
352 static __inline__ void
353 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
355 if (sysctl_tcp_ecn && th->ece && th->cwr)
356 inet_rsk(req)->ecn_ok = 1;
367 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
369 const struct tcphdr *th);
371 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
372 struct request_sock *req,
373 struct request_sock **prev);
374 extern int tcp_child_process(struct sock *parent,
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);
383 extern void tcp_close(struct sock *sk,
385 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
387 extern int tcp_getsockopt(struct sock *sk, int level,
391 extern int tcp_setsockopt(struct sock *sk, int level,
392 int optname, char __user *optval,
393 unsigned int optlen);
394 extern int compat_tcp_getsockopt(struct sock *sk,
395 int level, int optname,
396 char __user *optval, int __user *optlen);
397 extern int compat_tcp_setsockopt(struct sock *sk,
398 int level, int optname,
399 char __user *optval, unsigned int optlen);
400 extern void tcp_set_keepalive(struct sock *sk, int val);
401 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
403 size_t len, int nonblock,
404 int flags, int *addr_len);
406 extern void tcp_parse_options(struct sk_buff *skb,
407 struct tcp_options_received *opt_rx,
409 struct dst_entry *dst);
411 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
414 * TCP v4 functions exported for the inet6 API
417 extern void tcp_v4_send_check(struct sock *sk, int len,
418 struct sk_buff *skb);
420 extern int tcp_v4_conn_request(struct sock *sk,
421 struct sk_buff *skb);
423 extern struct sock * tcp_create_openreq_child(struct sock *sk,
424 struct request_sock *req,
425 struct sk_buff *skb);
427 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
429 struct request_sock *req,
430 struct dst_entry *dst);
432 extern int tcp_v4_do_rcv(struct sock *sk,
433 struct sk_buff *skb);
435 extern int tcp_v4_connect(struct sock *sk,
436 struct sockaddr *uaddr,
439 extern int tcp_connect(struct sock *sk);
441 extern struct sk_buff * tcp_make_synack(struct sock *sk,
442 struct dst_entry *dst,
443 struct request_sock *req);
445 extern int tcp_disconnect(struct sock *sk, int flags);
448 /* From syncookies.c */
449 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
450 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
451 struct ip_options *opt);
452 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
455 extern __u32 cookie_init_timestamp(struct request_sock *req);
456 extern void cookie_check_timestamp(struct tcp_options_received *tcp_opt);
458 /* From net/ipv6/syncookies.c */
459 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
460 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
465 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
467 extern int tcp_may_send_now(struct sock *sk);
468 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
469 extern void tcp_retransmit_timer(struct sock *sk);
470 extern void tcp_xmit_retransmit_queue(struct sock *);
471 extern void tcp_simple_retransmit(struct sock *);
472 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
473 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
475 extern void tcp_send_probe0(struct sock *);
476 extern void tcp_send_partial(struct sock *);
477 extern int tcp_write_wakeup(struct sock *);
478 extern void tcp_send_fin(struct sock *sk);
479 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
480 extern int tcp_send_synack(struct sock *);
481 extern void tcp_push_one(struct sock *, unsigned int mss_now);
482 extern void tcp_send_ack(struct sock *sk);
483 extern void tcp_send_delayed_ack(struct sock *sk);
486 extern void tcp_cwnd_application_limited(struct sock *sk);
489 extern void tcp_init_xmit_timers(struct sock *);
490 static inline void tcp_clear_xmit_timers(struct sock *sk)
492 inet_csk_clear_xmit_timers(sk);
495 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
496 extern unsigned int tcp_current_mss(struct sock *sk);
498 /* Bound MSS / TSO packet size with the half of the window */
499 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
501 if (tp->max_window && pktsize > (tp->max_window >> 1))
502 return max(tp->max_window >> 1, 68U - tp->tcp_header_len);
508 extern void tcp_get_info(struct sock *, struct tcp_info *);
510 /* Read 'sendfile()'-style from a TCP socket */
511 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
512 unsigned int, size_t);
513 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
514 sk_read_actor_t recv_actor);
516 extern void tcp_initialize_rcv_mss(struct sock *sk);
518 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
519 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
520 extern void tcp_mtup_init(struct sock *sk);
522 static inline void tcp_bound_rto(const struct sock *sk)
524 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
525 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
528 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
530 return (tp->srtt >> 3) + tp->rttvar;
533 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
535 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
536 ntohl(TCP_FLAG_ACK) |
540 static inline void tcp_fast_path_on(struct tcp_sock *tp)
542 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
545 static inline void tcp_fast_path_check(struct sock *sk)
547 struct tcp_sock *tp = tcp_sk(sk);
549 if (skb_queue_empty(&tp->out_of_order_queue) &&
551 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
553 tcp_fast_path_on(tp);
556 /* Compute the actual rto_min value */
557 static inline u32 tcp_rto_min(struct sock *sk)
559 struct dst_entry *dst = __sk_dst_get(sk);
560 u32 rto_min = TCP_RTO_MIN;
562 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
563 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
567 /* Compute the actual receive window we are currently advertising.
568 * Rcv_nxt can be after the window if our peer push more data
569 * than the offered window.
571 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
573 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
580 /* Choose a new window, without checks for shrinking, and without
581 * scaling applied to the result. The caller does these things
582 * if necessary. This is a "raw" window selection.
584 extern u32 __tcp_select_window(struct sock *sk);
586 /* TCP timestamps are only 32-bits, this causes a slight
587 * complication on 64-bit systems since we store a snapshot
588 * of jiffies in the buffer control blocks below. We decided
589 * to use only the low 32-bits of jiffies and hide the ugly
590 * casts with the following macro.
592 #define tcp_time_stamp ((__u32)(jiffies))
594 /* This is what the send packet queuing engine uses to pass
595 * TCP per-packet control information to the transmission
596 * code. We also store the host-order sequence numbers in
597 * here too. This is 36 bytes on 32-bit architectures,
598 * 40 bytes on 64-bit machines, if this grows please adjust
599 * skbuff.h:skbuff->cb[xxx] size appropriately.
603 struct inet_skb_parm h4;
604 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
605 struct inet6_skb_parm h6;
607 } header; /* For incoming frames */
608 __u32 seq; /* Starting sequence number */
609 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
610 __u32 when; /* used to compute rtt's */
611 __u8 flags; /* TCP header flags. */
613 /* NOTE: These must match up to the flags byte in a
616 #define TCPCB_FLAG_FIN 0x01
617 #define TCPCB_FLAG_SYN 0x02
618 #define TCPCB_FLAG_RST 0x04
619 #define TCPCB_FLAG_PSH 0x08
620 #define TCPCB_FLAG_ACK 0x10
621 #define TCPCB_FLAG_URG 0x20
622 #define TCPCB_FLAG_ECE 0x40
623 #define TCPCB_FLAG_CWR 0x80
625 __u8 sacked; /* State flags for SACK/FACK. */
626 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
627 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
628 #define TCPCB_LOST 0x04 /* SKB is lost */
629 #define TCPCB_TAGBITS 0x07 /* All tag bits */
631 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
632 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
634 __u32 ack_seq; /* Sequence number ACK'd */
637 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
639 /* Due to TSO, an SKB can be composed of multiple actual
640 * packets. To keep these tracked properly, we use this.
642 static inline int tcp_skb_pcount(const struct sk_buff *skb)
644 return skb_shinfo(skb)->gso_segs;
647 /* This is valid iff tcp_skb_pcount() > 1. */
648 static inline int tcp_skb_mss(const struct sk_buff *skb)
650 return skb_shinfo(skb)->gso_size;
653 /* Events passed to congestion control interface */
655 CA_EVENT_TX_START, /* first transmit when no packets in flight */
656 CA_EVENT_CWND_RESTART, /* congestion window restart */
657 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
658 CA_EVENT_FRTO, /* fast recovery timeout */
659 CA_EVENT_LOSS, /* loss timeout */
660 CA_EVENT_FAST_ACK, /* in sequence ack */
661 CA_EVENT_SLOW_ACK, /* other ack */
665 * Interface for adding new TCP congestion control handlers
667 #define TCP_CA_NAME_MAX 16
668 #define TCP_CA_MAX 128
669 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
671 #define TCP_CONG_NON_RESTRICTED 0x1
672 #define TCP_CONG_RTT_STAMP 0x2
674 struct tcp_congestion_ops {
675 struct list_head list;
678 /* initialize private data (optional) */
679 void (*init)(struct sock *sk);
680 /* cleanup private data (optional) */
681 void (*release)(struct sock *sk);
683 /* return slow start threshold (required) */
684 u32 (*ssthresh)(struct sock *sk);
685 /* lower bound for congestion window (optional) */
686 u32 (*min_cwnd)(const struct sock *sk);
687 /* do new cwnd calculation (required) */
688 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
689 /* call before changing ca_state (optional) */
690 void (*set_state)(struct sock *sk, u8 new_state);
691 /* call when cwnd event occurs (optional) */
692 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
693 /* new value of cwnd after loss (optional) */
694 u32 (*undo_cwnd)(struct sock *sk);
695 /* hook for packet ack accounting (optional) */
696 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
697 /* get info for inet_diag (optional) */
698 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
700 char name[TCP_CA_NAME_MAX];
701 struct module *owner;
704 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
705 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
707 extern void tcp_init_congestion_control(struct sock *sk);
708 extern void tcp_cleanup_congestion_control(struct sock *sk);
709 extern int tcp_set_default_congestion_control(const char *name);
710 extern void tcp_get_default_congestion_control(char *name);
711 extern void tcp_get_available_congestion_control(char *buf, size_t len);
712 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
713 extern int tcp_set_allowed_congestion_control(char *allowed);
714 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
715 extern void tcp_slow_start(struct tcp_sock *tp);
716 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
718 extern struct tcp_congestion_ops tcp_init_congestion_ops;
719 extern u32 tcp_reno_ssthresh(struct sock *sk);
720 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
721 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
722 extern struct tcp_congestion_ops tcp_reno;
724 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
726 struct inet_connection_sock *icsk = inet_csk(sk);
728 if (icsk->icsk_ca_ops->set_state)
729 icsk->icsk_ca_ops->set_state(sk, ca_state);
730 icsk->icsk_ca_state = ca_state;
733 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
735 const struct inet_connection_sock *icsk = inet_csk(sk);
737 if (icsk->icsk_ca_ops->cwnd_event)
738 icsk->icsk_ca_ops->cwnd_event(sk, event);
741 /* These functions determine how the current flow behaves in respect of SACK
742 * handling. SACK is negotiated with the peer, and therefore it can vary
743 * between different flows.
745 * tcp_is_sack - SACK enabled
746 * tcp_is_reno - No SACK
747 * tcp_is_fack - FACK enabled, implies SACK enabled
749 static inline int tcp_is_sack(const struct tcp_sock *tp)
751 return tp->rx_opt.sack_ok;
754 static inline int tcp_is_reno(const struct tcp_sock *tp)
756 return !tcp_is_sack(tp);
759 static inline int tcp_is_fack(const struct tcp_sock *tp)
761 return tp->rx_opt.sack_ok & 2;
764 static inline void tcp_enable_fack(struct tcp_sock *tp)
766 tp->rx_opt.sack_ok |= 2;
769 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
771 return tp->sacked_out + tp->lost_out;
774 /* This determines how many packets are "in the network" to the best
775 * of our knowledge. In many cases it is conservative, but where
776 * detailed information is available from the receiver (via SACK
777 * blocks etc.) we can make more aggressive calculations.
779 * Use this for decisions involving congestion control, use just
780 * tp->packets_out to determine if the send queue is empty or not.
782 * Read this equation as:
784 * "Packets sent once on transmission queue" MINUS
785 * "Packets left network, but not honestly ACKed yet" PLUS
786 * "Packets fast retransmitted"
788 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
790 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
793 #define TCP_INFINITE_SSTHRESH 0x7fffffff
795 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
797 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
800 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
801 * The exception is rate halving phase, when cwnd is decreasing towards
804 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
806 const struct tcp_sock *tp = tcp_sk(sk);
807 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
808 return tp->snd_ssthresh;
810 return max(tp->snd_ssthresh,
811 ((tp->snd_cwnd >> 1) +
812 (tp->snd_cwnd >> 2)));
815 /* Use define here intentionally to get WARN_ON location shown at the caller */
816 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
818 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
819 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
821 /* Slow start with delack produces 3 packets of burst, so that
822 * it is safe "de facto". This will be the default - same as
823 * the default reordering threshold - but if reordering increases,
824 * we must be able to allow cwnd to burst at least this much in order
825 * to not pull it back when holes are filled.
827 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
829 return tp->reordering;
832 /* Returns end sequence number of the receiver's advertised window */
833 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
835 return tp->snd_una + tp->snd_wnd;
837 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
839 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
840 const struct sk_buff *skb)
843 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
846 static inline void tcp_check_probe_timer(struct sock *sk)
848 struct tcp_sock *tp = tcp_sk(sk);
849 const struct inet_connection_sock *icsk = inet_csk(sk);
851 if (!tp->packets_out && !icsk->icsk_pending)
852 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
853 icsk->icsk_rto, TCP_RTO_MAX);
856 static inline void tcp_push_pending_frames(struct sock *sk)
858 struct tcp_sock *tp = tcp_sk(sk);
860 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
863 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
868 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
874 * Calculate(/check) TCP checksum
876 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
877 __be32 daddr, __wsum base)
879 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
882 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
884 return __skb_checksum_complete(skb);
887 static inline int tcp_checksum_complete(struct sk_buff *skb)
889 return !skb_csum_unnecessary(skb) &&
890 __tcp_checksum_complete(skb);
893 /* Prequeue for VJ style copy to user, combined with checksumming. */
895 static inline void tcp_prequeue_init(struct tcp_sock *tp)
897 tp->ucopy.task = NULL;
899 tp->ucopy.memory = 0;
900 skb_queue_head_init(&tp->ucopy.prequeue);
901 #ifdef CONFIG_NET_DMA
902 tp->ucopy.dma_chan = NULL;
903 tp->ucopy.wakeup = 0;
904 tp->ucopy.pinned_list = NULL;
905 tp->ucopy.dma_cookie = 0;
909 /* Packet is added to VJ-style prequeue for processing in process
910 * context, if a reader task is waiting. Apparently, this exciting
911 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
912 * failed somewhere. Latency? Burstiness? Well, at least now we will
913 * see, why it failed. 8)8) --ANK
915 * NOTE: is this not too big to inline?
917 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
919 struct tcp_sock *tp = tcp_sk(sk);
921 if (sysctl_tcp_low_latency || !tp->ucopy.task)
924 __skb_queue_tail(&tp->ucopy.prequeue, skb);
925 tp->ucopy.memory += skb->truesize;
926 if (tp->ucopy.memory > sk->sk_rcvbuf) {
927 struct sk_buff *skb1;
929 BUG_ON(sock_owned_by_user(sk));
931 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
932 sk_backlog_rcv(sk, skb1);
933 NET_INC_STATS_BH(sock_net(sk),
934 LINUX_MIB_TCPPREQUEUEDROPPED);
937 tp->ucopy.memory = 0;
938 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
939 wake_up_interruptible_poll(sk->sk_sleep,
940 POLLIN | POLLRDNORM | POLLRDBAND);
941 if (!inet_csk_ack_scheduled(sk))
942 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
943 (3 * tcp_rto_min(sk)) / 4,
953 static const char *statename[]={
954 "Unused","Established","Syn Sent","Syn Recv",
955 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
956 "Close Wait","Last ACK","Listen","Closing"
959 extern void tcp_set_state(struct sock *sk, int state);
961 extern void tcp_done(struct sock *sk);
963 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
966 rx_opt->num_sacks = 0;
969 /* Determine a window scaling and initial window to offer. */
970 extern void tcp_select_initial_window(int __space, __u32 mss,
971 __u32 *rcv_wnd, __u32 *window_clamp,
972 int wscale_ok, __u8 *rcv_wscale);
974 static inline int tcp_win_from_space(int space)
976 return sysctl_tcp_adv_win_scale<=0 ?
977 (space>>(-sysctl_tcp_adv_win_scale)) :
978 space - (space>>sysctl_tcp_adv_win_scale);
981 /* Note: caller must be prepared to deal with negative returns */
982 static inline int tcp_space(const struct sock *sk)
984 return tcp_win_from_space(sk->sk_rcvbuf -
985 atomic_read(&sk->sk_rmem_alloc));
988 static inline int tcp_full_space(const struct sock *sk)
990 return tcp_win_from_space(sk->sk_rcvbuf);
993 static inline void tcp_openreq_init(struct request_sock *req,
994 struct tcp_options_received *rx_opt,
997 struct inet_request_sock *ireq = inet_rsk(req);
999 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1001 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1002 req->mss = rx_opt->mss_clamp;
1003 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1004 ireq->tstamp_ok = rx_opt->tstamp_ok;
1005 ireq->sack_ok = rx_opt->sack_ok;
1006 ireq->snd_wscale = rx_opt->snd_wscale;
1007 ireq->wscale_ok = rx_opt->wscale_ok;
1010 ireq->rmt_port = tcp_hdr(skb)->source;
1011 ireq->loc_port = tcp_hdr(skb)->dest;
1014 extern void tcp_enter_memory_pressure(struct sock *sk);
1016 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1018 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1021 static inline int keepalive_time_when(const struct tcp_sock *tp)
1023 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1026 static inline int keepalive_probes(const struct tcp_sock *tp)
1028 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1031 static inline int tcp_fin_time(const struct sock *sk)
1033 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1034 const int rto = inet_csk(sk)->icsk_rto;
1036 if (fin_timeout < (rto << 2) - (rto >> 1))
1037 fin_timeout = (rto << 2) - (rto >> 1);
1042 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1045 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1047 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1053 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1056 if (tcp_paws_check(rx_opt, 0))
1059 /* RST segments are not recommended to carry timestamp,
1060 and, if they do, it is recommended to ignore PAWS because
1061 "their cleanup function should take precedence over timestamps."
1062 Certainly, it is mistake. It is necessary to understand the reasons
1063 of this constraint to relax it: if peer reboots, clock may go
1064 out-of-sync and half-open connections will not be reset.
1065 Actually, the problem would be not existing if all
1066 the implementations followed draft about maintaining clock
1067 via reboots. Linux-2.2 DOES NOT!
1069 However, we can relax time bounds for RST segments to MSL.
1071 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1076 #define TCP_CHECK_TIMER(sk) do { } while (0)
1078 static inline void tcp_mib_init(struct net *net)
1081 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1082 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1083 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1084 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1088 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1090 tp->lost_skb_hint = NULL;
1091 tp->scoreboard_skb_hint = NULL;
1094 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1096 tcp_clear_retrans_hints_partial(tp);
1097 tp->retransmit_skb_hint = NULL;
1103 /* - key database */
1104 struct tcp_md5sig_key {
1109 struct tcp4_md5sig_key {
1110 struct tcp_md5sig_key base;
1114 struct tcp6_md5sig_key {
1115 struct tcp_md5sig_key base;
1117 u32 scope_id; /* XXX */
1119 struct in6_addr addr;
1123 struct tcp_md5sig_info {
1124 struct tcp4_md5sig_key *keys4;
1125 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1126 struct tcp6_md5sig_key *keys6;
1134 /* - pseudo header */
1135 struct tcp4_pseudohdr {
1143 struct tcp6_pseudohdr {
1144 struct in6_addr saddr;
1145 struct in6_addr daddr;
1147 __be32 protocol; /* including padding */
1150 union tcp_md5sum_block {
1151 struct tcp4_pseudohdr ip4;
1152 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1153 struct tcp6_pseudohdr ip6;
1157 /* - pool: digest algorithm, hash description and scratch buffer */
1158 struct tcp_md5sig_pool {
1159 struct hash_desc md5_desc;
1160 union tcp_md5sum_block md5_blk;
1163 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1166 extern int tcp_v4_md5_hash_skb(char *md5_hash,
1167 struct tcp_md5sig_key *key,
1169 struct request_sock *req,
1170 struct sk_buff *skb);
1172 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1173 struct sock *addr_sk);
1175 extern int tcp_v4_md5_do_add(struct sock *sk,
1180 extern int tcp_v4_md5_do_del(struct sock *sk,
1183 #ifdef CONFIG_TCP_MD5SIG
1184 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1185 &(struct tcp_md5sig_key) { \
1186 .key = (twsk)->tw_md5_key, \
1187 .keylen = (twsk)->tw_md5_keylen, \
1190 #define tcp_twsk_md5_key(twsk) NULL
1193 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(struct sock *);
1194 extern void tcp_free_md5sig_pool(void);
1196 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1197 extern void __tcp_put_md5sig_pool(void);
1198 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1199 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1200 unsigned header_len);
1201 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1202 struct tcp_md5sig_key *key);
1205 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1207 int cpu = get_cpu();
1208 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1214 static inline void tcp_put_md5sig_pool(void)
1216 __tcp_put_md5sig_pool();
1220 /* write queue abstraction */
1221 static inline void tcp_write_queue_purge(struct sock *sk)
1223 struct sk_buff *skb;
1225 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1226 sk_wmem_free_skb(sk, skb);
1230 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1232 return skb_peek(&sk->sk_write_queue);
1235 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1237 return skb_peek_tail(&sk->sk_write_queue);
1240 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1242 return skb_queue_next(&sk->sk_write_queue, skb);
1245 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb)
1247 return skb_queue_prev(&sk->sk_write_queue, skb);
1250 #define tcp_for_write_queue(skb, sk) \
1251 skb_queue_walk(&(sk)->sk_write_queue, skb)
1253 #define tcp_for_write_queue_from(skb, sk) \
1254 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1256 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1257 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1259 /* This function calculates a "timeout" which is equivalent to the timeout of a
1260 * TCP connection after "boundary" unsucessful, exponentially backed-off
1261 * retransmissions with an initial RTO of TCP_RTO_MIN.
1263 static inline bool retransmits_timed_out(const struct sock *sk,
1264 unsigned int boundary)
1266 unsigned int timeout, linear_backoff_thresh;
1268 if (!inet_csk(sk)->icsk_retransmits)
1271 linear_backoff_thresh = ilog2(TCP_RTO_MAX/TCP_RTO_MIN);
1273 if (boundary <= linear_backoff_thresh)
1274 timeout = ((2 << boundary) - 1) * TCP_RTO_MIN;
1276 timeout = ((2 << linear_backoff_thresh) - 1) * TCP_RTO_MIN +
1277 (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
1279 return (tcp_time_stamp - tcp_sk(sk)->retrans_stamp) >= timeout;
1282 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1284 return sk->sk_send_head;
1287 static inline bool tcp_skb_is_last(const struct sock *sk,
1288 const struct sk_buff *skb)
1290 return skb_queue_is_last(&sk->sk_write_queue, skb);
1293 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1295 if (tcp_skb_is_last(sk, skb))
1296 sk->sk_send_head = NULL;
1298 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1301 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1303 if (sk->sk_send_head == skb_unlinked)
1304 sk->sk_send_head = NULL;
1307 static inline void tcp_init_send_head(struct sock *sk)
1309 sk->sk_send_head = NULL;
1312 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1314 __skb_queue_tail(&sk->sk_write_queue, skb);
1317 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1319 __tcp_add_write_queue_tail(sk, skb);
1321 /* Queue it, remembering where we must start sending. */
1322 if (sk->sk_send_head == NULL) {
1323 sk->sk_send_head = skb;
1325 if (tcp_sk(sk)->highest_sack == NULL)
1326 tcp_sk(sk)->highest_sack = skb;
1330 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1332 __skb_queue_head(&sk->sk_write_queue, skb);
1335 /* Insert buff after skb on the write queue of sk. */
1336 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1337 struct sk_buff *buff,
1340 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1343 /* Insert new before skb on the write queue of sk. */
1344 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1345 struct sk_buff *skb,
1348 __skb_queue_before(&sk->sk_write_queue, skb, new);
1350 if (sk->sk_send_head == skb)
1351 sk->sk_send_head = new;
1354 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1356 __skb_unlink(skb, &sk->sk_write_queue);
1359 static inline int tcp_write_queue_empty(struct sock *sk)
1361 return skb_queue_empty(&sk->sk_write_queue);
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;
1404 enum tcp_seq_states {
1405 TCP_SEQ_STATE_LISTENING,
1406 TCP_SEQ_STATE_OPENREQ,
1407 TCP_SEQ_STATE_ESTABLISHED,
1408 TCP_SEQ_STATE_TIME_WAIT,
1411 struct tcp_seq_afinfo {
1414 struct file_operations seq_fops;
1415 struct seq_operations seq_ops;
1418 struct tcp_iter_state {
1419 struct seq_net_private p;
1421 enum tcp_seq_states state;
1422 struct sock *syn_wait_sk;
1423 int bucket, sbucket, num, uid;
1426 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1427 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1429 extern struct request_sock_ops tcp_request_sock_ops;
1430 extern struct request_sock_ops tcp6_request_sock_ops;
1432 extern void tcp_v4_destroy_sock(struct sock *sk);
1434 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1435 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1436 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1437 struct sk_buff *skb);
1438 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1439 struct sk_buff *skb);
1440 extern int tcp_gro_complete(struct sk_buff *skb);
1441 extern int tcp4_gro_complete(struct sk_buff *skb);
1443 #ifdef CONFIG_PROC_FS
1444 extern int tcp4_proc_init(void);
1445 extern void tcp4_proc_exit(void);
1448 /* TCP af-specific functions */
1449 struct tcp_sock_af_ops {
1450 #ifdef CONFIG_TCP_MD5SIG
1451 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1452 struct sock *addr_sk);
1453 int (*calc_md5_hash) (char *location,
1454 struct tcp_md5sig_key *md5,
1456 struct request_sock *req,
1457 struct sk_buff *skb);
1458 int (*md5_add) (struct sock *sk,
1459 struct sock *addr_sk,
1462 int (*md5_parse) (struct sock *sk,
1463 char __user *optval,
1468 struct tcp_request_sock_ops {
1469 #ifdef CONFIG_TCP_MD5SIG
1470 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1471 struct request_sock *req);
1472 int (*calc_md5_hash) (char *location,
1473 struct tcp_md5sig_key *md5,
1475 struct request_sock *req,
1476 struct sk_buff *skb);
1480 extern void tcp_v4_init(void);
1481 extern void tcp_init(void);