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>
33 #include <net/inet_connection_sock.h>
34 #include <net/inet_timewait_sock.h>
35 #include <net/inet_hashtables.h>
36 #include <net/checksum.h>
37 #include <net/request_sock.h>
41 #include <net/tcp_states.h>
43 #include <linux/seq_file.h>
45 extern struct inet_hashinfo tcp_hashinfo;
47 extern atomic_t tcp_orphan_count;
48 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
50 #define MAX_TCP_HEADER (128 + MAX_HEADER)
53 * Never offer a window over 32767 without using window scaling. Some
54 * poor stacks do signed 16bit maths!
56 #define MAX_TCP_WINDOW 32767U
58 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
59 #define TCP_MIN_MSS 88U
61 /* Minimal RCV_MSS. */
62 #define TCP_MIN_RCVMSS 536U
64 /* The least MTU to use for probing */
65 #define TCP_BASE_MSS 512
67 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
68 #define TCP_FASTRETRANS_THRESH 3
70 /* Maximal reordering. */
71 #define TCP_MAX_REORDERING 127
73 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
74 #define TCP_MAX_QUICKACKS 16U
77 #define TCP_URG_VALID 0x0100
78 #define TCP_URG_NOTYET 0x0200
79 #define TCP_URG_READ 0x0400
81 #define TCP_RETR1 3 /*
82 * This is how many retries it does before it
83 * tries to figure out if the gateway is
84 * down. Minimal RFC value is 3; it corresponds
85 * to ~3sec-8min depending on RTO.
88 #define TCP_RETR2 15 /*
89 * This should take at least
90 * 90 minutes to time out.
91 * RFC1122 says that the limit is 100 sec.
92 * 15 is ~13-30min depending on RTO.
95 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
96 * connection: ~180sec is RFC minimum */
98 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
99 * connection: ~180sec is RFC minimum */
102 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
103 * socket. 7 is ~50sec-16min.
107 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
108 * state, about 60 seconds */
109 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
110 /* BSD style FIN_WAIT2 deadlock breaker.
111 * It used to be 3min, new value is 60sec,
112 * to combine FIN-WAIT-2 timeout with
116 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
118 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
119 #define TCP_ATO_MIN ((unsigned)(HZ/25))
121 #define TCP_DELACK_MIN 4U
122 #define TCP_ATO_MIN 4U
124 #define TCP_RTO_MAX ((unsigned)(120*HZ))
125 #define TCP_RTO_MIN ((unsigned)(HZ/5))
126 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
128 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
129 * for local resources.
132 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
133 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
134 #define TCP_KEEPALIVE_INTVL (75*HZ)
136 #define MAX_TCP_KEEPIDLE 32767
137 #define MAX_TCP_KEEPINTVL 32767
138 #define MAX_TCP_KEEPCNT 127
139 #define MAX_TCP_SYNCNT 127
141 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
143 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
144 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
145 * after this time. It should be equal
146 * (or greater than) TCP_TIMEWAIT_LEN
147 * to provide reliability equal to one
148 * provided by timewait state.
150 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
151 * timestamps. It must be less than
152 * minimal timewait lifetime.
158 #define TCPOPT_NOP 1 /* Padding */
159 #define TCPOPT_EOL 0 /* End of options */
160 #define TCPOPT_MSS 2 /* Segment size negotiating */
161 #define TCPOPT_WINDOW 3 /* Window scaling */
162 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
163 #define TCPOPT_SACK 5 /* SACK Block */
164 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
165 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
171 #define TCPOLEN_MSS 4
172 #define TCPOLEN_WINDOW 3
173 #define TCPOLEN_SACK_PERM 2
174 #define TCPOLEN_TIMESTAMP 10
175 #define TCPOLEN_MD5SIG 18
177 /* But this is what stacks really send out. */
178 #define TCPOLEN_TSTAMP_ALIGNED 12
179 #define TCPOLEN_WSCALE_ALIGNED 4
180 #define TCPOLEN_SACKPERM_ALIGNED 4
181 #define TCPOLEN_SACK_BASE 2
182 #define TCPOLEN_SACK_BASE_ALIGNED 4
183 #define TCPOLEN_SACK_PERBLOCK 8
184 #define TCPOLEN_MD5SIG_ALIGNED 20
186 /* Flags in tp->nonagle */
187 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
188 #define TCP_NAGLE_CORK 2 /* Socket is corked */
189 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
191 extern struct inet_timewait_death_row tcp_death_row;
193 /* sysctl variables for tcp */
194 extern int sysctl_tcp_timestamps;
195 extern int sysctl_tcp_window_scaling;
196 extern int sysctl_tcp_sack;
197 extern int sysctl_tcp_fin_timeout;
198 extern int sysctl_tcp_keepalive_time;
199 extern int sysctl_tcp_keepalive_probes;
200 extern int sysctl_tcp_keepalive_intvl;
201 extern int sysctl_tcp_syn_retries;
202 extern int sysctl_tcp_synack_retries;
203 extern int sysctl_tcp_retries1;
204 extern int sysctl_tcp_retries2;
205 extern int sysctl_tcp_orphan_retries;
206 extern int sysctl_tcp_syncookies;
207 extern int sysctl_tcp_retrans_collapse;
208 extern int sysctl_tcp_stdurg;
209 extern int sysctl_tcp_rfc1337;
210 extern int sysctl_tcp_abort_on_overflow;
211 extern int sysctl_tcp_max_orphans;
212 extern int sysctl_tcp_fack;
213 extern int sysctl_tcp_reordering;
214 extern int sysctl_tcp_ecn;
215 extern int sysctl_tcp_dsack;
216 extern int sysctl_tcp_mem[3];
217 extern int sysctl_tcp_wmem[3];
218 extern int sysctl_tcp_rmem[3];
219 extern int sysctl_tcp_app_win;
220 extern int sysctl_tcp_adv_win_scale;
221 extern int sysctl_tcp_tw_reuse;
222 extern int sysctl_tcp_frto;
223 extern int sysctl_tcp_frto_response;
224 extern int sysctl_tcp_low_latency;
225 extern int sysctl_tcp_dma_copybreak;
226 extern int sysctl_tcp_nometrics_save;
227 extern int sysctl_tcp_moderate_rcvbuf;
228 extern int sysctl_tcp_tso_win_divisor;
229 extern int sysctl_tcp_abc;
230 extern int sysctl_tcp_mtu_probing;
231 extern int sysctl_tcp_base_mss;
232 extern int sysctl_tcp_workaround_signed_windows;
233 extern int sysctl_tcp_slow_start_after_idle;
234 extern int sysctl_tcp_max_ssthresh;
236 extern atomic_t tcp_memory_allocated;
237 extern atomic_t tcp_sockets_allocated;
238 extern int tcp_memory_pressure;
241 * The next routines deal with comparing 32 bit unsigned ints
242 * and worry about wraparound (automatic with unsigned arithmetic).
245 static inline int before(__u32 seq1, __u32 seq2)
247 return (__s32)(seq1-seq2) < 0;
249 #define after(seq2, seq1) before(seq1, seq2)
251 /* is s2<=s1<=s3 ? */
252 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
254 return seq3 - seq2 >= seq1 - seq2;
258 extern struct proto tcp_prot;
260 DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
261 #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
262 #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
263 #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
264 #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
265 #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
266 #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
268 extern void tcp_v4_err(struct sk_buff *skb, u32);
270 extern void tcp_shutdown (struct sock *sk, int how);
272 extern int tcp_v4_rcv(struct sk_buff *skb);
274 extern int tcp_v4_remember_stamp(struct sock *sk);
276 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
278 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk,
279 struct msghdr *msg, size_t size);
280 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
282 extern int tcp_ioctl(struct sock *sk,
286 extern int tcp_rcv_state_process(struct sock *sk,
291 extern int tcp_rcv_established(struct sock *sk,
296 extern void tcp_rcv_space_adjust(struct sock *sk);
298 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
300 extern int tcp_twsk_unique(struct sock *sk,
301 struct sock *sktw, void *twp);
303 extern void tcp_twsk_destructor(struct sock *sk);
305 static inline void tcp_dec_quickack_mode(struct sock *sk,
306 const unsigned int pkts)
308 struct inet_connection_sock *icsk = inet_csk(sk);
310 if (icsk->icsk_ack.quick) {
311 if (pkts >= icsk->icsk_ack.quick) {
312 icsk->icsk_ack.quick = 0;
313 /* Leaving quickack mode we deflate ATO. */
314 icsk->icsk_ack.ato = TCP_ATO_MIN;
316 icsk->icsk_ack.quick -= pkts;
320 extern void tcp_enter_quickack_mode(struct sock *sk);
322 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
324 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
336 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
338 const struct tcphdr *th);
340 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
341 struct request_sock *req,
342 struct request_sock **prev);
343 extern int tcp_child_process(struct sock *parent,
345 struct sk_buff *skb);
346 extern int tcp_use_frto(struct sock *sk);
347 extern void tcp_enter_frto(struct sock *sk);
348 extern void tcp_enter_loss(struct sock *sk, int how);
349 extern void tcp_clear_retrans(struct tcp_sock *tp);
350 extern void tcp_update_metrics(struct sock *sk);
352 extern void tcp_close(struct sock *sk,
354 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
356 extern int tcp_getsockopt(struct sock *sk, int level,
360 extern int tcp_setsockopt(struct sock *sk, int level,
361 int optname, char __user *optval,
363 extern int compat_tcp_getsockopt(struct sock *sk,
364 int level, int optname,
365 char __user *optval, int __user *optlen);
366 extern int compat_tcp_setsockopt(struct sock *sk,
367 int level, int optname,
368 char __user *optval, int optlen);
369 extern void tcp_set_keepalive(struct sock *sk, int val);
370 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
372 size_t len, int nonblock,
373 int flags, int *addr_len);
375 extern void tcp_parse_options(struct sk_buff *skb,
376 struct tcp_options_received *opt_rx,
380 * TCP v4 functions exported for the inet6 API
383 extern void tcp_v4_send_check(struct sock *sk, int len,
384 struct sk_buff *skb);
386 extern int tcp_v4_conn_request(struct sock *sk,
387 struct sk_buff *skb);
389 extern struct sock * tcp_create_openreq_child(struct sock *sk,
390 struct request_sock *req,
391 struct sk_buff *skb);
393 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
395 struct request_sock *req,
396 struct dst_entry *dst);
398 extern int tcp_v4_do_rcv(struct sock *sk,
399 struct sk_buff *skb);
401 extern int tcp_v4_connect(struct sock *sk,
402 struct sockaddr *uaddr,
405 extern int tcp_connect(struct sock *sk);
407 extern struct sk_buff * tcp_make_synack(struct sock *sk,
408 struct dst_entry *dst,
409 struct request_sock *req);
411 extern int tcp_disconnect(struct sock *sk, int flags);
413 extern void tcp_unhash(struct sock *sk);
415 /* From syncookies.c */
416 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
417 struct ip_options *opt);
418 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
423 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
425 extern int tcp_may_send_now(struct sock *sk);
426 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
427 extern void tcp_xmit_retransmit_queue(struct sock *);
428 extern void tcp_simple_retransmit(struct sock *);
429 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
430 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
432 extern void tcp_send_probe0(struct sock *);
433 extern void tcp_send_partial(struct sock *);
434 extern int tcp_write_wakeup(struct sock *);
435 extern void tcp_send_fin(struct sock *sk);
436 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
437 extern int tcp_send_synack(struct sock *);
438 extern void tcp_push_one(struct sock *, unsigned int mss_now);
439 extern void tcp_send_ack(struct sock *sk);
440 extern void tcp_send_delayed_ack(struct sock *sk);
443 extern void tcp_cwnd_application_limited(struct sock *sk);
446 extern void tcp_init_xmit_timers(struct sock *);
447 static inline void tcp_clear_xmit_timers(struct sock *sk)
449 inet_csk_clear_xmit_timers(sk);
452 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
453 extern unsigned int tcp_current_mss(struct sock *sk, int large);
456 extern void tcp_get_info(struct sock *, struct tcp_info *);
458 /* Read 'sendfile()'-style from a TCP socket */
459 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
460 unsigned int, size_t);
461 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
462 sk_read_actor_t recv_actor);
464 extern void tcp_initialize_rcv_mss(struct sock *sk);
466 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
467 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
468 extern void tcp_mtup_init(struct sock *sk);
470 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
472 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
473 ntohl(TCP_FLAG_ACK) |
477 static inline void tcp_fast_path_on(struct tcp_sock *tp)
479 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
482 static inline void tcp_fast_path_check(struct sock *sk)
484 struct tcp_sock *tp = tcp_sk(sk);
486 if (skb_queue_empty(&tp->out_of_order_queue) &&
488 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
490 tcp_fast_path_on(tp);
493 /* Compute the actual receive window we are currently advertising.
494 * Rcv_nxt can be after the window if our peer push more data
495 * than the offered window.
497 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
499 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
506 /* Choose a new window, without checks for shrinking, and without
507 * scaling applied to the result. The caller does these things
508 * if necessary. This is a "raw" window selection.
510 extern u32 __tcp_select_window(struct sock *sk);
512 /* TCP timestamps are only 32-bits, this causes a slight
513 * complication on 64-bit systems since we store a snapshot
514 * of jiffies in the buffer control blocks below. We decided
515 * to use only the low 32-bits of jiffies and hide the ugly
516 * casts with the following macro.
518 #define tcp_time_stamp ((__u32)(jiffies))
520 /* This is what the send packet queuing engine uses to pass
521 * TCP per-packet control information to the transmission
522 * code. We also store the host-order sequence numbers in
523 * here too. This is 36 bytes on 32-bit architectures,
524 * 40 bytes on 64-bit machines, if this grows please adjust
525 * skbuff.h:skbuff->cb[xxx] size appropriately.
529 struct inet_skb_parm h4;
530 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
531 struct inet6_skb_parm h6;
533 } header; /* For incoming frames */
534 __u32 seq; /* Starting sequence number */
535 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
536 __u32 when; /* used to compute rtt's */
537 __u8 flags; /* TCP header flags. */
539 /* NOTE: These must match up to the flags byte in a
542 #define TCPCB_FLAG_FIN 0x01
543 #define TCPCB_FLAG_SYN 0x02
544 #define TCPCB_FLAG_RST 0x04
545 #define TCPCB_FLAG_PSH 0x08
546 #define TCPCB_FLAG_ACK 0x10
547 #define TCPCB_FLAG_URG 0x20
548 #define TCPCB_FLAG_ECE 0x40
549 #define TCPCB_FLAG_CWR 0x80
551 __u8 sacked; /* State flags for SACK/FACK. */
552 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
553 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
554 #define TCPCB_LOST 0x04 /* SKB is lost */
555 #define TCPCB_TAGBITS 0x07 /* All tag bits */
557 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
558 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
560 #define TCPCB_URG 0x20 /* Urgent pointer advanced here */
562 #define TCPCB_AT_TAIL (TCPCB_URG)
564 __u16 urg_ptr; /* Valid w/URG flags is set. */
565 __u32 ack_seq; /* Sequence number ACK'd */
568 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
570 #include <net/tcp_ecn.h>
572 /* Due to TSO, an SKB can be composed of multiple actual
573 * packets. To keep these tracked properly, we use this.
575 static inline int tcp_skb_pcount(const struct sk_buff *skb)
577 return skb_shinfo(skb)->gso_segs;
580 /* This is valid iff tcp_skb_pcount() > 1. */
581 static inline int tcp_skb_mss(const struct sk_buff *skb)
583 return skb_shinfo(skb)->gso_size;
586 static inline void tcp_dec_pcount_approx(__u32 *count,
587 const struct sk_buff *skb)
590 *count -= tcp_skb_pcount(skb);
596 static inline void tcp_packets_out_inc(struct sock *sk,
597 const struct sk_buff *skb)
599 struct tcp_sock *tp = tcp_sk(sk);
600 int orig = tp->packets_out;
602 tp->packets_out += tcp_skb_pcount(skb);
604 inet_csk_reset_xmit_timer(sk, ICSK_TIME_RETRANS,
605 inet_csk(sk)->icsk_rto, TCP_RTO_MAX);
608 static inline void tcp_packets_out_dec(struct tcp_sock *tp,
609 const struct sk_buff *skb)
611 tp->packets_out -= tcp_skb_pcount(skb);
614 /* Events passed to congestion control interface */
616 CA_EVENT_TX_START, /* first transmit when no packets in flight */
617 CA_EVENT_CWND_RESTART, /* congestion window restart */
618 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
619 CA_EVENT_FRTO, /* fast recovery timeout */
620 CA_EVENT_LOSS, /* loss timeout */
621 CA_EVENT_FAST_ACK, /* in sequence ack */
622 CA_EVENT_SLOW_ACK, /* other ack */
626 * Interface for adding new TCP congestion control handlers
628 #define TCP_CA_NAME_MAX 16
629 #define TCP_CA_MAX 128
630 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
632 #define TCP_CONG_NON_RESTRICTED 0x1
633 #define TCP_CONG_RTT_STAMP 0x2
635 struct tcp_congestion_ops {
636 struct list_head list;
639 /* initialize private data (optional) */
640 void (*init)(struct sock *sk);
641 /* cleanup private data (optional) */
642 void (*release)(struct sock *sk);
644 /* return slow start threshold (required) */
645 u32 (*ssthresh)(struct sock *sk);
646 /* lower bound for congestion window (optional) */
647 u32 (*min_cwnd)(const struct sock *sk);
648 /* do new cwnd calculation (required) */
649 void (*cong_avoid)(struct sock *sk, u32 ack,
650 u32 rtt, u32 in_flight, int good_ack);
651 /* call before changing ca_state (optional) */
652 void (*set_state)(struct sock *sk, u8 new_state);
653 /* call when cwnd event occurs (optional) */
654 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
655 /* new value of cwnd after loss (optional) */
656 u32 (*undo_cwnd)(struct sock *sk);
657 /* hook for packet ack accounting (optional) */
658 void (*pkts_acked)(struct sock *sk, u32 num_acked, ktime_t last);
659 /* get info for inet_diag (optional) */
660 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
662 char name[TCP_CA_NAME_MAX];
663 struct module *owner;
666 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
667 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
669 extern void tcp_init_congestion_control(struct sock *sk);
670 extern void tcp_cleanup_congestion_control(struct sock *sk);
671 extern int tcp_set_default_congestion_control(const char *name);
672 extern void tcp_get_default_congestion_control(char *name);
673 extern void tcp_get_available_congestion_control(char *buf, size_t len);
674 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
675 extern int tcp_set_allowed_congestion_control(char *allowed);
676 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
677 extern void tcp_slow_start(struct tcp_sock *tp);
679 extern struct tcp_congestion_ops tcp_init_congestion_ops;
680 extern u32 tcp_reno_ssthresh(struct sock *sk);
681 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack,
682 u32 rtt, u32 in_flight, int flag);
683 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
684 extern struct tcp_congestion_ops tcp_reno;
686 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
688 struct inet_connection_sock *icsk = inet_csk(sk);
690 if (icsk->icsk_ca_ops->set_state)
691 icsk->icsk_ca_ops->set_state(sk, ca_state);
692 icsk->icsk_ca_state = ca_state;
695 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
697 const struct inet_connection_sock *icsk = inet_csk(sk);
699 if (icsk->icsk_ca_ops->cwnd_event)
700 icsk->icsk_ca_ops->cwnd_event(sk, event);
703 /* This determines how many packets are "in the network" to the best
704 * of our knowledge. In many cases it is conservative, but where
705 * detailed information is available from the receiver (via SACK
706 * blocks etc.) we can make more aggressive calculations.
708 * Use this for decisions involving congestion control, use just
709 * tp->packets_out to determine if the send queue is empty or not.
711 * Read this equation as:
713 * "Packets sent once on transmission queue" MINUS
714 * "Packets left network, but not honestly ACKed yet" PLUS
715 * "Packets fast retransmitted"
717 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
719 return (tp->packets_out - tp->left_out + tp->retrans_out);
722 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
723 * The exception is rate halving phase, when cwnd is decreasing towards
726 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
728 const struct tcp_sock *tp = tcp_sk(sk);
729 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
730 return tp->snd_ssthresh;
732 return max(tp->snd_ssthresh,
733 ((tp->snd_cwnd >> 1) +
734 (tp->snd_cwnd >> 2)));
737 static inline void tcp_sync_left_out(struct tcp_sock *tp)
739 BUG_ON(tp->rx_opt.sack_ok &&
740 (tp->sacked_out + tp->lost_out > tp->packets_out));
741 tp->left_out = tp->sacked_out + tp->lost_out;
744 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
745 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
747 /* Slow start with delack produces 3 packets of burst, so that
748 * it is safe "de facto".
750 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
755 /* RFC2861 Check whether we are limited by application or congestion window
756 * This is the inverse of cwnd check in tcp_tso_should_defer
758 static inline int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight)
760 const struct tcp_sock *tp = tcp_sk(sk);
763 if (in_flight >= tp->snd_cwnd)
769 left = tp->snd_cwnd - in_flight;
770 if (sysctl_tcp_tso_win_divisor)
771 return left * sysctl_tcp_tso_win_divisor < tp->snd_cwnd;
773 return left <= tcp_max_burst(tp);
776 static inline void tcp_minshall_update(struct tcp_sock *tp, int mss,
777 const struct sk_buff *skb)
780 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
783 static inline void tcp_check_probe_timer(struct sock *sk)
785 struct tcp_sock *tp = tcp_sk(sk);
786 const struct inet_connection_sock *icsk = inet_csk(sk);
788 if (!tp->packets_out && !icsk->icsk_pending)
789 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
790 icsk->icsk_rto, TCP_RTO_MAX);
793 static inline void tcp_push_pending_frames(struct sock *sk)
795 struct tcp_sock *tp = tcp_sk(sk);
797 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle);
800 static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
805 static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
811 * Calculate(/check) TCP checksum
813 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
814 __be32 daddr, __wsum base)
816 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
819 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
821 return __skb_checksum_complete(skb);
824 static inline int tcp_checksum_complete(struct sk_buff *skb)
826 return !skb_csum_unnecessary(skb) &&
827 __tcp_checksum_complete(skb);
830 /* Prequeue for VJ style copy to user, combined with checksumming. */
832 static inline void tcp_prequeue_init(struct tcp_sock *tp)
834 tp->ucopy.task = NULL;
836 tp->ucopy.memory = 0;
837 skb_queue_head_init(&tp->ucopy.prequeue);
838 #ifdef CONFIG_NET_DMA
839 tp->ucopy.dma_chan = NULL;
840 tp->ucopy.wakeup = 0;
841 tp->ucopy.pinned_list = NULL;
842 tp->ucopy.dma_cookie = 0;
846 /* Packet is added to VJ-style prequeue for processing in process
847 * context, if a reader task is waiting. Apparently, this exciting
848 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
849 * failed somewhere. Latency? Burstiness? Well, at least now we will
850 * see, why it failed. 8)8) --ANK
852 * NOTE: is this not too big to inline?
854 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
856 struct tcp_sock *tp = tcp_sk(sk);
858 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
859 __skb_queue_tail(&tp->ucopy.prequeue, skb);
860 tp->ucopy.memory += skb->truesize;
861 if (tp->ucopy.memory > sk->sk_rcvbuf) {
862 struct sk_buff *skb1;
864 BUG_ON(sock_owned_by_user(sk));
866 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
867 sk->sk_backlog_rcv(sk, skb1);
868 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
871 tp->ucopy.memory = 0;
872 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
873 wake_up_interruptible(sk->sk_sleep);
874 if (!inet_csk_ack_scheduled(sk))
875 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
876 (3 * TCP_RTO_MIN) / 4,
888 static const char *statename[]={
889 "Unused","Established","Syn Sent","Syn Recv",
890 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
891 "Close Wait","Last ACK","Listen","Closing"
895 static inline void tcp_set_state(struct sock *sk, int state)
897 int oldstate = sk->sk_state;
900 case TCP_ESTABLISHED:
901 if (oldstate != TCP_ESTABLISHED)
902 TCP_INC_STATS(TCP_MIB_CURRESTAB);
906 if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED)
907 TCP_INC_STATS(TCP_MIB_ESTABRESETS);
909 sk->sk_prot->unhash(sk);
910 if (inet_csk(sk)->icsk_bind_hash &&
911 !(sk->sk_userlocks & SOCK_BINDPORT_LOCK))
912 inet_put_port(&tcp_hashinfo, sk);
915 if (oldstate==TCP_ESTABLISHED)
916 TCP_DEC_STATS(TCP_MIB_CURRESTAB);
919 /* Change state AFTER socket is unhashed to avoid closed
920 * socket sitting in hash tables.
922 sk->sk_state = state;
925 SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]);
929 extern void tcp_done(struct sock *sk);
931 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
934 rx_opt->eff_sacks = 0;
935 rx_opt->num_sacks = 0;
938 /* Determine a window scaling and initial window to offer. */
939 extern void tcp_select_initial_window(int __space, __u32 mss,
940 __u32 *rcv_wnd, __u32 *window_clamp,
941 int wscale_ok, __u8 *rcv_wscale);
943 static inline int tcp_win_from_space(int space)
945 return sysctl_tcp_adv_win_scale<=0 ?
946 (space>>(-sysctl_tcp_adv_win_scale)) :
947 space - (space>>sysctl_tcp_adv_win_scale);
950 /* Note: caller must be prepared to deal with negative returns */
951 static inline int tcp_space(const struct sock *sk)
953 return tcp_win_from_space(sk->sk_rcvbuf -
954 atomic_read(&sk->sk_rmem_alloc));
957 static inline int tcp_full_space(const struct sock *sk)
959 return tcp_win_from_space(sk->sk_rcvbuf);
962 static inline void tcp_openreq_init(struct request_sock *req,
963 struct tcp_options_received *rx_opt,
966 struct inet_request_sock *ireq = inet_rsk(req);
968 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
969 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
970 req->mss = rx_opt->mss_clamp;
971 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
972 ireq->tstamp_ok = rx_opt->tstamp_ok;
973 ireq->sack_ok = rx_opt->sack_ok;
974 ireq->snd_wscale = rx_opt->snd_wscale;
975 ireq->wscale_ok = rx_opt->wscale_ok;
978 ireq->rmt_port = tcp_hdr(skb)->source;
981 extern void tcp_enter_memory_pressure(void);
983 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
985 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
988 static inline int keepalive_time_when(const struct tcp_sock *tp)
990 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
993 static inline int tcp_fin_time(const struct sock *sk)
995 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
996 const int rto = inet_csk(sk)->icsk_rto;
998 if (fin_timeout < (rto << 2) - (rto >> 1))
999 fin_timeout = (rto << 2) - (rto >> 1);
1004 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1006 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1008 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1011 /* RST segments are not recommended to carry timestamp,
1012 and, if they do, it is recommended to ignore PAWS because
1013 "their cleanup function should take precedence over timestamps."
1014 Certainly, it is mistake. It is necessary to understand the reasons
1015 of this constraint to relax it: if peer reboots, clock may go
1016 out-of-sync and half-open connections will not be reset.
1017 Actually, the problem would be not existing if all
1018 the implementations followed draft about maintaining clock
1019 via reboots. Linux-2.2 DOES NOT!
1021 However, we can relax time bounds for RST segments to MSL.
1023 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1028 #define TCP_CHECK_TIMER(sk) do { } while (0)
1030 static inline void tcp_mib_init(void)
1033 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1034 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1035 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1036 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1040 static inline void clear_all_retrans_hints(struct tcp_sock *tp){
1041 tp->lost_skb_hint = NULL;
1042 tp->scoreboard_skb_hint = NULL;
1043 tp->retransmit_skb_hint = NULL;
1044 tp->forward_skb_hint = NULL;
1045 tp->fastpath_skb_hint = NULL;
1051 /* - key database */
1052 struct tcp_md5sig_key {
1057 struct tcp4_md5sig_key {
1063 struct tcp6_md5sig_key {
1067 u32 scope_id; /* XXX */
1069 struct in6_addr addr;
1073 struct tcp_md5sig_info {
1074 struct tcp4_md5sig_key *keys4;
1075 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1076 struct tcp6_md5sig_key *keys6;
1084 /* - pseudo header */
1085 struct tcp4_pseudohdr {
1093 struct tcp6_pseudohdr {
1094 struct in6_addr saddr;
1095 struct in6_addr daddr;
1097 __be32 protocol; /* including padding */
1100 union tcp_md5sum_block {
1101 struct tcp4_pseudohdr ip4;
1102 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1103 struct tcp6_pseudohdr ip6;
1107 /* - pool: digest algorithm, hash description and scratch buffer */
1108 struct tcp_md5sig_pool {
1109 struct hash_desc md5_desc;
1110 union tcp_md5sum_block md5_blk;
1113 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1116 extern int tcp_v4_calc_md5_hash(char *md5_hash,
1117 struct tcp_md5sig_key *key,
1119 struct dst_entry *dst,
1120 struct request_sock *req,
1122 int protocol, int tcplen);
1123 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1124 struct sock *addr_sk);
1126 extern int tcp_v4_md5_do_add(struct sock *sk,
1131 extern int tcp_v4_md5_do_del(struct sock *sk,
1134 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1135 extern void tcp_free_md5sig_pool(void);
1137 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1138 extern void __tcp_put_md5sig_pool(void);
1141 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1143 int cpu = get_cpu();
1144 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1150 static inline void tcp_put_md5sig_pool(void)
1152 __tcp_put_md5sig_pool();
1156 /* write queue abstraction */
1157 static inline void tcp_write_queue_purge(struct sock *sk)
1159 struct sk_buff *skb;
1161 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1162 sk_stream_free_skb(sk, skb);
1163 sk_stream_mem_reclaim(sk);
1166 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1168 struct sk_buff *skb = sk->sk_write_queue.next;
1169 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1174 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1176 struct sk_buff *skb = sk->sk_write_queue.prev;
1177 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1182 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1187 #define tcp_for_write_queue(skb, sk) \
1188 for (skb = (sk)->sk_write_queue.next; \
1189 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1192 #define tcp_for_write_queue_from(skb, sk) \
1193 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\
1196 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1198 return sk->sk_send_head;
1201 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1203 struct tcp_sock *tp = tcp_sk(sk);
1205 sk->sk_send_head = skb->next;
1206 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue)
1207 sk->sk_send_head = NULL;
1208 /* Don't override Nagle indefinately with F-RTO */
1209 if (tp->frto_counter == 2)
1210 tp->frto_counter = 3;
1213 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1215 if (sk->sk_send_head == skb_unlinked)
1216 sk->sk_send_head = NULL;
1219 static inline void tcp_init_send_head(struct sock *sk)
1221 sk->sk_send_head = NULL;
1224 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1226 __skb_queue_tail(&sk->sk_write_queue, skb);
1229 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1231 __tcp_add_write_queue_tail(sk, skb);
1233 /* Queue it, remembering where we must start sending. */
1234 if (sk->sk_send_head == NULL)
1235 sk->sk_send_head = skb;
1238 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1240 __skb_queue_head(&sk->sk_write_queue, skb);
1243 /* Insert buff after skb on the write queue of sk. */
1244 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1245 struct sk_buff *buff,
1248 __skb_append(skb, buff, &sk->sk_write_queue);
1251 /* Insert skb between prev and next on the write queue of sk. */
1252 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1253 struct sk_buff *skb,
1256 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue);
1259 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1261 __skb_unlink(skb, &sk->sk_write_queue);
1264 static inline int tcp_skb_is_last(const struct sock *sk,
1265 const struct sk_buff *skb)
1267 return skb->next == (struct sk_buff *)&sk->sk_write_queue;
1270 static inline int tcp_write_queue_empty(struct sock *sk)
1272 return skb_queue_empty(&sk->sk_write_queue);
1276 enum tcp_seq_states {
1277 TCP_SEQ_STATE_LISTENING,
1278 TCP_SEQ_STATE_OPENREQ,
1279 TCP_SEQ_STATE_ESTABLISHED,
1280 TCP_SEQ_STATE_TIME_WAIT,
1283 struct tcp_seq_afinfo {
1284 struct module *owner;
1287 int (*seq_show) (struct seq_file *m, void *v);
1288 struct file_operations *seq_fops;
1291 struct tcp_iter_state {
1293 enum tcp_seq_states state;
1294 struct sock *syn_wait_sk;
1295 int bucket, sbucket, num, uid;
1296 struct seq_operations seq_ops;
1299 extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo);
1300 extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo);
1302 extern struct request_sock_ops tcp_request_sock_ops;
1304 extern int tcp_v4_destroy_sock(struct sock *sk);
1306 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1307 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1309 #ifdef CONFIG_PROC_FS
1310 extern int tcp4_proc_init(void);
1311 extern void tcp4_proc_exit(void);
1314 /* TCP af-specific functions */
1315 struct tcp_sock_af_ops {
1316 #ifdef CONFIG_TCP_MD5SIG
1317 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1318 struct sock *addr_sk);
1319 int (*calc_md5_hash) (char *location,
1320 struct tcp_md5sig_key *md5,
1322 struct dst_entry *dst,
1323 struct request_sock *req,
1325 int protocol, int len);
1326 int (*md5_add) (struct sock *sk,
1327 struct sock *addr_sk,
1330 int (*md5_parse) (struct sock *sk,
1331 char __user *optval,
1336 struct tcp_request_sock_ops {
1337 #ifdef CONFIG_TCP_MD5SIG
1338 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1339 struct request_sock *req);
1343 extern void tcp_v4_init(struct net_proto_family *ops);
1344 extern void tcp_init(void);