blob: 29d3089038ab6dab1393abc5c00509d1010f366d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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.
5 *
6 * Definitions for the TCP protocol.
7 *
8 * Version: @(#)tcp.h 1.0.2 04/28/93
9 *
10 * Author: Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
11 *
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
16 */
17#ifndef _LINUX_TCP_H
18#define _LINUX_TCP_H
19
20#include <linux/types.h>
21#include <asm/byteorder.h>
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -080022#include <linux/socket.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023
24struct tcphdr {
Al Viro46a97322006-09-27 18:31:51 -070025 __be16 source;
26 __be16 dest;
27 __be32 seq;
28 __be32 ack_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#if defined(__LITTLE_ENDIAN_BITFIELD)
30 __u16 res1:4,
31 doff:4,
32 fin:1,
33 syn:1,
34 rst:1,
35 psh:1,
36 ack:1,
37 urg:1,
38 ece:1,
39 cwr:1;
40#elif defined(__BIG_ENDIAN_BITFIELD)
41 __u16 doff:4,
42 res1:4,
43 cwr:1,
44 ece:1,
45 urg:1,
46 ack:1,
47 psh:1,
48 rst:1,
49 syn:1,
50 fin:1;
51#else
52#error "Adjust your <asm/byteorder.h> defines"
53#endif
Al Viro46a97322006-09-27 18:31:51 -070054 __be16 window;
Al Viro9981a0e2006-11-14 21:24:30 -080055 __sum16 check;
Al Viro46a97322006-09-27 18:31:51 -070056 __be16 urg_ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -070057};
58
Linus Torvalds1da177e2005-04-16 15:20:36 -070059/*
60 * The union cast uses a gcc extension to avoid aliasing problems
61 * (union is compatible to any of its members)
62 * This means this part of the code is -fstrict-aliasing safe now.
63 */
64union tcp_word_hdr {
65 struct tcphdr hdr;
Al Viro46a97322006-09-27 18:31:51 -070066 __be32 words[5];
Linus Torvalds1da177e2005-04-16 15:20:36 -070067};
68
69#define tcp_flag_word(tp) ( ((union tcp_word_hdr *)(tp))->words [3])
70
71enum {
72 TCP_FLAG_CWR = __constant_htonl(0x00800000),
73 TCP_FLAG_ECE = __constant_htonl(0x00400000),
74 TCP_FLAG_URG = __constant_htonl(0x00200000),
75 TCP_FLAG_ACK = __constant_htonl(0x00100000),
76 TCP_FLAG_PSH = __constant_htonl(0x00080000),
77 TCP_FLAG_RST = __constant_htonl(0x00040000),
78 TCP_FLAG_SYN = __constant_htonl(0x00020000),
79 TCP_FLAG_FIN = __constant_htonl(0x00010000),
80 TCP_RESERVED_BITS = __constant_htonl(0x0F000000),
81 TCP_DATA_OFFSET = __constant_htonl(0xF0000000)
82};
83
84/* TCP socket options */
85#define TCP_NODELAY 1 /* Turn off Nagle's algorithm. */
86#define TCP_MAXSEG 2 /* Limit MSS */
87#define TCP_CORK 3 /* Never send partially complete segments */
88#define TCP_KEEPIDLE 4 /* Start keeplives after this period */
89#define TCP_KEEPINTVL 5 /* Interval between keepalives */
90#define TCP_KEEPCNT 6 /* Number of keepalives before death */
91#define TCP_SYNCNT 7 /* Number of SYN retransmits */
92#define TCP_LINGER2 8 /* Life time of orphaned FIN-WAIT-2 state */
93#define TCP_DEFER_ACCEPT 9 /* Wake up listener only when data arrive */
94#define TCP_WINDOW_CLAMP 10 /* Bound advertised window */
95#define TCP_INFO 11 /* Information about this connection. */
96#define TCP_QUICKACK 12 /* Block/reenable quick acks */
Stephen Hemminger5f8ef482005-06-23 20:37:36 -070097#define TCP_CONGESTION 13 /* Congestion control algorithm */
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -080098#define TCP_MD5SIG 14 /* TCP MD5 Signature (RFC2385) */
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
100#define TCPI_OPT_TIMESTAMPS 1
101#define TCPI_OPT_SACK 2
102#define TCPI_OPT_WSCALE 4
103#define TCPI_OPT_ECN 8
104
105enum tcp_ca_state
106{
107 TCP_CA_Open = 0,
108#define TCPF_CA_Open (1<<TCP_CA_Open)
109 TCP_CA_Disorder = 1,
110#define TCPF_CA_Disorder (1<<TCP_CA_Disorder)
111 TCP_CA_CWR = 2,
112#define TCPF_CA_CWR (1<<TCP_CA_CWR)
113 TCP_CA_Recovery = 3,
114#define TCPF_CA_Recovery (1<<TCP_CA_Recovery)
115 TCP_CA_Loss = 4
116#define TCPF_CA_Loss (1<<TCP_CA_Loss)
117};
118
119struct tcp_info
120{
121 __u8 tcpi_state;
122 __u8 tcpi_ca_state;
123 __u8 tcpi_retransmits;
124 __u8 tcpi_probes;
125 __u8 tcpi_backoff;
126 __u8 tcpi_options;
127 __u8 tcpi_snd_wscale : 4, tcpi_rcv_wscale : 4;
128
129 __u32 tcpi_rto;
130 __u32 tcpi_ato;
131 __u32 tcpi_snd_mss;
132 __u32 tcpi_rcv_mss;
133
134 __u32 tcpi_unacked;
135 __u32 tcpi_sacked;
136 __u32 tcpi_lost;
137 __u32 tcpi_retrans;
138 __u32 tcpi_fackets;
139
140 /* Times. */
141 __u32 tcpi_last_data_sent;
142 __u32 tcpi_last_ack_sent; /* Not remembered, sorry. */
143 __u32 tcpi_last_data_recv;
144 __u32 tcpi_last_ack_recv;
145
146 /* Metrics. */
147 __u32 tcpi_pmtu;
148 __u32 tcpi_rcv_ssthresh;
149 __u32 tcpi_rtt;
150 __u32 tcpi_rttvar;
151 __u32 tcpi_snd_ssthresh;
152 __u32 tcpi_snd_cwnd;
153 __u32 tcpi_advmss;
154 __u32 tcpi_reordering;
155
156 __u32 tcpi_rcv_rtt;
157 __u32 tcpi_rcv_space;
158
159 __u32 tcpi_total_retrans;
160};
161
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800162/* for TCP_MD5SIG socket option */
163#define TCP_MD5SIG_MAXKEYLEN 80
164
165struct tcp_md5sig {
166 struct __kernel_sockaddr_storage tcpm_addr; /* address associated */
167 __u16 __tcpm_pad1; /* zero */
168 __u16 tcpm_keylen; /* key length */
169 __u32 __tcpm_pad2; /* zero */
170 __u8 tcpm_key[TCP_MD5SIG_MAXKEYLEN]; /* key (binary) */
171};
172
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173#ifdef __KERNEL__
174
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175#include <linux/skbuff.h>
David Woodhousec8a553ad2006-06-22 14:28:09 -0700176#include <linux/dmaengine.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177#include <net/sock.h>
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700178#include <net/inet_connection_sock.h>
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700179#include <net/inet_timewait_sock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
181/* This defines a selective acknowledgement block. */
Al Viro269bd272006-09-27 18:32:28 -0700182struct tcp_sack_block_wire {
183 __be32 start_seq;
184 __be32 end_seq;
185};
186
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187struct tcp_sack_block {
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200188 u32 start_seq;
189 u32 end_seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190};
191
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192struct tcp_options_received {
193/* PAWS/RTTM data */
194 long ts_recent_stamp;/* Time we stored ts_recent (for aging) */
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200195 u32 ts_recent; /* Time stamp to echo next */
196 u32 rcv_tsval; /* Time stamp value */
197 u32 rcv_tsecr; /* Time stamp echo reply */
198 u16 saw_tstamp : 1, /* Saw TIMESTAMP on last packet */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 tstamp_ok : 1, /* TIMESTAMP seen on SYN packet */
200 dsack : 1, /* D-SACK is scheduled */
201 wscale_ok : 1, /* Wscale seen on SYN packet */
202 sack_ok : 4, /* SACK seen on SYN packet */
203 snd_wscale : 4, /* Window scaling received from sender */
204 rcv_wscale : 4; /* Window scaling to send to receiver */
205/* SACKs data */
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200206 u8 eff_sacks; /* Size of SACK array to send with next packet */
207 u8 num_sacks; /* Number of SACK blocks */
208 u16 user_mss; /* mss requested by user in ioctl */
209 u16 mss_clamp; /* Maximal mss, negotiated at connection setup */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210};
211
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700212struct tcp_request_sock {
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800213 struct inet_request_sock req;
214#ifdef CONFIG_TCP_MD5SIG
215 /* Only used by TCP MD5 Signature so far. */
216 struct tcp_request_sock_ops *af_specific;
217#endif
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200218 u32 rcv_isn;
219 u32 snt_isn;
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700220};
221
Arnaldo Carvalho de Melo60236fd2005-06-18 22:47:21 -0700222static inline struct tcp_request_sock *tcp_rsk(const struct request_sock *req)
Arnaldo Carvalho de Melo2e6599c2005-06-18 22:46:52 -0700223{
224 return (struct tcp_request_sock *)req;
225}
226
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227struct tcp_sock {
Arnaldo Carvalho de Melo463c84b2005-08-09 20:10:42 -0700228 /* inet_connection_sock has to be the first member of tcp_sock */
229 struct inet_connection_sock inet_conn;
Arnaldo Carvalho de Melo2ff52f22006-11-28 00:48:32 -0200230 u16 tcp_header_len; /* Bytes of tcp header to send */
231 u16 xmit_size_goal; /* Goal for segmenting output packets */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
233/*
234 * Header prediction flags
235 * 0x5?10 << 16 + snd_wnd in net byte order
236 */
Al Virodddc93c2006-09-27 18:32:46 -0700237 __be32 pred_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238
239/*
240 * RFC793 variables by their proper names. This means you can
241 * read the code and the spec side by side (and laugh ...)
242 * See RFC793 and RFC1122. The RFC writes these in capitals.
243 */
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200244 u32 rcv_nxt; /* What we want to receive next */
245 u32 snd_nxt; /* Next sequence we send */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200247 u32 snd_una; /* First byte we want an ack for */
248 u32 snd_sml; /* Last byte of the most recently transmitted small packet */
249 u32 rcv_tstamp; /* timestamp of last received ACK (for keepalives) */
250 u32 lsndtime; /* timestamp of last sent data packet (for restart window) */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
252 /* Data for direct copy to user */
253 struct {
254 struct sk_buff_head prequeue;
255 struct task_struct *task;
256 struct iovec *iov;
257 int memory;
258 int len;
Chris Leech97fc2f02006-05-23 17:55:33 -0700259#ifdef CONFIG_NET_DMA
260 /* members for async copy */
261 struct dma_chan *dma_chan;
262 int wakeup;
263 struct dma_pinned_list *pinned_list;
264 dma_cookie_t dma_cookie;
265#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 } ucopy;
267
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200268 u32 snd_wl1; /* Sequence for window update */
269 u32 snd_wnd; /* The window we expect to receive */
270 u32 max_window; /* Maximal window ever seen from peer */
271 u32 mss_cache; /* Cached effective mss, not including SACKS */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200273 u32 window_clamp; /* Maximal window to advertise */
274 u32 rcv_ssthresh; /* Current window clamp */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200276 u32 frto_highmark; /* snd_nxt when RTO occurred */
277 u8 reordering; /* Packet reordering metric. */
278 u8 frto_counter; /* Number of new acks after RTO */
279 u8 nonagle; /* Disable Nagle algorithm? */
280 u8 keepalive_probes; /* num of allowed keep alive probes */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281
282/* RTT measurement */
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200283 u32 srtt; /* smoothed round trip time << 3 */
284 u32 mdev; /* medium deviation */
285 u32 mdev_max; /* maximal mdev for the last rtt period */
286 u32 rttvar; /* smoothed mdev_max */
287 u32 rtt_seq; /* sequence number to update rttvar */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200289 u32 packets_out; /* Packets which are "in flight" */
290 u32 left_out; /* Packets which leaved network */
291 u32 retrans_out; /* Retransmitted packets out */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292/*
293 * Options received (usually on last packet, some only on SYN packets).
294 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 struct tcp_options_received rx_opt;
296
297/*
298 * Slow start and congestion control (see also Nagle, and Karn & Partridge)
299 */
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200300 u32 snd_ssthresh; /* Slow start size threshold */
301 u32 snd_cwnd; /* Sending congestion window */
302 u16 snd_cwnd_cnt; /* Linear increase counter */
303 u16 snd_cwnd_clamp; /* Do not allow snd_cwnd to grow above this */
304 u32 snd_cwnd_used;
305 u32 snd_cwnd_stamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 struct sk_buff_head out_of_order_queue; /* Out of order segments go here */
308
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200309 u32 rcv_wnd; /* Current receiver window */
310 u32 rcv_wup; /* rcv_nxt on last window update sent */
311 u32 write_seq; /* Tail(+1) of data held in tcp send buffer */
312 u32 pushed_seq; /* Last pushed seq, required to talk to windows */
313 u32 copied_seq; /* Head of yet unread data */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314
315/* SACKs data */
316 struct tcp_sack_block duplicate_sack[1]; /* D-SACK block */
317 struct tcp_sack_block selective_acks[4]; /* The SACKS themselves*/
318
Baruch Even6f746512007-02-04 23:36:42 -0800319 struct tcp_sack_block_wire recv_sack_cache[4];
Stephen Hemminger6a438bb2005-11-10 17:14:59 -0800320
321 /* from STCP, retrans queue hinting */
322 struct sk_buff* lost_skb_hint;
323
324 struct sk_buff *scoreboard_skb_hint;
325 struct sk_buff *retransmit_skb_hint;
326 struct sk_buff *forward_skb_hint;
327 struct sk_buff *fastpath_skb_hint;
328
329 int fastpath_cnt_hint;
330 int lost_cnt_hint;
331 int retransmit_cnt_hint;
332 int forward_cnt_hint;
333
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200334 u16 advmss; /* Advertised MSS */
335 u16 prior_ssthresh; /* ssthresh saved at recovery start */
336 u32 lost_out; /* Lost packets */
337 u32 sacked_out; /* SACK'd packets */
338 u32 fackets_out; /* FACK'd packets */
339 u32 high_seq; /* snd_nxt at onset of congestion */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200341 u32 retrans_stamp; /* Timestamp of the last retransmit,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 * also used in SYN-SENT to remember stamp of
343 * the first SYN. */
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200344 u32 undo_marker; /* tracking retrans started here. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 int undo_retrans; /* number of undoable retransmissions. */
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200346 u32 urg_seq; /* Seq of received urgent pointer */
347 u16 urg_data; /* Saved octet of OOB data and control flags */
348 u8 urg_mode; /* In urgent mode */
349 u8 ecn_flags; /* ECN status bits. */
350 u32 snd_up; /* Urgent pointer */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200352 u32 total_retrans; /* Total retransmits for entire connection */
353 u32 bytes_acked; /* Appropriate Byte Counting - RFC3465 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 unsigned int keepalive_time; /* time before keep alive takes place */
356 unsigned int keepalive_intvl; /* time interval between keep alive probes */
357 int linger2;
358
359 unsigned long last_synq_overflow;
360
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200361 u32 tso_deferred;
John Heffnerae8064a2006-10-18 20:36:48 -0700362
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363/* Receiver side RTT estimation */
364 struct {
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200365 u32 rtt;
366 u32 seq;
367 u32 time;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 } rcv_rtt_est;
369
370/* Receiver queue space */
371 struct {
372 int space;
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200373 u32 seq;
374 u32 time;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 } rcvq_space;
John Heffner0e7b1362006-03-20 21:32:58 -0800376
377/* TCP-specific MTU probe information. */
378 struct {
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200379 u32 probe_seq_start;
380 u32 probe_seq_end;
John Heffner0e7b1362006-03-20 21:32:58 -0800381 } mtu_probe;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800382
383#ifdef CONFIG_TCP_MD5SIG
384/* TCP AF-Specific parts; only used by MD5 Signature support so far */
385 struct tcp_sock_af_ops *af_specific;
386
387/* TCP MD5 Signagure Option information */
388 struct tcp_md5sig_info *md5sig_info;
389#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390};
391
392static inline struct tcp_sock *tcp_sk(const struct sock *sk)
393{
394 return (struct tcp_sock *)sk;
395}
396
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700397struct tcp_timewait_sock {
398 struct inet_timewait_sock tw_sk;
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200399 u32 tw_rcv_nxt;
400 u32 tw_snd_nxt;
401 u32 tw_rcv_wnd;
402 u32 tw_ts_recent;
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700403 long tw_ts_recent_stamp;
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800404#ifdef CONFIG_TCP_MD5SIG
Arnaldo Carvalho de Melo3a137d22006-11-28 01:12:38 -0200405 u16 tw_md5_keylen;
406 u8 tw_md5_key[TCP_MD5SIG_MAXKEYLEN];
YOSHIFUJI Hideakicfb6eeb2006-11-14 19:07:45 -0800407#endif
Arnaldo Carvalho de Melo8feaf0c02005-08-09 20:09:30 -0700408};
409
410static inline struct tcp_timewait_sock *tcp_twsk(const struct sock *sk)
411{
412 return (struct tcp_timewait_sock *)sk;
413}
414
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415#endif
416
417#endif /* _LINUX_TCP_H */