projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
tcp: Fix MD5 signatures for non-linear skbs
[linux-2.6-block.git] / include / net / tcp.h
CommitLineData
1da177e4
LT		 1/*
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 module.
7 *
8 * Version: @(#)tcp.h 1.0.5 05/23/93
9 *
02c30a84 10 * Authors: Ross Biro
1da177e4
LT		 11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 *
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.
17 */
18#ifndef _TCP_H
19#define _TCP_H
20
21#define TCP_DEBUG 1
22#define FASTRETRANS_DEBUG 1
23
1da177e4
LT		 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>
fb286bb2 29#include <linux/skbuff.h>
97fc2f08 30#include <linux/dmaengine.h>
cfb6eeb4 31#include <linux/crypto.h>
c6aefafb 32#include <linux/cryptohash.h>
3f421baa
ACM		 33
34#include <net/inet_connection_sock.h>
295ff7ed 35#include <net/inet_timewait_sock.h>
77d8bf9c 36#include <net/inet_hashtables.h>
1da177e4 37#include <net/checksum.h>
2e6599cb 38#include <net/request_sock.h>
1da177e4
LT		 39#include <net/sock.h>
40#include <net/snmp.h>
41#include <net/ip.h>
c752f073 42#include <net/tcp_states.h>
bdf1ee5d 43#include <net/inet_ecn.h>
c752f073 44
1da177e4
LT		 45#include <linux/seq_file.h>
46
6e04e021 47extern struct inet_hashinfo tcp_hashinfo;
1da177e4 48
1da177e4 49extern atomic_t tcp_orphan_count;
1da177e4 50extern void tcp_time_wait(struct sock *sk, int state, int timeo);
1da177e4 51
1da177e4
LT		 52#define MAX_TCP_HEADER (128 + MAX_HEADER)
53
54/*
55 * Never offer a window over 32767 without using window scaling. Some
56 * poor stacks do signed 16bit maths!
57 */
58#define MAX_TCP_WINDOW 32767U
59
60/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
61#define TCP_MIN_MSS 88U
62
63/* Minimal RCV_MSS. */
64#define TCP_MIN_RCVMSS 536U
65
5d424d5a
JH		 66/* The least MTU to use for probing */
67#define TCP_BASE_MSS 512
68
1da177e4
LT		 69/* After receiving this amount of duplicate ACKs fast retransmit starts. */
70#define TCP_FASTRETRANS_THRESH 3
71
72/* Maximal reordering. */
73#define TCP_MAX_REORDERING 127
74
75/* Maximal number of ACKs sent quickly to accelerate slow-start. */
76#define TCP_MAX_QUICKACKS 16U
77
78/* urg_data states */
79#define TCP_URG_VALID 0x0100
80#define TCP_URG_NOTYET 0x0200
81#define TCP_URG_READ 0x0400
82
83#define TCP_RETR1 3 /*
84 * This is how many retries it does before it
85 * tries to figure out if the gateway is
86 * down. Minimal RFC value is 3; it corresponds
87 * to ~3sec-8min depending on RTO.
88 */
89
90#define TCP_RETR2 15 /*
91 * This should take at least
92 * 90 minutes to time out.
93 * RFC1122 says that the limit is 100 sec.
94 * 15 is ~13-30min depending on RTO.
95 */
96
97#define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
caa20d9a 98 * connection: ~180sec is RFC minimum */
1da177e4
LT		 99
100#define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
caa20d9a 101 * connection: ~180sec is RFC minimum */
1da177e4
LT		 102
103
104#define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
105 * socket. 7 is ~50sec-16min.
106 */
107
108
109#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
110 * state, about 60 seconds */
111#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
112 /* BSD style FIN_WAIT2 deadlock breaker.
113 * It used to be 3min, new value is 60sec,
114 * to combine FIN-WAIT-2 timeout with
115 * TIME-WAIT timer.
116 */
117
118#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
119#if HZ >= 100
120#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
121#define TCP_ATO_MIN ((unsigned)(HZ/25))
122#else
123#define TCP_DELACK_MIN 4U
124#define TCP_ATO_MIN 4U
125#endif
126#define TCP_RTO_MAX ((unsigned)(120*HZ))
127#define TCP_RTO_MIN ((unsigned)(HZ/5))
128#define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
129
130#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
131 * for local resources.
132 */
133
134#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
135#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
136#define TCP_KEEPALIVE_INTVL (75*HZ)
137
138#define MAX_TCP_KEEPIDLE 32767
139#define MAX_TCP_KEEPINTVL 32767
140#define MAX_TCP_KEEPCNT 127
141#define MAX_TCP_SYNCNT 127
142
143#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
1da177e4
LT		 144
145#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
146#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
147 * after this time. It should be equal
148 * (or greater than) TCP_TIMEWAIT_LEN
149 * to provide reliability equal to one
150 * provided by timewait state.
151 */
152#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
153 * timestamps. It must be less than
154 * minimal timewait lifetime.
155 */
1da177e4
LT		 156/*
157 * TCP option
158 */
159
160#define TCPOPT_NOP 1 /* Padding */
161#define TCPOPT_EOL 0 /* End of options */
162#define TCPOPT_MSS 2 /* Segment size negotiating */
163#define TCPOPT_WINDOW 3 /* Window scaling */
164#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
165#define TCPOPT_SACK 5 /* SACK Block */
166#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
cfb6eeb4 167#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
1da177e4
LT		 168
169/*
170 * TCP option lengths
171 */
172
173#define TCPOLEN_MSS 4
174#define TCPOLEN_WINDOW 3
175#define TCPOLEN_SACK_PERM 2
176#define TCPOLEN_TIMESTAMP 10
cfb6eeb4 177#define TCPOLEN_MD5SIG 18
1da177e4
LT		 178
179/* But this is what stacks really send out. */
180#define TCPOLEN_TSTAMP_ALIGNED 12
181#define TCPOLEN_WSCALE_ALIGNED 4
182#define TCPOLEN_SACKPERM_ALIGNED 4
183#define TCPOLEN_SACK_BASE 2
184#define TCPOLEN_SACK_BASE_ALIGNED 4
185#define TCPOLEN_SACK_PERBLOCK 8
cfb6eeb4 186#define TCPOLEN_MD5SIG_ALIGNED 20
1da177e4 187
1da177e4
LT		 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 */
caa20d9a 191#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
1da177e4 192
295ff7ed
ACM		 193extern struct inet_timewait_death_row tcp_death_row;
194
1da177e4 195/* sysctl variables for tcp */
1da177e4
LT		 196extern int sysctl_tcp_timestamps;
197extern int sysctl_tcp_window_scaling;
198extern int sysctl_tcp_sack;
199extern int sysctl_tcp_fin_timeout;
1da177e4
LT		 200extern int sysctl_tcp_keepalive_time;
201extern int sysctl_tcp_keepalive_probes;
202extern int sysctl_tcp_keepalive_intvl;
203extern int sysctl_tcp_syn_retries;
204extern int sysctl_tcp_synack_retries;
205extern int sysctl_tcp_retries1;
206extern int sysctl_tcp_retries2;
207extern int sysctl_tcp_orphan_retries;
208extern int sysctl_tcp_syncookies;
209extern int sysctl_tcp_retrans_collapse;
210extern int sysctl_tcp_stdurg;
211extern int sysctl_tcp_rfc1337;
212extern int sysctl_tcp_abort_on_overflow;
213extern int sysctl_tcp_max_orphans;
1da177e4
LT		 214extern int sysctl_tcp_fack;
215extern int sysctl_tcp_reordering;
216extern int sysctl_tcp_ecn;
217extern int sysctl_tcp_dsack;
218extern int sysctl_tcp_mem[3];
219extern int sysctl_tcp_wmem[3];
220extern int sysctl_tcp_rmem[3];
221extern int sysctl_tcp_app_win;
222extern int sysctl_tcp_adv_win_scale;
223extern int sysctl_tcp_tw_reuse;
224extern int sysctl_tcp_frto;
3cfe3baa 225extern int sysctl_tcp_frto_response;
1da177e4 226extern int sysctl_tcp_low_latency;
95937825 227extern int sysctl_tcp_dma_copybreak;
1da177e4 228extern int sysctl_tcp_nometrics_save;
1da177e4
LT		 229extern int sysctl_tcp_moderate_rcvbuf;
230extern int sysctl_tcp_tso_win_divisor;
9772efb9 231extern int sysctl_tcp_abc;
5d424d5a
JH		 232extern int sysctl_tcp_mtu_probing;
233extern int sysctl_tcp_base_mss;
15d99e02 234extern int sysctl_tcp_workaround_signed_windows;
35089bb2 235extern int sysctl_tcp_slow_start_after_idle;
886236c1 236extern int sysctl_tcp_max_ssthresh;
1da177e4
LT		 237
238extern atomic_t tcp_memory_allocated;
239extern atomic_t tcp_sockets_allocated;
240extern int tcp_memory_pressure;
241
1da177e4
LT		 242/*
243 * The next routines deal with comparing 32 bit unsigned ints
244 * and worry about wraparound (automatic with unsigned arithmetic).
245 */
246
247static inline int before(__u32 seq1, __u32 seq2)
248{
0d630cc0 249 return (__s32)(seq1-seq2) < 0;
1da177e4 250}
9a036b9c 251#define after(seq2, seq1) before(seq1, seq2)
1da177e4
LT		 252
253/* is s2<=s1<=s3 ? */
254static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
255{
256 return seq3 - seq2 >= seq1 - seq2;
257}
258
e4fd5da3
PE		 259static inline int tcp_too_many_orphans(struct sock *sk, int num)
260{
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]);
264}
1da177e4
LT		 265
266extern struct proto tcp_prot;
267
57ef42d5
PE		 268#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
269#define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
270#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
271#define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
1da177e4 272
1da177e4
LT		 273extern void tcp_v4_err(struct sk_buff *skb, u32);
274
275extern void tcp_shutdown (struct sock *sk, int how);
276
277extern int tcp_v4_rcv(struct sk_buff *skb);
278
279extern int tcp_v4_remember_stamp(struct sock *sk);
280
8feaf0c0 281extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
1da177e4 282
3516ffb0 283extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
1da177e4
LT		 284 struct msghdr *msg, size_t size);
285extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
286
287extern int tcp_ioctl(struct sock *sk,
288 int cmd,
289 unsigned long arg);
290
291extern int tcp_rcv_state_process(struct sock *sk,
292 struct sk_buff *skb,
293 struct tcphdr *th,
294 unsigned len);
295
296extern int tcp_rcv_established(struct sock *sk,
297 struct sk_buff *skb,
298 struct tcphdr *th,
299 unsigned len);
300
301extern void tcp_rcv_space_adjust(struct sock *sk);
302
0e4b4992
CL		 303extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
304
6d6ee43e
ACM		 305extern int tcp_twsk_unique(struct sock *sk,
306 struct sock *sktw, void *twp);
307
cfb6eeb4
YH		 308extern void tcp_twsk_destructor(struct sock *sk);
309
9c55e01c
JA		 310extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
311 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
312
463c84b9
ACM		 313static inline void tcp_dec_quickack_mode(struct sock *sk,
314 const unsigned int pkts)
1da177e4 315{
463c84b9 316 struct inet_connection_sock *icsk = inet_csk(sk);
fc6415bc 317
463c84b9
ACM		 318 if (icsk->icsk_ack.quick) {
319 if (pkts >= icsk->icsk_ack.quick) {
320 icsk->icsk_ack.quick = 0;
fc6415bc 321 /* Leaving quickack mode we deflate ATO. */
463c84b9 322 icsk->icsk_ack.ato = TCP_ATO_MIN;
fc6415bc 323 } else
463c84b9 324 icsk->icsk_ack.quick -= pkts;
1da177e4
LT		 325 }
326}
327
463c84b9 328extern void tcp_enter_quickack_mode(struct sock *sk);
1da177e4 329
1da177e4
LT		 330static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
331{
332 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
333}
334
bdf1ee5d
IJ		 335#define TCP_ECN_OK 1
336#define TCP_ECN_QUEUE_CWR 2
337#define TCP_ECN_DEMAND_CWR 4
338
339static __inline__ void
340TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
341{
342 if (sysctl_tcp_ecn && th->ece && th->cwr)
343 inet_rsk(req)->ecn_ok = 1;
344}
345
1da177e4
LT		 346enum tcp_tw_status
347{
348 TCP_TW_SUCCESS = 0,
349 TCP_TW_RST = 1,
350 TCP_TW_ACK = 2,
351 TCP_TW_SYN = 3
352};
353
354
8feaf0c0 355extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
1da177e4 356 struct sk_buff *skb,
8feaf0c0 357 const struct tcphdr *th);
1da177e4
LT		 358
359extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
60236fdd
ACM		 360 struct request_sock *req,
361 struct request_sock **prev);
1da177e4
LT		 362extern int tcp_child_process(struct sock *parent,
363 struct sock *child,
364 struct sk_buff *skb);
46d0de4e 365extern int tcp_use_frto(struct sock *sk);
1da177e4
LT		 366extern void tcp_enter_frto(struct sock *sk);
367extern void tcp_enter_loss(struct sock *sk, int how);
368extern void tcp_clear_retrans(struct tcp_sock *tp);
369extern void tcp_update_metrics(struct sock *sk);
370
371extern void tcp_close(struct sock *sk,
372 long timeout);
1da177e4
LT		 373extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
374
375extern int tcp_getsockopt(struct sock *sk, int level,
376 int optname,
377 char __user *optval,
378 int __user *optlen);
379extern int tcp_setsockopt(struct sock *sk, int level,
380 int optname, char __user *optval,
381 int optlen);
3fdadf7d
DM		 382extern int compat_tcp_getsockopt(struct sock *sk,
383 int level, int optname,
384 char __user *optval, int __user *optlen);
385extern int compat_tcp_setsockopt(struct sock *sk,
386 int level, int optname,
387 char __user *optval, int optlen);
1da177e4
LT		 388extern void tcp_set_keepalive(struct sock *sk, int val);
389extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
390 struct msghdr *msg,
391 size_t len, int nonblock,
392 int flags, int *addr_len);
393
1da177e4
LT		 394extern void tcp_parse_options(struct sk_buff *skb,
395 struct tcp_options_received *opt_rx,
396 int estab);
397
7d5d5525
YH		 398extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
399
1da177e4
LT		 400/*
401 * TCP v4 functions exported for the inet6 API
402 */
403
8292a17a 404extern void tcp_v4_send_check(struct sock *sk, int len,
1da177e4
LT		 405 struct sk_buff *skb);
406
407extern int tcp_v4_conn_request(struct sock *sk,
408 struct sk_buff *skb);
409
410extern struct sock * tcp_create_openreq_child(struct sock *sk,
60236fdd 411 struct request_sock *req,
1da177e4
LT		 412 struct sk_buff *skb);
413
414extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
415 struct sk_buff *skb,
60236fdd 416 struct request_sock *req,
1da177e4
LT		 417 struct dst_entry *dst);
418
419extern int tcp_v4_do_rcv(struct sock *sk,
420 struct sk_buff *skb);
421
422extern int tcp_v4_connect(struct sock *sk,
423 struct sockaddr *uaddr,
424 int addr_len);
425
426extern int tcp_connect(struct sock *sk);
427
428extern struct sk_buff * tcp_make_synack(struct sock *sk,
429 struct dst_entry *dst,
60236fdd 430 struct request_sock *req);
1da177e4
LT		 431
432extern int tcp_disconnect(struct sock *sk, int flags);
433
1da177e4 434
1da177e4 435/* From syncookies.c */
2051f11f 436extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
1da177e4
LT		 437extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
438 struct ip_options *opt);
439extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
440 __u16 *mss);
441
4dfc2817
FW		 442extern __u32 cookie_init_timestamp(struct request_sock *req);
443extern void cookie_check_timestamp(struct tcp_options_received *tcp_opt);
444
c6aefafb
GG		 445/* From net/ipv6/syncookies.c */
446extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
447extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
448 __u16 *mss);
449
1da177e4
LT		 450/* tcp_output.c */
451
9e412ba7
IJ		 452extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
453 int nonagle);
454extern int tcp_may_send_now(struct sock *sk);
1da177e4
LT		 455extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
456extern void tcp_xmit_retransmit_queue(struct sock *);
457extern void tcp_simple_retransmit(struct sock *);
458extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
6475be16 459extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
1da177e4
LT		 460
461extern void tcp_send_probe0(struct sock *);
462extern void tcp_send_partial(struct sock *);
463extern int tcp_write_wakeup(struct sock *);
464extern void tcp_send_fin(struct sock *sk);
dd0fc66f 465extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
1da177e4 466extern int tcp_send_synack(struct sock *);
c1b4a7e6 467extern void tcp_push_one(struct sock *, unsigned int mss_now);
1da177e4
LT		 468extern void tcp_send_ack(struct sock *sk);
469extern void tcp_send_delayed_ack(struct sock *sk);
470
a762a980
DM		 471/* tcp_input.c */
472extern void tcp_cwnd_application_limited(struct sock *sk);
473
1da177e4
LT		 474/* tcp_timer.c */
475extern void tcp_init_xmit_timers(struct sock *);
463c84b9
ACM		 476static inline void tcp_clear_xmit_timers(struct sock *sk)
477{
478 inet_csk_clear_xmit_timers(sk);
479}
1da177e4 480
1da177e4
LT		 481extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
482extern unsigned int tcp_current_mss(struct sock *sk, int large);
483
17b085ea 484/* tcp.c */
1da177e4
LT		 485extern void tcp_get_info(struct sock *, struct tcp_info *);
486
487/* Read 'sendfile()'-style from a TCP socket */
488typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
489 unsigned int, size_t);
490extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
491 sk_read_actor_t recv_actor);
492
40efc6fa 493extern void tcp_initialize_rcv_mss(struct sock *sk);
1da177e4 494
5d424d5a
JH		 495extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
496extern int tcp_mss_to_mtu(struct sock *sk, int mss);
497extern void tcp_mtup_init(struct sock *sk);
498
40efc6fa 499static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
1da177e4
LT		 500{
501 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
502 ntohl(TCP_FLAG_ACK) |
503 snd_wnd);
504}
505
40efc6fa 506static inline void tcp_fast_path_on(struct tcp_sock *tp)
1da177e4
LT		 507{
508 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
509}
510
9e412ba7 511static inline void tcp_fast_path_check(struct sock *sk)
1da177e4 512{
9e412ba7
IJ		 513 struct tcp_sock *tp = tcp_sk(sk);
514
b03efcfb 515 if (skb_queue_empty(&tp->out_of_order_queue) &&
1da177e4
LT		 516 tp->rcv_wnd &&
517 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
518 !tp->urg_data)
519 tcp_fast_path_on(tp);
520}
521
522/* Compute the actual receive window we are currently advertising.
523 * Rcv_nxt can be after the window if our peer push more data
524 * than the offered window.
525 */
40efc6fa 526static inline u32 tcp_receive_window(const struct tcp_sock *tp)
1da177e4
LT		 527{
528 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
529
530 if (win < 0)
531 win = 0;
532 return (u32) win;
533}
534
535/* Choose a new window, without checks for shrinking, and without
536 * scaling applied to the result. The caller does these things
537 * if necessary. This is a "raw" window selection.
538 */
539extern u32 __tcp_select_window(struct sock *sk);
540
541/* TCP timestamps are only 32-bits, this causes a slight
542 * complication on 64-bit systems since we store a snapshot
31f34269
SH		 543 * of jiffies in the buffer control blocks below. We decided
544 * to use only the low 32-bits of jiffies and hide the ugly
1da177e4
LT		 545 * casts with the following macro.
546 */
547#define tcp_time_stamp ((__u32)(jiffies))
548
caa20d9a 549/* This is what the send packet queuing engine uses to pass
1da177e4
LT		 550 * TCP per-packet control information to the transmission
551 * code. We also store the host-order sequence numbers in
552 * here too. This is 36 bytes on 32-bit architectures,
553 * 40 bytes on 64-bit machines, if this grows please adjust
554 * skbuff.h:skbuff->cb[xxx] size appropriately.
555 */
556struct tcp_skb_cb {
557 union {
558 struct inet_skb_parm h4;
559#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
560 struct inet6_skb_parm h6;
561#endif
562 } header; /* For incoming frames */
563 __u32 seq; /* Starting sequence number */
564 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
565 __u32 when; /* used to compute rtt's */
566 __u8 flags; /* TCP header flags. */
567
568 /* NOTE: These must match up to the flags byte in a
569 * real TCP header.
570 */
571#define TCPCB_FLAG_FIN 0x01
572#define TCPCB_FLAG_SYN 0x02
573#define TCPCB_FLAG_RST 0x04
574#define TCPCB_FLAG_PSH 0x08
575#define TCPCB_FLAG_ACK 0x10
576#define TCPCB_FLAG_URG 0x20
577#define TCPCB_FLAG_ECE 0x40
578#define TCPCB_FLAG_CWR 0x80
579
580 __u8 sacked; /* State flags for SACK/FACK. */
581#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
582#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
583#define TCPCB_LOST 0x04 /* SKB is lost */
584#define TCPCB_TAGBITS 0x07 /* All tag bits */
585
586#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
587#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
588
1da177e4
LT		 589 __u16 urg_ptr; /* Valid w/URG flags is set. */
590 __u32 ack_seq; /* Sequence number ACK'd */
591};
592
593#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
594
1da177e4
LT		 595/* Due to TSO, an SKB can be composed of multiple actual
596 * packets. To keep these tracked properly, we use this.
597 */
598static inline int tcp_skb_pcount(const struct sk_buff *skb)
599{
7967168c 600 return skb_shinfo(skb)->gso_segs;
1da177e4
LT		 601}
602
603/* This is valid iff tcp_skb_pcount() > 1. */
604static inline int tcp_skb_mss(const struct sk_buff *skb)
605{
7967168c 606 return skb_shinfo(skb)->gso_size;
1da177e4
LT		 607}
608
af610b4c 609static inline void tcp_dec_pcount_approx_int(__u32 *count, const int decr)
1da177e4
LT		 610{
611 if (*count) {
af610b4c 612 *count -= decr;
1da177e4
LT		 613 if ((int)*count < 0)
614 *count = 0;
615 }
616}
617
af610b4c
IJ		 618static inline void tcp_dec_pcount_approx(__u32 *count,
619 const struct sk_buff *skb)
620{
621 tcp_dec_pcount_approx_int(count, tcp_skb_pcount(skb));
622}
623
317a76f9
SH		 624/* Events passed to congestion control interface */
625enum tcp_ca_event {
626 CA_EVENT_TX_START, /* first transmit when no packets in flight */
627 CA_EVENT_CWND_RESTART, /* congestion window restart */
628 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
629 CA_EVENT_FRTO, /* fast recovery timeout */
630 CA_EVENT_LOSS, /* loss timeout */
631 CA_EVENT_FAST_ACK, /* in sequence ack */
632 CA_EVENT_SLOW_ACK, /* other ack */
633};
634
635/*
636 * Interface for adding new TCP congestion control handlers
637 */
638#define TCP_CA_NAME_MAX 16
3ff825b2
SH		 639#define TCP_CA_MAX 128
640#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
641
164891aa
SH		 642#define TCP_CONG_NON_RESTRICTED 0x1
643#define TCP_CONG_RTT_STAMP 0x2
644
317a76f9
SH		 645struct tcp_congestion_ops {
646 struct list_head list;
164891aa 647 unsigned long flags;
317a76f9
SH		 648
649 /* initialize private data (optional) */
6687e988 650 void (*init)(struct sock *sk);
317a76f9 651 /* cleanup private data (optional) */
6687e988 652 void (*release)(struct sock *sk);
317a76f9
SH		 653
654 /* return slow start threshold (required) */
6687e988 655 u32 (*ssthresh)(struct sock *sk);
317a76f9 656 /* lower bound for congestion window (optional) */
72dc5b92 657 u32 (*min_cwnd)(const struct sock *sk);
317a76f9 658 /* do new cwnd calculation (required) */
c3a05c60 659 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
317a76f9 660 /* call before changing ca_state (optional) */
6687e988 661 void (*set_state)(struct sock *sk, u8 new_state);
317a76f9 662 /* call when cwnd event occurs (optional) */
6687e988 663 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
317a76f9 664 /* new value of cwnd after loss (optional) */
6687e988 665 u32 (*undo_cwnd)(struct sock *sk);
317a76f9 666 /* hook for packet ack accounting (optional) */
30cfd0ba 667 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
73c1f4a0 668 /* get info for inet_diag (optional) */
6687e988 669 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
317a76f9
SH		 670
671 char name[TCP_CA_NAME_MAX];
672 struct module *owner;
673};
674
675extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
676extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
677
6687e988
ACM		 678extern void tcp_init_congestion_control(struct sock *sk);
679extern void tcp_cleanup_congestion_control(struct sock *sk);
317a76f9
SH		 680extern int tcp_set_default_congestion_control(const char *name);
681extern void tcp_get_default_congestion_control(char *name);
3ff825b2 682extern void tcp_get_available_congestion_control(char *buf, size_t len);
ce7bc3bf
SH		 683extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
684extern int tcp_set_allowed_congestion_control(char *allowed);
6687e988 685extern int tcp_set_congestion_control(struct sock *sk, const char *name);
40efc6fa 686extern void tcp_slow_start(struct tcp_sock *tp);
317a76f9 687
5f8ef48d 688extern struct tcp_congestion_ops tcp_init_congestion_ops;
6687e988 689extern u32 tcp_reno_ssthresh(struct sock *sk);
c3a05c60 690extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
72dc5b92 691extern u32 tcp_reno_min_cwnd(const struct sock *sk);
a8acfbac 692extern struct tcp_congestion_ops tcp_reno;
317a76f9 693
6687e988 694static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
317a76f9 695{
6687e988
ACM		 696 struct inet_connection_sock *icsk = inet_csk(sk);
697
698 if (icsk->icsk_ca_ops->set_state)
699 icsk->icsk_ca_ops->set_state(sk, ca_state);
700 icsk->icsk_ca_state = ca_state;
317a76f9
SH		 701}
702
6687e988 703static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
317a76f9 704{
6687e988
ACM		 705 const struct inet_connection_sock *icsk = inet_csk(sk);
706
707 if (icsk->icsk_ca_ops->cwnd_event)
708 icsk->icsk_ca_ops->cwnd_event(sk, event);
317a76f9
SH		 709}
710
e60402d0
IJ		 711/* These functions determine how the current flow behaves in respect of SACK
712 * handling. SACK is negotiated with the peer, and therefore it can vary
713 * between different flows.
714 *
715 * tcp_is_sack - SACK enabled
716 * tcp_is_reno - No SACK
717 * tcp_is_fack - FACK enabled, implies SACK enabled
718 */
719static inline int tcp_is_sack(const struct tcp_sock *tp)
720{
721 return tp->rx_opt.sack_ok;
722}
723
724static inline int tcp_is_reno(const struct tcp_sock *tp)
725{
726 return !tcp_is_sack(tp);
727}
728
729static inline int tcp_is_fack(const struct tcp_sock *tp)
730{
731 return tp->rx_opt.sack_ok & 2;
732}
733
734static inline void tcp_enable_fack(struct tcp_sock *tp)
735{
736 tp->rx_opt.sack_ok |= 2;
737}
738
83ae4088
IJ		 739static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
740{
741 return tp->sacked_out + tp->lost_out;
742}
743
1da177e4
LT		 744/* This determines how many packets are "in the network" to the best
745 * of our knowledge. In many cases it is conservative, but where
746 * detailed information is available from the receiver (via SACK
747 * blocks etc.) we can make more aggressive calculations.
748 *
749 * Use this for decisions involving congestion control, use just
750 * tp->packets_out to determine if the send queue is empty or not.
751 *
752 * Read this equation as:
753 *
754 * "Packets sent once on transmission queue" MINUS
755 * "Packets left network, but not honestly ACKed yet" PLUS
756 * "Packets fast retransmitted"
757 */
40efc6fa 758static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1da177e4 759{
83ae4088 760 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1da177e4
LT		 761}
762
882bebaa
IJ		 763extern int tcp_limit_reno_sacked(struct tcp_sock *tp);
764
1da177e4
LT		 765/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
766 * The exception is rate halving phase, when cwnd is decreasing towards
767 * ssthresh.
768 */
6687e988 769static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1da177e4 770{
6687e988
ACM		 771 const struct tcp_sock *tp = tcp_sk(sk);
772 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
1da177e4
LT		 773 return tp->snd_ssthresh;
774 else
775 return max(tp->snd_ssthresh,
776 ((tp->snd_cwnd >> 1) +
777 (tp->snd_cwnd >> 2)));
778}
779
b9c4595b
IJ		 780/* Use define here intentionally to get WARN_ON location shown at the caller */
781#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1da177e4 782
3cfe3baa 783extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
1da177e4
LT		 784extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
785
786/* Slow start with delack produces 3 packets of burst, so that
dd9e0dda
JH		 787 * it is safe "de facto". This will be the default - same as
788 * the default reordering threshold - but if reordering increases,
789 * we must be able to allow cwnd to burst at least this much in order
790 * to not pull it back when holes are filled.
1da177e4
LT		 791 */
792static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
793{
dd9e0dda 794 return tp->reordering;
1da177e4
LT		 795}
796
90840def
IJ		 797/* Returns end sequence number of the receiver's advertised window */
798static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
799{
800 return tp->snd_una + tp->snd_wnd;
801}
cea14e0e 802extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
f4805ede 803
c1bd24b7 804static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
40efc6fa 805 const struct sk_buff *skb)
1da177e4
LT		 806{
807 if (skb->len < mss)
808 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
809}
810
9e412ba7 811static inline void tcp_check_probe_timer(struct sock *sk)
1da177e4 812{
9e412ba7 813 struct tcp_sock *tp = tcp_sk(sk);
463c84b9 814 const struct inet_connection_sock *icsk = inet_csk(sk);
9e412ba7 815
463c84b9 816 if (!tp->packets_out && !icsk->icsk_pending)
3f421baa
ACM		 817 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
818 icsk->icsk_rto, TCP_RTO_MAX);
1da177e4
LT		 819}
820
9e412ba7 821static inline void tcp_push_pending_frames(struct sock *sk)
1da177e4 822{
9e412ba7
IJ		 823 struct tcp_sock *tp = tcp_sk(sk);
824
825 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle);
1da177e4
LT		 826}
827
40efc6fa 828static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1da177e4
LT		 829{
830 tp->snd_wl1 = seq;
831}
832
40efc6fa 833static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1da177e4
LT		 834{
835 tp->snd_wl1 = seq;
836}
837
1da177e4
LT		 838/*
839 * Calculate(/check) TCP checksum
840 */
ba7808ea
FD		 841static inline __sum16 tcp_v4_check(int len, __be32 saddr,
842 __be32 daddr, __wsum base)
1da177e4
LT		 843{
844 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
845}
846
b51655b9 847static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1da177e4 848{
fb286bb2 849 return __skb_checksum_complete(skb);
1da177e4
LT		 850}
851
40efc6fa 852static inline int tcp_checksum_complete(struct sk_buff *skb)
1da177e4 853{
60476372 854 return !skb_csum_unnecessary(skb) &&
1da177e4
LT		 855 __tcp_checksum_complete(skb);
856}
857
858/* Prequeue for VJ style copy to user, combined with checksumming. */
859
40efc6fa 860static inline void tcp_prequeue_init(struct tcp_sock *tp)
1da177e4
LT		 861{
862 tp->ucopy.task = NULL;
863 tp->ucopy.len = 0;
864 tp->ucopy.memory = 0;
865 skb_queue_head_init(&tp->ucopy.prequeue);
97fc2f08
CL		 866#ifdef CONFIG_NET_DMA
867 tp->ucopy.dma_chan = NULL;
868 tp->ucopy.wakeup = 0;
869 tp->ucopy.pinned_list = NULL;
870 tp->ucopy.dma_cookie = 0;
871#endif
1da177e4
LT		 872}
873
874/* Packet is added to VJ-style prequeue for processing in process
875 * context, if a reader task is waiting. Apparently, this exciting
876 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
877 * failed somewhere. Latency? Burstiness? Well, at least now we will
878 * see, why it failed. 8)8) --ANK
879 *
880 * NOTE: is this not too big to inline?
881 */
40efc6fa 882static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1da177e4
LT		 883{
884 struct tcp_sock *tp = tcp_sk(sk);
885
886 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
887 __skb_queue_tail(&tp->ucopy.prequeue, skb);
888 tp->ucopy.memory += skb->truesize;
889 if (tp->ucopy.memory > sk->sk_rcvbuf) {
890 struct sk_buff *skb1;
891
892 BUG_ON(sock_owned_by_user(sk));
893
894 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
895 sk->sk_backlog_rcv(sk, skb1);
de0744af 896 NET_INC_STATS_BH(sock_net(sk), LINUX_MIB_TCPPREQUEUEDROPPED);
1da177e4
LT		 897 }
898
899 tp->ucopy.memory = 0;
900 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
901 wake_up_interruptible(sk->sk_sleep);
463c84b9
ACM		 902 if (!inet_csk_ack_scheduled(sk))
903 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
3f421baa
ACM		 904 (3 * TCP_RTO_MIN) / 4,
905 TCP_RTO_MAX);
1da177e4
LT		 906 }
907 return 1;
908 }
909 return 0;
910}
911
912
913#undef STATE_TRACE
914
915#ifdef STATE_TRACE
916static const char *statename[]={
917 "Unused","Established","Syn Sent","Syn Recv",
918 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
919 "Close Wait","Last ACK","Listen","Closing"
920};
921#endif
490d5046 922extern void tcp_set_state(struct sock *sk, int state);
1da177e4 923
4ac02bab 924extern void tcp_done(struct sock *sk);
1da177e4 925
40efc6fa 926static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1da177e4
LT		 927{
928 rx_opt->dsack = 0;
929 rx_opt->eff_sacks = 0;
930 rx_opt->num_sacks = 0;
931}
932
1da177e4
LT		 933/* Determine a window scaling and initial window to offer. */
934extern void tcp_select_initial_window(int __space, __u32 mss,
935 __u32 *rcv_wnd, __u32 *window_clamp,
936 int wscale_ok, __u8 *rcv_wscale);
937
938static inline int tcp_win_from_space(int space)
939{
940 return sysctl_tcp_adv_win_scale<=0 ?
941 (space>>(-sysctl_tcp_adv_win_scale)) :
942 space - (space>>sysctl_tcp_adv_win_scale);
943}
944
945/* Note: caller must be prepared to deal with negative returns */
946static inline int tcp_space(const struct sock *sk)
947{
948 return tcp_win_from_space(sk->sk_rcvbuf -
949 atomic_read(&sk->sk_rmem_alloc));
950}
951
952static inline int tcp_full_space(const struct sock *sk)
953{
954 return tcp_win_from_space(sk->sk_rcvbuf);
955}
956
40efc6fa
SH		 957static inline void tcp_openreq_init(struct request_sock *req,
958 struct tcp_options_received *rx_opt,
959 struct sk_buff *skb)
1da177e4 960{
2e6599cb
ACM		 961 struct inet_request_sock *ireq = inet_rsk(req);
962
1da177e4 963 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
4dfc2817 964 req->cookie_ts = 0;
2e6599cb 965 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1da177e4
LT		 966 req->mss = rx_opt->mss_clamp;
967 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
2e6599cb
ACM		 968 ireq->tstamp_ok = rx_opt->tstamp_ok;
969 ireq->sack_ok = rx_opt->sack_ok;
970 ireq->snd_wscale = rx_opt->snd_wscale;
971 ireq->wscale_ok = rx_opt->wscale_ok;
972 ireq->acked = 0;
973 ireq->ecn_ok = 0;
aa8223c7 974 ireq->rmt_port = tcp_hdr(skb)->source;
1da177e4
LT		 975}
976
5c52ba17 977extern void tcp_enter_memory_pressure(struct sock *sk);
1da177e4 978
1da177e4
LT		 979static inline int keepalive_intvl_when(const struct tcp_sock *tp)
980{
981 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
982}
983
984static inline int keepalive_time_when(const struct tcp_sock *tp)
985{
986 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
987}
988
463c84b9 989static inline int tcp_fin_time(const struct sock *sk)
1da177e4 990{
463c84b9
ACM		 991 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
992 const int rto = inet_csk(sk)->icsk_rto;
1da177e4 993
463c84b9
ACM		 994 if (fin_timeout < (rto << 2) - (rto >> 1))
995 fin_timeout = (rto << 2) - (rto >> 1);
1da177e4
LT		 996
997 return fin_timeout;
998}
999
1000static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
1001{
1002 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
1003 return 0;
9d729f72 1004 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1da177e4
LT		 1005 return 0;
1006
1007 /* RST segments are not recommended to carry timestamp,
1008 and, if they do, it is recommended to ignore PAWS because
1009 "their cleanup function should take precedence over timestamps."
1010 Certainly, it is mistake. It is necessary to understand the reasons
1011 of this constraint to relax it: if peer reboots, clock may go
1012 out-of-sync and half-open connections will not be reset.
1013 Actually, the problem would be not existing if all
1014 the implementations followed draft about maintaining clock
1015 via reboots. Linux-2.2 DOES NOT!
1016
1017 However, we can relax time bounds for RST segments to MSL.
1018 */
9d729f72 1019 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1da177e4
LT		 1020 return 0;
1021 return 1;
1022}
1023
1da177e4
LT		 1024#define TCP_CHECK_TIMER(sk) do { } while (0)
1025
a9c19329 1026static inline void tcp_mib_init(struct net *net)
1da177e4
LT		 1027{
1028 /* See RFC 2012 */
cf1100a7
PE		 1029 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1030 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1031 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1032 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1da177e4
LT		 1033}
1034
5af4ec23 1035/* from STCP */
0800f170
DM		 1036static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1037{
6a438bbe
SH		 1038 tp->lost_skb_hint = NULL;
1039 tp->scoreboard_skb_hint = NULL;
1040 tp->retransmit_skb_hint = NULL;
1041 tp->forward_skb_hint = NULL;
b7689205
IJ		 1042}
1043
0800f170
DM		 1044static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1045{
b7689205 1046 tcp_clear_retrans_hints_partial(tp);
6a438bbe
SH		 1047}
1048
cfb6eeb4
YH		 1049/* MD5 Signature */
1050struct crypto_hash;
1051
1052/* - key database */
1053struct tcp_md5sig_key {
1054 u8 *key;
1055 u8 keylen;
1056};
1057
1058struct tcp4_md5sig_key {
f8ab18d2 1059 struct tcp_md5sig_key base;
cfb6eeb4
YH		 1060 __be32 addr;
1061};
1062
1063struct tcp6_md5sig_key {
f8ab18d2 1064 struct tcp_md5sig_key base;
cfb6eeb4
YH		 1065#if 0
1066 u32 scope_id; /* XXX */
1067#endif
1068 struct in6_addr addr;
1069};
1070
1071/* - sock block */
1072struct tcp_md5sig_info {
1073 struct tcp4_md5sig_key *keys4;
1074#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1075 struct tcp6_md5sig_key *keys6;
1076 u32 entries6;
1077 u32 alloced6;
1078#endif
1079 u32 entries4;
1080 u32 alloced4;
1081};
1082
1083/* - pseudo header */
1084struct tcp4_pseudohdr {
1085 __be32 saddr;
1086 __be32 daddr;
1087 __u8 pad;
1088 __u8 protocol;
1089 __be16 len;
1090};
1091
1092struct tcp6_pseudohdr {
1093 struct in6_addr saddr;
1094 struct in6_addr daddr;
1095 __be32 len;
1096 __be32 protocol; /* including padding */
1097};
1098
1099union tcp_md5sum_block {
1100 struct tcp4_pseudohdr ip4;
1101#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1102 struct tcp6_pseudohdr ip6;
1103#endif
1104};
1105
1106/* - pool: digest algorithm, hash description and scratch buffer */
1107struct tcp_md5sig_pool {
1108 struct hash_desc md5_desc;
1109 union tcp_md5sum_block md5_blk;
1110};
1111
1112#define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1113
1114/* - functions */
49a72dfb
AL		 1115extern int tcp_v4_md5_hash_skb(char *md5_hash,
1116 struct tcp_md5sig_key *key,
1117 struct sock *sk,
1118 struct request_sock *req,
1119 struct sk_buff *skb);
1120
cfb6eeb4
YH		 1121extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1122 struct sock *addr_sk);
1123
1124extern int tcp_v4_md5_do_add(struct sock *sk,
1125 __be32 addr,
1126 u8 *newkey,
1127 u8 newkeylen);
1128
1129extern int tcp_v4_md5_do_del(struct sock *sk,
8e5200f5 1130 __be32 addr);
cfb6eeb4 1131
9501f972
YH		 1132#ifdef CONFIG_TCP_MD5SIG
1133#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1134 &(struct tcp_md5sig_key) { \
1135 .key = (twsk)->tw_md5_key, \
1136 .keylen = (twsk)->tw_md5_keylen, \
1137 } : NULL)
1138#else
1139#define tcp_twsk_md5_key(twsk) NULL
1140#endif
1141
cfb6eeb4
YH		 1142extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1143extern void tcp_free_md5sig_pool(void);
1144
1145extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1146extern void __tcp_put_md5sig_pool(void);
49a72dfb
AL		 1147extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1148extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1149 unsigned header_len);
1150extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1151 struct tcp_md5sig_key *key);
cfb6eeb4
YH		 1152
1153static inline
1154struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1155{
1156 int cpu = get_cpu();
1157 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1158 if (!ret)
1159 put_cpu();
1160 return ret;
1161}
1162
1163static inline void tcp_put_md5sig_pool(void)
1164{
1165 __tcp_put_md5sig_pool();
1166 put_cpu();
1167}
1168
fe067e8a
DM		 1169/* write queue abstraction */
1170static inline void tcp_write_queue_purge(struct sock *sk)
1171{
1172 struct sk_buff *skb;
1173
1174 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
3ab224be
HA		 1175 sk_wmem_free_skb(sk, skb);
1176 sk_mem_reclaim(sk);
fe067e8a
DM		 1177}
1178
1179static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1180{
1181 struct sk_buff *skb = sk->sk_write_queue.next;
1182 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1183 return NULL;
1184 return skb;
1185}
1186
1187static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1188{
1189 struct sk_buff *skb = sk->sk_write_queue.prev;
1190 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1191 return NULL;
1192 return skb;
1193}
1194
1195static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1196{
1197 return skb->next;
1198}
1199
1200#define tcp_for_write_queue(skb, sk) \
1201 for (skb = (sk)->sk_write_queue.next; \
1202 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1203 skb = skb->next)
1204
1205#define tcp_for_write_queue_from(skb, sk) \
1206 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\
1207 skb = skb->next)
1208
234b6860
IJ		 1209#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1210 for (tmp = skb->next; \
1211 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1212 skb = tmp, tmp = skb->next)
1213
fe067e8a
DM		 1214static inline struct sk_buff *tcp_send_head(struct sock *sk)
1215{
1216 return sk->sk_send_head;
1217}
1218
1219static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1220{
1221 sk->sk_send_head = skb->next;
1222 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue)
1223 sk->sk_send_head = NULL;
1224}
1225
1226static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1227{
1228 if (sk->sk_send_head == skb_unlinked)
1229 sk->sk_send_head = NULL;
1230}
1231
1232static inline void tcp_init_send_head(struct sock *sk)
1233{
1234 sk->sk_send_head = NULL;
1235}
1236
1237static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1238{
1239 __skb_queue_tail(&sk->sk_write_queue, skb);
1240}
1241
1242static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1243{
1244 __tcp_add_write_queue_tail(sk, skb);
1245
1246 /* Queue it, remembering where we must start sending. */
6859d494 1247 if (sk->sk_send_head == NULL) {
fe067e8a 1248 sk->sk_send_head = skb;
6859d494
IJ		 1249
1250 if (tcp_sk(sk)->highest_sack == NULL)
1251 tcp_sk(sk)->highest_sack = skb;
1252 }
fe067e8a
DM		 1253}
1254
1255static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1256{
1257 __skb_queue_head(&sk->sk_write_queue, skb);
1258}
1259
1260/* Insert buff after skb on the write queue of sk. */
1261static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1262 struct sk_buff *buff,
1263 struct sock *sk)
1264{
7de6c033 1265 __skb_queue_after(&sk->sk_write_queue, skb, buff);
fe067e8a
DM		 1266}
1267
1268/* Insert skb between prev and next on the write queue of sk. */
1269static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1270 struct sk_buff *skb,
1271 struct sock *sk)
1272{
1273 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue);