]> bbs.cooldavid.org Git - net-next-2.6.git/blame - include/net/tcp.h
git://bbs.cooldavid.org / net-next-2.6.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
tcp: Combat per-cpu skew in orphan tests.
[net-next-2.6.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>
435cf559 33#include <linux/kref.h>
3f421baa
ACM		 34
35#include <net/inet_connection_sock.h>
295ff7ed 36#include <net/inet_timewait_sock.h>
77d8bf9c 37#include <net/inet_hashtables.h>
1da177e4 38#include <net/checksum.h>
2e6599cb 39#include <net/request_sock.h>
1da177e4
LT		 40#include <net/sock.h>
41#include <net/snmp.h>
42#include <net/ip.h>
c752f073 43#include <net/tcp_states.h>
bdf1ee5d 44#include <net/inet_ecn.h>
0c266898 45#include <net/dst.h>
c752f073 46
1da177e4
LT		 47#include <linux/seq_file.h>
48
6e04e021 49extern struct inet_hashinfo tcp_hashinfo;
1da177e4 50
dd24c001 51extern struct percpu_counter tcp_orphan_count;
1da177e4 52extern void tcp_time_wait(struct sock *sk, int state, int timeo);
1da177e4 53
1da177e4 54#define MAX_TCP_HEADER (128 + MAX_HEADER)
33ad798c 55#define MAX_TCP_OPTION_SPACE 40
1da177e4
LT		 56
57/*
58 * Never offer a window over 32767 without using window scaling. Some
59 * poor stacks do signed 16bit maths!
60 */
61#define MAX_TCP_WINDOW 32767U
62
63/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
64#define TCP_MIN_MSS 88U
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) */
435cf559 168#define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
1da177e4
LT		 169
170/*
171 * TCP option lengths
172 */
173
174#define TCPOLEN_MSS 4
175#define TCPOLEN_WINDOW 3
176#define TCPOLEN_SACK_PERM 2
177#define TCPOLEN_TIMESTAMP 10
cfb6eeb4 178#define TCPOLEN_MD5SIG 18
435cf559
WAS		 179#define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
180#define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
181#define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
182#define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
1da177e4
LT		 183
184/* But this is what stacks really send out. */
185#define TCPOLEN_TSTAMP_ALIGNED 12
186#define TCPOLEN_WSCALE_ALIGNED 4
187#define TCPOLEN_SACKPERM_ALIGNED 4
188#define TCPOLEN_SACK_BASE 2
189#define TCPOLEN_SACK_BASE_ALIGNED 4
190#define TCPOLEN_SACK_PERBLOCK 8
cfb6eeb4 191#define TCPOLEN_MD5SIG_ALIGNED 20
33ad798c 192#define TCPOLEN_MSS_ALIGNED 4
1da177e4 193
1da177e4
LT		 194/* Flags in tp->nonagle */
195#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
196#define TCP_NAGLE_CORK 2 /* Socket is corked */
caa20d9a 197#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
1da177e4 198
36e31b0a
AP		 199/* TCP thin-stream limits */
200#define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
201
295ff7ed
ACM		 202extern struct inet_timewait_death_row tcp_death_row;
203
1da177e4 204/* sysctl variables for tcp */
1da177e4
LT		 205extern int sysctl_tcp_timestamps;
206extern int sysctl_tcp_window_scaling;
207extern int sysctl_tcp_sack;
208extern int sysctl_tcp_fin_timeout;
1da177e4
LT		 209extern int sysctl_tcp_keepalive_time;
210extern int sysctl_tcp_keepalive_probes;
211extern int sysctl_tcp_keepalive_intvl;
212extern int sysctl_tcp_syn_retries;
213extern int sysctl_tcp_synack_retries;
214extern int sysctl_tcp_retries1;
215extern int sysctl_tcp_retries2;
216extern int sysctl_tcp_orphan_retries;
217extern int sysctl_tcp_syncookies;
218extern int sysctl_tcp_retrans_collapse;
219extern int sysctl_tcp_stdurg;
220extern int sysctl_tcp_rfc1337;
221extern int sysctl_tcp_abort_on_overflow;
222extern int sysctl_tcp_max_orphans;
1da177e4
LT		 223extern int sysctl_tcp_fack;
224extern int sysctl_tcp_reordering;
225extern int sysctl_tcp_ecn;
226extern int sysctl_tcp_dsack;
227extern int sysctl_tcp_mem[3];
228extern int sysctl_tcp_wmem[3];
229extern int sysctl_tcp_rmem[3];
230extern int sysctl_tcp_app_win;
231extern int sysctl_tcp_adv_win_scale;
232extern int sysctl_tcp_tw_reuse;
233extern int sysctl_tcp_frto;
3cfe3baa 234extern int sysctl_tcp_frto_response;
1da177e4 235extern int sysctl_tcp_low_latency;
95937825 236extern int sysctl_tcp_dma_copybreak;
1da177e4 237extern int sysctl_tcp_nometrics_save;
1da177e4
LT		 238extern int sysctl_tcp_moderate_rcvbuf;
239extern int sysctl_tcp_tso_win_divisor;
9772efb9 240extern int sysctl_tcp_abc;
5d424d5a
JH		 241extern int sysctl_tcp_mtu_probing;
242extern int sysctl_tcp_base_mss;
15d99e02 243extern int sysctl_tcp_workaround_signed_windows;
35089bb2 244extern int sysctl_tcp_slow_start_after_idle;
886236c1 245extern int sysctl_tcp_max_ssthresh;
519855c5 246extern int sysctl_tcp_cookie_size;
36e31b0a 247extern int sysctl_tcp_thin_linear_timeouts;
7e380175 248extern int sysctl_tcp_thin_dupack;
1da177e4
LT		 249
250extern atomic_t tcp_memory_allocated;
1748376b 251extern struct percpu_counter tcp_sockets_allocated;
1da177e4
LT		 252extern int tcp_memory_pressure;
253
1da177e4
LT		 254/*
255 * The next routines deal with comparing 32 bit unsigned ints
256 * and worry about wraparound (automatic with unsigned arithmetic).
257 */
258
259static inline int before(__u32 seq1, __u32 seq2)
260{
0d630cc0 261 return (__s32)(seq1-seq2) < 0;
1da177e4 262}
9a036b9c 263#define after(seq2, seq1) before(seq1, seq2)
1da177e4
LT		 264
265/* is s2<=s1<=s3 ? */
266static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
267{
268 return seq3 - seq2 >= seq1 - seq2;
269}
270
ad1af0fe 271static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
e4fd5da3 272{
ad1af0fe
DM		 273 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
274 int orphans = percpu_counter_read_positive(ocp);
275
276 if (orphans << shift > sysctl_tcp_max_orphans) {
277 orphans = percpu_counter_sum_positive(ocp);
278 if (orphans << shift > sysctl_tcp_max_orphans)
279 return true;
280 }
281
282 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
283 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2])
284 return true;
285 return false;
e4fd5da3 286}
1da177e4 287
a0f82f64
FW		 288/* syncookies: remember time of last synqueue overflow */
289static inline void tcp_synq_overflow(struct sock *sk)
290{
291 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
292}
293
294/* syncookies: no recent synqueue overflow on this listening socket? */
295static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
296{
297 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
298 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT);
299}
300
1da177e4
LT		 301extern struct proto tcp_prot;
302
57ef42d5
PE		 303#define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
304#define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
305#define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
306#define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
aa2ea058 307#define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
1da177e4 308
53d3176b
CG		 309extern void tcp_v4_err(struct sk_buff *skb, u32);
310
311extern void tcp_shutdown (struct sock *sk, int how);
312
313extern int tcp_v4_rcv(struct sk_buff *skb);
314
315extern int tcp_v4_remember_stamp(struct sock *sk);
316extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
7ba42910
CG		 317extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
318 size_t size);
319extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
320 size_t size, int flags);
53d3176b
CG		 321extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
322extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
323 struct tcphdr *th, unsigned len);
324extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
325 struct tcphdr *th, unsigned len);
326extern void tcp_rcv_space_adjust(struct sock *sk);
327extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
328extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
329extern void tcp_twsk_destructor(struct sock *sk);
330extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
331 struct pipe_inode_info *pipe, size_t len,
332 unsigned int flags);
9c55e01c 333
463c84b9
ACM		 334static inline void tcp_dec_quickack_mode(struct sock *sk,
335 const unsigned int pkts)
1da177e4 336{
463c84b9 337 struct inet_connection_sock *icsk = inet_csk(sk);
fc6415bc 338
463c84b9
ACM		 339 if (icsk->icsk_ack.quick) {
340 if (pkts >= icsk->icsk_ack.quick) {
341 icsk->icsk_ack.quick = 0;
fc6415bc 342 /* Leaving quickack mode we deflate ATO. */
463c84b9 343 icsk->icsk_ack.ato = TCP_ATO_MIN;
fc6415bc 344 } else
463c84b9 345 icsk->icsk_ack.quick -= pkts;
1da177e4
LT		 346 }
347}
348
463c84b9 349extern void tcp_enter_quickack_mode(struct sock *sk);
1da177e4 350
bdf1ee5d
IJ		 351#define TCP_ECN_OK 1
352#define TCP_ECN_QUEUE_CWR 2
353#define TCP_ECN_DEMAND_CWR 4
354
355static __inline__ void
356TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
357{
358 if (sysctl_tcp_ecn && th->ece && th->cwr)
359 inet_rsk(req)->ecn_ok = 1;
360}
361
fd2c3ef7 362enum tcp_tw_status {
1da177e4
LT		 363 TCP_TW_SUCCESS = 0,
364 TCP_TW_RST = 1,
365 TCP_TW_ACK = 2,
366 TCP_TW_SYN = 3
367};
368
369
53d3176b
CG		 370extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
371 struct sk_buff *skb,
372 const struct tcphdr *th);
373extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
374 struct request_sock *req,
375 struct request_sock **prev);
376extern int tcp_child_process(struct sock *parent, struct sock *child,
377 struct sk_buff *skb);
378extern int tcp_use_frto(struct sock *sk);
379extern void tcp_enter_frto(struct sock *sk);
380extern void tcp_enter_loss(struct sock *sk, int how);
381extern void tcp_clear_retrans(struct tcp_sock *tp);
382extern void tcp_update_metrics(struct sock *sk);
383extern void tcp_close(struct sock *sk, long timeout);
384extern unsigned int tcp_poll(struct file * file, struct socket *sock,
385 struct poll_table_struct *wait);
386extern int tcp_getsockopt(struct sock *sk, int level, int optname,
387 char __user *optval, int __user *optlen);
388extern int tcp_setsockopt(struct sock *sk, int level, int optname,
389 char __user *optval, unsigned int optlen);
390extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
391 char __user *optval, int __user *optlen);
392extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
393 char __user *optval, unsigned int optlen);
394extern void tcp_set_keepalive(struct sock *sk, int val);
395extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
396extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
397 size_t len, int nonblock, int flags, int *addr_len);
398extern void tcp_parse_options(struct sk_buff *skb,
399 struct tcp_options_received *opt_rx, u8 **hvpp,
400 int estab);
401extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
7d5d5525 402
1da177e4
LT		 403/*
404 * TCP v4 functions exported for the inet6 API
405 */
406
53d3176b
CG		 407extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
408extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
409extern struct sock * tcp_create_openreq_child(struct sock *sk,
410 struct request_sock *req,
1da177e4 411 struct sk_buff *skb);
53d3176b
CG		 412extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
413 struct request_sock *req,
414 struct dst_entry *dst);
415extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
416extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
417 int addr_len);
418extern int tcp_connect(struct sock *sk);
419extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
420 struct request_sock *req,
421 struct request_values *rvp);
422extern int tcp_disconnect(struct sock *sk, int flags);
1da177e4 423
1da177e4 424
1da177e4 425/* From syncookies.c */
2051f11f 426extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
1da177e4
LT		 427extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
428 struct ip_options *opt);
429extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
430 __u16 *mss);
431
4dfc2817 432extern __u32 cookie_init_timestamp(struct request_sock *req);
172d69e6 433extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *);
4dfc2817 434
c6aefafb
GG		 435/* From net/ipv6/syncookies.c */
436extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
437extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
438 __u16 *mss);
439
1da177e4
LT		 440/* tcp_output.c */
441
9e412ba7
IJ		 442extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
443 int nonagle);
444extern int tcp_may_send_now(struct sock *sk);
1da177e4 445extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
f1ecd5d9 446extern void tcp_retransmit_timer(struct sock *sk);
1da177e4
LT		 447extern void tcp_xmit_retransmit_queue(struct sock *);
448extern void tcp_simple_retransmit(struct sock *);
449extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
6475be16 450extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
1da177e4
LT		 451
452extern void tcp_send_probe0(struct sock *);
453extern void tcp_send_partial(struct sock *);
53d3176b 454extern int tcp_write_wakeup(struct sock *);
1da177e4 455extern void tcp_send_fin(struct sock *sk);
dd0fc66f 456extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
53d3176b 457extern int tcp_send_synack(struct sock *);
c1b4a7e6 458extern void tcp_push_one(struct sock *, unsigned int mss_now);
1da177e4
LT		 459extern void tcp_send_ack(struct sock *sk);
460extern void tcp_send_delayed_ack(struct sock *sk);
461
a762a980
DM		 462/* tcp_input.c */
463extern void tcp_cwnd_application_limited(struct sock *sk);
464
1da177e4
LT		 465/* tcp_timer.c */
466extern void tcp_init_xmit_timers(struct sock *);
463c84b9
ACM		 467static inline void tcp_clear_xmit_timers(struct sock *sk)
468{
469 inet_csk_clear_xmit_timers(sk);
470}
1da177e4 471
1da177e4 472extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
0c54b85f
IJ		 473extern unsigned int tcp_current_mss(struct sock *sk);
474
475/* Bound MSS / TSO packet size with the half of the window */
476static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
477{
478 if (tp->max_window && pktsize > (tp->max_window >> 1))
479 return max(tp->max_window >> 1, 68U - tp->tcp_header_len);
480 else
481 return pktsize;
482}
1da177e4 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
f1ecd5d9
DL		 499static inline void tcp_bound_rto(const struct sock *sk)
500{
501 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
502 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
503}
504
505static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
506{
507 return (tp->srtt >> 3) + tp->rttvar;
508}
509
40efc6fa 510static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
1da177e4
LT		 511{
512 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
513 ntohl(TCP_FLAG_ACK) |
514 snd_wnd);
515}
516
40efc6fa 517static inline void tcp_fast_path_on(struct tcp_sock *tp)
1da177e4
LT		 518{
519 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
520}
521
9e412ba7 522static inline void tcp_fast_path_check(struct sock *sk)
1da177e4 523{
9e412ba7
IJ		 524 struct tcp_sock *tp = tcp_sk(sk);
525
b03efcfb 526 if (skb_queue_empty(&tp->out_of_order_queue) &&
1da177e4
LT		 527 tp->rcv_wnd &&
528 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
529 !tp->urg_data)
530 tcp_fast_path_on(tp);
531}
532
0c266898
SS		 533/* Compute the actual rto_min value */
534static inline u32 tcp_rto_min(struct sock *sk)
535{
536 struct dst_entry *dst = __sk_dst_get(sk);
537 u32 rto_min = TCP_RTO_MIN;
538
539 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
540 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
541 return rto_min;
542}
543
1da177e4
LT		 544/* Compute the actual receive window we are currently advertising.
545 * Rcv_nxt can be after the window if our peer push more data
546 * than the offered window.
547 */
40efc6fa 548static inline u32 tcp_receive_window(const struct tcp_sock *tp)
1da177e4
LT		 549{
550 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
551
552 if (win < 0)
553 win = 0;
554 return (u32) win;
555}
556
557/* Choose a new window, without checks for shrinking, and without
558 * scaling applied to the result. The caller does these things
559 * if necessary. This is a "raw" window selection.
560 */
53d3176b 561extern u32 __tcp_select_window(struct sock *sk);
1da177e4
LT		 562
563/* TCP timestamps are only 32-bits, this causes a slight
564 * complication on 64-bit systems since we store a snapshot
31f34269
SH		 565 * of jiffies in the buffer control blocks below. We decided
566 * to use only the low 32-bits of jiffies and hide the ugly
1da177e4
LT		 567 * casts with the following macro.
568 */
569#define tcp_time_stamp ((__u32)(jiffies))
570
a3433f35
CG		 571#define tcp_flag_byte(th) (((u_int8_t *)th)[13])
572
573#define TCPHDR_FIN 0x01
574#define TCPHDR_SYN 0x02
575#define TCPHDR_RST 0x04
576#define TCPHDR_PSH 0x08
577#define TCPHDR_ACK 0x10
578#define TCPHDR_URG 0x20
579#define TCPHDR_ECE 0x40
580#define TCPHDR_CWR 0x80
581
caa20d9a 582/* This is what the send packet queuing engine uses to pass
f86586fa
ED		 583 * TCP per-packet control information to the transmission code.
584 * We also store the host-order sequence numbers in here too.
585 * This is 44 bytes if IPV6 is enabled.
586 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
1da177e4
LT		 587 */
588struct tcp_skb_cb {
589 union {
590 struct inet_skb_parm h4;
591#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
592 struct inet6_skb_parm h6;
593#endif
594 } header; /* For incoming frames */
595 __u32 seq; /* Starting sequence number */
596 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
597 __u32 when; /* used to compute rtt's */
598 __u8 flags; /* TCP header flags. */
1da177e4
LT		 599 __u8 sacked; /* State flags for SACK/FACK. */
600#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
601#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
602#define TCPCB_LOST 0x04 /* SKB is lost */
603#define TCPCB_TAGBITS 0x07 /* All tag bits */
604
605#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
606#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
607
1da177e4
LT		 608 __u32 ack_seq; /* Sequence number ACK'd */
609};
610
611#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
612
1da177e4
LT		 613/* Due to TSO, an SKB can be composed of multiple actual
614 * packets. To keep these tracked properly, we use this.
615 */
616static inline int tcp_skb_pcount(const struct sk_buff *skb)
617{
7967168c 618 return skb_shinfo(skb)->gso_segs;
1da177e4
LT		 619}
620
621/* This is valid iff tcp_skb_pcount() > 1. */
622static inline int tcp_skb_mss(const struct sk_buff *skb)
623{
7967168c 624 return skb_shinfo(skb)->gso_size;
1da177e4
LT		 625}
626
317a76f9
SH		 627/* Events passed to congestion control interface */
628enum tcp_ca_event {
629 CA_EVENT_TX_START, /* first transmit when no packets in flight */
630 CA_EVENT_CWND_RESTART, /* congestion window restart */
631 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
632 CA_EVENT_FRTO, /* fast recovery timeout */
633 CA_EVENT_LOSS, /* loss timeout */
634 CA_EVENT_FAST_ACK, /* in sequence ack */
635 CA_EVENT_SLOW_ACK, /* other ack */
636};
637
638/*
639 * Interface for adding new TCP congestion control handlers
640 */
641#define TCP_CA_NAME_MAX 16
3ff825b2
SH		 642#define TCP_CA_MAX 128
643#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
644
164891aa
SH		 645#define TCP_CONG_NON_RESTRICTED 0x1
646#define TCP_CONG_RTT_STAMP 0x2
647
317a76f9
SH		 648struct tcp_congestion_ops {
649 struct list_head list;
164891aa 650 unsigned long flags;
317a76f9
SH		 651
652 /* initialize private data (optional) */
6687e988 653 void (*init)(struct sock *sk);
317a76f9 654 /* cleanup private data (optional) */
6687e988 655 void (*release)(struct sock *sk);
317a76f9
SH		 656
657 /* return slow start threshold (required) */
6687e988 658 u32 (*ssthresh)(struct sock *sk);
317a76f9 659 /* lower bound for congestion window (optional) */
72dc5b92 660 u32 (*min_cwnd)(const struct sock *sk);
317a76f9 661 /* do new cwnd calculation (required) */
c3a05c60 662 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
317a76f9 663 /* call before changing ca_state (optional) */
6687e988 664 void (*set_state)(struct sock *sk, u8 new_state);
317a76f9 665 /* call when cwnd event occurs (optional) */
6687e988 666 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
317a76f9 667 /* new value of cwnd after loss (optional) */
6687e988 668 u32 (*undo_cwnd)(struct sock *sk);
317a76f9 669 /* hook for packet ack accounting (optional) */
30cfd0ba 670 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
73c1f4a0 671 /* get info for inet_diag (optional) */
6687e988 672 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
317a76f9
SH		 673
674 char name[TCP_CA_NAME_MAX];
675 struct module *owner;
676};
677
678extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
679extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
680
6687e988
ACM		 681extern void tcp_init_congestion_control(struct sock *sk);
682extern void tcp_cleanup_congestion_control(struct sock *sk);
317a76f9
SH		 683extern int tcp_set_default_congestion_control(const char *name);
684extern void tcp_get_default_congestion_control(char *name);
3ff825b2 685extern void tcp_get_available_congestion_control(char *buf, size_t len);
ce7bc3bf
SH		 686extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
687extern int tcp_set_allowed_congestion_control(char *allowed);
6687e988 688extern int tcp_set_congestion_control(struct sock *sk, const char *name);
40efc6fa 689extern void tcp_slow_start(struct tcp_sock *tp);
758ce5c8 690extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
317a76f9 691
5f8ef48d 692extern struct tcp_congestion_ops tcp_init_congestion_ops;
6687e988 693extern u32 tcp_reno_ssthresh(struct sock *sk);
c3a05c60 694extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
72dc5b92 695extern u32 tcp_reno_min_cwnd(const struct sock *sk);
a8acfbac 696extern struct tcp_congestion_ops tcp_reno;
317a76f9 697
6687e988 698static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
317a76f9 699{
6687e988
ACM		 700 struct inet_connection_sock *icsk = inet_csk(sk);
701
702 if (icsk->icsk_ca_ops->set_state)
703 icsk->icsk_ca_ops->set_state(sk, ca_state);
704 icsk->icsk_ca_state = ca_state;
317a76f9
SH		 705}
706
6687e988 707static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
317a76f9 708{
6687e988
ACM		 709 const struct inet_connection_sock *icsk = inet_csk(sk);
710
711 if (icsk->icsk_ca_ops->cwnd_event)
712 icsk->icsk_ca_ops->cwnd_event(sk, event);
317a76f9
SH		 713}
714
e60402d0
IJ		 715/* These functions determine how the current flow behaves in respect of SACK
716 * handling. SACK is negotiated with the peer, and therefore it can vary
717 * between different flows.
718 *
719 * tcp_is_sack - SACK enabled
720 * tcp_is_reno - No SACK
721 * tcp_is_fack - FACK enabled, implies SACK enabled
722 */
723static inline int tcp_is_sack(const struct tcp_sock *tp)
724{
725 return tp->rx_opt.sack_ok;
726}
727
728static inline int tcp_is_reno(const struct tcp_sock *tp)
729{
730 return !tcp_is_sack(tp);
731}
732
733static inline int tcp_is_fack(const struct tcp_sock *tp)
734{
735 return tp->rx_opt.sack_ok & 2;
736}
737
738static inline void tcp_enable_fack(struct tcp_sock *tp)
739{
740 tp->rx_opt.sack_ok |= 2;
741}
742
83ae4088
IJ		 743static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
744{
745 return tp->sacked_out + tp->lost_out;
746}
747
1da177e4
LT		 748/* This determines how many packets are "in the network" to the best
749 * of our knowledge. In many cases it is conservative, but where
750 * detailed information is available from the receiver (via SACK
751 * blocks etc.) we can make more aggressive calculations.
752 *
753 * Use this for decisions involving congestion control, use just
754 * tp->packets_out to determine if the send queue is empty or not.
755 *
756 * Read this equation as:
757 *
758 * "Packets sent once on transmission queue" MINUS
759 * "Packets left network, but not honestly ACKed yet" PLUS
760 * "Packets fast retransmitted"
761 */
40efc6fa 762static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1da177e4 763{
83ae4088 764 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1da177e4
LT		 765}
766
0b6a05c1
IJ		 767#define TCP_INFINITE_SSTHRESH 0x7fffffff
768
769static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
770{
771 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
772}
773
1da177e4
LT		 774/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
775 * The exception is rate halving phase, when cwnd is decreasing towards
776 * ssthresh.
777 */
6687e988 778static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1da177e4 779{
6687e988
ACM		 780 const struct tcp_sock *tp = tcp_sk(sk);
781 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
1da177e4
LT		 782 return tp->snd_ssthresh;
783 else
784 return max(tp->snd_ssthresh,
785 ((tp->snd_cwnd >> 1) +
786 (tp->snd_cwnd >> 2)));
787}
788
b9c4595b
IJ		 789/* Use define here intentionally to get WARN_ON location shown at the caller */
790#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1da177e4 791
3cfe3baa 792extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
1da177e4
LT		 793extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
794
795/* Slow start with delack produces 3 packets of burst, so that
dd9e0dda
JH		 796 * it is safe "de facto". This will be the default - same as
797 * the default reordering threshold - but if reordering increases,
798 * we must be able to allow cwnd to burst at least this much in order
799 * to not pull it back when holes are filled.
1da177e4
LT		 800 */
801static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
802{
dd9e0dda 803 return tp->reordering;
1da177e4
LT		 804}
805
90840def
IJ		 806/* Returns end sequence number of the receiver's advertised window */
807static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
808{
809 return tp->snd_una + tp->snd_wnd;
810}
cea14e0e 811extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
f4805ede 812
c1bd24b7 813static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
40efc6fa 814 const struct sk_buff *skb)
1da177e4
LT		 815{
816 if (skb->len < mss)
817 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
818}
819
9e412ba7 820static inline void tcp_check_probe_timer(struct sock *sk)
1da177e4 821{
9e412ba7 822 struct tcp_sock *tp = tcp_sk(sk);
463c84b9 823 const struct inet_connection_sock *icsk = inet_csk(sk);
9e412ba7 824
463c84b9 825 if (!tp->packets_out && !icsk->icsk_pending)
3f421baa
ACM		 826 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
827 icsk->icsk_rto, TCP_RTO_MAX);
1da177e4
LT		 828}
829
ee7537b6 830static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1da177e4
LT		 831{
832 tp->snd_wl1 = seq;
833}
834
ee7537b6 835static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1da177e4
LT		 836{
837 tp->snd_wl1 = seq;
838}
839
1da177e4
LT		 840/*
841 * Calculate(/check) TCP checksum
842 */
ba7808ea
FD		 843static inline __sum16 tcp_v4_check(int len, __be32 saddr,
844 __be32 daddr, __wsum base)
1da177e4
LT		 845{
846 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
847}
848
b51655b9 849static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1da177e4 850{
fb286bb2 851 return __skb_checksum_complete(skb);
1da177e4
LT		 852}
853
40efc6fa 854static inline int tcp_checksum_complete(struct sk_buff *skb)
1da177e4 855{
60476372 856 return !skb_csum_unnecessary(skb) &&
1da177e4
LT		 857 __tcp_checksum_complete(skb);
858}
859
860/* Prequeue for VJ style copy to user, combined with checksumming. */
861
40efc6fa 862static inline void tcp_prequeue_init(struct tcp_sock *tp)
1da177e4
LT		 863{
864 tp->ucopy.task = NULL;
865 tp->ucopy.len = 0;
866 tp->ucopy.memory = 0;
867 skb_queue_head_init(&tp->ucopy.prequeue);
97fc2f08
CL		 868#ifdef CONFIG_NET_DMA
869 tp->ucopy.dma_chan = NULL;
870 tp->ucopy.wakeup = 0;
871 tp->ucopy.pinned_list = NULL;
872 tp->ucopy.dma_cookie = 0;
873#endif
1da177e4
LT		 874}
875
876/* Packet is added to VJ-style prequeue for processing in process
877 * context, if a reader task is waiting. Apparently, this exciting
878 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
879 * failed somewhere. Latency? Burstiness? Well, at least now we will
880 * see, why it failed. 8)8) --ANK
881 *
882 * NOTE: is this not too big to inline?
883 */
40efc6fa 884static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1da177e4
LT		 885{
886 struct tcp_sock *tp = tcp_sk(sk);
887
f5f8d86b
ED		 888 if (sysctl_tcp_low_latency || !tp->ucopy.task)
889 return 0;
890
891 __skb_queue_tail(&tp->ucopy.prequeue, skb);
892 tp->ucopy.memory += skb->truesize;
893 if (tp->ucopy.memory > sk->sk_rcvbuf) {
894 struct sk_buff *skb1;
895
896 BUG_ON(sock_owned_by_user(sk));
897
898 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
899 sk_backlog_rcv(sk, skb1);
900 NET_INC_STATS_BH(sock_net(sk),
901 LINUX_MIB_TCPPREQUEUEDROPPED);
1da177e4 902 }
f5f8d86b
ED		 903
904 tp->ucopy.memory = 0;
905 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
aa395145 906 wake_up_interruptible_sync_poll(sk_sleep(sk),
7aedec2a 907 POLLIN | POLLRDNORM | POLLRDBAND);
f5f8d86b
ED		 908 if (!inet_csk_ack_scheduled(sk))
909 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
22f6dacd 910 (3 * tcp_rto_min(sk)) / 4,
f5f8d86b 911 TCP_RTO_MAX);
1da177e4 912 }
f5f8d86b 913 return 1;
1da177e4
LT		 914}
915
916
917#undef STATE_TRACE
918
919#ifdef STATE_TRACE
920static const char *statename[]={
921 "Unused","Established","Syn Sent","Syn Recv",
922 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
923 "Close Wait","Last ACK","Listen","Closing"
924};
925#endif
490d5046 926extern void tcp_set_state(struct sock *sk, int state);
1da177e4 927
4ac02bab 928extern void tcp_done(struct sock *sk);
1da177e4 929
40efc6fa 930static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1da177e4
LT		 931{
932 rx_opt->dsack = 0;
1da177e4
LT		 933 rx_opt->num_sacks = 0;
934}
935
1da177e4
LT		 936/* Determine a window scaling and initial window to offer. */
937extern void tcp_select_initial_window(int __space, __u32 mss,
938 __u32 *rcv_wnd, __u32 *window_clamp,
31d12926 939 int wscale_ok, __u8 *rcv_wscale,
940 __u32 init_rcv_wnd);
1da177e4
LT		 941
942static inline int tcp_win_from_space(int space)
943{
944 return sysctl_tcp_adv_win_scale<=0 ?
945 (space>>(-sysctl_tcp_adv_win_scale)) :
946 space - (space>>sysctl_tcp_adv_win_scale);
947}
948
949/* Note: caller must be prepared to deal with negative returns */
950static inline int tcp_space(const struct sock *sk)
951{
952 return tcp_win_from_space(sk->sk_rcvbuf -
953 atomic_read(&sk->sk_rmem_alloc));
954}
955
956static inline int tcp_full_space(const struct sock *sk)
957{
958 return tcp_win_from_space(sk->sk_rcvbuf);
959}
960
40efc6fa
SH		 961static inline void tcp_openreq_init(struct request_sock *req,
962 struct tcp_options_received *rx_opt,
963 struct sk_buff *skb)
1da177e4 964{
2e6599cb
ACM		 965 struct inet_request_sock *ireq = inet_rsk(req);
966
1da177e4 967 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
4dfc2817 968 req->cookie_ts = 0;
2e6599cb 969 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1da177e4
LT		 970 req->mss = rx_opt->mss_clamp;
971 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
2e6599cb
ACM		 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;
976 ireq->acked = 0;
977 ireq->ecn_ok = 0;
aa8223c7 978 ireq->rmt_port = tcp_hdr(skb)->source;
a3116ac5 979 ireq->loc_port = tcp_hdr(skb)->dest;
1da177e4
LT		 980}
981
5c52ba17 982extern void tcp_enter_memory_pressure(struct sock *sk);
1da177e4 983
1da177e4
LT		 984static inline int keepalive_intvl_when(const struct tcp_sock *tp)
985{
986 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
987}
988
989static inline int keepalive_time_when(const struct tcp_sock *tp)
990{
991 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
992}
993
df19a626
ED		 994static inline int keepalive_probes(const struct tcp_sock *tp)
995{
996 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
997}
998
6c37e5de
FL		 999static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1000{
1001 const struct inet_connection_sock *icsk = &tp->inet_conn;
1002
1003 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1004 tcp_time_stamp - tp->rcv_tstamp);
1005}
1006
463c84b9 1007static inline int tcp_fin_time(const struct sock *sk)
1da177e4 1008{
463c84b9
ACM		 1009 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1010 const int rto = inet_csk(sk)->icsk_rto;
1da177e4 1011
463c84b9
ACM		 1012 if (fin_timeout < (rto << 2) - (rto >> 1))
1013 fin_timeout = (rto << 2) - (rto >> 1);
1da177e4
LT		 1014
1015 return fin_timeout;
1016}
1017
c887e6d2
IJ		 1018static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1019 int paws_win)
1da177e4 1020{
c887e6d2
IJ		 1021 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1022 return 1;
1023 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1024 return 1;
1025
1026 return 0;
1027}
1028
1029static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1030 int rst)
1031{
1032 if (tcp_paws_check(rx_opt, 0))
1da177e4
LT		 1033 return 0;
1034
1035 /* RST segments are not recommended to carry timestamp,
1036 and, if they do, it is recommended to ignore PAWS because
1037 "their cleanup function should take precedence over timestamps."
1038 Certainly, it is mistake. It is necessary to understand the reasons
1039 of this constraint to relax it: if peer reboots, clock may go
1040 out-of-sync and half-open connections will not be reset.
1041 Actually, the problem would be not existing if all
1042 the implementations followed draft about maintaining clock
1043 via reboots. Linux-2.2 DOES NOT!
1044
1045 However, we can relax time bounds for RST segments to MSL.
1046 */
9d729f72 1047 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1da177e4
LT		 1048 return 0;
1049 return 1;
1050}
1051
1da177e4
LT		 1052#define TCP_CHECK_TIMER(sk) do { } while (0)
1053
a9c19329 1054static inline void tcp_mib_init(struct net *net)
1da177e4
LT		 1055{
1056 /* See RFC 2012 */
cf1100a7
PE		 1057 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1058 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1059 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1060 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1da177e4
LT		 1061}
1062
5af4ec23 1063/* from STCP */
ef9da47c 1064static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
0800f170 1065{
6a438bbe
SH		 1066 tp->lost_skb_hint = NULL;
1067 tp->scoreboard_skb_hint = NULL;
ef9da47c
IJ		 1068}
1069
1070static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1071{
1072 tcp_clear_retrans_hints_partial(tp);
6a438bbe 1073 tp->retransmit_skb_hint = NULL;
b7689205
IJ		 1074}
1075
cfb6eeb4
YH		 1076/* MD5 Signature */
1077struct crypto_hash;
1078
1079/* - key database */
1080struct tcp_md5sig_key {
1081 u8 *key;
1082 u8 keylen;
1083};
1084
1085struct tcp4_md5sig_key {
f8ab18d2 1086 struct tcp_md5sig_key base;
cfb6eeb4
YH		 1087 __be32 addr;
1088};
1089
1090struct tcp6_md5sig_key {
f8ab18d2 1091 struct tcp_md5sig_key base;
cfb6eeb4
YH		 1092#if 0
1093 u32 scope_id; /* XXX */
1094#endif
1095 struct in6_addr addr;
1096};
1097
1098/* - sock block */
1099struct tcp_md5sig_info {
1100 struct tcp4_md5sig_key *keys4;
1101#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1102 struct tcp6_md5sig_key *keys6;
1103 u32 entries6;
1104 u32 alloced6;
1105#endif
1106 u32 entries4;
1107 u32 alloced4;
1108};
1109
1110/* - pseudo header */
1111struct tcp4_pseudohdr {
1112 __be32 saddr;
1113 __be32 daddr;
1114 __u8 pad;
1115 __u8 protocol;
1116 __be16 len;
1117};
1118
1119struct tcp6_pseudohdr {
1120 struct in6_addr saddr;
1121 struct in6_addr daddr;
1122 __be32 len;
1123 __be32 protocol; /* including padding */
1124};
1125
1126union tcp_md5sum_block {
1127 struct tcp4_pseudohdr ip4;
1128#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1129 struct tcp6_pseudohdr ip6;
1130#endif
1131};
1132
1133/* - pool: digest algorithm, hash description and scratch buffer */
1134struct tcp_md5sig_pool {
1135 struct hash_desc md5_desc;
1136 union tcp_md5sum_block md5_blk;
1137};
1138
1139#define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1140
1141/* - functions */
53d3176b
CG		 1142extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1143 struct sock *sk, struct request_sock *req,
1144 struct sk_buff *skb);
1145extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk,
1146 struct sock *addr_sk);
1147extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey,
1148 u8 newkeylen);
1149extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr);
cfb6eeb4 1150
9501f972
YH		 1151#ifdef CONFIG_TCP_MD5SIG
1152#define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1153 &(struct tcp_md5sig_key) { \
1154 .key = (twsk)->tw_md5_key, \
1155 .keylen = (twsk)->tw_md5_keylen, \
1156 } : NULL)
1157#else
1158#define tcp_twsk_md5_key(twsk) NULL
1159#endif
1160
7d720c3e 1161extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *);
53d3176b 1162extern void tcp_free_md5sig_pool(void);
cfb6eeb4 1163
35790c04 1164extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
53d3176b 1165extern void tcp_put_md5sig_pool(void);
35790c04 1166
49a72dfb
AL		 1167extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1168extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1169 unsigned header_len);
1170extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1171 struct tcp_md5sig_key *key);
cfb6eeb4 1172
fe067e8a
DM		 1173/* write queue abstraction */
1174static inline void tcp_write_queue_purge(struct sock *sk)
1175{
1176 struct sk_buff *skb;
1177
1178 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
3ab224be
HA		 1179 sk_wmem_free_skb(sk, skb);
1180 sk_mem_reclaim(sk);
8818a9d8 1181 tcp_clear_all_retrans_hints(tcp_sk(sk));
fe067e8a
DM		 1182}
1183
1184static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1185{
cd07a8ea 1186 return skb_peek(&sk->sk_write_queue);
fe067e8a
DM		 1187}
1188
1189static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1190{
cd07a8ea 1191 return skb_peek_tail(&sk->sk_write_queue);
fe067e8a
DM		 1192}
1193
1194static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1195{
cd07a8ea 1196 return skb_queue_next(&sk->sk_write_queue, skb);
fe067e8a
DM		 1197}
1198
832d11c5
IJ		 1199static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb)
1200{
1201 return skb_queue_prev(&sk->sk_write_queue, skb);
1202}
1203
fe067e8a 1204#define tcp_for_write_queue(skb, sk) \
cd07a8ea 1205 skb_queue_walk(&(sk)->sk_write_queue, skb)
fe067e8a
DM		 1206
1207#define tcp_for_write_queue_from(skb, sk) \
cd07a8ea 1208 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
fe067e8a 1209
234b6860 1210#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
cd07a8ea 1211 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
234b6860 1212
fe067e8a
DM		 1213static inline struct sk_buff *tcp_send_head(struct sock *sk)
1214{
1215 return sk->sk_send_head;
1216}
1217
cd07a8ea
DM		 1218static inline bool tcp_skb_is_last(const struct sock *sk,
1219 const struct sk_buff *skb)
1220{
1221 return skb_queue_is_last(&sk->sk_write_queue, skb);
1222}
1223
fe067e8a
DM		 1224static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1225{
cd07a8ea 1226 if (tcp_skb_is_last(sk, skb))
fe067e8a 1227 sk->sk_send_head = NULL;
cd07a8ea
DM		 1228 else
1229 sk->sk_send_head = tcp_write_queue_next(sk, skb);
fe067e8a
DM		 1230}
1231
1232static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1233{
1234 if (sk->sk_send_head == skb_unlinked)
1235 sk->sk_send_head = NULL;
1236}
1237
1238static inline void tcp_init_send_head(struct sock *sk)
1239{
1240 sk->sk_send_head = NULL;
1241}
1242
1243static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1244{
1245 __skb_queue_tail(&sk->sk_write_queue, skb);
1246}
1247
1248static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1249{
1250 __tcp_add_write_queue_tail(sk, skb);
1251
1252 /* Queue it, remembering where we must start sending. */
6859d494 1253 if (sk->sk_send_head == NULL) {
fe067e8a 1254 sk->sk_send_head = skb;
6859d494
IJ		 1255
1256 if (tcp_sk(sk)->highest_sack == NULL)
1257 tcp_sk(sk)->highest_sack = skb;
1258 }
fe067e8a
DM		 1259}
1260
1261static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1262{
1263 __skb_queue_head(&sk->sk_write_queue, skb);
1264}
1265
1266/* Insert buff after skb on the write queue of sk. */
1267static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1268 struct sk_buff *buff,
1269 struct sock *sk)
1270{
7de6c033 1271 __skb_queue_after(&sk->sk_write_queue, skb, buff);
fe067e8a
DM		 1272}
1273
43f59c89 1274/* Insert new before skb on the write queue of sk. */
fe067e8a
DM		 1275static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1276 struct sk_buff *skb,
1277 struct sock *sk)
1278{
43f59c89 1279 __skb_queue_before(&sk->sk_write_queue, skb, new);