projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[TCP]: lower stack usage in cookie_hash() function
[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>
3f421baa
ACM		 32
33#include <net/inet_connection_sock.h>
295ff7ed 34#include <net/inet_timewait_sock.h>
77d8bf9c 35#include <net/inet_hashtables.h>
1da177e4 36#include <net/checksum.h>
2e6599cb 37#include <net/request_sock.h>
1da177e4
LT		 38#include <net/sock.h>
39#include <net/snmp.h>
40#include <net/ip.h>
c752f073 41#include <net/tcp_states.h>
bdf1ee5d 42#include <net/inet_ecn.h>
c752f073 43
1da177e4
LT		 44#include <linux/seq_file.h>
45
6e04e021 46extern struct inet_hashinfo tcp_hashinfo;
1da177e4 47
1da177e4 48extern atomic_t tcp_orphan_count;
1da177e4 49extern void tcp_time_wait(struct sock *sk, int state, int timeo);
1da177e4 50
1da177e4
LT		 51#define MAX_TCP_HEADER (128 + MAX_HEADER)
52
53/*
54 * Never offer a window over 32767 without using window scaling. Some
55 * poor stacks do signed 16bit maths!
56 */
57#define MAX_TCP_WINDOW 32767U
58
59/* Minimal accepted MSS. It is (60+60+8) - (20+20). */
60#define TCP_MIN_MSS 88U
61
62/* Minimal RCV_MSS. */
63#define TCP_MIN_RCVMSS 536U
64
5d424d5a
JH		 65/* The least MTU to use for probing */
66#define TCP_BASE_MSS 512
67
1da177e4
LT		 68/* After receiving this amount of duplicate ACKs fast retransmit starts. */
69#define TCP_FASTRETRANS_THRESH 3
70
71/* Maximal reordering. */
72#define TCP_MAX_REORDERING 127
73
74/* Maximal number of ACKs sent quickly to accelerate slow-start. */
75#define TCP_MAX_QUICKACKS 16U
76
77/* urg_data states */
78#define TCP_URG_VALID 0x0100
79#define TCP_URG_NOTYET 0x0200
80#define TCP_URG_READ 0x0400
81
82#define TCP_RETR1 3 /*
83 * This is how many retries it does before it
84 * tries to figure out if the gateway is
85 * down. Minimal RFC value is 3; it corresponds
86 * to ~3sec-8min depending on RTO.
87 */
88
89#define TCP_RETR2 15 /*
90 * This should take at least
91 * 90 minutes to time out.
92 * RFC1122 says that the limit is 100 sec.
93 * 15 is ~13-30min depending on RTO.
94 */
95
96#define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
caa20d9a 97 * connection: ~180sec is RFC minimum */
1da177e4
LT		 98
99#define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
caa20d9a 100 * connection: ~180sec is RFC minimum */
1da177e4
LT		 101
102
103#define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
104 * socket. 7 is ~50sec-16min.
105 */
106
107
108#define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
109 * state, about 60 seconds */
110#define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
111 /* BSD style FIN_WAIT2 deadlock breaker.
112 * It used to be 3min, new value is 60sec,
113 * to combine FIN-WAIT-2 timeout with
114 * TIME-WAIT timer.
115 */
116
117#define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
118#if HZ >= 100
119#define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
120#define TCP_ATO_MIN ((unsigned)(HZ/25))
121#else
122#define TCP_DELACK_MIN 4U
123#define TCP_ATO_MIN 4U
124#endif
125#define TCP_RTO_MAX ((unsigned)(120*HZ))
126#define TCP_RTO_MIN ((unsigned)(HZ/5))
127#define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
128
129#define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
130 * for local resources.
131 */
132
133#define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
134#define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
135#define TCP_KEEPALIVE_INTVL (75*HZ)
136
137#define MAX_TCP_KEEPIDLE 32767
138#define MAX_TCP_KEEPINTVL 32767
139#define MAX_TCP_KEEPCNT 127
140#define MAX_TCP_SYNCNT 127
141
142#define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
1da177e4
LT		 143
144#define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
145#define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
146 * after this time. It should be equal
147 * (or greater than) TCP_TIMEWAIT_LEN
148 * to provide reliability equal to one
149 * provided by timewait state.
150 */
151#define TCP_PAWS_WINDOW 1 /* Replay window for per-host
152 * timestamps. It must be less than
153 * minimal timewait lifetime.
154 */
1da177e4
LT		 155/*
156 * TCP option
157 */
158
159#define TCPOPT_NOP 1 /* Padding */
160#define TCPOPT_EOL 0 /* End of options */
161#define TCPOPT_MSS 2 /* Segment size negotiating */
162#define TCPOPT_WINDOW 3 /* Window scaling */
163#define TCPOPT_SACK_PERM 4 /* SACK Permitted */
164#define TCPOPT_SACK 5 /* SACK Block */
165#define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
cfb6eeb4 166#define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
1da177e4
LT		 167
168/*
169 * TCP option lengths
170 */
171
172#define TCPOLEN_MSS 4
173#define TCPOLEN_WINDOW 3
174#define TCPOLEN_SACK_PERM 2
175#define TCPOLEN_TIMESTAMP 10
cfb6eeb4 176#define TCPOLEN_MD5SIG 18
1da177e4
LT		 177
178/* But this is what stacks really send out. */
179#define TCPOLEN_TSTAMP_ALIGNED 12
180#define TCPOLEN_WSCALE_ALIGNED 4
181#define TCPOLEN_SACKPERM_ALIGNED 4
182#define TCPOLEN_SACK_BASE 2
183#define TCPOLEN_SACK_BASE_ALIGNED 4
184#define TCPOLEN_SACK_PERBLOCK 8
cfb6eeb4 185#define TCPOLEN_MD5SIG_ALIGNED 20
1da177e4 186
1da177e4
LT		 187/* Flags in tp->nonagle */
188#define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
189#define TCP_NAGLE_CORK 2 /* Socket is corked */
caa20d9a 190#define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
1da177e4 191
295ff7ed
ACM		 192extern struct inet_timewait_death_row tcp_death_row;
193
1da177e4 194/* sysctl variables for tcp */
1da177e4
LT		 195extern int sysctl_tcp_timestamps;
196extern int sysctl_tcp_window_scaling;
197extern int sysctl_tcp_sack;
198extern int sysctl_tcp_fin_timeout;
1da177e4
LT		 199extern int sysctl_tcp_keepalive_time;
200extern int sysctl_tcp_keepalive_probes;
201extern int sysctl_tcp_keepalive_intvl;
202extern int sysctl_tcp_syn_retries;
203extern int sysctl_tcp_synack_retries;
204extern int sysctl_tcp_retries1;
205extern int sysctl_tcp_retries2;
206extern int sysctl_tcp_orphan_retries;
207extern int sysctl_tcp_syncookies;
208extern int sysctl_tcp_retrans_collapse;
209extern int sysctl_tcp_stdurg;
210extern int sysctl_tcp_rfc1337;
211extern int sysctl_tcp_abort_on_overflow;
212extern int sysctl_tcp_max_orphans;
1da177e4
LT		 213extern int sysctl_tcp_fack;
214extern int sysctl_tcp_reordering;
215extern int sysctl_tcp_ecn;
216extern int sysctl_tcp_dsack;
217extern int sysctl_tcp_mem[3];
218extern int sysctl_tcp_wmem[3];
219extern int sysctl_tcp_rmem[3];
220extern int sysctl_tcp_app_win;
221extern int sysctl_tcp_adv_win_scale;
222extern int sysctl_tcp_tw_reuse;
223extern int sysctl_tcp_frto;
3cfe3baa 224extern int sysctl_tcp_frto_response;
1da177e4 225extern int sysctl_tcp_low_latency;
95937825 226extern int sysctl_tcp_dma_copybreak;
1da177e4 227extern int sysctl_tcp_nometrics_save;
1da177e4
LT		 228extern int sysctl_tcp_moderate_rcvbuf;
229extern int sysctl_tcp_tso_win_divisor;
9772efb9 230extern int sysctl_tcp_abc;
5d424d5a
JH		 231extern int sysctl_tcp_mtu_probing;
232extern int sysctl_tcp_base_mss;
15d99e02 233extern int sysctl_tcp_workaround_signed_windows;
35089bb2 234extern int sysctl_tcp_slow_start_after_idle;
886236c1 235extern int sysctl_tcp_max_ssthresh;
1da177e4
LT		 236
237extern atomic_t tcp_memory_allocated;
238extern atomic_t tcp_sockets_allocated;
239extern int tcp_memory_pressure;
240
1da177e4
LT		 241/*
242 * The next routines deal with comparing 32 bit unsigned ints
243 * and worry about wraparound (automatic with unsigned arithmetic).
244 */
245
246static inline int before(__u32 seq1, __u32 seq2)
247{
0d630cc0 248 return (__s32)(seq1-seq2) < 0;
1da177e4 249}
9a036b9c 250#define after(seq2, seq1) before(seq1, seq2)
1da177e4
LT		 251
252/* is s2<=s1<=s3 ? */
253static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
254{
255 return seq3 - seq2 >= seq1 - seq2;
256}
257
e4fd5da3
PE		 258static inline int tcp_too_many_orphans(struct sock *sk, int num)
259{
260 return (num > sysctl_tcp_max_orphans) ||
261 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
262 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
263}
1da177e4
LT		 264
265extern struct proto tcp_prot;
266
267DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics);
268#define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field)
269#define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field)
270#define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field)
271#define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field)
272#define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val)
273#define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val)
274
1da177e4
LT		 275extern void tcp_v4_err(struct sk_buff *skb, u32);
276
277extern void tcp_shutdown (struct sock *sk, int how);
278
279extern int tcp_v4_rcv(struct sk_buff *skb);
280
281extern int tcp_v4_remember_stamp(struct sock *sk);
282
8feaf0c0 283extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
1da177e4 284
3516ffb0 285extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
1da177e4
LT		 286 struct msghdr *msg, size_t size);
287extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
288
289extern int tcp_ioctl(struct sock *sk,
290 int cmd,
291 unsigned long arg);
292
293extern int tcp_rcv_state_process(struct sock *sk,
294 struct sk_buff *skb,
295 struct tcphdr *th,
296 unsigned len);
297
298extern int tcp_rcv_established(struct sock *sk,
299 struct sk_buff *skb,
300 struct tcphdr *th,
301 unsigned len);
302
303extern void tcp_rcv_space_adjust(struct sock *sk);
304
0e4b4992
CL		 305extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
306
6d6ee43e
ACM		 307extern int tcp_twsk_unique(struct sock *sk,
308 struct sock *sktw, void *twp);
309
cfb6eeb4
YH		 310extern void tcp_twsk_destructor(struct sock *sk);
311
9c55e01c
JA		 312extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
313 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
314
463c84b9
ACM		 315static inline void tcp_dec_quickack_mode(struct sock *sk,
316 const unsigned int pkts)
1da177e4 317{
463c84b9 318 struct inet_connection_sock *icsk = inet_csk(sk);
fc6415bc 319
463c84b9
ACM		 320 if (icsk->icsk_ack.quick) {
321 if (pkts >= icsk->icsk_ack.quick) {
322 icsk->icsk_ack.quick = 0;
fc6415bc 323 /* Leaving quickack mode we deflate ATO. */
463c84b9 324 icsk->icsk_ack.ato = TCP_ATO_MIN;
fc6415bc 325 } else
463c84b9 326 icsk->icsk_ack.quick -= pkts;
1da177e4
LT		 327 }
328}
329
463c84b9 330extern void tcp_enter_quickack_mode(struct sock *sk);
1da177e4 331
1da177e4
LT		 332static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
333{
334 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
335}
336
bdf1ee5d
IJ		 337#define TCP_ECN_OK 1
338#define TCP_ECN_QUEUE_CWR 2
339#define TCP_ECN_DEMAND_CWR 4
340
341static __inline__ void
342TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
343{
344 if (sysctl_tcp_ecn && th->ece && th->cwr)
345 inet_rsk(req)->ecn_ok = 1;
346}
347
1da177e4
LT		 348enum tcp_tw_status
349{
350 TCP_TW_SUCCESS = 0,
351 TCP_TW_RST = 1,
352 TCP_TW_ACK = 2,
353 TCP_TW_SYN = 3
354};
355
356
8feaf0c0 357extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
1da177e4 358 struct sk_buff *skb,
8feaf0c0 359 const struct tcphdr *th);
1da177e4
LT		 360
361extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
60236fdd
ACM		 362 struct request_sock *req,
363 struct request_sock **prev);
1da177e4
LT		 364extern int tcp_child_process(struct sock *parent,
365 struct sock *child,
366 struct sk_buff *skb);
46d0de4e 367extern int tcp_use_frto(struct sock *sk);
1da177e4
LT		 368extern void tcp_enter_frto(struct sock *sk);
369extern void tcp_enter_loss(struct sock *sk, int how);
370extern void tcp_clear_retrans(struct tcp_sock *tp);
371extern void tcp_update_metrics(struct sock *sk);
372
373extern void tcp_close(struct sock *sk,
374 long timeout);
1da177e4
LT		 375extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
376
377extern int tcp_getsockopt(struct sock *sk, int level,
378 int optname,
379 char __user *optval,
380 int __user *optlen);
381extern int tcp_setsockopt(struct sock *sk, int level,
382 int optname, char __user *optval,
383 int optlen);
3fdadf7d
DM		 384extern int compat_tcp_getsockopt(struct sock *sk,
385 int level, int optname,
386 char __user *optval, int __user *optlen);
387extern int compat_tcp_setsockopt(struct sock *sk,
388 int level, int optname,
389 char __user *optval, int optlen);
1da177e4
LT		 390extern void tcp_set_keepalive(struct sock *sk, int val);
391extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
392 struct msghdr *msg,
393 size_t len, int nonblock,
394 int flags, int *addr_len);
395
1da177e4
LT		 396extern void tcp_parse_options(struct sk_buff *skb,
397 struct tcp_options_received *opt_rx,
398 int estab);
399
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
434extern void tcp_unhash(struct sock *sk);
435
1da177e4
LT		 436/* From syncookies.c */
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
442/* tcp_output.c */
443
9e412ba7
IJ		 444extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
445 int nonagle);
446extern int tcp_may_send_now(struct sock *sk);
1da177e4
LT		 447extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
448extern void tcp_xmit_retransmit_queue(struct sock *);
449extern void tcp_simple_retransmit(struct sock *);
450extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
6475be16 451extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
1da177e4
LT		 452
453extern void tcp_send_probe0(struct sock *);
454extern void tcp_send_partial(struct sock *);
455extern int tcp_write_wakeup(struct sock *);
456extern void tcp_send_fin(struct sock *sk);
dd0fc66f 457extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
1da177e4 458extern int tcp_send_synack(struct sock *);
c1b4a7e6 459extern void tcp_push_one(struct sock *, unsigned int mss_now);
1da177e4
LT		 460extern void tcp_send_ack(struct sock *sk);
461extern void tcp_send_delayed_ack(struct sock *sk);
462
a762a980
DM		 463/* tcp_input.c */
464extern void tcp_cwnd_application_limited(struct sock *sk);
465
1da177e4
LT		 466/* tcp_timer.c */
467extern void tcp_init_xmit_timers(struct sock *);
463c84b9
ACM		 468static inline void tcp_clear_xmit_timers(struct sock *sk)
469{
470 inet_csk_clear_xmit_timers(sk);
471}
1da177e4 472
1da177e4
LT		 473extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
474extern unsigned int tcp_current_mss(struct sock *sk, int large);
475
17b085ea 476/* tcp.c */
1da177e4
LT		 477extern void tcp_get_info(struct sock *, struct tcp_info *);
478
479/* Read 'sendfile()'-style from a TCP socket */
480typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
481 unsigned int, size_t);
482extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
483 sk_read_actor_t recv_actor);
484
40efc6fa 485extern void tcp_initialize_rcv_mss(struct sock *sk);
1da177e4 486
5d424d5a
JH		 487extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
488extern int tcp_mss_to_mtu(struct sock *sk, int mss);
489extern void tcp_mtup_init(struct sock *sk);
490
40efc6fa 491static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
1da177e4
LT		 492{
493 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
494 ntohl(TCP_FLAG_ACK) |
495 snd_wnd);
496}
497
40efc6fa 498static inline void tcp_fast_path_on(struct tcp_sock *tp)
1da177e4
LT		 499{
500 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
501}
502
9e412ba7 503static inline void tcp_fast_path_check(struct sock *sk)
1da177e4 504{
9e412ba7
IJ		 505 struct tcp_sock *tp = tcp_sk(sk);
506
b03efcfb 507 if (skb_queue_empty(&tp->out_of_order_queue) &&
1da177e4
LT		 508 tp->rcv_wnd &&
509 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
510 !tp->urg_data)
511 tcp_fast_path_on(tp);
512}
513
514/* Compute the actual receive window we are currently advertising.
515 * Rcv_nxt can be after the window if our peer push more data
516 * than the offered window.
517 */
40efc6fa 518static inline u32 tcp_receive_window(const struct tcp_sock *tp)
1da177e4
LT		 519{
520 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
521
522 if (win < 0)
523 win = 0;
524 return (u32) win;
525}
526
527/* Choose a new window, without checks for shrinking, and without
528 * scaling applied to the result. The caller does these things
529 * if necessary. This is a "raw" window selection.
530 */
531extern u32 __tcp_select_window(struct sock *sk);
532
533/* TCP timestamps are only 32-bits, this causes a slight
534 * complication on 64-bit systems since we store a snapshot
31f34269
SH		 535 * of jiffies in the buffer control blocks below. We decided
536 * to use only the low 32-bits of jiffies and hide the ugly
1da177e4
LT		 537 * casts with the following macro.
538 */
539#define tcp_time_stamp ((__u32)(jiffies))
540
caa20d9a 541/* This is what the send packet queuing engine uses to pass
1da177e4
LT		 542 * TCP per-packet control information to the transmission
543 * code. We also store the host-order sequence numbers in
544 * here too. This is 36 bytes on 32-bit architectures,
545 * 40 bytes on 64-bit machines, if this grows please adjust
546 * skbuff.h:skbuff->cb[xxx] size appropriately.
547 */
548struct tcp_skb_cb {
549 union {
550 struct inet_skb_parm h4;
551#if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
552 struct inet6_skb_parm h6;
553#endif
554 } header; /* For incoming frames */
555 __u32 seq; /* Starting sequence number */
556 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
557 __u32 when; /* used to compute rtt's */
558 __u8 flags; /* TCP header flags. */
559
560 /* NOTE: These must match up to the flags byte in a
561 * real TCP header.
562 */
563#define TCPCB_FLAG_FIN 0x01
564#define TCPCB_FLAG_SYN 0x02
565#define TCPCB_FLAG_RST 0x04
566#define TCPCB_FLAG_PSH 0x08
567#define TCPCB_FLAG_ACK 0x10
568#define TCPCB_FLAG_URG 0x20
569#define TCPCB_FLAG_ECE 0x40
570#define TCPCB_FLAG_CWR 0x80
571
572 __u8 sacked; /* State flags for SACK/FACK. */
573#define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
574#define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
575#define TCPCB_LOST 0x04 /* SKB is lost */
576#define TCPCB_TAGBITS 0x07 /* All tag bits */
577
578#define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
579#define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
580
1da177e4
LT		 581 __u16 urg_ptr; /* Valid w/URG flags is set. */
582 __u32 ack_seq; /* Sequence number ACK'd */
583};
584
585#define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
586
1da177e4
LT		 587/* Due to TSO, an SKB can be composed of multiple actual
588 * packets. To keep these tracked properly, we use this.
589 */
590static inline int tcp_skb_pcount(const struct sk_buff *skb)
591{
7967168c 592 return skb_shinfo(skb)->gso_segs;
1da177e4
LT		 593}
594
595/* This is valid iff tcp_skb_pcount() > 1. */
596static inline int tcp_skb_mss(const struct sk_buff *skb)
597{
7967168c 598 return skb_shinfo(skb)->gso_size;
1da177e4
LT		 599}
600
af610b4c 601static inline void tcp_dec_pcount_approx_int(__u32 *count, const int decr)
1da177e4
LT		 602{
603 if (*count) {
af610b4c 604 *count -= decr;
1da177e4
LT		 605 if ((int)*count < 0)
606 *count = 0;
607 }
608}
609
af610b4c
IJ		 610static inline void tcp_dec_pcount_approx(__u32 *count,
611 const struct sk_buff *skb)
612{
613 tcp_dec_pcount_approx_int(count, tcp_skb_pcount(skb));
614}
615
317a76f9
SH		 616/* Events passed to congestion control interface */
617enum tcp_ca_event {
618 CA_EVENT_TX_START, /* first transmit when no packets in flight */
619 CA_EVENT_CWND_RESTART, /* congestion window restart */
620 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
621 CA_EVENT_FRTO, /* fast recovery timeout */
622 CA_EVENT_LOSS, /* loss timeout */
623 CA_EVENT_FAST_ACK, /* in sequence ack */
624 CA_EVENT_SLOW_ACK, /* other ack */
625};
626
627/*
628 * Interface for adding new TCP congestion control handlers
629 */
630#define TCP_CA_NAME_MAX 16
3ff825b2
SH		 631#define TCP_CA_MAX 128
632#define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
633
164891aa
SH		 634#define TCP_CONG_NON_RESTRICTED 0x1
635#define TCP_CONG_RTT_STAMP 0x2
636
317a76f9
SH		 637struct tcp_congestion_ops {
638 struct list_head list;
164891aa 639 unsigned long flags;
317a76f9
SH		 640
641 /* initialize private data (optional) */
6687e988 642 void (*init)(struct sock *sk);
317a76f9 643 /* cleanup private data (optional) */
6687e988 644 void (*release)(struct sock *sk);
317a76f9
SH		 645
646 /* return slow start threshold (required) */
6687e988 647 u32 (*ssthresh)(struct sock *sk);
317a76f9 648 /* lower bound for congestion window (optional) */
72dc5b92 649 u32 (*min_cwnd)(const struct sock *sk);
317a76f9 650 /* do new cwnd calculation (required) */
c3a05c60 651 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
317a76f9 652 /* call before changing ca_state (optional) */
6687e988 653 void (*set_state)(struct sock *sk, u8 new_state);
317a76f9 654 /* call when cwnd event occurs (optional) */
6687e988 655 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
317a76f9 656 /* new value of cwnd after loss (optional) */
6687e988 657 u32 (*undo_cwnd)(struct sock *sk);
317a76f9 658 /* hook for packet ack accounting (optional) */
30cfd0ba 659 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
73c1f4a0 660 /* get info for inet_diag (optional) */
6687e988 661 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
317a76f9
SH		 662
663 char name[TCP_CA_NAME_MAX];
664 struct module *owner;
665};
666
667extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
668extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
669
6687e988
ACM		 670extern void tcp_init_congestion_control(struct sock *sk);
671extern void tcp_cleanup_congestion_control(struct sock *sk);
317a76f9
SH		 672extern int tcp_set_default_congestion_control(const char *name);
673extern void tcp_get_default_congestion_control(char *name);
3ff825b2 674extern void tcp_get_available_congestion_control(char *buf, size_t len);
ce7bc3bf
SH		 675extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
676extern int tcp_set_allowed_congestion_control(char *allowed);
6687e988 677extern int tcp_set_congestion_control(struct sock *sk, const char *name);
40efc6fa 678extern void tcp_slow_start(struct tcp_sock *tp);
317a76f9 679
5f8ef48d 680extern struct tcp_congestion_ops tcp_init_congestion_ops;
6687e988 681extern u32 tcp_reno_ssthresh(struct sock *sk);
c3a05c60 682extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
72dc5b92 683extern u32 tcp_reno_min_cwnd(const struct sock *sk);
a8acfbac 684extern struct tcp_congestion_ops tcp_reno;
317a76f9 685
6687e988 686static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
317a76f9 687{
6687e988
ACM		 688 struct inet_connection_sock *icsk = inet_csk(sk);
689
690 if (icsk->icsk_ca_ops->set_state)
691 icsk->icsk_ca_ops->set_state(sk, ca_state);
692 icsk->icsk_ca_state = ca_state;
317a76f9
SH		 693}
694
6687e988 695static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
317a76f9 696{
6687e988
ACM		 697 const struct inet_connection_sock *icsk = inet_csk(sk);
698
699 if (icsk->icsk_ca_ops->cwnd_event)
700 icsk->icsk_ca_ops->cwnd_event(sk, event);
317a76f9
SH		 701}
702
e60402d0
IJ		 703/* These functions determine how the current flow behaves in respect of SACK
704 * handling. SACK is negotiated with the peer, and therefore it can vary
705 * between different flows.
706 *
707 * tcp_is_sack - SACK enabled
708 * tcp_is_reno - No SACK
709 * tcp_is_fack - FACK enabled, implies SACK enabled
710 */
711static inline int tcp_is_sack(const struct tcp_sock *tp)
712{
713 return tp->rx_opt.sack_ok;
714}
715
716static inline int tcp_is_reno(const struct tcp_sock *tp)
717{
718 return !tcp_is_sack(tp);
719}
720
721static inline int tcp_is_fack(const struct tcp_sock *tp)
722{
723 return tp->rx_opt.sack_ok & 2;
724}
725
726static inline void tcp_enable_fack(struct tcp_sock *tp)
727{
728 tp->rx_opt.sack_ok |= 2;
729}
730
83ae4088
IJ		 731static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
732{
733 return tp->sacked_out + tp->lost_out;
734}
735
1da177e4
LT		 736/* This determines how many packets are "in the network" to the best
737 * of our knowledge. In many cases it is conservative, but where
738 * detailed information is available from the receiver (via SACK
739 * blocks etc.) we can make more aggressive calculations.
740 *
741 * Use this for decisions involving congestion control, use just
742 * tp->packets_out to determine if the send queue is empty or not.
743 *
744 * Read this equation as:
745 *
746 * "Packets sent once on transmission queue" MINUS
747 * "Packets left network, but not honestly ACKed yet" PLUS
748 * "Packets fast retransmitted"
749 */
40efc6fa 750static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1da177e4 751{
83ae4088 752 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1da177e4
LT		 753}
754
1da177e4
LT		 755/* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
756 * The exception is rate halving phase, when cwnd is decreasing towards
757 * ssthresh.
758 */
6687e988 759static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1da177e4 760{
6687e988
ACM		 761 const struct tcp_sock *tp = tcp_sk(sk);
762 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
1da177e4
LT		 763 return tp->snd_ssthresh;
764 else
765 return max(tp->snd_ssthresh,
766 ((tp->snd_cwnd >> 1) +
767 (tp->snd_cwnd >> 2)));
768}
769
b9c4595b
IJ		 770/* Use define here intentionally to get WARN_ON location shown at the caller */
771#define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1da177e4 772
3cfe3baa 773extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
1da177e4
LT		 774extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
775
776/* Slow start with delack produces 3 packets of burst, so that
777 * it is safe "de facto".
778 */
779static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
780{
781 return 3;
782}
783
90840def
IJ		 784/* Returns end sequence number of the receiver's advertised window */
785static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
786{
787 return tp->snd_una + tp->snd_wnd;
788}
cea14e0e 789extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
f4805ede 790
c1bd24b7 791static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
40efc6fa 792 const struct sk_buff *skb)
1da177e4
LT		 793{
794 if (skb->len < mss)
795 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
796}
797
9e412ba7 798static inline void tcp_check_probe_timer(struct sock *sk)
1da177e4 799{
9e412ba7 800 struct tcp_sock *tp = tcp_sk(sk);
463c84b9 801 const struct inet_connection_sock *icsk = inet_csk(sk);
9e412ba7 802
463c84b9 803 if (!tp->packets_out && !icsk->icsk_pending)
3f421baa
ACM		 804 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
805 icsk->icsk_rto, TCP_RTO_MAX);
1da177e4
LT		 806}
807
9e412ba7 808static inline void tcp_push_pending_frames(struct sock *sk)
1da177e4 809{
9e412ba7
IJ		 810 struct tcp_sock *tp = tcp_sk(sk);
811
812 __tcp_push_pending_frames(sk, tcp_current_mss(sk, 1), tp->nonagle);
1da177e4
LT		 813}
814
40efc6fa 815static inline void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1da177e4
LT		 816{
817 tp->snd_wl1 = seq;
818}
819
40efc6fa 820static inline void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq)
1da177e4
LT		 821{
822 tp->snd_wl1 = seq;
823}
824
1da177e4
LT		 825/*
826 * Calculate(/check) TCP checksum
827 */
ba7808ea
FD		 828static inline __sum16 tcp_v4_check(int len, __be32 saddr,
829 __be32 daddr, __wsum base)
1da177e4
LT		 830{
831 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
832}
833
b51655b9 834static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1da177e4 835{
fb286bb2 836 return __skb_checksum_complete(skb);
1da177e4
LT		 837}
838
40efc6fa 839static inline int tcp_checksum_complete(struct sk_buff *skb)
1da177e4 840{
60476372 841 return !skb_csum_unnecessary(skb) &&
1da177e4
LT		 842 __tcp_checksum_complete(skb);
843}
844
845/* Prequeue for VJ style copy to user, combined with checksumming. */
846
40efc6fa 847static inline void tcp_prequeue_init(struct tcp_sock *tp)
1da177e4
LT		 848{
849 tp->ucopy.task = NULL;
850 tp->ucopy.len = 0;
851 tp->ucopy.memory = 0;
852 skb_queue_head_init(&tp->ucopy.prequeue);
97fc2f08
CL		 853#ifdef CONFIG_NET_DMA
854 tp->ucopy.dma_chan = NULL;
855 tp->ucopy.wakeup = 0;
856 tp->ucopy.pinned_list = NULL;
857 tp->ucopy.dma_cookie = 0;
858#endif
1da177e4
LT		 859}
860
861/* Packet is added to VJ-style prequeue for processing in process
862 * context, if a reader task is waiting. Apparently, this exciting
863 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
864 * failed somewhere. Latency? Burstiness? Well, at least now we will
865 * see, why it failed. 8)8) --ANK
866 *
867 * NOTE: is this not too big to inline?
868 */
40efc6fa 869static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
1da177e4
LT		 870{
871 struct tcp_sock *tp = tcp_sk(sk);
872
873 if (!sysctl_tcp_low_latency && tp->ucopy.task) {
874 __skb_queue_tail(&tp->ucopy.prequeue, skb);
875 tp->ucopy.memory += skb->truesize;
876 if (tp->ucopy.memory > sk->sk_rcvbuf) {
877 struct sk_buff *skb1;
878
879 BUG_ON(sock_owned_by_user(sk));
880
881 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
882 sk->sk_backlog_rcv(sk, skb1);
883 NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED);
884 }
885
886 tp->ucopy.memory = 0;
887 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
888 wake_up_interruptible(sk->sk_sleep);
463c84b9
ACM		 889 if (!inet_csk_ack_scheduled(sk))
890 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
3f421baa
ACM		 891 (3 * TCP_RTO_MIN) / 4,
892 TCP_RTO_MAX);
1da177e4
LT		 893 }
894 return 1;
895 }
896 return 0;
897}
898
899
900#undef STATE_TRACE
901
902#ifdef STATE_TRACE
903static const char *statename[]={
904 "Unused","Established","Syn Sent","Syn Recv",
905 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
906 "Close Wait","Last ACK","Listen","Closing"
907};
908#endif
490d5046 909extern void tcp_set_state(struct sock *sk, int state);
1da177e4 910
4ac02bab 911extern void tcp_done(struct sock *sk);
1da177e4 912
40efc6fa 913static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1da177e4
LT		 914{
915 rx_opt->dsack = 0;
916 rx_opt->eff_sacks = 0;
917 rx_opt->num_sacks = 0;
918}
919
1da177e4
LT		 920/* Determine a window scaling and initial window to offer. */
921extern void tcp_select_initial_window(int __space, __u32 mss,
922 __u32 *rcv_wnd, __u32 *window_clamp,
923 int wscale_ok, __u8 *rcv_wscale);
924
925static inline int tcp_win_from_space(int space)
926{
927 return sysctl_tcp_adv_win_scale<=0 ?
928 (space>>(-sysctl_tcp_adv_win_scale)) :
929 space - (space>>sysctl_tcp_adv_win_scale);
930}
931
932/* Note: caller must be prepared to deal with negative returns */
933static inline int tcp_space(const struct sock *sk)
934{
935 return tcp_win_from_space(sk->sk_rcvbuf -
936 atomic_read(&sk->sk_rmem_alloc));
937}
938
939static inline int tcp_full_space(const struct sock *sk)
940{
941 return tcp_win_from_space(sk->sk_rcvbuf);
942}
943
40efc6fa
SH		 944static inline void tcp_openreq_init(struct request_sock *req,
945 struct tcp_options_received *rx_opt,
946 struct sk_buff *skb)
1da177e4 947{
2e6599cb
ACM		 948 struct inet_request_sock *ireq = inet_rsk(req);
949
1da177e4 950 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
2e6599cb 951 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1da177e4
LT		 952 req->mss = rx_opt->mss_clamp;
953 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
2e6599cb
ACM		 954 ireq->tstamp_ok = rx_opt->tstamp_ok;
955 ireq->sack_ok = rx_opt->sack_ok;
956 ireq->snd_wscale = rx_opt->snd_wscale;
957 ireq->wscale_ok = rx_opt->wscale_ok;
958 ireq->acked = 0;
959 ireq->ecn_ok = 0;
aa8223c7 960 ireq->rmt_port = tcp_hdr(skb)->source;
1da177e4
LT		 961}
962
963extern void tcp_enter_memory_pressure(void);
964
1da177e4
LT		 965static inline int keepalive_intvl_when(const struct tcp_sock *tp)
966{
967 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
968}
969
970static inline int keepalive_time_when(const struct tcp_sock *tp)
971{
972 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
973}
974
463c84b9 975static inline int tcp_fin_time(const struct sock *sk)
1da177e4 976{
463c84b9
ACM		 977 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
978 const int rto = inet_csk(sk)->icsk_rto;
1da177e4 979
463c84b9
ACM		 980 if (fin_timeout < (rto << 2) - (rto >> 1))
981 fin_timeout = (rto << 2) - (rto >> 1);
1da177e4
LT		 982
983 return fin_timeout;
984}
985
986static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst)
987{
988 if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0)
989 return 0;
9d729f72 990 if (get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)
1da177e4
LT		 991 return 0;
992
993 /* RST segments are not recommended to carry timestamp,
994 and, if they do, it is recommended to ignore PAWS because
995 "their cleanup function should take precedence over timestamps."
996 Certainly, it is mistake. It is necessary to understand the reasons
997 of this constraint to relax it: if peer reboots, clock may go
998 out-of-sync and half-open connections will not be reset.
999 Actually, the problem would be not existing if all
1000 the implementations followed draft about maintaining clock
1001 via reboots. Linux-2.2 DOES NOT!
1002
1003 However, we can relax time bounds for RST segments to MSL.
1004 */
9d729f72 1005 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1da177e4
LT		 1006 return 0;
1007 return 1;
1008}
1009
1da177e4
LT		 1010#define TCP_CHECK_TIMER(sk) do { } while (0)
1011
1da177e4
LT		 1012static inline void tcp_mib_init(void)
1013{
1014 /* See RFC 2012 */
1015 TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1);
1016 TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1017 TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1018 TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1);
1019}
1020
5af4ec23 1021/* from STCP */
0800f170
DM		 1022static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1023{
6a438bbe
SH		 1024 tp->lost_skb_hint = NULL;
1025 tp->scoreboard_skb_hint = NULL;
1026 tp->retransmit_skb_hint = NULL;
1027 tp->forward_skb_hint = NULL;
b7689205
IJ		 1028}
1029
0800f170
DM		 1030static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1031{
b7689205 1032 tcp_clear_retrans_hints_partial(tp);
6a438bbe
SH		 1033}
1034
cfb6eeb4
YH		 1035/* MD5 Signature */
1036struct crypto_hash;
1037
1038/* - key database */
1039struct tcp_md5sig_key {
1040 u8 *key;
1041 u8 keylen;
1042};
1043
1044struct tcp4_md5sig_key {
f8ab18d2 1045 struct tcp_md5sig_key base;
cfb6eeb4
YH		 1046 __be32 addr;
1047};
1048
1049struct tcp6_md5sig_key {
f8ab18d2 1050 struct tcp_md5sig_key base;
cfb6eeb4
YH		 1051#if 0
1052 u32 scope_id; /* XXX */
1053#endif
1054 struct in6_addr addr;
1055};
1056
1057/* - sock block */
1058struct tcp_md5sig_info {
1059 struct tcp4_md5sig_key *keys4;
1060#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1061 struct tcp6_md5sig_key *keys6;
1062 u32 entries6;
1063 u32 alloced6;
1064#endif
1065 u32 entries4;
1066 u32 alloced4;
1067};
1068
1069/* - pseudo header */
1070struct tcp4_pseudohdr {
1071 __be32 saddr;
1072 __be32 daddr;
1073 __u8 pad;
1074 __u8 protocol;
1075 __be16 len;
1076};
1077
1078struct tcp6_pseudohdr {
1079 struct in6_addr saddr;
1080 struct in6_addr daddr;
1081 __be32 len;
1082 __be32 protocol; /* including padding */
1083};
1084
1085union tcp_md5sum_block {
1086 struct tcp4_pseudohdr ip4;
1087#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1088 struct tcp6_pseudohdr ip6;
1089#endif
1090};
1091
1092/* - pool: digest algorithm, hash description and scratch buffer */
1093struct tcp_md5sig_pool {
1094 struct hash_desc md5_desc;
1095 union tcp_md5sum_block md5_blk;
1096};
1097
1098#define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1099
1100/* - functions */
1101extern int tcp_v4_calc_md5_hash(char *md5_hash,
1102 struct tcp_md5sig_key *key,
1103 struct sock *sk,
1104 struct dst_entry *dst,
1105 struct request_sock *req,
1106 struct tcphdr *th,
9cb5734e
YH		 1107 int protocol,
1108 unsigned int tcplen);
cfb6eeb4
YH		 1109extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1110 struct sock *addr_sk);
1111
1112extern int tcp_v4_md5_do_add(struct sock *sk,
1113 __be32 addr,
1114 u8 *newkey,
1115 u8 newkeylen);
1116
1117extern int tcp_v4_md5_do_del(struct sock *sk,
8e5200f5 1118 __be32 addr);
cfb6eeb4
YH		 1119
1120extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1121extern void tcp_free_md5sig_pool(void);
1122
1123extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1124extern void __tcp_put_md5sig_pool(void);
1125
1126static inline
1127struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1128{
1129 int cpu = get_cpu();
1130 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1131 if (!ret)
1132 put_cpu();
1133 return ret;
1134}
1135
1136static inline void tcp_put_md5sig_pool(void)
1137{
1138 __tcp_put_md5sig_pool();
1139 put_cpu();
1140}
1141
fe067e8a
DM		 1142/* write queue abstraction */
1143static inline void tcp_write_queue_purge(struct sock *sk)
1144{
1145 struct sk_buff *skb;
1146
1147 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
3ab224be
HA		 1148 sk_wmem_free_skb(sk, skb);
1149 sk_mem_reclaim(sk);
fe067e8a
DM		 1150}
1151
1152static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1153{
1154 struct sk_buff *skb = sk->sk_write_queue.next;
1155 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1156 return NULL;
1157 return skb;
1158}
1159
1160static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1161{
1162 struct sk_buff *skb = sk->sk_write_queue.prev;
1163 if (skb == (struct sk_buff *) &sk->sk_write_queue)
1164 return NULL;
1165 return skb;
1166}
1167
1168static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1169{
1170 return skb->next;
1171}
1172
1173#define tcp_for_write_queue(skb, sk) \
1174 for (skb = (sk)->sk_write_queue.next; \
1175 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1176 skb = skb->next)
1177
1178#define tcp_for_write_queue_from(skb, sk) \
1179 for (; (skb != (struct sk_buff *)&(sk)->sk_write_queue);\
1180 skb = skb->next)
1181
234b6860
IJ		 1182#define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1183 for (tmp = skb->next; \
1184 (skb != (struct sk_buff *)&(sk)->sk_write_queue); \
1185 skb = tmp, tmp = skb->next)
1186
fe067e8a
DM		 1187static inline struct sk_buff *tcp_send_head(struct sock *sk)
1188{
1189 return sk->sk_send_head;
1190}
1191
1192static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1193{
1194 sk->sk_send_head = skb->next;
1195 if (sk->sk_send_head == (struct sk_buff *)&sk->sk_write_queue)
1196 sk->sk_send_head = NULL;
1197}
1198
1199static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1200{
1201 if (sk->sk_send_head == skb_unlinked)
1202 sk->sk_send_head = NULL;
1203}
1204
1205static inline void tcp_init_send_head(struct sock *sk)
1206{
1207 sk->sk_send_head = NULL;
1208}
1209
1210static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1211{
1212 __skb_queue_tail(&sk->sk_write_queue, skb);
1213}
1214
1215static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1216{
1217 __tcp_add_write_queue_tail(sk, skb);
1218
1219 /* Queue it, remembering where we must start sending. */
6859d494 1220 if (sk->sk_send_head == NULL) {
fe067e8a 1221 sk->sk_send_head = skb;
6859d494
IJ		 1222
1223 if (tcp_sk(sk)->highest_sack == NULL)
1224 tcp_sk(sk)->highest_sack = skb;
1225 }
fe067e8a
DM		 1226}
1227
1228static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1229{
1230 __skb_queue_head(&sk->sk_write_queue, skb);
1231}
1232
1233/* Insert buff after skb on the write queue of sk. */
1234static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1235 struct sk_buff *buff,
1236 struct sock *sk)
1237{
1238 __skb_append(skb, buff, &sk->sk_write_queue);
1239}
1240
1241/* Insert skb between prev and next on the write queue of sk. */
1242static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1243 struct sk_buff *skb,
1244 struct sock *sk)
1245{
1246 __skb_insert(new, skb->prev, skb, &sk->sk_write_queue);