?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
conn rcv_lock converted to spinlock, struct cor_sock created, kernel_packet skb_clone...
[cor_2_6_31.git] / include / net / tcp.h
blob88af843064710bf43e25cb6932aa40479c19e07e
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 *
10 * Authors: Ross Biro
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
24 #include <linux/list.h>
25 #include <linux/tcp.h>
26 #include <linux/slab.h>
27 #include <linux/cache.h>
28 #include <linux/percpu.h>
29 #include <linux/skbuff.h>
30 #include <linux/dmaengine.h>
31 #include <linux/crypto.h>
32 #include <linux/cryptohash.h>
33
34 #include <net/inet_connection_sock.h>
35 #include <net/inet_timewait_sock.h>
36 #include <net/inet_hashtables.h>
37 #include <net/checksum.h>
38 #include <net/request_sock.h>
39 #include <net/sock.h>
40 #include <net/snmp.h>
41 #include <net/ip.h>
42 #include <net/tcp_states.h>
43 #include <net/inet_ecn.h>
44 #include <net/dst.h>
45
46 #include <linux/seq_file.h>
47
48 extern struct inet_hashinfo tcp_hashinfo;
49
50 extern struct percpu_counter tcp_orphan_count;
51 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
52
53 #define MAX_TCP_HEADER (128 + MAX_HEADER)
54 #define MAX_TCP_OPTION_SPACE 40
55
56 /*
57 * Never offer a window over 32767 without using window scaling. Some
58 * poor stacks do signed 16bit maths!
59 */
60 #define MAX_TCP_WINDOW 32767U
61
62 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
63 #define TCP_MIN_MSS 88U
64
65 /* Minimal RCV_MSS. */
66 #define TCP_MIN_RCVMSS 536U
67
68 /* The least MTU to use for probing */
69 #define TCP_BASE_MSS 512
70
71 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
72 #define TCP_FASTRETRANS_THRESH 3
73
74 /* Maximal reordering. */
75 #define TCP_MAX_REORDERING 127
76
77 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
78 #define TCP_MAX_QUICKACKS 16U
79
80 /* urg_data states */
81 #define TCP_URG_VALID 0x0100
82 #define TCP_URG_NOTYET 0x0200
83 #define TCP_URG_READ 0x0400
84
85 #define TCP_RETR1 3 /*
86 * This is how many retries it does before it
87 * tries to figure out if the gateway is
88 * down. Minimal RFC value is 3; it corresponds
89 * to ~3sec-8min depending on RTO.
90 */
91
92 #define TCP_RETR2 15 /*
93 * This should take at least
94 * 90 minutes to time out.
95 * RFC1122 says that the limit is 100 sec.
96 * 15 is ~13-30min depending on RTO.
97 */
98
99 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
100 * connection: ~180sec is RFC minimum */
101
102 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
103 * connection: ~180sec is RFC minimum */
104
105
106 #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned
107 * socket. 7 is ~50sec-16min.
108 */
109
110
111 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
112 * state, about 60 seconds */
113 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
114 /* BSD style FIN_WAIT2 deadlock breaker.
115 * It used to be 3min, new value is 60sec,
116 * to combine FIN-WAIT-2 timeout with
117 * TIME-WAIT timer.
118 */
119
120 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
121 #if HZ >= 100
122 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
123 #define TCP_ATO_MIN ((unsigned)(HZ/25))
124 #else
125 #define TCP_DELACK_MIN 4U
126 #define TCP_ATO_MIN 4U
127 #endif
128 #define TCP_RTO_MAX ((unsigned)(120*HZ))
129 #define TCP_RTO_MIN ((unsigned)(HZ/5))
130 #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */
131
132 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
133 * for local resources.
134 */
135
136 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
137 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
138 #define TCP_KEEPALIVE_INTVL (75*HZ)
139
140 #define MAX_TCP_KEEPIDLE 32767
141 #define MAX_TCP_KEEPINTVL 32767
142 #define MAX_TCP_KEEPCNT 127
143 #define MAX_TCP_SYNCNT 127
144
145 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
146
147 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
148 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
149 * after this time. It should be equal
150 * (or greater than) TCP_TIMEWAIT_LEN
151 * to provide reliability equal to one
152 * provided by timewait state.
153 */
154 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
155 * timestamps. It must be less than
156 * minimal timewait lifetime.
157 */
158 /*
159 * TCP option
160 */
161
162 #define TCPOPT_NOP 1 /* Padding */
163 #define TCPOPT_EOL 0 /* End of options */
164 #define TCPOPT_MSS 2 /* Segment size negotiating */
165 #define TCPOPT_WINDOW 3 /* Window scaling */
166 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
167 #define TCPOPT_SACK 5 /* SACK Block */
168 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
169 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
170
171 /*
172 * TCP option lengths
173 */
174
175 #define TCPOLEN_MSS 4
176 #define TCPOLEN_WINDOW 3
177 #define TCPOLEN_SACK_PERM 2
178 #define TCPOLEN_TIMESTAMP 10
179 #define TCPOLEN_MD5SIG 18
180
181 /* But this is what stacks really send out. */
182 #define TCPOLEN_TSTAMP_ALIGNED 12
183 #define TCPOLEN_WSCALE_ALIGNED 4
184 #define TCPOLEN_SACKPERM_ALIGNED 4
185 #define TCPOLEN_SACK_BASE 2
186 #define TCPOLEN_SACK_BASE_ALIGNED 4
187 #define TCPOLEN_SACK_PERBLOCK 8
188 #define TCPOLEN_MD5SIG_ALIGNED 20
189 #define TCPOLEN_MSS_ALIGNED 4
190
191 /* Flags in tp->nonagle */
192 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
193 #define TCP_NAGLE_CORK 2 /* Socket is corked */
194 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
195
196 extern struct inet_timewait_death_row tcp_death_row;
197
198 /* sysctl variables for tcp */
199 extern int sysctl_tcp_timestamps;
200 extern int sysctl_tcp_window_scaling;
201 extern int sysctl_tcp_sack;
202 extern int sysctl_tcp_fin_timeout;
203 extern int sysctl_tcp_keepalive_time;
204 extern int sysctl_tcp_keepalive_probes;
205 extern int sysctl_tcp_keepalive_intvl;
206 extern int sysctl_tcp_syn_retries;
207 extern int sysctl_tcp_synack_retries;
208 extern int sysctl_tcp_retries1;
209 extern int sysctl_tcp_retries2;
210 extern int sysctl_tcp_orphan_retries;
211 extern int sysctl_tcp_syncookies;
212 extern int sysctl_tcp_retrans_collapse;
213 extern int sysctl_tcp_stdurg;
214 extern int sysctl_tcp_rfc1337;
215 extern int sysctl_tcp_abort_on_overflow;
216 extern int sysctl_tcp_max_orphans;
217 extern int sysctl_tcp_fack;
218 extern int sysctl_tcp_reordering;
219 extern int sysctl_tcp_ecn;
220 extern int sysctl_tcp_dsack;
221 extern int sysctl_tcp_mem[3];
222 extern int sysctl_tcp_wmem[3];
223 extern int sysctl_tcp_rmem[3];
224 extern int sysctl_tcp_app_win;
225 extern int sysctl_tcp_adv_win_scale;
226 extern int sysctl_tcp_tw_reuse;
227 extern int sysctl_tcp_frto;
228 extern int sysctl_tcp_frto_response;
229 extern int sysctl_tcp_low_latency;
230 extern int sysctl_tcp_dma_copybreak;
231 extern int sysctl_tcp_nometrics_save;
232 extern int sysctl_tcp_moderate_rcvbuf;
233 extern int sysctl_tcp_tso_win_divisor;
234 extern int sysctl_tcp_abc;
235 extern int sysctl_tcp_mtu_probing;
236 extern int sysctl_tcp_base_mss;
237 extern int sysctl_tcp_workaround_signed_windows;
238 extern int sysctl_tcp_slow_start_after_idle;
239 extern int sysctl_tcp_max_ssthresh;
240
241 extern atomic_t tcp_memory_allocated;
242 extern struct percpu_counter tcp_sockets_allocated;
243 extern int tcp_memory_pressure;
244
245 /*
246 * The next routines deal with comparing 32 bit unsigned ints
247 * and worry about wraparound (automatic with unsigned arithmetic).
248 */
249
250 static inline int before(__u32 seq1, __u32 seq2)
251 {
252 return (__s32)(seq1-seq2) < 0;
253 }
254 #define after(seq2, seq1) before(seq1, seq2)
255
256 /* is s2<=s1<=s3 ? */
257 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
258 {
259 return seq3 - seq2 >= seq1 - seq2;
260 }
261
262 static inline int tcp_too_many_orphans(struct sock *sk, int num)
263 {
264 return (num > sysctl_tcp_max_orphans) ||
265 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
266 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
267 }
268
269 /* syncookies: remember time of last synqueue overflow */
270 static inline void tcp_synq_overflow(struct sock *sk)
271 {
272 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
273 }
274
275 /* syncookies: no recent synqueue overflow on this listening socket? */
276 static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
277 {
278 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
279 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT);
280 }
281
282 extern struct proto tcp_prot;
283
284 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
285 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
286 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
287 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
288
289 extern void tcp_v4_err(struct sk_buff *skb, u32);
290
291 extern void tcp_shutdown (struct sock *sk, int how);
292
293 extern int tcp_v4_rcv(struct sk_buff *skb);
294
295 extern int tcp_v4_remember_stamp(struct sock *sk);
296
297 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
298
299 extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
300 struct msghdr *msg, size_t size);
301 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
302
303 extern int tcp_ioctl(struct sock *sk,
304 int cmd,
305 unsigned long arg);
306
307 extern int tcp_rcv_state_process(struct sock *sk,
308 struct sk_buff *skb,
309 struct tcphdr *th,
310 unsigned len);
311
312 extern int tcp_rcv_established(struct sock *sk,
313 struct sk_buff *skb,
314 struct tcphdr *th,
315 unsigned len);
316
317 extern void tcp_rcv_space_adjust(struct sock *sk);
318
319 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
320
321 extern int tcp_twsk_unique(struct sock *sk,
322 struct sock *sktw, void *twp);
323
324 extern void tcp_twsk_destructor(struct sock *sk);
325
326 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
327 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
328
329 static inline void tcp_dec_quickack_mode(struct sock *sk,
330 const unsigned int pkts)
331 {
332 struct inet_connection_sock *icsk = inet_csk(sk);
333
334 if (icsk->icsk_ack.quick) {
335 if (pkts >= icsk->icsk_ack.quick) {
336 icsk->icsk_ack.quick = 0;
337 /* Leaving quickack mode we deflate ATO. */
338 icsk->icsk_ack.ato = TCP_ATO_MIN;
339 } else
340 icsk->icsk_ack.quick -= pkts;
341 }
342 }
343
344 extern void tcp_enter_quickack_mode(struct sock *sk);
345
346 static inline void tcp_clear_options(struct tcp_options_received *rx_opt)
347 {
348 rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0;
349 }
350
351 #define TCP_ECN_OK 1
352 #define TCP_ECN_QUEUE_CWR 2
353 #define TCP_ECN_DEMAND_CWR 4
354
355 static __inline__ void
356 TCP_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
362 enum tcp_tw_status
363 {
364 TCP_TW_SUCCESS = 0,
365 TCP_TW_RST = 1,
366 TCP_TW_ACK = 2,
367 TCP_TW_SYN = 3
368 };
369
370
371 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
372 struct sk_buff *skb,
373 const struct tcphdr *th);
374
375 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
376 struct request_sock *req,
377 struct request_sock **prev);
378 extern int tcp_child_process(struct sock *parent,
379 struct sock *child,
380 struct sk_buff *skb);
381 extern int tcp_use_frto(struct sock *sk);
382 extern void tcp_enter_frto(struct sock *sk);
383 extern void tcp_enter_loss(struct sock *sk, int how);
384 extern void tcp_clear_retrans(struct tcp_sock *tp);
385 extern void tcp_update_metrics(struct sock *sk);
386
387 extern void tcp_close(struct sock *sk,
388 long timeout);
389 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
390
391 extern int tcp_getsockopt(struct sock *sk, int level,
392 int optname,
393 char __user *optval,
394 int __user *optlen);
395 extern int tcp_setsockopt(struct sock *sk, int level,
396 int optname, char __user *optval,
397 int optlen);
398 extern int compat_tcp_getsockopt(struct sock *sk,
399 int level, int optname,
400 char __user *optval, int __user *optlen);
401 extern int compat_tcp_setsockopt(struct sock *sk,
402 int level, int optname,
403 char __user *optval, int optlen);
404 extern void tcp_set_keepalive(struct sock *sk, int val);
405 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
406 struct msghdr *msg,
407 size_t len, int nonblock,
408 int flags, int *addr_len);
409
410 extern void tcp_parse_options(struct sk_buff *skb,
411 struct tcp_options_received *opt_rx,
412 int estab);
413
414 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
415
416 /*
417 * TCP v4 functions exported for the inet6 API
418 */
419
420 extern void tcp_v4_send_check(struct sock *sk, int len,
421 struct sk_buff *skb);
422
423 extern int tcp_v4_conn_request(struct sock *sk,
424 struct sk_buff *skb);
425
426 extern struct sock * tcp_create_openreq_child(struct sock *sk,
427 struct request_sock *req,
428 struct sk_buff *skb);
429
430 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk,
431 struct sk_buff *skb,
432 struct request_sock *req,
433 struct dst_entry *dst);
434
435 extern int tcp_v4_do_rcv(struct sock *sk,
436 struct sk_buff *skb);
437
438 extern int tcp_v4_connect(struct sock *sk,
439 struct sockaddr *uaddr,
440 int addr_len);
441
442 extern int tcp_connect(struct sock *sk);
443
444 extern struct sk_buff * tcp_make_synack(struct sock *sk,
445 struct dst_entry *dst,
446 struct request_sock *req);
447
448 extern int tcp_disconnect(struct sock *sk, int flags);
449
450
451 /* From syncookies.c */
452 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
453 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
454 struct ip_options *opt);
455 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
456 __u16 *mss);
457
458 extern __u32 cookie_init_timestamp(struct request_sock *req);
459 extern void cookie_check_timestamp(struct tcp_options_received *tcp_opt);
460
461 /* From net/ipv6/syncookies.c */
462 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
463 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
464 __u16 *mss);
465
466 /* tcp_output.c */
467
468 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
469 int nonagle);
470 extern int tcp_may_send_now(struct sock *sk);
471 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
472 extern void tcp_xmit_retransmit_queue(struct sock *);
473 extern void tcp_simple_retransmit(struct sock *);
474 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
475 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
476
477 extern void tcp_send_probe0(struct sock *);
478 extern void tcp_send_partial(struct sock *);
479 extern int tcp_write_wakeup(struct sock *);
480 extern void tcp_send_fin(struct sock *sk);
481 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
482 extern int tcp_send_synack(struct sock *);
483 extern void tcp_push_one(struct sock *, unsigned int mss_now);
484 extern void tcp_send_ack(struct sock *sk);
485 extern void tcp_send_delayed_ack(struct sock *sk);
486
487 /* tcp_input.c */
488 extern void tcp_cwnd_application_limited(struct sock *sk);
489
490 /* tcp_timer.c */
491 extern void tcp_init_xmit_timers(struct sock *);
492 static inline void tcp_clear_xmit_timers(struct sock *sk)
493 {
494 inet_csk_clear_xmit_timers(sk);
495 }
496
497 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
498 extern unsigned int tcp_current_mss(struct sock *sk);
499
500 /* Bound MSS / TSO packet size with the half of the window */
501 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
502 {
503 if (tp->max_window && pktsize > (tp->max_window >> 1))
504 return max(tp->max_window >> 1, 68U - tp->tcp_header_len);
505 else
506 return pktsize;
507 }
508
509 /* tcp.c */
510 extern void tcp_get_info(struct sock *, struct tcp_info *);
511
512 /* Read 'sendfile()'-style from a TCP socket */
513 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
514 unsigned int, size_t);
515 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
516 sk_read_actor_t recv_actor);
517
518 extern void tcp_initialize_rcv_mss(struct sock *sk);
519
520 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
521 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
522 extern void tcp_mtup_init(struct sock *sk);
523
524 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
525 {
526 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
527 ntohl(TCP_FLAG_ACK) |
528 snd_wnd);
529 }
530
531 static inline void tcp_fast_path_on(struct tcp_sock *tp)
532 {
533 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
534 }
535
536 static inline void tcp_fast_path_check(struct sock *sk)
537 {
538 struct tcp_sock *tp = tcp_sk(sk);
539
540 if (skb_queue_empty(&tp->out_of_order_queue) &&
541 tp->rcv_wnd &&
542 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
543 !tp->urg_data)
544 tcp_fast_path_on(tp);
545 }
546
547 /* Compute the actual rto_min value */
548 static inline u32 tcp_rto_min(struct sock *sk)
549 {
550 struct dst_entry *dst = __sk_dst_get(sk);
551 u32 rto_min = TCP_RTO_MIN;
552
553 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
554 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
555 return rto_min;
556 }
557
558 /* Compute the actual receive window we are currently advertising.
559 * Rcv_nxt can be after the window if our peer push more data
560 * than the offered window.
561 */
562 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
563 {
564 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
565
566 if (win < 0)
567 win = 0;
568 return (u32) win;
569 }
570
571 /* Choose a new window, without checks for shrinking, and without
572 * scaling applied to the result. The caller does these things
573 * if necessary. This is a "raw" window selection.
574 */
575 extern u32 __tcp_select_window(struct sock *sk);
576
577 /* TCP timestamps are only 32-bits, this causes a slight
578 * complication on 64-bit systems since we store a snapshot
579 * of jiffies in the buffer control blocks below. We decided
580 * to use only the low 32-bits of jiffies and hide the ugly
581 * casts with the following macro.
582 */
583 #define tcp_time_stamp ((__u32)(jiffies))
584
585 /* This is what the send packet queuing engine uses to pass
586 * TCP per-packet control information to the transmission
587 * code. We also store the host-order sequence numbers in
588 * here too. This is 36 bytes on 32-bit architectures,
589 * 40 bytes on 64-bit machines, if this grows please adjust
590 * skbuff.h:skbuff->cb[xxx] size appropriately.
591 */
592 struct tcp_skb_cb {
593 union {
594 struct inet_skb_parm h4;
595 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
596 struct inet6_skb_parm h6;
597 #endif
598 } header; /* For incoming frames */
599 __u32 seq; /* Starting sequence number */
600 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
601 __u32 when; /* used to compute rtt's */
602 __u8 flags; /* TCP header flags. */
603
604 /* NOTE: These must match up to the flags byte in a
605 * real TCP header.
606 */
607 #define TCPCB_FLAG_FIN 0x01
608 #define TCPCB_FLAG_SYN 0x02
609 #define TCPCB_FLAG_RST 0x04
610 #define TCPCB_FLAG_PSH 0x08
611 #define TCPCB_FLAG_ACK 0x10
612 #define TCPCB_FLAG_URG 0x20
613 #define TCPCB_FLAG_ECE 0x40
614 #define TCPCB_FLAG_CWR 0x80
615
616 __u8 sacked; /* State flags for SACK/FACK. */
617 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
618 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
619 #define TCPCB_LOST 0x04 /* SKB is lost */
620 #define TCPCB_TAGBITS 0x07 /* All tag bits */
621
622 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
623 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
624
625 __u32 ack_seq; /* Sequence number ACK'd */
626 };
627
628 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
629
630 /* Due to TSO, an SKB can be composed of multiple actual
631 * packets. To keep these tracked properly, we use this.
632 */
633 static inline int tcp_skb_pcount(const struct sk_buff *skb)
634 {
635 return skb_shinfo(skb)->gso_segs;
636 }
637
638 /* This is valid iff tcp_skb_pcount() > 1. */
639 static inline int tcp_skb_mss(const struct sk_buff *skb)
640 {
641 return skb_shinfo(skb)->gso_size;
642 }
643
644 /* Events passed to congestion control interface */
645 enum tcp_ca_event {
646 CA_EVENT_TX_START, /* first transmit when no packets in flight */
647 CA_EVENT_CWND_RESTART, /* congestion window restart */
648 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
649 CA_EVENT_FRTO, /* fast recovery timeout */
650 CA_EVENT_LOSS, /* loss timeout */
651 CA_EVENT_FAST_ACK, /* in sequence ack */
652 CA_EVENT_SLOW_ACK, /* other ack */
653 };
654
655 /*
656 * Interface for adding new TCP congestion control handlers
657 */
658 #define TCP_CA_NAME_MAX 16
659 #define TCP_CA_MAX 128
660 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
661
662 #define TCP_CONG_NON_RESTRICTED 0x1
663 #define TCP_CONG_RTT_STAMP 0x2
664
665 struct tcp_congestion_ops {
666 struct list_head list;
667 unsigned long flags;
668
669 /* initialize private data (optional) */
670 void (*init)(struct sock *sk);
671 /* cleanup private data (optional) */
672 void (*release)(struct sock *sk);
673
674 /* return slow start threshold (required) */
675 u32 (*ssthresh)(struct sock *sk);
676 /* lower bound for congestion window (optional) */
677 u32 (*min_cwnd)(const struct sock *sk);
678 /* do new cwnd calculation (required) */
679 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
680 /* call before changing ca_state (optional) */
681 void (*set_state)(struct sock *sk, u8 new_state);
682 /* call when cwnd event occurs (optional) */
683 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
684 /* new value of cwnd after loss (optional) */
685 u32 (*undo_cwnd)(struct sock *sk);
686 /* hook for packet ack accounting (optional) */
687 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
688 /* get info for inet_diag (optional) */
689 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
690
691 char name[TCP_CA_NAME_MAX];
692 struct module *owner;
693 };
694
695 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
696 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
697
698 extern void tcp_init_congestion_control(struct sock *sk);
699 extern void tcp_cleanup_congestion_control(struct sock *sk);
700 extern int tcp_set_default_congestion_control(const char *name);
701 extern void tcp_get_default_congestion_control(char *name);
702 extern void tcp_get_available_congestion_control(char *buf, size_t len);
703 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
704 extern int tcp_set_allowed_congestion_control(char *allowed);
705 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
706 extern void tcp_slow_start(struct tcp_sock *tp);
707 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
708
709 extern struct tcp_congestion_ops tcp_init_congestion_ops;
710 extern u32 tcp_reno_ssthresh(struct sock *sk);
711 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
712 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
713 extern struct tcp_congestion_ops tcp_reno;
714
715 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
716 {
717 struct inet_connection_sock *icsk = inet_csk(sk);
718
719 if (icsk->icsk_ca_ops->set_state)
720 icsk->icsk_ca_ops->set_state(sk, ca_state);
721 icsk->icsk_ca_state = ca_state;
722 }
723
724 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
725 {
726 const struct inet_connection_sock *icsk = inet_csk(sk);
727
728 if (icsk->icsk_ca_ops->cwnd_event)
729 icsk->icsk_ca_ops->cwnd_event(sk, event);
730 }
731
732 /* These functions determine how the current flow behaves in respect of SACK
733 * handling. SACK is negotiated with the peer, and therefore it can vary
734 * between different flows.
735 *
736 * tcp_is_sack - SACK enabled
737 * tcp_is_reno - No SACK
738 * tcp_is_fack - FACK enabled, implies SACK enabled
739 */
740 static inline int tcp_is_sack(const struct tcp_sock *tp)
741 {
742 return tp->rx_opt.sack_ok;
743 }
744
745 static inline int tcp_is_reno(const struct tcp_sock *tp)
746 {
747 return !tcp_is_sack(tp);
748 }
749
750 static inline int tcp_is_fack(const struct tcp_sock *tp)
751 {
752 return tp->rx_opt.sack_ok & 2;
753 }
754
755 static inline void tcp_enable_fack(struct tcp_sock *tp)
756 {
757 tp->rx_opt.sack_ok |= 2;
758 }
759
760 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
761 {
762 return tp->sacked_out + tp->lost_out;
763 }
764
765 /* This determines how many packets are "in the network" to the best
766 * of our knowledge. In many cases it is conservative, but where
767 * detailed information is available from the receiver (via SACK
768 * blocks etc.) we can make more aggressive calculations.
769 *
770 * Use this for decisions involving congestion control, use just
771 * tp->packets_out to determine if the send queue is empty or not.
772 *
773 * Read this equation as:
774 *
775 * "Packets sent once on transmission queue" MINUS
776 * "Packets left network, but not honestly ACKed yet" PLUS
777 * "Packets fast retransmitted"
778 */
779 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
780 {
781 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
782 }
783
784 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
785 * The exception is rate halving phase, when cwnd is decreasing towards
786 * ssthresh.
787 */
788 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
789 {
790 const struct tcp_sock *tp = tcp_sk(sk);
791 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
792 return tp->snd_ssthresh;
793 else
794 return max(tp->snd_ssthresh,
795 ((tp->snd_cwnd >> 1) +
796 (tp->snd_cwnd >> 2)));
797 }
798
799 /* Use define here intentionally to get WARN_ON location shown at the caller */
800 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
801
802 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
803 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
804
805 /* Slow start with delack produces 3 packets of burst, so that
806 * it is safe "de facto". This will be the default - same as
807 * the default reordering threshold - but if reordering increases,
808 * we must be able to allow cwnd to burst at least this much in order
809 * to not pull it back when holes are filled.
810 */
811 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
812 {
813 return tp->reordering;
814 }
815
816 /* Returns end sequence number of the receiver's advertised window */
817 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
818 {
819 return tp->snd_una + tp->snd_wnd;
820 }
821 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
822
823 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
824 const struct sk_buff *skb)
825 {
826 if (skb->len < mss)
827 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
828 }
829
830 static inline void tcp_check_probe_timer(struct sock *sk)
831 {
832 struct tcp_sock *tp = tcp_sk(sk);
833 const struct inet_connection_sock *icsk = inet_csk(sk);
834
835 if (!tp->packets_out && !icsk->icsk_pending)
836 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
837 icsk->icsk_rto, TCP_RTO_MAX);
838 }
839
840 static inline void tcp_push_pending_frames(struct sock *sk)
841 {
842 struct tcp_sock *tp = tcp_sk(sk);
843
844 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
845 }
846
847 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
848 {
849 tp->snd_wl1 = seq;
850 }
851
852 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
853 {
854 tp->snd_wl1 = seq;
855 }
856
857 /*
858 * Calculate(/check) TCP checksum
859 */
860 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
861 __be32 daddr, __wsum base)
862 {
863 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
864 }
865
866 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
867 {
868 return __skb_checksum_complete(skb);
869 }
870
871 static inline int tcp_checksum_complete(struct sk_buff *skb)
872 {
873 return !skb_csum_unnecessary(skb) &&
874 __tcp_checksum_complete(skb);
875 }
876
877 /* Prequeue for VJ style copy to user, combined with checksumming. */
878
879 static inline void tcp_prequeue_init(struct tcp_sock *tp)
880 {
881 tp->ucopy.task = NULL;
882 tp->ucopy.len = 0;
883 tp->ucopy.memory = 0;
884 skb_queue_head_init(&tp->ucopy.prequeue);
885 #ifdef CONFIG_NET_DMA
886 tp->ucopy.dma_chan = NULL;
887 tp->ucopy.wakeup = 0;
888 tp->ucopy.pinned_list = NULL;
889 tp->ucopy.dma_cookie = 0;
890 #endif
891 }
892
893 /* Packet is added to VJ-style prequeue for processing in process
894 * context, if a reader task is waiting. Apparently, this exciting
895 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
896 * failed somewhere. Latency? Burstiness? Well, at least now we will
897 * see, why it failed. 8)8) --ANK
898 *
899 * NOTE: is this not too big to inline?
900 */
901 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
902 {
903 struct tcp_sock *tp = tcp_sk(sk);
904
905 if (sysctl_tcp_low_latency || !tp->ucopy.task)
906 return 0;
907
908 __skb_queue_tail(&tp->ucopy.prequeue, skb);
909 tp->ucopy.memory += skb->truesize;
910 if (tp->ucopy.memory > sk->sk_rcvbuf) {
911 struct sk_buff *skb1;
912
913 BUG_ON(sock_owned_by_user(sk));
914
915 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
916 sk_backlog_rcv(sk, skb1);
917 NET_INC_STATS_BH(sock_net(sk),
918 LINUX_MIB_TCPPREQUEUEDROPPED);
919 }
920
921 tp->ucopy.memory = 0;
922 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
923 wake_up_interruptible_poll(sk->sk_sleep,
924 POLLIN | POLLRDNORM | POLLRDBAND);
925 if (!inet_csk_ack_scheduled(sk))
926 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
927 (3 * tcp_rto_min(sk)) / 4,
928 TCP_RTO_MAX);
929 }
930 return 1;
931 }
932
933
934 #undef STATE_TRACE
935
936 #ifdef STATE_TRACE
937 static const char *statename[]={
938 "Unused","Established","Syn Sent","Syn Recv",
939 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
940 "Close Wait","Last ACK","Listen","Closing"
941 };
942 #endif
943 extern void tcp_set_state(struct sock *sk, int state);
944
945 extern void tcp_done(struct sock *sk);
946
947 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
948 {
949 rx_opt->dsack = 0;
950 rx_opt->num_sacks = 0;
951 }
952
953 /* Determine a window scaling and initial window to offer. */
954 extern void tcp_select_initial_window(int __space, __u32 mss,
955 __u32 *rcv_wnd, __u32 *window_clamp,
956 int wscale_ok, __u8 *rcv_wscale);
957
958 static inline int tcp_win_from_space(int space)
959 {
960 return sysctl_tcp_adv_win_scale<=0 ?
961 (space>>(-sysctl_tcp_adv_win_scale)) :
962 space - (space>>sysctl_tcp_adv_win_scale);
963 }
964
965 /* Note: caller must be prepared to deal with negative returns */
966 static inline int tcp_space(const struct sock *sk)
967 {
968 return tcp_win_from_space(sk->sk_rcvbuf -
969 atomic_read(&sk->sk_rmem_alloc));
970 }
971
972 static inline int tcp_full_space(const struct sock *sk)
973 {
974 return tcp_win_from_space(sk->sk_rcvbuf);
975 }
976
977 static inline void tcp_openreq_init(struct request_sock *req,
978 struct tcp_options_received *rx_opt,
979 struct sk_buff *skb)
980 {
981 struct inet_request_sock *ireq = inet_rsk(req);
982
983 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
984 req->cookie_ts = 0;
985 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
986 req->mss = rx_opt->mss_clamp;
987 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
988 ireq->tstamp_ok = rx_opt->tstamp_ok;
989 ireq->sack_ok = rx_opt->sack_ok;
990 ireq->snd_wscale = rx_opt->snd_wscale;
991 ireq->wscale_ok = rx_opt->wscale_ok;
992 ireq->acked = 0;
993 ireq->ecn_ok = 0;
994 ireq->rmt_port = tcp_hdr(skb)->source;
995 ireq->loc_port = tcp_hdr(skb)->dest;
996 }
997
998 extern void tcp_enter_memory_pressure(struct sock *sk);
999
1000 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1001 {
1002 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1003 }
1004
1005 static inline int keepalive_time_when(const struct tcp_sock *tp)
1006 {
1007 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1008 }
1009
1010 static inline int tcp_fin_time(const struct sock *sk)
1011 {
1012 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1013 const int rto = inet_csk(sk)->icsk_rto;
1014
1015 if (fin_timeout < (rto << 2) - (rto >> 1))
1016 fin_timeout = (rto << 2) - (rto >> 1);
1017
1018 return fin_timeout;
1019 }
1020
1021 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1022 int paws_win)
1023 {
1024 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1025 return 1;
1026 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1027 return 1;
1028
1029 return 0;
1030 }
1031
1032 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1033 int rst)
1034 {
1035 if (tcp_paws_check(rx_opt, 0))
1036 return 0;
1037
1038 /* RST segments are not recommended to carry timestamp,
1039 and, if they do, it is recommended to ignore PAWS because
1040 "their cleanup function should take precedence over timestamps."
1041 Certainly, it is mistake. It is necessary to understand the reasons
1042 of this constraint to relax it: if peer reboots, clock may go
1043 out-of-sync and half-open connections will not be reset.
1044 Actually, the problem would be not existing if all
1045 the implementations followed draft about maintaining clock
1046 via reboots. Linux-2.2 DOES NOT!
1047
1048 However, we can relax time bounds for RST segments to MSL.
1049 */
1050 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1051 return 0;
1052 return 1;
1053 }
1054
1055 #define TCP_CHECK_TIMER(sk) do { } while (0)
1056
1057 static inline void tcp_mib_init(struct net *net)
1058 {
1059 /* See RFC 2012 */
1060 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1061 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1062 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1063 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1064 }
1065
1066 /* from STCP */
1067 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1068 {
1069 tp->lost_skb_hint = NULL;
1070 tp->scoreboard_skb_hint = NULL;
1071 }
1072
1073 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1074 {
1075 tcp_clear_retrans_hints_partial(tp);
1076 tp->retransmit_skb_hint = NULL;
1077 }
1078
1079 /* MD5 Signature */
1080 struct crypto_hash;
1081
1082 /* - key database */
1083 struct tcp_md5sig_key {
1084 u8 *key;
1085 u8 keylen;
1086 };
1087
1088 struct tcp4_md5sig_key {
1089 struct tcp_md5sig_key base;
1090 __be32 addr;
1091 };
1092
1093 struct tcp6_md5sig_key {
1094 struct tcp_md5sig_key base;
1095 #if 0
1096 u32 scope_id; /* XXX */
1097 #endif
1098 struct in6_addr addr;
1099 };
1100
1101 /* - sock block */
1102 struct tcp_md5sig_info {
1103 struct tcp4_md5sig_key *keys4;
1104 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1105 struct tcp6_md5sig_key *keys6;
1106 u32 entries6;
1107 u32 alloced6;
1108 #endif
1109 u32 entries4;
1110 u32 alloced4;
1111 };
1112
1113 /* - pseudo header */
1114 struct tcp4_pseudohdr {
1115 __be32 saddr;
1116 __be32 daddr;
1117 __u8 pad;
1118 __u8 protocol;
1119 __be16 len;
1120 };
1121
1122 struct tcp6_pseudohdr {
1123 struct in6_addr saddr;
1124 struct in6_addr daddr;
1125 __be32 len;
1126 __be32 protocol; /* including padding */
1127 };
1128
1129 union tcp_md5sum_block {
1130 struct tcp4_pseudohdr ip4;
1131 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1132 struct tcp6_pseudohdr ip6;
1133 #endif
1134 };
1135
1136 /* - pool: digest algorithm, hash description and scratch buffer */
1137 struct tcp_md5sig_pool {
1138 struct hash_desc md5_desc;
1139 union tcp_md5sum_block md5_blk;
1140 };
1141
1142 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1143
1144 /* - functions */
1145 extern int tcp_v4_md5_hash_skb(char *md5_hash,
1146 struct tcp_md5sig_key *key,
1147 struct sock *sk,
1148 struct request_sock *req,
1149 struct sk_buff *skb);
1150
1151 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1152 struct sock *addr_sk);
1153
1154 extern int tcp_v4_md5_do_add(struct sock *sk,
1155 __be32 addr,
1156 u8 *newkey,
1157 u8 newkeylen);
1158
1159 extern int tcp_v4_md5_do_del(struct sock *sk,
1160 __be32 addr);
1161
1162 #ifdef CONFIG_TCP_MD5SIG
1163 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1164 &(struct tcp_md5sig_key) { \
1165 .key = (twsk)->tw_md5_key, \
1166 .keylen = (twsk)->tw_md5_keylen, \
1167 } : NULL)
1168 #else
1169 #define tcp_twsk_md5_key(twsk) NULL
1170 #endif
1171
1172 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(void);
1173 extern void tcp_free_md5sig_pool(void);
1174
1175 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1176 extern void __tcp_put_md5sig_pool(void);
1177 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1178 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1179 unsigned header_len);
1180 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1181 struct tcp_md5sig_key *key);
1182
1183 static inline
1184 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1185 {
1186 int cpu = get_cpu();
1187 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1188 if (!ret)
1189 put_cpu();
1190 return ret;
1191 }
1192
1193 static inline void tcp_put_md5sig_pool(void)
1194 {
1195 __tcp_put_md5sig_pool();
1196 put_cpu();
1197 }
1198
1199 /* write queue abstraction */
1200 static inline void tcp_write_queue_purge(struct sock *sk)
1201 {
1202 struct sk_buff *skb;
1203
1204 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1205 sk_wmem_free_skb(sk, skb);
1206 sk_mem_reclaim(sk);
1207 }
1208
1209 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1210 {
1211 return skb_peek(&sk->sk_write_queue);
1212 }
1213
1214 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1215 {
1216 return skb_peek_tail(&sk->sk_write_queue);
1217 }
1218
1219 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1220 {
1221 return skb_queue_next(&sk->sk_write_queue, skb);
1222 }
1223
1224 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb)
1225 {
1226 return skb_queue_prev(&sk->sk_write_queue, skb);
1227 }
1228
1229 #define tcp_for_write_queue(skb, sk) \
1230 skb_queue_walk(&(sk)->sk_write_queue, skb)
1231
1232 #define tcp_for_write_queue_from(skb, sk) \
1233 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1234
1235 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1236 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1237
1238 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1239 {
1240 return sk->sk_send_head;
1241 }
1242
1243 static inline bool tcp_skb_is_last(const struct sock *sk,
1244 const struct sk_buff *skb)
1245 {
1246 return skb_queue_is_last(&sk->sk_write_queue, skb);
1247 }
1248
1249 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1250 {
1251 if (tcp_skb_is_last(sk, skb))
1252 sk->sk_send_head = NULL;
1253 else
1254 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1255 }
1256
1257 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1258 {
1259 if (sk->sk_send_head == skb_unlinked)
1260 sk->sk_send_head = NULL;
1261 }
1262
1263 static inline void tcp_init_send_head(struct sock *sk)
1264 {
1265 sk->sk_send_head = NULL;
1266 }
1267
1268 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1269 {
1270 __skb_queue_tail(&sk->sk_write_queue, skb);
1271 }
1272
1273 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1274 {
1275 __tcp_add_write_queue_tail(sk, skb);
1276
1277 /* Queue it, remembering where we must start sending. */
1278 if (sk->sk_send_head == NULL) {
1279 sk->sk_send_head = skb;
1280
1281 if (tcp_sk(sk)->highest_sack == NULL)
1282 tcp_sk(sk)->highest_sack = skb;
1283 }
1284 }
1285
1286 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1287 {
1288 __skb_queue_head(&sk->sk_write_queue, skb);
1289 }
1290
1291 /* Insert buff after skb on the write queue of sk. */
1292 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1293 struct sk_buff *buff,
1294 struct sock *sk)
1295 {
1296 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1297 }
1298
1299 /* Insert new before skb on the write queue of sk. */
1300 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1301 struct sk_buff *skb,
1302 struct sock *sk)
1303 {
1304 __skb_queue_before(&sk->sk_write_queue, skb, new);
1305
1306 if (sk->sk_send_head == skb)
1307 sk->sk_send_head = new;
1308 }
1309
1310 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1311 {
1312 __skb_unlink(skb, &sk->sk_write_queue);
1313 }
1314
1315 static inline int tcp_write_queue_empty(struct sock *sk)
1316 {
1317 return skb_queue_empty(&sk->sk_write_queue);
1318 }
1319
1320 /* Start sequence of the highest skb with SACKed bit, valid only if
1321 * sacked > 0 or when the caller has ensured validity by itself.
1322 */
1323 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1324 {
1325 if (!tp->sacked_out)
1326 return tp->snd_una;
1327
1328 if (tp->highest_sack == NULL)
1329 return tp->snd_nxt;
1330
1331 return TCP_SKB_CB(tp->highest_sack)->seq;
1332 }
1333
1334 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1335 {
1336 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1337 tcp_write_queue_next(sk, skb);
1338 }
1339
1340 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1341 {
1342 return tcp_sk(sk)->highest_sack;
1343 }
1344
1345 static inline void tcp_highest_sack_reset(struct sock *sk)
1346 {
1347 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1348 }
1349
1350 /* Called when old skb is about to be deleted (to be combined with new skb) */
1351 static inline void tcp_highest_sack_combine(struct sock *sk,
1352 struct sk_buff *old,
1353 struct sk_buff *new)
1354 {
1355 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1356 tcp_sk(sk)->highest_sack = new;
1357 }
1358
1359 /* /proc */
1360 enum tcp_seq_states {
1361 TCP_SEQ_STATE_LISTENING,
1362 TCP_SEQ_STATE_OPENREQ,
1363 TCP_SEQ_STATE_ESTABLISHED,
1364 TCP_SEQ_STATE_TIME_WAIT,
1365 };
1366
1367 struct tcp_seq_afinfo {
1368 char *name;
1369 sa_family_t family;
1370 struct file_operations seq_fops;
1371 struct seq_operations seq_ops;
1372 };
1373
1374 struct tcp_iter_state {
1375 struct seq_net_private p;
1376 sa_family_t family;
1377 enum tcp_seq_states state;
1378 struct sock *syn_wait_sk;
1379 int bucket, sbucket, num, uid;
1380 };
1381
1382 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1383 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1384
1385 extern struct request_sock_ops tcp_request_sock_ops;
1386 extern struct request_sock_ops tcp6_request_sock_ops;
1387
1388 extern void tcp_v4_destroy_sock(struct sock *sk);
1389
1390 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1391 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1392 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1393 struct sk_buff *skb);
1394 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1395 struct sk_buff *skb);
1396 extern int tcp_gro_complete(struct sk_buff *skb);
1397 extern int tcp4_gro_complete(struct sk_buff *skb);
1398
1399 #ifdef CONFIG_PROC_FS
1400 extern int tcp4_proc_init(void);
1401 extern void tcp4_proc_exit(void);
1402 #endif
1403
1404 /* TCP af-specific functions */
1405 struct tcp_sock_af_ops {
1406 #ifdef CONFIG_TCP_MD5SIG
1407 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1408 struct sock *addr_sk);
1409 int (*calc_md5_hash) (char *location,
1410 struct tcp_md5sig_key *md5,
1411 struct sock *sk,
1412 struct request_sock *req,
1413 struct sk_buff *skb);
1414 int (*md5_add) (struct sock *sk,
1415 struct sock *addr_sk,
1416 u8 *newkey,
1417 u8 len);
1418 int (*md5_parse) (struct sock *sk,
1419 char __user *optval,
1420 int optlen);
1421 #endif
1422 };
1423
1424 struct tcp_request_sock_ops {
1425 #ifdef CONFIG_TCP_MD5SIG
1426 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1427 struct request_sock *req);
1428 int (*calc_md5_hash) (char *location,
1429 struct tcp_md5sig_key *md5,
1430 struct sock *sk,
1431 struct request_sock *req,
1432 struct sk_buff *skb);
1433 #endif
1434 };
1435
1436 extern void tcp_v4_init(void);
1437 extern void tcp_init(void);
1438
1439 #endif /* _TCP_H */
connection oriented routing - ARCHIVE NO UPDATES HERE