?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
x86, msr: Add support for non-contiguous cpumasks
[linux-2.6.git] / include / net / tcp.h
blob5740b85bc5a0978349eb042caf17ec6255c98158
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 #include <linux/kref.h>
34
35 #include <net/inet_connection_sock.h>
36 #include <net/inet_timewait_sock.h>
37 #include <net/inet_hashtables.h>
38 #include <net/checksum.h>
39 #include <net/request_sock.h>
40 #include <net/sock.h>
41 #include <net/snmp.h>
42 #include <net/ip.h>
43 #include <net/tcp_states.h>
44 #include <net/inet_ecn.h>
45 #include <net/dst.h>
46
47 #include <linux/seq_file.h>
48
49 extern struct inet_hashinfo tcp_hashinfo;
50
51 extern struct percpu_counter tcp_orphan_count;
52 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
53
54 #define MAX_TCP_HEADER (128 + MAX_HEADER)
55 #define MAX_TCP_OPTION_SPACE 40
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
66 /* The least MTU to use for probing */
67 #define TCP_BASE_MSS 512
68
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
98 * connection: ~180sec is RFC minimum */
99
100 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
101 * connection: ~180sec is RFC minimum */
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 */
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 */
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 */
167 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
168 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
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
178 #define TCPOLEN_MD5SIG 18
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)
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
191 #define TCPOLEN_MD5SIG_ALIGNED 20
192 #define TCPOLEN_MSS_ALIGNED 4
193
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 */
197 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
198
199 extern struct inet_timewait_death_row tcp_death_row;
200
201 /* sysctl variables for tcp */
202 extern int sysctl_tcp_timestamps;
203 extern int sysctl_tcp_window_scaling;
204 extern int sysctl_tcp_sack;
205 extern int sysctl_tcp_fin_timeout;
206 extern int sysctl_tcp_keepalive_time;
207 extern int sysctl_tcp_keepalive_probes;
208 extern int sysctl_tcp_keepalive_intvl;
209 extern int sysctl_tcp_syn_retries;
210 extern int sysctl_tcp_synack_retries;
211 extern int sysctl_tcp_retries1;
212 extern int sysctl_tcp_retries2;
213 extern int sysctl_tcp_orphan_retries;
214 extern int sysctl_tcp_syncookies;
215 extern int sysctl_tcp_retrans_collapse;
216 extern int sysctl_tcp_stdurg;
217 extern int sysctl_tcp_rfc1337;
218 extern int sysctl_tcp_abort_on_overflow;
219 extern int sysctl_tcp_max_orphans;
220 extern int sysctl_tcp_fack;
221 extern int sysctl_tcp_reordering;
222 extern int sysctl_tcp_ecn;
223 extern int sysctl_tcp_dsack;
224 extern int sysctl_tcp_mem[3];
225 extern int sysctl_tcp_wmem[3];
226 extern int sysctl_tcp_rmem[3];
227 extern int sysctl_tcp_app_win;
228 extern int sysctl_tcp_adv_win_scale;
229 extern int sysctl_tcp_tw_reuse;
230 extern int sysctl_tcp_frto;
231 extern int sysctl_tcp_frto_response;
232 extern int sysctl_tcp_low_latency;
233 extern int sysctl_tcp_dma_copybreak;
234 extern int sysctl_tcp_nometrics_save;
235 extern int sysctl_tcp_moderate_rcvbuf;
236 extern int sysctl_tcp_tso_win_divisor;
237 extern int sysctl_tcp_abc;
238 extern int sysctl_tcp_mtu_probing;
239 extern int sysctl_tcp_base_mss;
240 extern int sysctl_tcp_workaround_signed_windows;
241 extern int sysctl_tcp_slow_start_after_idle;
242 extern int sysctl_tcp_max_ssthresh;
243 extern int sysctl_tcp_cookie_size;
244
245 extern atomic_t tcp_memory_allocated;
246 extern struct percpu_counter tcp_sockets_allocated;
247 extern int tcp_memory_pressure;
248
249 /*
250 * The next routines deal with comparing 32 bit unsigned ints
251 * and worry about wraparound (automatic with unsigned arithmetic).
252 */
253
254 static inline int before(__u32 seq1, __u32 seq2)
255 {
256 return (__s32)(seq1-seq2) < 0;
257 }
258 #define after(seq2, seq1) before(seq1, seq2)
259
260 /* is s2<=s1<=s3 ? */
261 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
262 {
263 return seq3 - seq2 >= seq1 - seq2;
264 }
265
266 static inline int tcp_too_many_orphans(struct sock *sk, int num)
267 {
268 return (num > sysctl_tcp_max_orphans) ||
269 (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
270 atomic_read(&tcp_memory_allocated) > sysctl_tcp_mem[2]);
271 }
272
273 /* syncookies: remember time of last synqueue overflow */
274 static inline void tcp_synq_overflow(struct sock *sk)
275 {
276 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
277 }
278
279 /* syncookies: no recent synqueue overflow on this listening socket? */
280 static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
281 {
282 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
283 return time_after(jiffies, last_overflow + TCP_TIMEOUT_INIT);
284 }
285
286 extern struct proto tcp_prot;
287
288 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
289 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
290 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
291 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
292
293 extern void tcp_v4_err(struct sk_buff *skb, u32);
294
295 extern void tcp_shutdown (struct sock *sk, int how);
296
297 extern int tcp_v4_rcv(struct sk_buff *skb);
298
299 extern int tcp_v4_remember_stamp(struct sock *sk);
300
301 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
302
303 extern int tcp_sendmsg(struct kiocb *iocb, struct socket *sock,
304 struct msghdr *msg, size_t size);
305 extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags);
306
307 extern int tcp_ioctl(struct sock *sk,
308 int cmd,
309 unsigned long arg);
310
311 extern int tcp_rcv_state_process(struct sock *sk,
312 struct sk_buff *skb,
313 struct tcphdr *th,
314 unsigned len);
315
316 extern int tcp_rcv_established(struct sock *sk,
317 struct sk_buff *skb,
318 struct tcphdr *th,
319 unsigned len);
320
321 extern void tcp_rcv_space_adjust(struct sock *sk);
322
323 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
324
325 extern int tcp_twsk_unique(struct sock *sk,
326 struct sock *sktw, void *twp);
327
328 extern void tcp_twsk_destructor(struct sock *sk);
329
330 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
331 struct pipe_inode_info *pipe, size_t len, unsigned int flags);
332
333 static inline void tcp_dec_quickack_mode(struct sock *sk,
334 const unsigned int pkts)
335 {
336 struct inet_connection_sock *icsk = inet_csk(sk);
337
338 if (icsk->icsk_ack.quick) {
339 if (pkts >= icsk->icsk_ack.quick) {
340 icsk->icsk_ack.quick = 0;
341 /* Leaving quickack mode we deflate ATO. */
342 icsk->icsk_ack.ato = TCP_ATO_MIN;
343 } else
344 icsk->icsk_ack.quick -= pkts;
345 }
346 }
347
348 extern void tcp_enter_quickack_mode(struct sock *sk);
349
350 #define TCP_ECN_OK 1
351 #define TCP_ECN_QUEUE_CWR 2
352 #define TCP_ECN_DEMAND_CWR 4
353
354 static __inline__ void
355 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
356 {
357 if (sysctl_tcp_ecn && th->ece && th->cwr)
358 inet_rsk(req)->ecn_ok = 1;
359 }
360
361 enum tcp_tw_status {
362 TCP_TW_SUCCESS = 0,
363 TCP_TW_RST = 1,
364 TCP_TW_ACK = 2,
365 TCP_TW_SYN = 3
366 };
367
368
369 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
370 struct sk_buff *skb,
371 const struct tcphdr *th);
372
373 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
374 struct request_sock *req,
375 struct request_sock **prev);
376 extern int tcp_child_process(struct sock *parent,
377 struct sock *child,
378 struct sk_buff *skb);
379 extern int tcp_use_frto(struct sock *sk);
380 extern void tcp_enter_frto(struct sock *sk);
381 extern void tcp_enter_loss(struct sock *sk, int how);
382 extern void tcp_clear_retrans(struct tcp_sock *tp);
383 extern void tcp_update_metrics(struct sock *sk);
384
385 extern void tcp_close(struct sock *sk,
386 long timeout);
387 extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait);
388
389 extern int tcp_getsockopt(struct sock *sk, int level,
390 int optname,
391 char __user *optval,
392 int __user *optlen);
393 extern int tcp_setsockopt(struct sock *sk, int level,
394 int optname, char __user *optval,
395 unsigned int optlen);
396 extern int compat_tcp_getsockopt(struct sock *sk,
397 int level, int optname,
398 char __user *optval, int __user *optlen);
399 extern int compat_tcp_setsockopt(struct sock *sk,
400 int level, int optname,
401 char __user *optval, unsigned int optlen);
402 extern void tcp_set_keepalive(struct sock *sk, int val);
403 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk,
404 struct msghdr *msg,
405 size_t len, int nonblock,
406 int flags, int *addr_len);
407
408 extern void tcp_parse_options(struct sk_buff *skb,
409 struct tcp_options_received *opt_rx,
410 u8 **hvpp,
411 int estab,
412 struct dst_entry *dst);
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 struct request_values *rvp);
448
449 extern int tcp_disconnect(struct sock *sk, int flags);
450
451
452 /* From syncookies.c */
453 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
454 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
455 struct ip_options *opt);
456 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
457 __u16 *mss);
458
459 extern __u32 cookie_init_timestamp(struct request_sock *req);
460 extern void cookie_check_timestamp(struct tcp_options_received *tcp_opt);
461
462 /* From net/ipv6/syncookies.c */
463 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
464 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
465 __u16 *mss);
466
467 /* tcp_output.c */
468
469 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
470 int nonagle);
471 extern int tcp_may_send_now(struct sock *sk);
472 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
473 extern void tcp_retransmit_timer(struct sock *sk);
474 extern void tcp_xmit_retransmit_queue(struct sock *);
475 extern void tcp_simple_retransmit(struct sock *);
476 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
477 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
478
479 extern void tcp_send_probe0(struct sock *);
480 extern void tcp_send_partial(struct sock *);
481 extern int tcp_write_wakeup(struct sock *);
482 extern void tcp_send_fin(struct sock *sk);
483 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
484 extern int tcp_send_synack(struct sock *);
485 extern void tcp_push_one(struct sock *, unsigned int mss_now);
486 extern void tcp_send_ack(struct sock *sk);
487 extern void tcp_send_delayed_ack(struct sock *sk);
488
489 /* tcp_input.c */
490 extern void tcp_cwnd_application_limited(struct sock *sk);
491
492 /* tcp_timer.c */
493 extern void tcp_init_xmit_timers(struct sock *);
494 static inline void tcp_clear_xmit_timers(struct sock *sk)
495 {
496 inet_csk_clear_xmit_timers(sk);
497 }
498
499 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
500 extern unsigned int tcp_current_mss(struct sock *sk);
501
502 /* Bound MSS / TSO packet size with the half of the window */
503 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
504 {
505 if (tp->max_window && pktsize > (tp->max_window >> 1))
506 return max(tp->max_window >> 1, 68U - tp->tcp_header_len);
507 else
508 return pktsize;
509 }
510
511 /* tcp.c */
512 extern void tcp_get_info(struct sock *, struct tcp_info *);
513
514 /* Read 'sendfile()'-style from a TCP socket */
515 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
516 unsigned int, size_t);
517 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
518 sk_read_actor_t recv_actor);
519
520 extern void tcp_initialize_rcv_mss(struct sock *sk);
521
522 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
523 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
524 extern void tcp_mtup_init(struct sock *sk);
525
526 static inline void tcp_bound_rto(const struct sock *sk)
527 {
528 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
529 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
530 }
531
532 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
533 {
534 return (tp->srtt >> 3) + tp->rttvar;
535 }
536
537 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
538 {
539 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
540 ntohl(TCP_FLAG_ACK) |
541 snd_wnd);
542 }
543
544 static inline void tcp_fast_path_on(struct tcp_sock *tp)
545 {
546 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
547 }
548
549 static inline void tcp_fast_path_check(struct sock *sk)
550 {
551 struct tcp_sock *tp = tcp_sk(sk);
552
553 if (skb_queue_empty(&tp->out_of_order_queue) &&
554 tp->rcv_wnd &&
555 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
556 !tp->urg_data)
557 tcp_fast_path_on(tp);
558 }
559
560 /* Compute the actual rto_min value */
561 static inline u32 tcp_rto_min(struct sock *sk)
562 {
563 struct dst_entry *dst = __sk_dst_get(sk);
564 u32 rto_min = TCP_RTO_MIN;
565
566 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
567 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
568 return rto_min;
569 }
570
571 /* Compute the actual receive window we are currently advertising.
572 * Rcv_nxt can be after the window if our peer push more data
573 * than the offered window.
574 */
575 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
576 {
577 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
578
579 if (win < 0)
580 win = 0;
581 return (u32) win;
582 }
583
584 /* Choose a new window, without checks for shrinking, and without
585 * scaling applied to the result. The caller does these things
586 * if necessary. This is a "raw" window selection.
587 */
588 extern u32 __tcp_select_window(struct sock *sk);
589
590 /* TCP timestamps are only 32-bits, this causes a slight
591 * complication on 64-bit systems since we store a snapshot
592 * of jiffies in the buffer control blocks below. We decided
593 * to use only the low 32-bits of jiffies and hide the ugly
594 * casts with the following macro.
595 */
596 #define tcp_time_stamp ((__u32)(jiffies))
597
598 /* This is what the send packet queuing engine uses to pass
599 * TCP per-packet control information to the transmission
600 * code. We also store the host-order sequence numbers in
601 * here too. This is 36 bytes on 32-bit architectures,
602 * 40 bytes on 64-bit machines, if this grows please adjust
603 * skbuff.h:skbuff->cb[xxx] size appropriately.
604 */
605 struct tcp_skb_cb {
606 union {
607 struct inet_skb_parm h4;
608 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
609 struct inet6_skb_parm h6;
610 #endif
611 } header; /* For incoming frames */
612 __u32 seq; /* Starting sequence number */
613 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
614 __u32 when; /* used to compute rtt's */
615 __u8 flags; /* TCP header flags. */
616
617 /* NOTE: These must match up to the flags byte in a
618 * real TCP header.
619 */
620 #define TCPCB_FLAG_FIN 0x01
621 #define TCPCB_FLAG_SYN 0x02
622 #define TCPCB_FLAG_RST 0x04
623 #define TCPCB_FLAG_PSH 0x08
624 #define TCPCB_FLAG_ACK 0x10
625 #define TCPCB_FLAG_URG 0x20
626 #define TCPCB_FLAG_ECE 0x40
627 #define TCPCB_FLAG_CWR 0x80
628
629 __u8 sacked; /* State flags for SACK/FACK. */
630 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
631 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
632 #define TCPCB_LOST 0x04 /* SKB is lost */
633 #define TCPCB_TAGBITS 0x07 /* All tag bits */
634
635 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
636 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
637
638 __u32 ack_seq; /* Sequence number ACK'd */
639 };
640
641 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
642
643 /* Due to TSO, an SKB can be composed of multiple actual
644 * packets. To keep these tracked properly, we use this.
645 */
646 static inline int tcp_skb_pcount(const struct sk_buff *skb)
647 {
648 return skb_shinfo(skb)->gso_segs;
649 }
650
651 /* This is valid iff tcp_skb_pcount() > 1. */
652 static inline int tcp_skb_mss(const struct sk_buff *skb)
653 {
654 return skb_shinfo(skb)->gso_size;
655 }
656
657 /* Events passed to congestion control interface */
658 enum tcp_ca_event {
659 CA_EVENT_TX_START, /* first transmit when no packets in flight */
660 CA_EVENT_CWND_RESTART, /* congestion window restart */
661 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
662 CA_EVENT_FRTO, /* fast recovery timeout */
663 CA_EVENT_LOSS, /* loss timeout */
664 CA_EVENT_FAST_ACK, /* in sequence ack */
665 CA_EVENT_SLOW_ACK, /* other ack */
666 };
667
668 /*
669 * Interface for adding new TCP congestion control handlers
670 */
671 #define TCP_CA_NAME_MAX 16
672 #define TCP_CA_MAX 128
673 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
674
675 #define TCP_CONG_NON_RESTRICTED 0x1
676 #define TCP_CONG_RTT_STAMP 0x2
677
678 struct tcp_congestion_ops {
679 struct list_head list;
680 unsigned long flags;
681
682 /* initialize private data (optional) */
683 void (*init)(struct sock *sk);
684 /* cleanup private data (optional) */
685 void (*release)(struct sock *sk);
686
687 /* return slow start threshold (required) */
688 u32 (*ssthresh)(struct sock *sk);
689 /* lower bound for congestion window (optional) */
690 u32 (*min_cwnd)(const struct sock *sk);
691 /* do new cwnd calculation (required) */
692 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
693 /* call before changing ca_state (optional) */
694 void (*set_state)(struct sock *sk, u8 new_state);
695 /* call when cwnd event occurs (optional) */
696 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
697 /* new value of cwnd after loss (optional) */
698 u32 (*undo_cwnd)(struct sock *sk);
699 /* hook for packet ack accounting (optional) */
700 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
701 /* get info for inet_diag (optional) */
702 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
703
704 char name[TCP_CA_NAME_MAX];
705 struct module *owner;
706 };
707
708 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
709 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
710
711 extern void tcp_init_congestion_control(struct sock *sk);
712 extern void tcp_cleanup_congestion_control(struct sock *sk);
713 extern int tcp_set_default_congestion_control(const char *name);
714 extern void tcp_get_default_congestion_control(char *name);
715 extern void tcp_get_available_congestion_control(char *buf, size_t len);
716 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
717 extern int tcp_set_allowed_congestion_control(char *allowed);
718 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
719 extern void tcp_slow_start(struct tcp_sock *tp);
720 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
721
722 extern struct tcp_congestion_ops tcp_init_congestion_ops;
723 extern u32 tcp_reno_ssthresh(struct sock *sk);
724 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
725 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
726 extern struct tcp_congestion_ops tcp_reno;
727
728 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
729 {
730 struct inet_connection_sock *icsk = inet_csk(sk);
731
732 if (icsk->icsk_ca_ops->set_state)
733 icsk->icsk_ca_ops->set_state(sk, ca_state);
734 icsk->icsk_ca_state = ca_state;
735 }
736
737 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
738 {
739 const struct inet_connection_sock *icsk = inet_csk(sk);
740
741 if (icsk->icsk_ca_ops->cwnd_event)
742 icsk->icsk_ca_ops->cwnd_event(sk, event);
743 }
744
745 /* These functions determine how the current flow behaves in respect of SACK
746 * handling. SACK is negotiated with the peer, and therefore it can vary
747 * between different flows.
748 *
749 * tcp_is_sack - SACK enabled
750 * tcp_is_reno - No SACK
751 * tcp_is_fack - FACK enabled, implies SACK enabled
752 */
753 static inline int tcp_is_sack(const struct tcp_sock *tp)
754 {
755 return tp->rx_opt.sack_ok;
756 }
757
758 static inline int tcp_is_reno(const struct tcp_sock *tp)
759 {
760 return !tcp_is_sack(tp);
761 }
762
763 static inline int tcp_is_fack(const struct tcp_sock *tp)
764 {
765 return tp->rx_opt.sack_ok & 2;
766 }
767
768 static inline void tcp_enable_fack(struct tcp_sock *tp)
769 {
770 tp->rx_opt.sack_ok |= 2;
771 }
772
773 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
774 {
775 return tp->sacked_out + tp->lost_out;
776 }
777
778 /* This determines how many packets are "in the network" to the best
779 * of our knowledge. In many cases it is conservative, but where
780 * detailed information is available from the receiver (via SACK
781 * blocks etc.) we can make more aggressive calculations.
782 *
783 * Use this for decisions involving congestion control, use just
784 * tp->packets_out to determine if the send queue is empty or not.
785 *
786 * Read this equation as:
787 *
788 * "Packets sent once on transmission queue" MINUS
789 * "Packets left network, but not honestly ACKed yet" PLUS
790 * "Packets fast retransmitted"
791 */
792 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
793 {
794 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
795 }
796
797 #define TCP_INFINITE_SSTHRESH 0x7fffffff
798
799 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
800 {
801 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
802 }
803
804 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
805 * The exception is rate halving phase, when cwnd is decreasing towards
806 * ssthresh.
807 */
808 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
809 {
810 const struct tcp_sock *tp = tcp_sk(sk);
811 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
812 return tp->snd_ssthresh;
813 else
814 return max(tp->snd_ssthresh,
815 ((tp->snd_cwnd >> 1) +
816 (tp->snd_cwnd >> 2)));
817 }
818
819 /* Use define here intentionally to get WARN_ON location shown at the caller */
820 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
821
822 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
823 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
824
825 /* Slow start with delack produces 3 packets of burst, so that
826 * it is safe "de facto". This will be the default - same as
827 * the default reordering threshold - but if reordering increases,
828 * we must be able to allow cwnd to burst at least this much in order
829 * to not pull it back when holes are filled.
830 */
831 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
832 {
833 return tp->reordering;
834 }
835
836 /* Returns end sequence number of the receiver's advertised window */
837 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
838 {
839 return tp->snd_una + tp->snd_wnd;
840 }
841 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
842
843 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
844 const struct sk_buff *skb)
845 {
846 if (skb->len < mss)
847 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
848 }
849
850 static inline void tcp_check_probe_timer(struct sock *sk)
851 {
852 struct tcp_sock *tp = tcp_sk(sk);
853 const struct inet_connection_sock *icsk = inet_csk(sk);
854
855 if (!tp->packets_out && !icsk->icsk_pending)
856 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
857 icsk->icsk_rto, TCP_RTO_MAX);
858 }
859
860 static inline void tcp_push_pending_frames(struct sock *sk)
861 {
862 struct tcp_sock *tp = tcp_sk(sk);
863
864 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
865 }
866
867 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
868 {
869 tp->snd_wl1 = seq;
870 }
871
872 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
873 {
874 tp->snd_wl1 = seq;
875 }
876
877 /*
878 * Calculate(/check) TCP checksum
879 */
880 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
881 __be32 daddr, __wsum base)
882 {
883 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
884 }
885
886 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
887 {
888 return __skb_checksum_complete(skb);
889 }
890
891 static inline int tcp_checksum_complete(struct sk_buff *skb)
892 {
893 return !skb_csum_unnecessary(skb) &&
894 __tcp_checksum_complete(skb);
895 }
896
897 /* Prequeue for VJ style copy to user, combined with checksumming. */
898
899 static inline void tcp_prequeue_init(struct tcp_sock *tp)
900 {
901 tp->ucopy.task = NULL;
902 tp->ucopy.len = 0;
903 tp->ucopy.memory = 0;
904 skb_queue_head_init(&tp->ucopy.prequeue);
905 #ifdef CONFIG_NET_DMA
906 tp->ucopy.dma_chan = NULL;
907 tp->ucopy.wakeup = 0;
908 tp->ucopy.pinned_list = NULL;
909 tp->ucopy.dma_cookie = 0;
910 #endif
911 }
912
913 /* Packet is added to VJ-style prequeue for processing in process
914 * context, if a reader task is waiting. Apparently, this exciting
915 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
916 * failed somewhere. Latency? Burstiness? Well, at least now we will
917 * see, why it failed. 8)8) --ANK
918 *
919 * NOTE: is this not too big to inline?
920 */
921 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
922 {
923 struct tcp_sock *tp = tcp_sk(sk);
924
925 if (sysctl_tcp_low_latency || !tp->ucopy.task)
926 return 0;
927
928 __skb_queue_tail(&tp->ucopy.prequeue, skb);
929 tp->ucopy.memory += skb->truesize;
930 if (tp->ucopy.memory > sk->sk_rcvbuf) {
931 struct sk_buff *skb1;
932
933 BUG_ON(sock_owned_by_user(sk));
934
935 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
936 sk_backlog_rcv(sk, skb1);
937 NET_INC_STATS_BH(sock_net(sk),
938 LINUX_MIB_TCPPREQUEUEDROPPED);
939 }
940
941 tp->ucopy.memory = 0;
942 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
943 wake_up_interruptible_poll(sk->sk_sleep,
944 POLLIN | POLLRDNORM | POLLRDBAND);
945 if (!inet_csk_ack_scheduled(sk))
946 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
947 (3 * tcp_rto_min(sk)) / 4,
948 TCP_RTO_MAX);
949 }
950 return 1;
951 }
952
953
954 #undef STATE_TRACE
955
956 #ifdef STATE_TRACE
957 static const char *statename[]={
958 "Unused","Established","Syn Sent","Syn Recv",
959 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
960 "Close Wait","Last ACK","Listen","Closing"
961 };
962 #endif
963 extern void tcp_set_state(struct sock *sk, int state);
964
965 extern void tcp_done(struct sock *sk);
966
967 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
968 {
969 rx_opt->dsack = 0;
970 rx_opt->num_sacks = 0;
971 }
972
973 /* Determine a window scaling and initial window to offer. */
974 extern void tcp_select_initial_window(int __space, __u32 mss,
975 __u32 *rcv_wnd, __u32 *window_clamp,
976 int wscale_ok, __u8 *rcv_wscale);
977
978 static inline int tcp_win_from_space(int space)
979 {
980 return sysctl_tcp_adv_win_scale<=0 ?
981 (space>>(-sysctl_tcp_adv_win_scale)) :
982 space - (space>>sysctl_tcp_adv_win_scale);
983 }
984
985 /* Note: caller must be prepared to deal with negative returns */
986 static inline int tcp_space(const struct sock *sk)
987 {
988 return tcp_win_from_space(sk->sk_rcvbuf -
989 atomic_read(&sk->sk_rmem_alloc));
990 }
991
992 static inline int tcp_full_space(const struct sock *sk)
993 {
994 return tcp_win_from_space(sk->sk_rcvbuf);
995 }
996
997 static inline void tcp_openreq_init(struct request_sock *req,
998 struct tcp_options_received *rx_opt,
999 struct sk_buff *skb)
1000 {
1001 struct inet_request_sock *ireq = inet_rsk(req);
1002
1003 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1004 req->cookie_ts = 0;
1005 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1006 req->mss = rx_opt->mss_clamp;
1007 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1008 ireq->tstamp_ok = rx_opt->tstamp_ok;
1009 ireq->sack_ok = rx_opt->sack_ok;
1010 ireq->snd_wscale = rx_opt->snd_wscale;
1011 ireq->wscale_ok = rx_opt->wscale_ok;
1012 ireq->acked = 0;
1013 ireq->ecn_ok = 0;
1014 ireq->rmt_port = tcp_hdr(skb)->source;
1015 ireq->loc_port = tcp_hdr(skb)->dest;
1016 }
1017
1018 extern void tcp_enter_memory_pressure(struct sock *sk);
1019
1020 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1021 {
1022 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1023 }
1024
1025 static inline int keepalive_time_when(const struct tcp_sock *tp)
1026 {
1027 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1028 }
1029
1030 static inline int keepalive_probes(const struct tcp_sock *tp)
1031 {
1032 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1033 }
1034
1035 static inline int tcp_fin_time(const struct sock *sk)
1036 {
1037 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1038 const int rto = inet_csk(sk)->icsk_rto;
1039
1040 if (fin_timeout < (rto << 2) - (rto >> 1))
1041 fin_timeout = (rto << 2) - (rto >> 1);
1042
1043 return fin_timeout;
1044 }
1045
1046 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1047 int paws_win)
1048 {
1049 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1050 return 1;
1051 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1052 return 1;
1053
1054 return 0;
1055 }
1056
1057 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1058 int rst)
1059 {
1060 if (tcp_paws_check(rx_opt, 0))
1061 return 0;
1062
1063 /* RST segments are not recommended to carry timestamp,
1064 and, if they do, it is recommended to ignore PAWS because
1065 "their cleanup function should take precedence over timestamps."
1066 Certainly, it is mistake. It is necessary to understand the reasons
1067 of this constraint to relax it: if peer reboots, clock may go
1068 out-of-sync and half-open connections will not be reset.
1069 Actually, the problem would be not existing if all
1070 the implementations followed draft about maintaining clock
1071 via reboots. Linux-2.2 DOES NOT!
1072
1073 However, we can relax time bounds for RST segments to MSL.
1074 */
1075 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1076 return 0;
1077 return 1;
1078 }
1079
1080 #define TCP_CHECK_TIMER(sk) do { } while (0)
1081
1082 static inline void tcp_mib_init(struct net *net)
1083 {
1084 /* See RFC 2012 */
1085 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1086 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1087 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1088 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1089 }
1090
1091 /* from STCP */
1092 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1093 {
1094 tp->lost_skb_hint = NULL;
1095 tp->scoreboard_skb_hint = NULL;
1096 }
1097
1098 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1099 {
1100 tcp_clear_retrans_hints_partial(tp);
1101 tp->retransmit_skb_hint = NULL;
1102 }
1103
1104 /* MD5 Signature */
1105 struct crypto_hash;
1106
1107 /* - key database */
1108 struct tcp_md5sig_key {
1109 u8 *key;
1110 u8 keylen;
1111 };
1112
1113 struct tcp4_md5sig_key {
1114 struct tcp_md5sig_key base;
1115 __be32 addr;
1116 };
1117
1118 struct tcp6_md5sig_key {
1119 struct tcp_md5sig_key base;
1120 #if 0
1121 u32 scope_id; /* XXX */
1122 #endif
1123 struct in6_addr addr;
1124 };
1125
1126 /* - sock block */
1127 struct tcp_md5sig_info {
1128 struct tcp4_md5sig_key *keys4;
1129 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1130 struct tcp6_md5sig_key *keys6;
1131 u32 entries6;
1132 u32 alloced6;
1133 #endif
1134 u32 entries4;
1135 u32 alloced4;
1136 };
1137
1138 /* - pseudo header */
1139 struct tcp4_pseudohdr {
1140 __be32 saddr;
1141 __be32 daddr;
1142 __u8 pad;
1143 __u8 protocol;
1144 __be16 len;
1145 };
1146
1147 struct tcp6_pseudohdr {
1148 struct in6_addr saddr;
1149 struct in6_addr daddr;
1150 __be32 len;
1151 __be32 protocol; /* including padding */
1152 };
1153
1154 union tcp_md5sum_block {
1155 struct tcp4_pseudohdr ip4;
1156 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1157 struct tcp6_pseudohdr ip6;
1158 #endif
1159 };
1160
1161 /* - pool: digest algorithm, hash description and scratch buffer */
1162 struct tcp_md5sig_pool {
1163 struct hash_desc md5_desc;
1164 union tcp_md5sum_block md5_blk;
1165 };
1166
1167 #define TCP_MD5SIG_MAXKEYS (~(u32)0) /* really?! */
1168
1169 /* - functions */
1170 extern int tcp_v4_md5_hash_skb(char *md5_hash,
1171 struct tcp_md5sig_key *key,
1172 struct sock *sk,
1173 struct request_sock *req,
1174 struct sk_buff *skb);
1175
1176 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1177 struct sock *addr_sk);
1178
1179 extern int tcp_v4_md5_do_add(struct sock *sk,
1180 __be32 addr,
1181 u8 *newkey,
1182 u8 newkeylen);
1183
1184 extern int tcp_v4_md5_do_del(struct sock *sk,
1185 __be32 addr);
1186
1187 #ifdef CONFIG_TCP_MD5SIG
1188 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1189 &(struct tcp_md5sig_key) { \
1190 .key = (twsk)->tw_md5_key, \
1191 .keylen = (twsk)->tw_md5_keylen, \
1192 } : NULL)
1193 #else
1194 #define tcp_twsk_md5_key(twsk) NULL
1195 #endif
1196
1197 extern struct tcp_md5sig_pool **tcp_alloc_md5sig_pool(struct sock *);
1198 extern void tcp_free_md5sig_pool(void);
1199
1200 extern struct tcp_md5sig_pool *__tcp_get_md5sig_pool(int cpu);
1201 extern void __tcp_put_md5sig_pool(void);
1202 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1203 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1204 unsigned header_len);
1205 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1206 struct tcp_md5sig_key *key);
1207
1208 static inline
1209 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void)
1210 {
1211 int cpu = get_cpu();
1212 struct tcp_md5sig_pool *ret = __tcp_get_md5sig_pool(cpu);
1213 if (!ret)
1214 put_cpu();
1215 return ret;
1216 }
1217
1218 static inline void tcp_put_md5sig_pool(void)
1219 {
1220 __tcp_put_md5sig_pool();
1221 put_cpu();
1222 }
1223
1224 /* write queue abstraction */
1225 static inline void tcp_write_queue_purge(struct sock *sk)
1226 {
1227 struct sk_buff *skb;
1228
1229 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1230 sk_wmem_free_skb(sk, skb);
1231 sk_mem_reclaim(sk);
1232 tcp_clear_all_retrans_hints(tcp_sk(sk));
1233 }
1234
1235 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1236 {
1237 return skb_peek(&sk->sk_write_queue);
1238 }
1239
1240 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1241 {
1242 return skb_peek_tail(&sk->sk_write_queue);
1243 }
1244
1245 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1246 {
1247 return skb_queue_next(&sk->sk_write_queue, skb);
1248 }
1249
1250 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb)
1251 {
1252 return skb_queue_prev(&sk->sk_write_queue, skb);
1253 }
1254
1255 #define tcp_for_write_queue(skb, sk) \
1256 skb_queue_walk(&(sk)->sk_write_queue, skb)
1257
1258 #define tcp_for_write_queue_from(skb, sk) \
1259 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1260
1261 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1262 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1263
1264 /* This function calculates a "timeout" which is equivalent to the timeout of a
1265 * TCP connection after "boundary" unsuccessful, exponentially backed-off
1266 * retransmissions with an initial RTO of TCP_RTO_MIN.
1267 */
1268 static inline bool retransmits_timed_out(const struct sock *sk,
1269 unsigned int boundary)
1270 {
1271 unsigned int timeout, linear_backoff_thresh;
1272
1273 if (!inet_csk(sk)->icsk_retransmits)
1274 return false;
1275
1276 linear_backoff_thresh = ilog2(TCP_RTO_MAX/TCP_RTO_MIN);
1277
1278 if (boundary <= linear_backoff_thresh)
1279 timeout = ((2 << boundary) - 1) * TCP_RTO_MIN;
1280 else
1281 timeout = ((2 << linear_backoff_thresh) - 1) * TCP_RTO_MIN +
1282 (boundary - linear_backoff_thresh) * TCP_RTO_MAX;
1283
1284 return (tcp_time_stamp - tcp_sk(sk)->retrans_stamp) >= timeout;
1285 }
1286
1287 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1288 {
1289 return sk->sk_send_head;
1290 }
1291
1292 static inline bool tcp_skb_is_last(const struct sock *sk,
1293 const struct sk_buff *skb)
1294 {
1295 return skb_queue_is_last(&sk->sk_write_queue, skb);
1296 }
1297
1298 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1299 {
1300 if (tcp_skb_is_last(sk, skb))
1301 sk->sk_send_head = NULL;
1302 else
1303 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1304 }
1305
1306 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1307 {
1308 if (sk->sk_send_head == skb_unlinked)
1309 sk->sk_send_head = NULL;
1310 }
1311
1312 static inline void tcp_init_send_head(struct sock *sk)
1313 {
1314 sk->sk_send_head = NULL;
1315 }
1316
1317 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1318 {
1319 __skb_queue_tail(&sk->sk_write_queue, skb);
1320 }
1321
1322 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1323 {
1324 __tcp_add_write_queue_tail(sk, skb);
1325
1326 /* Queue it, remembering where we must start sending. */
1327 if (sk->sk_send_head == NULL) {
1328 sk->sk_send_head = skb;
1329
1330 if (tcp_sk(sk)->highest_sack == NULL)
1331 tcp_sk(sk)->highest_sack = skb;
1332 }
1333 }
1334
1335 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1336 {
1337 __skb_queue_head(&sk->sk_write_queue, skb);
1338 }
1339
1340 /* Insert buff after skb on the write queue of sk. */
1341 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1342 struct sk_buff *buff,
1343 struct sock *sk)
1344 {
1345 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1346 }
1347
1348 /* Insert new before skb on the write queue of sk. */
1349 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1350 struct sk_buff *skb,
1351 struct sock *sk)
1352 {
1353 __skb_queue_before(&sk->sk_write_queue, skb, new);
1354
1355 if (sk->sk_send_head == skb)
1356 sk->sk_send_head = new;
1357 }
1358
1359 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1360 {
1361 __skb_unlink(skb, &sk->sk_write_queue);
1362 }
1363
1364 static inline int tcp_write_queue_empty(struct sock *sk)
1365 {
1366 return skb_queue_empty(&sk->sk_write_queue);
1367 }
1368
1369 /* Start sequence of the highest skb with SACKed bit, valid only if
1370 * sacked > 0 or when the caller has ensured validity by itself.
1371 */
1372 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1373 {
1374 if (!tp->sacked_out)
1375 return tp->snd_una;
1376
1377 if (tp->highest_sack == NULL)
1378 return tp->snd_nxt;
1379
1380 return TCP_SKB_CB(tp->highest_sack)->seq;
1381 }
1382
1383 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1384 {
1385 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1386 tcp_write_queue_next(sk, skb);
1387 }
1388
1389 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1390 {
1391 return tcp_sk(sk)->highest_sack;
1392 }
1393
1394 static inline void tcp_highest_sack_reset(struct sock *sk)
1395 {
1396 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1397 }
1398
1399 /* Called when old skb is about to be deleted (to be combined with new skb) */
1400 static inline void tcp_highest_sack_combine(struct sock *sk,
1401 struct sk_buff *old,
1402 struct sk_buff *new)
1403 {
1404 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1405 tcp_sk(sk)->highest_sack = new;
1406 }
1407
1408 /* /proc */
1409 enum tcp_seq_states {
1410 TCP_SEQ_STATE_LISTENING,
1411 TCP_SEQ_STATE_OPENREQ,
1412 TCP_SEQ_STATE_ESTABLISHED,
1413 TCP_SEQ_STATE_TIME_WAIT,
1414 };
1415
1416 struct tcp_seq_afinfo {
1417 char *name;
1418 sa_family_t family;
1419 struct file_operations seq_fops;
1420 struct seq_operations seq_ops;
1421 };
1422
1423 struct tcp_iter_state {
1424 struct seq_net_private p;
1425 sa_family_t family;
1426 enum tcp_seq_states state;
1427 struct sock *syn_wait_sk;
1428 int bucket, sbucket, num, uid;
1429 };
1430
1431 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1432 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1433
1434 extern struct request_sock_ops tcp_request_sock_ops;
1435 extern struct request_sock_ops tcp6_request_sock_ops;
1436
1437 extern void tcp_v4_destroy_sock(struct sock *sk);
1438
1439 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1440 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, int features);
1441 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1442 struct sk_buff *skb);
1443 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1444 struct sk_buff *skb);
1445 extern int tcp_gro_complete(struct sk_buff *skb);
1446 extern int tcp4_gro_complete(struct sk_buff *skb);
1447
1448 #ifdef CONFIG_PROC_FS
1449 extern int tcp4_proc_init(void);
1450 extern void tcp4_proc_exit(void);
1451 #endif
1452
1453 /* TCP af-specific functions */
1454 struct tcp_sock_af_ops {
1455 #ifdef CONFIG_TCP_MD5SIG
1456 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1457 struct sock *addr_sk);
1458 int (*calc_md5_hash) (char *location,
1459 struct tcp_md5sig_key *md5,
1460 struct sock *sk,
1461 struct request_sock *req,
1462 struct sk_buff *skb);
1463 int (*md5_add) (struct sock *sk,
1464 struct sock *addr_sk,
1465 u8 *newkey,
1466 u8 len);
1467 int (*md5_parse) (struct sock *sk,
1468 char __user *optval,
1469 int optlen);
1470 #endif
1471 };
1472
1473 struct tcp_request_sock_ops {
1474 #ifdef CONFIG_TCP_MD5SIG
1475 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1476 struct request_sock *req);
1477 int (*calc_md5_hash) (char *location,
1478 struct tcp_md5sig_key *md5,
1479 struct sock *sk,
1480 struct request_sock *req,
1481 struct sk_buff *skb);
1482 #endif
1483 };
1484
1485 /* Using SHA1 for now, define some constants.
1486 */
1487 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1488 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1489 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1490
1491 extern int tcp_cookie_generator(u32 *bakery);
1492
1493 /**
1494 * struct tcp_cookie_values - each socket needs extra space for the
1495 * cookies, together with (optional) space for any SYN data.
1496 *
1497 * A tcp_sock contains a pointer to the current value, and this is
1498 * cloned to the tcp_timewait_sock.
1499 *
1500 * @cookie_pair: variable data from the option exchange.
1501 *
1502 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1503 * indicates default (sysctl_tcp_cookie_size).
1504 * After cookie sent, remembers size of cookie.
1505 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1506 *
1507 * @s_data_desired: user specified tcpct_s_data_desired. When the
1508 * constant payload is specified (@s_data_constant),
1509 * holds its length instead.
1510 * Range 0 to TCP_MSS_DESIRED.
1511 *
1512 * @s_data_payload: constant data that is to be included in the
1513 * payload of SYN or SYNACK segments when the
1514 * cookie option is present.
1515 */
1516 struct tcp_cookie_values {
1517 struct kref kref;
1518 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE];
1519 u8 cookie_pair_size;
1520 u8 cookie_desired;
1521 u16 s_data_desired:11,
1522 s_data_constant:1,
1523 s_data_in:1,
1524 s_data_out:1,
1525 s_data_unused:2;
1526 u8 s_data_payload[0];
1527 };
1528
1529 static inline void tcp_cookie_values_release(struct kref *kref)
1530 {
1531 kfree(container_of(kref, struct tcp_cookie_values, kref));
1532 }
1533
1534 /* The length of constant payload data. Note that s_data_desired is
1535 * overloaded, depending on s_data_constant: either the length of constant
1536 * data (returned here) or the limit on variable data.
1537 */
1538 static inline int tcp_s_data_size(const struct tcp_sock *tp)
1539 {
1540 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1541 ? tp->cookie_values->s_data_desired
1542 : 0;
1543 }
1544
1545 /**
1546 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1547 *
1548 * As tcp_request_sock has already been extended in other places, the
1549 * only remaining method is to pass stack values along as function
1550 * parameters. These parameters are not needed after sending SYNACK.
1551 *
1552 * @cookie_bakery: cryptographic secret and message workspace.
1553 *
1554 * @cookie_plus: bytes in authenticator/cookie option, copied from
1555 * struct tcp_options_received (above).
1556 */
1557 struct tcp_extend_values {
1558 struct request_values rv;
1559 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS];
1560 u8 cookie_plus:6,
1561 cookie_out_never:1,
1562 cookie_in_always:1;
1563 };
1564
1565 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1566 {
1567 return (struct tcp_extend_values *)rvp;
1568 }
1569
1570 extern void tcp_v4_init(void);
1571 extern void tcp_init(void);
1572
1573 #endif /* _TCP_H */
Linus' kernel tree