?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
writeback: dirty rate control
[linux-2.6.git] / include / net / tcp.h
blobacc620a4a45f318f745c7164363c6fdf787df144
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 /* Offer an initial receive window of 10 mss. */
64 #define TCP_DEFAULT_INIT_RCVWND 10
65
66 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
67 #define TCP_MIN_MSS 88U
68
69 /* The least MTU to use for probing */
70 #define TCP_BASE_MSS 512
71
72 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
73 #define TCP_FASTRETRANS_THRESH 3
74
75 /* Maximal reordering. */
76 #define TCP_MAX_REORDERING 127
77
78 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
79 #define TCP_MAX_QUICKACKS 16U
80
81 /* urg_data states */
82 #define TCP_URG_VALID 0x0100
83 #define TCP_URG_NOTYET 0x0200
84 #define TCP_URG_READ 0x0400
85
86 #define TCP_RETR1 3 /*
87 * This is how many retries it does before it
88 * tries to figure out if the gateway is
89 * down. Minimal RFC value is 3; it corresponds
90 * to ~3sec-8min depending on RTO.
91 */
92
93 #define TCP_RETR2 15 /*
94 * This should take at least
95 * 90 minutes to time out.
96 * RFC1122 says that the limit is 100 sec.
97 * 15 is ~13-30min depending on RTO.
98 */
99
100 #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a
101 * connection: ~180sec is RFC minimum */
102
103 #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a
104 * connection: ~180sec is RFC minimum */
105
106 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
107 * state, about 60 seconds */
108 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
109 /* BSD style FIN_WAIT2 deadlock breaker.
110 * It used to be 3min, new value is 60sec,
111 * to combine FIN-WAIT-2 timeout with
112 * TIME-WAIT timer.
113 */
114
115 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
116 #if HZ >= 100
117 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
118 #define TCP_ATO_MIN ((unsigned)(HZ/25))
119 #else
120 #define TCP_DELACK_MIN 4U
121 #define TCP_ATO_MIN 4U
122 #endif
123 #define TCP_RTO_MAX ((unsigned)(120*HZ))
124 #define TCP_RTO_MIN ((unsigned)(HZ/5))
125 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC2988bis initial RTO value */
126 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
127 * used as a fallback RTO for the
128 * initial data transmission if no
129 * valid RTT sample has been acquired,
130 * most likely due to retrans in 3WHS.
131 */
132
133 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
134 * for local resources.
135 */
136
137 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
138 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
139 #define TCP_KEEPALIVE_INTVL (75*HZ)
140
141 #define MAX_TCP_KEEPIDLE 32767
142 #define MAX_TCP_KEEPINTVL 32767
143 #define MAX_TCP_KEEPCNT 127
144 #define MAX_TCP_SYNCNT 127
145
146 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
147
148 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
149 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
150 * after this time. It should be equal
151 * (or greater than) TCP_TIMEWAIT_LEN
152 * to provide reliability equal to one
153 * provided by timewait state.
154 */
155 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
156 * timestamps. It must be less than
157 * minimal timewait lifetime.
158 */
159 /*
160 * TCP option
161 */
162
163 #define TCPOPT_NOP 1 /* Padding */
164 #define TCPOPT_EOL 0 /* End of options */
165 #define TCPOPT_MSS 2 /* Segment size negotiating */
166 #define TCPOPT_WINDOW 3 /* Window scaling */
167 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
168 #define TCPOPT_SACK 5 /* SACK Block */
169 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
170 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
171 #define TCPOPT_COOKIE 253 /* Cookie extension (experimental) */
172
173 /*
174 * TCP option lengths
175 */
176
177 #define TCPOLEN_MSS 4
178 #define TCPOLEN_WINDOW 3
179 #define TCPOLEN_SACK_PERM 2
180 #define TCPOLEN_TIMESTAMP 10
181 #define TCPOLEN_MD5SIG 18
182 #define TCPOLEN_COOKIE_BASE 2 /* Cookie-less header extension */
183 #define TCPOLEN_COOKIE_PAIR 3 /* Cookie pair header extension */
184 #define TCPOLEN_COOKIE_MIN (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MIN)
185 #define TCPOLEN_COOKIE_MAX (TCPOLEN_COOKIE_BASE+TCP_COOKIE_MAX)
186
187 /* But this is what stacks really send out. */
188 #define TCPOLEN_TSTAMP_ALIGNED 12
189 #define TCPOLEN_WSCALE_ALIGNED 4
190 #define TCPOLEN_SACKPERM_ALIGNED 4
191 #define TCPOLEN_SACK_BASE 2
192 #define TCPOLEN_SACK_BASE_ALIGNED 4
193 #define TCPOLEN_SACK_PERBLOCK 8
194 #define TCPOLEN_MD5SIG_ALIGNED 20
195 #define TCPOLEN_MSS_ALIGNED 4
196
197 /* Flags in tp->nonagle */
198 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
199 #define TCP_NAGLE_CORK 2 /* Socket is corked */
200 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
201
202 /* TCP thin-stream limits */
203 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
204
205 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
206 #define TCP_INIT_CWND 10
207
208 extern struct inet_timewait_death_row tcp_death_row;
209
210 /* sysctl variables for tcp */
211 extern int sysctl_tcp_timestamps;
212 extern int sysctl_tcp_window_scaling;
213 extern int sysctl_tcp_sack;
214 extern int sysctl_tcp_fin_timeout;
215 extern int sysctl_tcp_keepalive_time;
216 extern int sysctl_tcp_keepalive_probes;
217 extern int sysctl_tcp_keepalive_intvl;
218 extern int sysctl_tcp_syn_retries;
219 extern int sysctl_tcp_synack_retries;
220 extern int sysctl_tcp_retries1;
221 extern int sysctl_tcp_retries2;
222 extern int sysctl_tcp_orphan_retries;
223 extern int sysctl_tcp_syncookies;
224 extern int sysctl_tcp_retrans_collapse;
225 extern int sysctl_tcp_stdurg;
226 extern int sysctl_tcp_rfc1337;
227 extern int sysctl_tcp_abort_on_overflow;
228 extern int sysctl_tcp_max_orphans;
229 extern int sysctl_tcp_fack;
230 extern int sysctl_tcp_reordering;
231 extern int sysctl_tcp_ecn;
232 extern int sysctl_tcp_dsack;
233 extern long sysctl_tcp_mem[3];
234 extern int sysctl_tcp_wmem[3];
235 extern int sysctl_tcp_rmem[3];
236 extern int sysctl_tcp_app_win;
237 extern int sysctl_tcp_adv_win_scale;
238 extern int sysctl_tcp_tw_reuse;
239 extern int sysctl_tcp_frto;
240 extern int sysctl_tcp_frto_response;
241 extern int sysctl_tcp_low_latency;
242 extern int sysctl_tcp_dma_copybreak;
243 extern int sysctl_tcp_nometrics_save;
244 extern int sysctl_tcp_moderate_rcvbuf;
245 extern int sysctl_tcp_tso_win_divisor;
246 extern int sysctl_tcp_abc;
247 extern int sysctl_tcp_mtu_probing;
248 extern int sysctl_tcp_base_mss;
249 extern int sysctl_tcp_workaround_signed_windows;
250 extern int sysctl_tcp_slow_start_after_idle;
251 extern int sysctl_tcp_max_ssthresh;
252 extern int sysctl_tcp_cookie_size;
253 extern int sysctl_tcp_thin_linear_timeouts;
254 extern int sysctl_tcp_thin_dupack;
255
256 extern atomic_long_t tcp_memory_allocated;
257 extern struct percpu_counter tcp_sockets_allocated;
258 extern int tcp_memory_pressure;
259
260 /*
261 * The next routines deal with comparing 32 bit unsigned ints
262 * and worry about wraparound (automatic with unsigned arithmetic).
263 */
264
265 static inline int before(__u32 seq1, __u32 seq2)
266 {
267 return (__s32)(seq1-seq2) < 0;
268 }
269 #define after(seq2, seq1) before(seq1, seq2)
270
271 /* is s2<=s1<=s3 ? */
272 static inline int between(__u32 seq1, __u32 seq2, __u32 seq3)
273 {
274 return seq3 - seq2 >= seq1 - seq2;
275 }
276
277 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
278 {
279 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
280 int orphans = percpu_counter_read_positive(ocp);
281
282 if (orphans << shift > sysctl_tcp_max_orphans) {
283 orphans = percpu_counter_sum_positive(ocp);
284 if (orphans << shift > sysctl_tcp_max_orphans)
285 return true;
286 }
287
288 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
289 atomic_long_read(&tcp_memory_allocated) > sysctl_tcp_mem[2])
290 return true;
291 return false;
292 }
293
294 /* syncookies: remember time of last synqueue overflow */
295 static inline void tcp_synq_overflow(struct sock *sk)
296 {
297 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
298 }
299
300 /* syncookies: no recent synqueue overflow on this listening socket? */
301 static inline int tcp_synq_no_recent_overflow(const struct sock *sk)
302 {
303 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
304 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
305 }
306
307 extern struct proto tcp_prot;
308
309 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
310 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
311 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
312 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
313 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
314
315 extern void tcp_v4_err(struct sk_buff *skb, u32);
316
317 extern void tcp_shutdown (struct sock *sk, int how);
318
319 extern int tcp_v4_rcv(struct sk_buff *skb);
320
321 extern struct inet_peer *tcp_v4_get_peer(struct sock *sk, bool *release_it);
322 extern void *tcp_v4_tw_get_peer(struct sock *sk);
323 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
324 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
325 size_t size);
326 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
327 size_t size, int flags);
328 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
329 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
330 struct tcphdr *th, unsigned len);
331 extern int tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
332 struct tcphdr *th, unsigned len);
333 extern void tcp_rcv_space_adjust(struct sock *sk);
334 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
335 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
336 extern void tcp_twsk_destructor(struct sock *sk);
337 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
338 struct pipe_inode_info *pipe, size_t len,
339 unsigned int flags);
340
341 static inline void tcp_dec_quickack_mode(struct sock *sk,
342 const unsigned int pkts)
343 {
344 struct inet_connection_sock *icsk = inet_csk(sk);
345
346 if (icsk->icsk_ack.quick) {
347 if (pkts >= icsk->icsk_ack.quick) {
348 icsk->icsk_ack.quick = 0;
349 /* Leaving quickack mode we deflate ATO. */
350 icsk->icsk_ack.ato = TCP_ATO_MIN;
351 } else
352 icsk->icsk_ack.quick -= pkts;
353 }
354 }
355
356 #define TCP_ECN_OK 1
357 #define TCP_ECN_QUEUE_CWR 2
358 #define TCP_ECN_DEMAND_CWR 4
359
360 static __inline__ void
361 TCP_ECN_create_request(struct request_sock *req, struct tcphdr *th)
362 {
363 if (sysctl_tcp_ecn && th->ece && th->cwr)
364 inet_rsk(req)->ecn_ok = 1;
365 }
366
367 enum tcp_tw_status {
368 TCP_TW_SUCCESS = 0,
369 TCP_TW_RST = 1,
370 TCP_TW_ACK = 2,
371 TCP_TW_SYN = 3
372 };
373
374
375 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
376 struct sk_buff *skb,
377 const struct tcphdr *th);
378 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
379 struct request_sock *req,
380 struct request_sock **prev);
381 extern int tcp_child_process(struct sock *parent, struct sock *child,
382 struct sk_buff *skb);
383 extern int tcp_use_frto(struct sock *sk);
384 extern void tcp_enter_frto(struct sock *sk);
385 extern void tcp_enter_loss(struct sock *sk, int how);
386 extern void tcp_clear_retrans(struct tcp_sock *tp);
387 extern void tcp_update_metrics(struct sock *sk);
388 extern void tcp_close(struct sock *sk, long timeout);
389 extern unsigned int tcp_poll(struct file * file, struct socket *sock,
390 struct poll_table_struct *wait);
391 extern int tcp_getsockopt(struct sock *sk, int level, int optname,
392 char __user *optval, int __user *optlen);
393 extern int tcp_setsockopt(struct sock *sk, int level, int optname,
394 char __user *optval, unsigned int optlen);
395 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
396 char __user *optval, int __user *optlen);
397 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
398 char __user *optval, unsigned int optlen);
399 extern void tcp_set_keepalive(struct sock *sk, int val);
400 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
401 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
402 size_t len, int nonblock, int flags, int *addr_len);
403 extern void tcp_parse_options(struct sk_buff *skb,
404 struct tcp_options_received *opt_rx, u8 **hvpp,
405 int estab);
406 extern u8 *tcp_parse_md5sig_option(struct tcphdr *th);
407
408 /*
409 * TCP v4 functions exported for the inet6 API
410 */
411
412 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
413 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
414 extern struct sock * tcp_create_openreq_child(struct sock *sk,
415 struct request_sock *req,
416 struct sk_buff *skb);
417 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
418 struct request_sock *req,
419 struct dst_entry *dst);
420 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
421 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
422 int addr_len);
423 extern int tcp_connect(struct sock *sk);
424 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
425 struct request_sock *req,
426 struct request_values *rvp);
427 extern int tcp_disconnect(struct sock *sk, int flags);
428
429
430 /* From syncookies.c */
431 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
432 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
433 struct ip_options *opt);
434 #ifdef CONFIG_SYN_COOKIES
435 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
436 __u16 *mss);
437 #else
438 static inline __u32 cookie_v4_init_sequence(struct sock *sk,
439 struct sk_buff *skb,
440 __u16 *mss)
441 {
442 return 0;
443 }
444 #endif
445
446 extern __u32 cookie_init_timestamp(struct request_sock *req);
447 extern bool cookie_check_timestamp(struct tcp_options_received *opt, bool *);
448
449 /* From net/ipv6/syncookies.c */
450 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
451 #ifdef CONFIG_SYN_COOKIES
452 extern __u32 cookie_v6_init_sequence(struct sock *sk, struct sk_buff *skb,
453 __u16 *mss);
454 #else
455 static inline __u32 cookie_v6_init_sequence(struct sock *sk,
456 struct sk_buff *skb,
457 __u16 *mss)
458 {
459 return 0;
460 }
461 #endif
462 /* tcp_output.c */
463
464 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
465 int nonagle);
466 extern int tcp_may_send_now(struct sock *sk);
467 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
468 extern void tcp_retransmit_timer(struct sock *sk);
469 extern void tcp_xmit_retransmit_queue(struct sock *);
470 extern void tcp_simple_retransmit(struct sock *);
471 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
472 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
473
474 extern void tcp_send_probe0(struct sock *);
475 extern void tcp_send_partial(struct sock *);
476 extern int tcp_write_wakeup(struct sock *);
477 extern void tcp_send_fin(struct sock *sk);
478 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
479 extern int tcp_send_synack(struct sock *);
480 extern int tcp_syn_flood_action(struct sock *sk,
481 const struct sk_buff *skb,
482 const char *proto);
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 int cutoff;
504
505 /* When peer uses tiny windows, there is no use in packetizing
506 * to sub-MSS pieces for the sake of SWS or making sure there
507 * are enough packets in the pipe for fast recovery.
508 *
509 * On the other hand, for extremely large MSS devices, handling
510 * smaller than MSS windows in this way does make sense.
511 */
512 if (tp->max_window >= 512)
513 cutoff = (tp->max_window >> 1);
514 else
515 cutoff = tp->max_window;
516
517 if (cutoff && pktsize > cutoff)
518 return max_t(int, cutoff, 68U - tp->tcp_header_len);
519 else
520 return pktsize;
521 }
522
523 /* tcp.c */
524 extern void tcp_get_info(struct sock *, struct tcp_info *);
525
526 /* Read 'sendfile()'-style from a TCP socket */
527 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
528 unsigned int, size_t);
529 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
530 sk_read_actor_t recv_actor);
531
532 extern void tcp_initialize_rcv_mss(struct sock *sk);
533
534 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
535 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
536 extern void tcp_mtup_init(struct sock *sk);
537 extern void tcp_valid_rtt_meas(struct sock *sk, u32 seq_rtt);
538
539 static inline void tcp_bound_rto(const struct sock *sk)
540 {
541 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
542 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
543 }
544
545 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
546 {
547 return (tp->srtt >> 3) + tp->rttvar;
548 }
549
550 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
551 {
552 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
553 ntohl(TCP_FLAG_ACK) |
554 snd_wnd);
555 }
556
557 static inline void tcp_fast_path_on(struct tcp_sock *tp)
558 {
559 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
560 }
561
562 static inline void tcp_fast_path_check(struct sock *sk)
563 {
564 struct tcp_sock *tp = tcp_sk(sk);
565
566 if (skb_queue_empty(&tp->out_of_order_queue) &&
567 tp->rcv_wnd &&
568 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
569 !tp->urg_data)
570 tcp_fast_path_on(tp);
571 }
572
573 /* Compute the actual rto_min value */
574 static inline u32 tcp_rto_min(struct sock *sk)
575 {
576 struct dst_entry *dst = __sk_dst_get(sk);
577 u32 rto_min = TCP_RTO_MIN;
578
579 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
580 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
581 return rto_min;
582 }
583
584 /* Compute the actual receive window we are currently advertising.
585 * Rcv_nxt can be after the window if our peer push more data
586 * than the offered window.
587 */
588 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
589 {
590 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
591
592 if (win < 0)
593 win = 0;
594 return (u32) win;
595 }
596
597 /* Choose a new window, without checks for shrinking, and without
598 * scaling applied to the result. The caller does these things
599 * if necessary. This is a "raw" window selection.
600 */
601 extern u32 __tcp_select_window(struct sock *sk);
602
603 /* TCP timestamps are only 32-bits, this causes a slight
604 * complication on 64-bit systems since we store a snapshot
605 * of jiffies in the buffer control blocks below. We decided
606 * to use only the low 32-bits of jiffies and hide the ugly
607 * casts with the following macro.
608 */
609 #define tcp_time_stamp ((__u32)(jiffies))
610
611 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
612
613 #define TCPHDR_FIN 0x01
614 #define TCPHDR_SYN 0x02
615 #define TCPHDR_RST 0x04
616 #define TCPHDR_PSH 0x08
617 #define TCPHDR_ACK 0x10
618 #define TCPHDR_URG 0x20
619 #define TCPHDR_ECE 0x40
620 #define TCPHDR_CWR 0x80
621
622 /* This is what the send packet queuing engine uses to pass
623 * TCP per-packet control information to the transmission code.
624 * We also store the host-order sequence numbers in here too.
625 * This is 44 bytes if IPV6 is enabled.
626 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
627 */
628 struct tcp_skb_cb {
629 union {
630 struct inet_skb_parm h4;
631 #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE)
632 struct inet6_skb_parm h6;
633 #endif
634 } header; /* For incoming frames */
635 __u32 seq; /* Starting sequence number */
636 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
637 __u32 when; /* used to compute rtt's */
638 __u8 flags; /* TCP header flags. */
639 __u8 sacked; /* State flags for SACK/FACK. */
640 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
641 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
642 #define TCPCB_LOST 0x04 /* SKB is lost */
643 #define TCPCB_TAGBITS 0x07 /* All tag bits */
644
645 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
646 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
647
648 __u32 ack_seq; /* Sequence number ACK'd */
649 };
650
651 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
652
653 /* Due to TSO, an SKB can be composed of multiple actual
654 * packets. To keep these tracked properly, we use this.
655 */
656 static inline int tcp_skb_pcount(const struct sk_buff *skb)
657 {
658 return skb_shinfo(skb)->gso_segs;
659 }
660
661 /* This is valid iff tcp_skb_pcount() > 1. */
662 static inline int tcp_skb_mss(const struct sk_buff *skb)
663 {
664 return skb_shinfo(skb)->gso_size;
665 }
666
667 /* Events passed to congestion control interface */
668 enum tcp_ca_event {
669 CA_EVENT_TX_START, /* first transmit when no packets in flight */
670 CA_EVENT_CWND_RESTART, /* congestion window restart */
671 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
672 CA_EVENT_FRTO, /* fast recovery timeout */
673 CA_EVENT_LOSS, /* loss timeout */
674 CA_EVENT_FAST_ACK, /* in sequence ack */
675 CA_EVENT_SLOW_ACK, /* other ack */
676 };
677
678 /*
679 * Interface for adding new TCP congestion control handlers
680 */
681 #define TCP_CA_NAME_MAX 16
682 #define TCP_CA_MAX 128
683 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
684
685 #define TCP_CONG_NON_RESTRICTED 0x1
686 #define TCP_CONG_RTT_STAMP 0x2
687
688 struct tcp_congestion_ops {
689 struct list_head list;
690 unsigned long flags;
691
692 /* initialize private data (optional) */
693 void (*init)(struct sock *sk);
694 /* cleanup private data (optional) */
695 void (*release)(struct sock *sk);
696
697 /* return slow start threshold (required) */
698 u32 (*ssthresh)(struct sock *sk);
699 /* lower bound for congestion window (optional) */
700 u32 (*min_cwnd)(const struct sock *sk);
701 /* do new cwnd calculation (required) */
702 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
703 /* call before changing ca_state (optional) */
704 void (*set_state)(struct sock *sk, u8 new_state);
705 /* call when cwnd event occurs (optional) */
706 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
707 /* new value of cwnd after loss (optional) */
708 u32 (*undo_cwnd)(struct sock *sk);
709 /* hook for packet ack accounting (optional) */
710 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
711 /* get info for inet_diag (optional) */
712 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
713
714 char name[TCP_CA_NAME_MAX];
715 struct module *owner;
716 };
717
718 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
719 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
720
721 extern void tcp_init_congestion_control(struct sock *sk);
722 extern void tcp_cleanup_congestion_control(struct sock *sk);
723 extern int tcp_set_default_congestion_control(const char *name);
724 extern void tcp_get_default_congestion_control(char *name);
725 extern void tcp_get_available_congestion_control(char *buf, size_t len);
726 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
727 extern int tcp_set_allowed_congestion_control(char *allowed);
728 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
729 extern void tcp_slow_start(struct tcp_sock *tp);
730 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
731
732 extern struct tcp_congestion_ops tcp_init_congestion_ops;
733 extern u32 tcp_reno_ssthresh(struct sock *sk);
734 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
735 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
736 extern struct tcp_congestion_ops tcp_reno;
737
738 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
739 {
740 struct inet_connection_sock *icsk = inet_csk(sk);
741
742 if (icsk->icsk_ca_ops->set_state)
743 icsk->icsk_ca_ops->set_state(sk, ca_state);
744 icsk->icsk_ca_state = ca_state;
745 }
746
747 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
748 {
749 const struct inet_connection_sock *icsk = inet_csk(sk);
750
751 if (icsk->icsk_ca_ops->cwnd_event)
752 icsk->icsk_ca_ops->cwnd_event(sk, event);
753 }
754
755 /* These functions determine how the current flow behaves in respect of SACK
756 * handling. SACK is negotiated with the peer, and therefore it can vary
757 * between different flows.
758 *
759 * tcp_is_sack - SACK enabled
760 * tcp_is_reno - No SACK
761 * tcp_is_fack - FACK enabled, implies SACK enabled
762 */
763 static inline int tcp_is_sack(const struct tcp_sock *tp)
764 {
765 return tp->rx_opt.sack_ok;
766 }
767
768 static inline int tcp_is_reno(const struct tcp_sock *tp)
769 {
770 return !tcp_is_sack(tp);
771 }
772
773 static inline int tcp_is_fack(const struct tcp_sock *tp)
774 {
775 return tp->rx_opt.sack_ok & 2;
776 }
777
778 static inline void tcp_enable_fack(struct tcp_sock *tp)
779 {
780 tp->rx_opt.sack_ok |= 2;
781 }
782
783 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
784 {
785 return tp->sacked_out + tp->lost_out;
786 }
787
788 /* This determines how many packets are "in the network" to the best
789 * of our knowledge. In many cases it is conservative, but where
790 * detailed information is available from the receiver (via SACK
791 * blocks etc.) we can make more aggressive calculations.
792 *
793 * Use this for decisions involving congestion control, use just
794 * tp->packets_out to determine if the send queue is empty or not.
795 *
796 * Read this equation as:
797 *
798 * "Packets sent once on transmission queue" MINUS
799 * "Packets left network, but not honestly ACKed yet" PLUS
800 * "Packets fast retransmitted"
801 */
802 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
803 {
804 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
805 }
806
807 #define TCP_INFINITE_SSTHRESH 0x7fffffff
808
809 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
810 {
811 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
812 }
813
814 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
815 * The exception is rate halving phase, when cwnd is decreasing towards
816 * ssthresh.
817 */
818 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
819 {
820 const struct tcp_sock *tp = tcp_sk(sk);
821 if ((1 << inet_csk(sk)->icsk_ca_state) & (TCPF_CA_CWR | TCPF_CA_Recovery))
822 return tp->snd_ssthresh;
823 else
824 return max(tp->snd_ssthresh,
825 ((tp->snd_cwnd >> 1) +
826 (tp->snd_cwnd >> 2)));
827 }
828
829 /* Use define here intentionally to get WARN_ON location shown at the caller */
830 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
831
832 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
833 extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst);
834
835 /* Slow start with delack produces 3 packets of burst, so that
836 * it is safe "de facto". This will be the default - same as
837 * the default reordering threshold - but if reordering increases,
838 * we must be able to allow cwnd to burst at least this much in order
839 * to not pull it back when holes are filled.
840 */
841 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
842 {
843 return tp->reordering;
844 }
845
846 /* Returns end sequence number of the receiver's advertised window */
847 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
848 {
849 return tp->snd_una + tp->snd_wnd;
850 }
851 extern int tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
852
853 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
854 const struct sk_buff *skb)
855 {
856 if (skb->len < mss)
857 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
858 }
859
860 static inline void tcp_check_probe_timer(struct sock *sk)
861 {
862 struct tcp_sock *tp = tcp_sk(sk);
863 const struct inet_connection_sock *icsk = inet_csk(sk);
864
865 if (!tp->packets_out && !icsk->icsk_pending)
866 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
867 icsk->icsk_rto, TCP_RTO_MAX);
868 }
869
870 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
871 {
872 tp->snd_wl1 = seq;
873 }
874
875 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
876 {
877 tp->snd_wl1 = seq;
878 }
879
880 /*
881 * Calculate(/check) TCP checksum
882 */
883 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
884 __be32 daddr, __wsum base)
885 {
886 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
887 }
888
889 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
890 {
891 return __skb_checksum_complete(skb);
892 }
893
894 static inline int tcp_checksum_complete(struct sk_buff *skb)
895 {
896 return !skb_csum_unnecessary(skb) &&
897 __tcp_checksum_complete(skb);
898 }
899
900 /* Prequeue for VJ style copy to user, combined with checksumming. */
901
902 static inline void tcp_prequeue_init(struct tcp_sock *tp)
903 {
904 tp->ucopy.task = NULL;
905 tp->ucopy.len = 0;
906 tp->ucopy.memory = 0;
907 skb_queue_head_init(&tp->ucopy.prequeue);
908 #ifdef CONFIG_NET_DMA
909 tp->ucopy.dma_chan = NULL;
910 tp->ucopy.wakeup = 0;
911 tp->ucopy.pinned_list = NULL;
912 tp->ucopy.dma_cookie = 0;
913 #endif
914 }
915
916 /* Packet is added to VJ-style prequeue for processing in process
917 * context, if a reader task is waiting. Apparently, this exciting
918 * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93)
919 * failed somewhere. Latency? Burstiness? Well, at least now we will
920 * see, why it failed. 8)8) --ANK
921 *
922 * NOTE: is this not too big to inline?
923 */
924 static inline int tcp_prequeue(struct sock *sk, struct sk_buff *skb)
925 {
926 struct tcp_sock *tp = tcp_sk(sk);
927
928 if (sysctl_tcp_low_latency || !tp->ucopy.task)
929 return 0;
930
931 __skb_queue_tail(&tp->ucopy.prequeue, skb);
932 tp->ucopy.memory += skb->truesize;
933 if (tp->ucopy.memory > sk->sk_rcvbuf) {
934 struct sk_buff *skb1;
935
936 BUG_ON(sock_owned_by_user(sk));
937
938 while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) {
939 sk_backlog_rcv(sk, skb1);
940 NET_INC_STATS_BH(sock_net(sk),
941 LINUX_MIB_TCPPREQUEUEDROPPED);
942 }
943
944 tp->ucopy.memory = 0;
945 } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) {
946 wake_up_interruptible_sync_poll(sk_sleep(sk),
947 POLLIN | POLLRDNORM | POLLRDBAND);
948 if (!inet_csk_ack_scheduled(sk))
949 inet_csk_reset_xmit_timer(sk, ICSK_TIME_DACK,
950 (3 * tcp_rto_min(sk)) / 4,
951 TCP_RTO_MAX);
952 }
953 return 1;
954 }
955
956
957 #undef STATE_TRACE
958
959 #ifdef STATE_TRACE
960 static const char *statename[]={
961 "Unused","Established","Syn Sent","Syn Recv",
962 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
963 "Close Wait","Last ACK","Listen","Closing"
964 };
965 #endif
966 extern void tcp_set_state(struct sock *sk, int state);
967
968 extern void tcp_done(struct sock *sk);
969
970 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
971 {
972 rx_opt->dsack = 0;
973 rx_opt->num_sacks = 0;
974 }
975
976 /* Determine a window scaling and initial window to offer. */
977 extern void tcp_select_initial_window(int __space, __u32 mss,
978 __u32 *rcv_wnd, __u32 *window_clamp,
979 int wscale_ok, __u8 *rcv_wscale,
980 __u32 init_rcv_wnd);
981
982 static inline int tcp_win_from_space(int space)
983 {
984 return sysctl_tcp_adv_win_scale<=0 ?
985 (space>>(-sysctl_tcp_adv_win_scale)) :
986 space - (space>>sysctl_tcp_adv_win_scale);
987 }
988
989 /* Note: caller must be prepared to deal with negative returns */
990 static inline int tcp_space(const struct sock *sk)
991 {
992 return tcp_win_from_space(sk->sk_rcvbuf -
993 atomic_read(&sk->sk_rmem_alloc));
994 }
995
996 static inline int tcp_full_space(const struct sock *sk)
997 {
998 return tcp_win_from_space(sk->sk_rcvbuf);
999 }
1000
1001 static inline void tcp_openreq_init(struct request_sock *req,
1002 struct tcp_options_received *rx_opt,
1003 struct sk_buff *skb)
1004 {
1005 struct inet_request_sock *ireq = inet_rsk(req);
1006
1007 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1008 req->cookie_ts = 0;
1009 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1010 req->mss = rx_opt->mss_clamp;
1011 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1012 ireq->tstamp_ok = rx_opt->tstamp_ok;
1013 ireq->sack_ok = rx_opt->sack_ok;
1014 ireq->snd_wscale = rx_opt->snd_wscale;
1015 ireq->wscale_ok = rx_opt->wscale_ok;
1016 ireq->acked = 0;
1017 ireq->ecn_ok = 0;
1018 ireq->rmt_port = tcp_hdr(skb)->source;
1019 ireq->loc_port = tcp_hdr(skb)->dest;
1020 }
1021
1022 extern void tcp_enter_memory_pressure(struct sock *sk);
1023
1024 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1025 {
1026 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1027 }
1028
1029 static inline int keepalive_time_when(const struct tcp_sock *tp)
1030 {
1031 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1032 }
1033
1034 static inline int keepalive_probes(const struct tcp_sock *tp)
1035 {
1036 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1037 }
1038
1039 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1040 {
1041 const struct inet_connection_sock *icsk = &tp->inet_conn;
1042
1043 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1044 tcp_time_stamp - tp->rcv_tstamp);
1045 }
1046
1047 static inline int tcp_fin_time(const struct sock *sk)
1048 {
1049 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1050 const int rto = inet_csk(sk)->icsk_rto;
1051
1052 if (fin_timeout < (rto << 2) - (rto >> 1))
1053 fin_timeout = (rto << 2) - (rto >> 1);
1054
1055 return fin_timeout;
1056 }
1057
1058 static inline int tcp_paws_check(const struct tcp_options_received *rx_opt,
1059 int paws_win)
1060 {
1061 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1062 return 1;
1063 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1064 return 1;
1065 /*
1066 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1067 * then following tcp messages have valid values. Ignore 0 value,
1068 * or else 'negative' tsval might forbid us to accept their packets.
1069 */
1070 if (!rx_opt->ts_recent)
1071 return 1;
1072 return 0;
1073 }
1074
1075 static inline int tcp_paws_reject(const struct tcp_options_received *rx_opt,
1076 int rst)
1077 {
1078 if (tcp_paws_check(rx_opt, 0))
1079 return 0;
1080
1081 /* RST segments are not recommended to carry timestamp,
1082 and, if they do, it is recommended to ignore PAWS because
1083 "their cleanup function should take precedence over timestamps."
1084 Certainly, it is mistake. It is necessary to understand the reasons
1085 of this constraint to relax it: if peer reboots, clock may go
1086 out-of-sync and half-open connections will not be reset.
1087 Actually, the problem would be not existing if all
1088 the implementations followed draft about maintaining clock
1089 via reboots. Linux-2.2 DOES NOT!
1090
1091 However, we can relax time bounds for RST segments to MSL.
1092 */
1093 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1094 return 0;
1095 return 1;
1096 }
1097
1098 static inline void tcp_mib_init(struct net *net)
1099 {
1100 /* See RFC 2012 */
1101 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1102 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1103 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1104 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1105 }
1106
1107 /* from STCP */
1108 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1109 {
1110 tp->lost_skb_hint = NULL;
1111 tp->scoreboard_skb_hint = NULL;
1112 }
1113
1114 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1115 {
1116 tcp_clear_retrans_hints_partial(tp);
1117 tp->retransmit_skb_hint = NULL;
1118 }
1119
1120 /* MD5 Signature */
1121 struct crypto_hash;
1122
1123 /* - key database */
1124 struct tcp_md5sig_key {
1125 u8 *key;
1126 u8 keylen;
1127 };
1128
1129 struct tcp4_md5sig_key {
1130 struct tcp_md5sig_key base;
1131 __be32 addr;
1132 };
1133
1134 struct tcp6_md5sig_key {
1135 struct tcp_md5sig_key base;
1136 #if 0
1137 u32 scope_id; /* XXX */
1138 #endif
1139 struct in6_addr addr;
1140 };
1141
1142 /* - sock block */
1143 struct tcp_md5sig_info {
1144 struct tcp4_md5sig_key *keys4;
1145 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1146 struct tcp6_md5sig_key *keys6;
1147 u32 entries6;
1148 u32 alloced6;
1149 #endif
1150 u32 entries4;
1151 u32 alloced4;
1152 };
1153
1154 /* - pseudo header */
1155 struct tcp4_pseudohdr {
1156 __be32 saddr;
1157 __be32 daddr;
1158 __u8 pad;
1159 __u8 protocol;
1160 __be16 len;
1161 };
1162
1163 struct tcp6_pseudohdr {
1164 struct in6_addr saddr;
1165 struct in6_addr daddr;
1166 __be32 len;
1167 __be32 protocol; /* including padding */
1168 };
1169
1170 union tcp_md5sum_block {
1171 struct tcp4_pseudohdr ip4;
1172 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1173 struct tcp6_pseudohdr ip6;
1174 #endif
1175 };
1176
1177 /* - pool: digest algorithm, hash description and scratch buffer */
1178 struct tcp_md5sig_pool {
1179 struct hash_desc md5_desc;
1180 union tcp_md5sum_block md5_blk;
1181 };
1182
1183 /* - functions */
1184 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1185 struct sock *sk, struct request_sock *req,
1186 struct sk_buff *skb);
1187 extern struct tcp_md5sig_key * tcp_v4_md5_lookup(struct sock *sk,
1188 struct sock *addr_sk);
1189 extern int tcp_v4_md5_do_add(struct sock *sk, __be32 addr, u8 *newkey,
1190 u8 newkeylen);
1191 extern int tcp_v4_md5_do_del(struct sock *sk, __be32 addr);
1192
1193 #ifdef CONFIG_TCP_MD5SIG
1194 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_keylen ? \
1195 &(struct tcp_md5sig_key) { \
1196 .key = (twsk)->tw_md5_key, \
1197 .keylen = (twsk)->tw_md5_keylen, \
1198 } : NULL)
1199 #else
1200 #define tcp_twsk_md5_key(twsk) NULL
1201 #endif
1202
1203 extern struct tcp_md5sig_pool * __percpu *tcp_alloc_md5sig_pool(struct sock *);
1204 extern void tcp_free_md5sig_pool(void);
1205
1206 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1207 extern void tcp_put_md5sig_pool(void);
1208
1209 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, struct tcphdr *);
1210 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, struct sk_buff *,
1211 unsigned header_len);
1212 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1213 struct tcp_md5sig_key *key);
1214
1215 /* write queue abstraction */
1216 static inline void tcp_write_queue_purge(struct sock *sk)
1217 {
1218 struct sk_buff *skb;
1219
1220 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1221 sk_wmem_free_skb(sk, skb);
1222 sk_mem_reclaim(sk);
1223 tcp_clear_all_retrans_hints(tcp_sk(sk));
1224 }
1225
1226 static inline struct sk_buff *tcp_write_queue_head(struct sock *sk)
1227 {
1228 return skb_peek(&sk->sk_write_queue);
1229 }
1230
1231 static inline struct sk_buff *tcp_write_queue_tail(struct sock *sk)
1232 {
1233 return skb_peek_tail(&sk->sk_write_queue);
1234 }
1235
1236 static inline struct sk_buff *tcp_write_queue_next(struct sock *sk, struct sk_buff *skb)
1237 {
1238 return skb_queue_next(&sk->sk_write_queue, skb);
1239 }
1240
1241 static inline struct sk_buff *tcp_write_queue_prev(struct sock *sk, struct sk_buff *skb)
1242 {
1243 return skb_queue_prev(&sk->sk_write_queue, skb);
1244 }
1245
1246 #define tcp_for_write_queue(skb, sk) \
1247 skb_queue_walk(&(sk)->sk_write_queue, skb)
1248
1249 #define tcp_for_write_queue_from(skb, sk) \
1250 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1251
1252 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1253 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1254
1255 static inline struct sk_buff *tcp_send_head(struct sock *sk)
1256 {
1257 return sk->sk_send_head;
1258 }
1259
1260 static inline bool tcp_skb_is_last(const struct sock *sk,
1261 const struct sk_buff *skb)
1262 {
1263 return skb_queue_is_last(&sk->sk_write_queue, skb);
1264 }
1265
1266 static inline void tcp_advance_send_head(struct sock *sk, struct sk_buff *skb)
1267 {
1268 if (tcp_skb_is_last(sk, skb))
1269 sk->sk_send_head = NULL;
1270 else
1271 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1272 }
1273
1274 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1275 {
1276 if (sk->sk_send_head == skb_unlinked)
1277 sk->sk_send_head = NULL;
1278 }
1279
1280 static inline void tcp_init_send_head(struct sock *sk)
1281 {
1282 sk->sk_send_head = NULL;
1283 }
1284
1285 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1286 {
1287 __skb_queue_tail(&sk->sk_write_queue, skb);
1288 }
1289
1290 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1291 {
1292 __tcp_add_write_queue_tail(sk, skb);
1293
1294 /* Queue it, remembering where we must start sending. */
1295 if (sk->sk_send_head == NULL) {
1296 sk->sk_send_head = skb;
1297
1298 if (tcp_sk(sk)->highest_sack == NULL)
1299 tcp_sk(sk)->highest_sack = skb;
1300 }
1301 }
1302
1303 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1304 {
1305 __skb_queue_head(&sk->sk_write_queue, skb);
1306 }
1307
1308 /* Insert buff after skb on the write queue of sk. */
1309 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1310 struct sk_buff *buff,
1311 struct sock *sk)
1312 {
1313 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1314 }
1315
1316 /* Insert new before skb on the write queue of sk. */
1317 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1318 struct sk_buff *skb,
1319 struct sock *sk)
1320 {
1321 __skb_queue_before(&sk->sk_write_queue, skb, new);
1322
1323 if (sk->sk_send_head == skb)
1324 sk->sk_send_head = new;
1325 }
1326
1327 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1328 {
1329 __skb_unlink(skb, &sk->sk_write_queue);
1330 }
1331
1332 static inline int tcp_write_queue_empty(struct sock *sk)
1333 {
1334 return skb_queue_empty(&sk->sk_write_queue);
1335 }
1336
1337 static inline void tcp_push_pending_frames(struct sock *sk)
1338 {
1339 if (tcp_send_head(sk)) {
1340 struct tcp_sock *tp = tcp_sk(sk);
1341
1342 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1343 }
1344 }
1345
1346 /* Start sequence of the highest skb with SACKed bit, valid only if
1347 * sacked > 0 or when the caller has ensured validity by itself.
1348 */
1349 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1350 {
1351 if (!tp->sacked_out)
1352 return tp->snd_una;
1353
1354 if (tp->highest_sack == NULL)
1355 return tp->snd_nxt;
1356
1357 return TCP_SKB_CB(tp->highest_sack)->seq;
1358 }
1359
1360 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1361 {
1362 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1363 tcp_write_queue_next(sk, skb);
1364 }
1365
1366 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1367 {
1368 return tcp_sk(sk)->highest_sack;
1369 }
1370
1371 static inline void tcp_highest_sack_reset(struct sock *sk)
1372 {
1373 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1374 }
1375
1376 /* Called when old skb is about to be deleted (to be combined with new skb) */
1377 static inline void tcp_highest_sack_combine(struct sock *sk,
1378 struct sk_buff *old,
1379 struct sk_buff *new)
1380 {
1381 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1382 tcp_sk(sk)->highest_sack = new;
1383 }
1384
1385 /* Determines whether this is a thin stream (which may suffer from
1386 * increased latency). Used to trigger latency-reducing mechanisms.
1387 */
1388 static inline unsigned int tcp_stream_is_thin(struct tcp_sock *tp)
1389 {
1390 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1391 }
1392
1393 /* /proc */
1394 enum tcp_seq_states {
1395 TCP_SEQ_STATE_LISTENING,
1396 TCP_SEQ_STATE_OPENREQ,
1397 TCP_SEQ_STATE_ESTABLISHED,
1398 TCP_SEQ_STATE_TIME_WAIT,
1399 };
1400
1401 struct tcp_seq_afinfo {
1402 char *name;
1403 sa_family_t family;
1404 struct file_operations seq_fops;
1405 struct seq_operations seq_ops;
1406 };
1407
1408 struct tcp_iter_state {
1409 struct seq_net_private p;
1410 sa_family_t family;
1411 enum tcp_seq_states state;
1412 struct sock *syn_wait_sk;
1413 int bucket, offset, sbucket, num, uid;
1414 loff_t last_pos;
1415 };
1416
1417 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1418 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1419
1420 extern struct request_sock_ops tcp_request_sock_ops;
1421 extern struct request_sock_ops tcp6_request_sock_ops;
1422
1423 extern void tcp_v4_destroy_sock(struct sock *sk);
1424
1425 extern int tcp_v4_gso_send_check(struct sk_buff *skb);
1426 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb, u32 features);
1427 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1428 struct sk_buff *skb);
1429 extern struct sk_buff **tcp4_gro_receive(struct sk_buff **head,
1430 struct sk_buff *skb);
1431 extern int tcp_gro_complete(struct sk_buff *skb);
1432 extern int tcp4_gro_complete(struct sk_buff *skb);
1433
1434 #ifdef CONFIG_PROC_FS
1435 extern int tcp4_proc_init(void);
1436 extern void tcp4_proc_exit(void);
1437 #endif
1438
1439 /* TCP af-specific functions */
1440 struct tcp_sock_af_ops {
1441 #ifdef CONFIG_TCP_MD5SIG
1442 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1443 struct sock *addr_sk);
1444 int (*calc_md5_hash) (char *location,
1445 struct tcp_md5sig_key *md5,
1446 struct sock *sk,
1447 struct request_sock *req,
1448 struct sk_buff *skb);
1449 int (*md5_add) (struct sock *sk,
1450 struct sock *addr_sk,
1451 u8 *newkey,
1452 u8 len);
1453 int (*md5_parse) (struct sock *sk,
1454 char __user *optval,
1455 int optlen);
1456 #endif
1457 };
1458
1459 struct tcp_request_sock_ops {
1460 #ifdef CONFIG_TCP_MD5SIG
1461 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1462 struct request_sock *req);
1463 int (*calc_md5_hash) (char *location,
1464 struct tcp_md5sig_key *md5,
1465 struct sock *sk,
1466 struct request_sock *req,
1467 struct sk_buff *skb);
1468 #endif
1469 };
1470
1471 /* Using SHA1 for now, define some constants.
1472 */
1473 #define COOKIE_DIGEST_WORDS (SHA_DIGEST_WORDS)
1474 #define COOKIE_MESSAGE_WORDS (SHA_MESSAGE_BYTES / 4)
1475 #define COOKIE_WORKSPACE_WORDS (COOKIE_DIGEST_WORDS + COOKIE_MESSAGE_WORDS)
1476
1477 extern int tcp_cookie_generator(u32 *bakery);
1478
1479 /**
1480 * struct tcp_cookie_values - each socket needs extra space for the
1481 * cookies, together with (optional) space for any SYN data.
1482 *
1483 * A tcp_sock contains a pointer to the current value, and this is
1484 * cloned to the tcp_timewait_sock.
1485 *
1486 * @cookie_pair: variable data from the option exchange.
1487 *
1488 * @cookie_desired: user specified tcpct_cookie_desired. Zero
1489 * indicates default (sysctl_tcp_cookie_size).
1490 * After cookie sent, remembers size of cookie.
1491 * Range 0, TCP_COOKIE_MIN to TCP_COOKIE_MAX.
1492 *
1493 * @s_data_desired: user specified tcpct_s_data_desired. When the
1494 * constant payload is specified (@s_data_constant),
1495 * holds its length instead.
1496 * Range 0 to TCP_MSS_DESIRED.
1497 *
1498 * @s_data_payload: constant data that is to be included in the
1499 * payload of SYN or SYNACK segments when the
1500 * cookie option is present.
1501 */
1502 struct tcp_cookie_values {
1503 struct kref kref;
1504 u8 cookie_pair[TCP_COOKIE_PAIR_SIZE];
1505 u8 cookie_pair_size;
1506 u8 cookie_desired;
1507 u16 s_data_desired:11,
1508 s_data_constant:1,
1509 s_data_in:1,
1510 s_data_out:1,
1511 s_data_unused:2;
1512 u8 s_data_payload[0];
1513 };
1514
1515 static inline void tcp_cookie_values_release(struct kref *kref)
1516 {
1517 kfree(container_of(kref, struct tcp_cookie_values, kref));
1518 }
1519
1520 /* The length of constant payload data. Note that s_data_desired is
1521 * overloaded, depending on s_data_constant: either the length of constant
1522 * data (returned here) or the limit on variable data.
1523 */
1524 static inline int tcp_s_data_size(const struct tcp_sock *tp)
1525 {
1526 return (tp->cookie_values != NULL && tp->cookie_values->s_data_constant)
1527 ? tp->cookie_values->s_data_desired
1528 : 0;
1529 }
1530
1531 /**
1532 * struct tcp_extend_values - tcp_ipv?.c to tcp_output.c workspace.
1533 *
1534 * As tcp_request_sock has already been extended in other places, the
1535 * only remaining method is to pass stack values along as function
1536 * parameters. These parameters are not needed after sending SYNACK.
1537 *
1538 * @cookie_bakery: cryptographic secret and message workspace.
1539 *
1540 * @cookie_plus: bytes in authenticator/cookie option, copied from
1541 * struct tcp_options_received (above).
1542 */
1543 struct tcp_extend_values {
1544 struct request_values rv;
1545 u32 cookie_bakery[COOKIE_WORKSPACE_WORDS];
1546 u8 cookie_plus:6,
1547 cookie_out_never:1,
1548 cookie_in_always:1;
1549 };
1550
1551 static inline struct tcp_extend_values *tcp_xv(struct request_values *rvp)
1552 {
1553 return (struct tcp_extend_values *)rvp;
1554 }
1555
1556 extern void tcp_v4_init(void);
1557 extern void tcp_init(void);
1558
1559 #endif /* _TCP_H */
Linus' kernel tree