Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * INET An implementation of the TCP/IP protocol suite for the LINUX | |
3 | * operating system. INET is implemented using the BSD Socket | |
4 | * interface as the means of communication with the user level. | |
5 | * | |
6 | * Definitions for the TCP module. | |
7 | * | |
8 | * Version: @(#)tcp.h 1.0.5 05/23/93 | |
9 | * | |
02c30a84 | 10 | * Authors: Ross Biro |
1da177e4 LT |
11 | * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG> |
12 | * | |
13 | * This program is free software; you can redistribute it and/or | |
14 | * modify it under the terms of the GNU General Public License | |
15 | * as published by the Free Software Foundation; either version | |
16 | * 2 of the License, or (at your option) any later version. | |
17 | */ | |
18 | #ifndef _TCP_H | |
19 | #define _TCP_H | |
20 | ||
21 | #define TCP_DEBUG 1 | |
22 | #define FASTRETRANS_DEBUG 1 | |
23 | ||
24 | /* Cancel timers, when they are not required. */ | |
25 | #undef TCP_CLEAR_TIMERS | |
26 | ||
27 | #include <linux/config.h> | |
28 | #include <linux/list.h> | |
29 | #include <linux/tcp.h> | |
30 | #include <linux/slab.h> | |
31 | #include <linux/cache.h> | |
32 | #include <linux/percpu.h> | |
77d8bf9c | 33 | #include <net/inet_hashtables.h> |
1da177e4 | 34 | #include <net/checksum.h> |
2e6599cb | 35 | #include <net/request_sock.h> |
1da177e4 LT |
36 | #include <net/sock.h> |
37 | #include <net/snmp.h> | |
38 | #include <net/ip.h> | |
c752f073 ACM |
39 | #include <net/tcp_states.h> |
40 | ||
1da177e4 LT |
41 | #include <linux/seq_file.h> |
42 | ||
6e04e021 | 43 | extern struct inet_hashinfo tcp_hashinfo; |
1da177e4 | 44 | |
1da177e4 LT |
45 | extern atomic_t tcp_orphan_count; |
46 | extern int tcp_tw_count; | |
47 | extern void tcp_time_wait(struct sock *sk, int state, int timeo); | |
8feaf0c0 | 48 | extern void tcp_tw_deschedule(struct inet_timewait_sock *tw); |
1da177e4 | 49 | |
1da177e4 LT |
50 | #define MAX_TCP_HEADER (128 + MAX_HEADER) |
51 | ||
52 | /* | |
53 | * Never offer a window over 32767 without using window scaling. Some | |
54 | * poor stacks do signed 16bit maths! | |
55 | */ | |
56 | #define MAX_TCP_WINDOW 32767U | |
57 | ||
58 | /* Minimal accepted MSS. It is (60+60+8) - (20+20). */ | |
59 | #define TCP_MIN_MSS 88U | |
60 | ||
61 | /* Minimal RCV_MSS. */ | |
62 | #define TCP_MIN_RCVMSS 536U | |
63 | ||
64 | /* After receiving this amount of duplicate ACKs fast retransmit starts. */ | |
65 | #define TCP_FASTRETRANS_THRESH 3 | |
66 | ||
67 | /* Maximal reordering. */ | |
68 | #define TCP_MAX_REORDERING 127 | |
69 | ||
70 | /* Maximal number of ACKs sent quickly to accelerate slow-start. */ | |
71 | #define TCP_MAX_QUICKACKS 16U | |
72 | ||
73 | /* urg_data states */ | |
74 | #define TCP_URG_VALID 0x0100 | |
75 | #define TCP_URG_NOTYET 0x0200 | |
76 | #define TCP_URG_READ 0x0400 | |
77 | ||
78 | #define TCP_RETR1 3 /* | |
79 | * This is how many retries it does before it | |
80 | * tries to figure out if the gateway is | |
81 | * down. Minimal RFC value is 3; it corresponds | |
82 | * to ~3sec-8min depending on RTO. | |
83 | */ | |
84 | ||
85 | #define TCP_RETR2 15 /* | |
86 | * This should take at least | |
87 | * 90 minutes to time out. | |
88 | * RFC1122 says that the limit is 100 sec. | |
89 | * 15 is ~13-30min depending on RTO. | |
90 | */ | |
91 | ||
92 | #define TCP_SYN_RETRIES 5 /* number of times to retry active opening a | |
93 | * connection: ~180sec is RFC minumum */ | |
94 | ||
95 | #define TCP_SYNACK_RETRIES 5 /* number of times to retry passive opening a | |
96 | * connection: ~180sec is RFC minumum */ | |
97 | ||
98 | ||
99 | #define TCP_ORPHAN_RETRIES 7 /* number of times to retry on an orphaned | |
100 | * socket. 7 is ~50sec-16min. | |
101 | */ | |
102 | ||
103 | ||
104 | #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT | |
105 | * state, about 60 seconds */ | |
106 | #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN | |
107 | /* BSD style FIN_WAIT2 deadlock breaker. | |
108 | * It used to be 3min, new value is 60sec, | |
109 | * to combine FIN-WAIT-2 timeout with | |
110 | * TIME-WAIT timer. | |
111 | */ | |
112 | ||
113 | #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */ | |
114 | #if HZ >= 100 | |
115 | #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */ | |
116 | #define TCP_ATO_MIN ((unsigned)(HZ/25)) | |
117 | #else | |
118 | #define TCP_DELACK_MIN 4U | |
119 | #define TCP_ATO_MIN 4U | |
120 | #endif | |
121 | #define TCP_RTO_MAX ((unsigned)(120*HZ)) | |
122 | #define TCP_RTO_MIN ((unsigned)(HZ/5)) | |
123 | #define TCP_TIMEOUT_INIT ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value */ | |
124 | ||
125 | #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes | |
126 | * for local resources. | |
127 | */ | |
128 | ||
129 | #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */ | |
130 | #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */ | |
131 | #define TCP_KEEPALIVE_INTVL (75*HZ) | |
132 | ||
133 | #define MAX_TCP_KEEPIDLE 32767 | |
134 | #define MAX_TCP_KEEPINTVL 32767 | |
135 | #define MAX_TCP_KEEPCNT 127 | |
136 | #define MAX_TCP_SYNCNT 127 | |
137 | ||
138 | #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */ | |
139 | #define TCP_SYNQ_HSIZE 512 /* Size of SYNACK hash table */ | |
140 | ||
141 | #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24) | |
142 | #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated | |
143 | * after this time. It should be equal | |
144 | * (or greater than) TCP_TIMEWAIT_LEN | |
145 | * to provide reliability equal to one | |
146 | * provided by timewait state. | |
147 | */ | |
148 | #define TCP_PAWS_WINDOW 1 /* Replay window for per-host | |
149 | * timestamps. It must be less than | |
150 | * minimal timewait lifetime. | |
151 | */ | |
152 | ||
153 | #define TCP_TW_RECYCLE_SLOTS_LOG 5 | |
154 | #define TCP_TW_RECYCLE_SLOTS (1<<TCP_TW_RECYCLE_SLOTS_LOG) | |
155 | ||
156 | /* If time > 4sec, it is "slow" path, no recycling is required, | |
157 | so that we select tick to get range about 4 seconds. | |
158 | */ | |
159 | ||
160 | #if HZ <= 16 || HZ > 4096 | |
161 | # error Unsupported: HZ <= 16 or HZ > 4096 | |
162 | #elif HZ <= 32 | |
163 | # define TCP_TW_RECYCLE_TICK (5+2-TCP_TW_RECYCLE_SLOTS_LOG) | |
164 | #elif HZ <= 64 | |
165 | # define TCP_TW_RECYCLE_TICK (6+2-TCP_TW_RECYCLE_SLOTS_LOG) | |
166 | #elif HZ <= 128 | |
167 | # define TCP_TW_RECYCLE_TICK (7+2-TCP_TW_RECYCLE_SLOTS_LOG) | |
168 | #elif HZ <= 256 | |
169 | # define TCP_TW_RECYCLE_TICK (8+2-TCP_TW_RECYCLE_SLOTS_LOG) | |
170 | #elif HZ <= 512 | |
171 | # define TCP_TW_RECYCLE_TICK (9+2-TCP_TW_RECYCLE_SLOTS_LOG) | |
172 | #elif HZ <= 1024 | |
173 | # define TCP_TW_RECYCLE_TICK (10+2-TCP_TW_RECYCLE_SLOTS_LOG) | |
174 | #elif HZ <= 2048 | |
175 | # define TCP_TW_RECYCLE_TICK (11+2-TCP_TW_RECYCLE_SLOTS_LOG) | |
176 | #else | |
177 | # define TCP_TW_RECYCLE_TICK (12+2-TCP_TW_RECYCLE_SLOTS_LOG) | |
178 | #endif | |
1da177e4 LT |
179 | /* |
180 | * TCP option | |
181 | */ | |
182 | ||
183 | #define TCPOPT_NOP 1 /* Padding */ | |
184 | #define TCPOPT_EOL 0 /* End of options */ | |
185 | #define TCPOPT_MSS 2 /* Segment size negotiating */ | |
186 | #define TCPOPT_WINDOW 3 /* Window scaling */ | |
187 | #define TCPOPT_SACK_PERM 4 /* SACK Permitted */ | |
188 | #define TCPOPT_SACK 5 /* SACK Block */ | |
189 | #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */ | |
190 | ||
191 | /* | |
192 | * TCP option lengths | |
193 | */ | |
194 | ||
195 | #define TCPOLEN_MSS 4 | |
196 | #define TCPOLEN_WINDOW 3 | |
197 | #define TCPOLEN_SACK_PERM 2 | |
198 | #define TCPOLEN_TIMESTAMP 10 | |
199 | ||
200 | /* But this is what stacks really send out. */ | |
201 | #define TCPOLEN_TSTAMP_ALIGNED 12 | |
202 | #define TCPOLEN_WSCALE_ALIGNED 4 | |
203 | #define TCPOLEN_SACKPERM_ALIGNED 4 | |
204 | #define TCPOLEN_SACK_BASE 2 | |
205 | #define TCPOLEN_SACK_BASE_ALIGNED 4 | |
206 | #define TCPOLEN_SACK_PERBLOCK 8 | |
207 | ||
208 | #define TCP_TIME_RETRANS 1 /* Retransmit timer */ | |
209 | #define TCP_TIME_DACK 2 /* Delayed ack timer */ | |
210 | #define TCP_TIME_PROBE0 3 /* Zero window probe timer */ | |
211 | #define TCP_TIME_KEEPOPEN 4 /* Keepalive timer */ | |
212 | ||
213 | /* Flags in tp->nonagle */ | |
214 | #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */ | |
215 | #define TCP_NAGLE_CORK 2 /* Socket is corked */ | |
216 | #define TCP_NAGLE_PUSH 4 /* Cork is overriden for already queued data */ | |
217 | ||
218 | /* sysctl variables for tcp */ | |
1da177e4 LT |
219 | extern int sysctl_tcp_timestamps; |
220 | extern int sysctl_tcp_window_scaling; | |
221 | extern int sysctl_tcp_sack; | |
222 | extern int sysctl_tcp_fin_timeout; | |
223 | extern int sysctl_tcp_tw_recycle; | |
224 | extern int sysctl_tcp_keepalive_time; | |
225 | extern int sysctl_tcp_keepalive_probes; | |
226 | extern int sysctl_tcp_keepalive_intvl; | |
227 | extern int sysctl_tcp_syn_retries; | |
228 | extern int sysctl_tcp_synack_retries; | |
229 | extern int sysctl_tcp_retries1; | |
230 | extern int sysctl_tcp_retries2; | |
231 | extern int sysctl_tcp_orphan_retries; | |
232 | extern int sysctl_tcp_syncookies; | |
233 | extern int sysctl_tcp_retrans_collapse; | |
234 | extern int sysctl_tcp_stdurg; | |
235 | extern int sysctl_tcp_rfc1337; | |
236 | extern int sysctl_tcp_abort_on_overflow; | |
237 | extern int sysctl_tcp_max_orphans; | |
238 | extern int sysctl_tcp_max_tw_buckets; | |
239 | extern int sysctl_tcp_fack; | |
240 | extern int sysctl_tcp_reordering; | |
241 | extern int sysctl_tcp_ecn; | |
242 | extern int sysctl_tcp_dsack; | |
243 | extern int sysctl_tcp_mem[3]; | |
244 | extern int sysctl_tcp_wmem[3]; | |
245 | extern int sysctl_tcp_rmem[3]; | |
246 | extern int sysctl_tcp_app_win; | |
247 | extern int sysctl_tcp_adv_win_scale; | |
248 | extern int sysctl_tcp_tw_reuse; | |
249 | extern int sysctl_tcp_frto; | |
250 | extern int sysctl_tcp_low_latency; | |
1da177e4 | 251 | extern int sysctl_tcp_nometrics_save; |
1da177e4 LT |
252 | extern int sysctl_tcp_moderate_rcvbuf; |
253 | extern int sysctl_tcp_tso_win_divisor; | |
254 | ||
255 | extern atomic_t tcp_memory_allocated; | |
256 | extern atomic_t tcp_sockets_allocated; | |
257 | extern int tcp_memory_pressure; | |
258 | ||
1da177e4 LT |
259 | #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) |
260 | #define TCP_INET_FAMILY(fam) ((fam) == AF_INET) | |
261 | #else | |
262 | #define TCP_INET_FAMILY(fam) 1 | |
263 | #endif | |
264 | ||
265 | /* | |
266 | * Pointers to address related TCP functions | |
267 | * (i.e. things that depend on the address family) | |
268 | */ | |
269 | ||
270 | struct tcp_func { | |
271 | int (*queue_xmit) (struct sk_buff *skb, | |
272 | int ipfragok); | |
273 | ||
274 | void (*send_check) (struct sock *sk, | |
275 | struct tcphdr *th, | |
276 | int len, | |
277 | struct sk_buff *skb); | |
278 | ||
279 | int (*rebuild_header) (struct sock *sk); | |
280 | ||
281 | int (*conn_request) (struct sock *sk, | |
282 | struct sk_buff *skb); | |
283 | ||
284 | struct sock * (*syn_recv_sock) (struct sock *sk, | |
285 | struct sk_buff *skb, | |
60236fdd | 286 | struct request_sock *req, |
1da177e4 LT |
287 | struct dst_entry *dst); |
288 | ||
289 | int (*remember_stamp) (struct sock *sk); | |
290 | ||
291 | __u16 net_header_len; | |
292 | ||
293 | int (*setsockopt) (struct sock *sk, | |
294 | int level, | |
295 | int optname, | |
296 | char __user *optval, | |
297 | int optlen); | |
298 | ||
299 | int (*getsockopt) (struct sock *sk, | |
300 | int level, | |
301 | int optname, | |
302 | char __user *optval, | |
303 | int __user *optlen); | |
304 | ||
305 | ||
306 | void (*addr2sockaddr) (struct sock *sk, | |
307 | struct sockaddr *); | |
308 | ||
309 | int sockaddr_len; | |
310 | }; | |
311 | ||
312 | /* | |
313 | * The next routines deal with comparing 32 bit unsigned ints | |
314 | * and worry about wraparound (automatic with unsigned arithmetic). | |
315 | */ | |
316 | ||
317 | static inline int before(__u32 seq1, __u32 seq2) | |
318 | { | |
319 | return (__s32)(seq1-seq2) < 0; | |
320 | } | |
321 | ||
322 | static inline int after(__u32 seq1, __u32 seq2) | |
323 | { | |
324 | return (__s32)(seq2-seq1) < 0; | |
325 | } | |
326 | ||
327 | ||
328 | /* is s2<=s1<=s3 ? */ | |
329 | static inline int between(__u32 seq1, __u32 seq2, __u32 seq3) | |
330 | { | |
331 | return seq3 - seq2 >= seq1 - seq2; | |
332 | } | |
333 | ||
334 | ||
335 | extern struct proto tcp_prot; | |
336 | ||
337 | DECLARE_SNMP_STAT(struct tcp_mib, tcp_statistics); | |
338 | #define TCP_INC_STATS(field) SNMP_INC_STATS(tcp_statistics, field) | |
339 | #define TCP_INC_STATS_BH(field) SNMP_INC_STATS_BH(tcp_statistics, field) | |
340 | #define TCP_INC_STATS_USER(field) SNMP_INC_STATS_USER(tcp_statistics, field) | |
341 | #define TCP_DEC_STATS(field) SNMP_DEC_STATS(tcp_statistics, field) | |
342 | #define TCP_ADD_STATS_BH(field, val) SNMP_ADD_STATS_BH(tcp_statistics, field, val) | |
343 | #define TCP_ADD_STATS_USER(field, val) SNMP_ADD_STATS_USER(tcp_statistics, field, val) | |
344 | ||
1da177e4 LT |
345 | extern void tcp_v4_err(struct sk_buff *skb, u32); |
346 | ||
347 | extern void tcp_shutdown (struct sock *sk, int how); | |
348 | ||
349 | extern int tcp_v4_rcv(struct sk_buff *skb); | |
350 | ||
351 | extern int tcp_v4_remember_stamp(struct sock *sk); | |
352 | ||
8feaf0c0 | 353 | extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw); |
1da177e4 LT |
354 | |
355 | extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, | |
356 | struct msghdr *msg, size_t size); | |
357 | extern ssize_t tcp_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags); | |
358 | ||
359 | extern int tcp_ioctl(struct sock *sk, | |
360 | int cmd, | |
361 | unsigned long arg); | |
362 | ||
363 | extern int tcp_rcv_state_process(struct sock *sk, | |
364 | struct sk_buff *skb, | |
365 | struct tcphdr *th, | |
366 | unsigned len); | |
367 | ||
368 | extern int tcp_rcv_established(struct sock *sk, | |
369 | struct sk_buff *skb, | |
370 | struct tcphdr *th, | |
371 | unsigned len); | |
372 | ||
373 | extern void tcp_rcv_space_adjust(struct sock *sk); | |
374 | ||
375 | enum tcp_ack_state_t | |
376 | { | |
377 | TCP_ACK_SCHED = 1, | |
378 | TCP_ACK_TIMER = 2, | |
379 | TCP_ACK_PUSHED= 4 | |
380 | }; | |
381 | ||
382 | static inline void tcp_schedule_ack(struct tcp_sock *tp) | |
383 | { | |
384 | tp->ack.pending |= TCP_ACK_SCHED; | |
385 | } | |
386 | ||
387 | static inline int tcp_ack_scheduled(struct tcp_sock *tp) | |
388 | { | |
389 | return tp->ack.pending&TCP_ACK_SCHED; | |
390 | } | |
391 | ||
fc6415bc | 392 | static __inline__ void tcp_dec_quickack_mode(struct tcp_sock *tp, unsigned int pkts) |
1da177e4 | 393 | { |
fc6415bc DM |
394 | if (tp->ack.quick) { |
395 | if (pkts >= tp->ack.quick) { | |
396 | tp->ack.quick = 0; | |
397 | ||
398 | /* Leaving quickack mode we deflate ATO. */ | |
399 | tp->ack.ato = TCP_ATO_MIN; | |
400 | } else | |
401 | tp->ack.quick -= pkts; | |
1da177e4 LT |
402 | } |
403 | } | |
404 | ||
405 | extern void tcp_enter_quickack_mode(struct tcp_sock *tp); | |
406 | ||
407 | static __inline__ void tcp_delack_init(struct tcp_sock *tp) | |
408 | { | |
409 | memset(&tp->ack, 0, sizeof(tp->ack)); | |
410 | } | |
411 | ||
412 | static inline void tcp_clear_options(struct tcp_options_received *rx_opt) | |
413 | { | |
414 | rx_opt->tstamp_ok = rx_opt->sack_ok = rx_opt->wscale_ok = rx_opt->snd_wscale = 0; | |
415 | } | |
416 | ||
417 | enum tcp_tw_status | |
418 | { | |
419 | TCP_TW_SUCCESS = 0, | |
420 | TCP_TW_RST = 1, | |
421 | TCP_TW_ACK = 2, | |
422 | TCP_TW_SYN = 3 | |
423 | }; | |
424 | ||
425 | ||
8feaf0c0 | 426 | extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw, |
1da177e4 | 427 | struct sk_buff *skb, |
8feaf0c0 | 428 | const struct tcphdr *th); |
1da177e4 LT |
429 | |
430 | extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb, | |
60236fdd ACM |
431 | struct request_sock *req, |
432 | struct request_sock **prev); | |
1da177e4 LT |
433 | extern int tcp_child_process(struct sock *parent, |
434 | struct sock *child, | |
435 | struct sk_buff *skb); | |
436 | extern void tcp_enter_frto(struct sock *sk); | |
437 | extern void tcp_enter_loss(struct sock *sk, int how); | |
438 | extern void tcp_clear_retrans(struct tcp_sock *tp); | |
439 | extern void tcp_update_metrics(struct sock *sk); | |
440 | ||
441 | extern void tcp_close(struct sock *sk, | |
442 | long timeout); | |
443 | extern struct sock * tcp_accept(struct sock *sk, int flags, int *err); | |
444 | extern unsigned int tcp_poll(struct file * file, struct socket *sock, struct poll_table_struct *wait); | |
445 | ||
446 | extern int tcp_getsockopt(struct sock *sk, int level, | |
447 | int optname, | |
448 | char __user *optval, | |
449 | int __user *optlen); | |
450 | extern int tcp_setsockopt(struct sock *sk, int level, | |
451 | int optname, char __user *optval, | |
452 | int optlen); | |
453 | extern void tcp_set_keepalive(struct sock *sk, int val); | |
454 | extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, | |
455 | struct msghdr *msg, | |
456 | size_t len, int nonblock, | |
457 | int flags, int *addr_len); | |
458 | ||
459 | extern int tcp_listen_start(struct sock *sk); | |
460 | ||
461 | extern void tcp_parse_options(struct sk_buff *skb, | |
462 | struct tcp_options_received *opt_rx, | |
463 | int estab); | |
464 | ||
465 | /* | |
466 | * TCP v4 functions exported for the inet6 API | |
467 | */ | |
468 | ||
1da177e4 LT |
469 | extern void tcp_v4_send_check(struct sock *sk, |
470 | struct tcphdr *th, int len, | |
471 | struct sk_buff *skb); | |
472 | ||
473 | extern int tcp_v4_conn_request(struct sock *sk, | |
474 | struct sk_buff *skb); | |
475 | ||
476 | extern struct sock * tcp_create_openreq_child(struct sock *sk, | |
60236fdd | 477 | struct request_sock *req, |
1da177e4 LT |
478 | struct sk_buff *skb); |
479 | ||
480 | extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, | |
481 | struct sk_buff *skb, | |
60236fdd | 482 | struct request_sock *req, |
1da177e4 LT |
483 | struct dst_entry *dst); |
484 | ||
485 | extern int tcp_v4_do_rcv(struct sock *sk, | |
486 | struct sk_buff *skb); | |
487 | ||
488 | extern int tcp_v4_connect(struct sock *sk, | |
489 | struct sockaddr *uaddr, | |
490 | int addr_len); | |
491 | ||
492 | extern int tcp_connect(struct sock *sk); | |
493 | ||
494 | extern struct sk_buff * tcp_make_synack(struct sock *sk, | |
495 | struct dst_entry *dst, | |
60236fdd | 496 | struct request_sock *req); |
1da177e4 LT |
497 | |
498 | extern int tcp_disconnect(struct sock *sk, int flags); | |
499 | ||
500 | extern void tcp_unhash(struct sock *sk); | |
501 | ||
502 | extern int tcp_v4_hash_connecting(struct sock *sk); | |
503 | ||
504 | ||
505 | /* From syncookies.c */ | |
506 | extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb, | |
507 | struct ip_options *opt); | |
508 | extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb, | |
509 | __u16 *mss); | |
510 | ||
511 | /* tcp_output.c */ | |
512 | ||
f6302d1d | 513 | extern void __tcp_push_pending_frames(struct sock *sk, struct tcp_sock *tp, |
a2e2a59c | 514 | unsigned int cur_mss, int nonagle); |
f6302d1d | 515 | extern int tcp_may_send_now(struct sock *sk, struct tcp_sock *tp); |
1da177e4 LT |
516 | extern int tcp_retransmit_skb(struct sock *, struct sk_buff *); |
517 | extern void tcp_xmit_retransmit_queue(struct sock *); | |
518 | extern void tcp_simple_retransmit(struct sock *); | |
519 | extern int tcp_trim_head(struct sock *, struct sk_buff *, u32); | |
520 | ||
521 | extern void tcp_send_probe0(struct sock *); | |
522 | extern void tcp_send_partial(struct sock *); | |
523 | extern int tcp_write_wakeup(struct sock *); | |
524 | extern void tcp_send_fin(struct sock *sk); | |
86a76caf VF |
525 | extern void tcp_send_active_reset(struct sock *sk, |
526 | unsigned int __nocast priority); | |
1da177e4 | 527 | extern int tcp_send_synack(struct sock *); |
c1b4a7e6 | 528 | extern void tcp_push_one(struct sock *, unsigned int mss_now); |
1da177e4 LT |
529 | extern void tcp_send_ack(struct sock *sk); |
530 | extern void tcp_send_delayed_ack(struct sock *sk); | |
531 | ||
a762a980 DM |
532 | /* tcp_input.c */ |
533 | extern void tcp_cwnd_application_limited(struct sock *sk); | |
534 | ||
1da177e4 LT |
535 | /* tcp_timer.c */ |
536 | extern void tcp_init_xmit_timers(struct sock *); | |
537 | extern void tcp_clear_xmit_timers(struct sock *); | |
538 | ||
539 | extern void tcp_delete_keepalive_timer(struct sock *); | |
540 | extern void tcp_reset_keepalive_timer(struct sock *, unsigned long); | |
541 | extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu); | |
542 | extern unsigned int tcp_current_mss(struct sock *sk, int large); | |
543 | ||
544 | #ifdef TCP_DEBUG | |
545 | extern const char tcp_timer_bug_msg[]; | |
546 | #endif | |
547 | ||
548 | /* tcp_diag.c */ | |
549 | extern void tcp_get_info(struct sock *, struct tcp_info *); | |
550 | ||
551 | /* Read 'sendfile()'-style from a TCP socket */ | |
552 | typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *, | |
553 | unsigned int, size_t); | |
554 | extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc, | |
555 | sk_read_actor_t recv_actor); | |
556 | ||
557 | static inline void tcp_clear_xmit_timer(struct sock *sk, int what) | |
558 | { | |
559 | struct tcp_sock *tp = tcp_sk(sk); | |
560 | ||
561 | switch (what) { | |
562 | case TCP_TIME_RETRANS: | |
563 | case TCP_TIME_PROBE0: | |
564 | tp->pending = 0; | |
565 | ||
566 | #ifdef TCP_CLEAR_TIMERS | |
567 | sk_stop_timer(sk, &tp->retransmit_timer); | |
568 | #endif | |
569 | break; | |
570 | case TCP_TIME_DACK: | |
571 | tp->ack.blocked = 0; | |
572 | tp->ack.pending = 0; | |
573 | ||
574 | #ifdef TCP_CLEAR_TIMERS | |
575 | sk_stop_timer(sk, &tp->delack_timer); | |
576 | #endif | |
577 | break; | |
578 | default: | |
579 | #ifdef TCP_DEBUG | |
580 | printk(tcp_timer_bug_msg); | |
581 | #endif | |
582 | return; | |
583 | }; | |
584 | ||
585 | } | |
586 | ||
587 | /* | |
588 | * Reset the retransmission timer | |
589 | */ | |
590 | static inline void tcp_reset_xmit_timer(struct sock *sk, int what, unsigned long when) | |
591 | { | |
592 | struct tcp_sock *tp = tcp_sk(sk); | |
593 | ||
594 | if (when > TCP_RTO_MAX) { | |
595 | #ifdef TCP_DEBUG | |
596 | printk(KERN_DEBUG "reset_xmit_timer sk=%p %d when=0x%lx, caller=%p\n", sk, what, when, current_text_addr()); | |
597 | #endif | |
598 | when = TCP_RTO_MAX; | |
599 | } | |
600 | ||
601 | switch (what) { | |
602 | case TCP_TIME_RETRANS: | |
603 | case TCP_TIME_PROBE0: | |
604 | tp->pending = what; | |
605 | tp->timeout = jiffies+when; | |
606 | sk_reset_timer(sk, &tp->retransmit_timer, tp->timeout); | |
607 | break; | |
608 | ||
609 | case TCP_TIME_DACK: | |
610 | tp->ack.pending |= TCP_ACK_TIMER; | |
611 | tp->ack.timeout = jiffies+when; | |
612 | sk_reset_timer(sk, &tp->delack_timer, tp->ack.timeout); | |
613 | break; | |
614 | ||
615 | default: | |
616 | #ifdef TCP_DEBUG | |
617 | printk(tcp_timer_bug_msg); | |
618 | #endif | |
619 | return; | |
620 | }; | |
621 | } | |
622 | ||
623 | /* Initialize RCV_MSS value. | |
624 | * RCV_MSS is an our guess about MSS used by the peer. | |
625 | * We haven't any direct information about the MSS. | |
626 | * It's better to underestimate the RCV_MSS rather than overestimate. | |
627 | * Overestimations make us ACKing less frequently than needed. | |
628 | * Underestimations are more easy to detect and fix by tcp_measure_rcv_mss(). | |
629 | */ | |
630 | ||
631 | static inline void tcp_initialize_rcv_mss(struct sock *sk) | |
632 | { | |
633 | struct tcp_sock *tp = tcp_sk(sk); | |
c1b4a7e6 | 634 | unsigned int hint = min_t(unsigned int, tp->advmss, tp->mss_cache); |
1da177e4 LT |
635 | |
636 | hint = min(hint, tp->rcv_wnd/2); | |
637 | hint = min(hint, TCP_MIN_RCVMSS); | |
638 | hint = max(hint, TCP_MIN_MSS); | |
639 | ||
640 | tp->ack.rcv_mss = hint; | |
641 | } | |
642 | ||
643 | static __inline__ void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd) | |
644 | { | |
645 | tp->pred_flags = htonl((tp->tcp_header_len << 26) | | |
646 | ntohl(TCP_FLAG_ACK) | | |
647 | snd_wnd); | |
648 | } | |
649 | ||
650 | static __inline__ void tcp_fast_path_on(struct tcp_sock *tp) | |
651 | { | |
652 | __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale); | |
653 | } | |
654 | ||
655 | static inline void tcp_fast_path_check(struct sock *sk, struct tcp_sock *tp) | |
656 | { | |
b03efcfb | 657 | if (skb_queue_empty(&tp->out_of_order_queue) && |
1da177e4 LT |
658 | tp->rcv_wnd && |
659 | atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf && | |
660 | !tp->urg_data) | |
661 | tcp_fast_path_on(tp); | |
662 | } | |
663 | ||
664 | /* Compute the actual receive window we are currently advertising. | |
665 | * Rcv_nxt can be after the window if our peer push more data | |
666 | * than the offered window. | |
667 | */ | |
668 | static __inline__ u32 tcp_receive_window(const struct tcp_sock *tp) | |
669 | { | |
670 | s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt; | |
671 | ||
672 | if (win < 0) | |
673 | win = 0; | |
674 | return (u32) win; | |
675 | } | |
676 | ||
677 | /* Choose a new window, without checks for shrinking, and without | |
678 | * scaling applied to the result. The caller does these things | |
679 | * if necessary. This is a "raw" window selection. | |
680 | */ | |
681 | extern u32 __tcp_select_window(struct sock *sk); | |
682 | ||
683 | /* TCP timestamps are only 32-bits, this causes a slight | |
684 | * complication on 64-bit systems since we store a snapshot | |
685 | * of jiffies in the buffer control blocks below. We decidely | |
686 | * only use of the low 32-bits of jiffies and hide the ugly | |
687 | * casts with the following macro. | |
688 | */ | |
689 | #define tcp_time_stamp ((__u32)(jiffies)) | |
690 | ||
691 | /* This is what the send packet queueing engine uses to pass | |
692 | * TCP per-packet control information to the transmission | |
693 | * code. We also store the host-order sequence numbers in | |
694 | * here too. This is 36 bytes on 32-bit architectures, | |
695 | * 40 bytes on 64-bit machines, if this grows please adjust | |
696 | * skbuff.h:skbuff->cb[xxx] size appropriately. | |
697 | */ | |
698 | struct tcp_skb_cb { | |
699 | union { | |
700 | struct inet_skb_parm h4; | |
701 | #if defined(CONFIG_IPV6) || defined (CONFIG_IPV6_MODULE) | |
702 | struct inet6_skb_parm h6; | |
703 | #endif | |
704 | } header; /* For incoming frames */ | |
705 | __u32 seq; /* Starting sequence number */ | |
706 | __u32 end_seq; /* SEQ + FIN + SYN + datalen */ | |
707 | __u32 when; /* used to compute rtt's */ | |
708 | __u8 flags; /* TCP header flags. */ | |
709 | ||
710 | /* NOTE: These must match up to the flags byte in a | |
711 | * real TCP header. | |
712 | */ | |
713 | #define TCPCB_FLAG_FIN 0x01 | |
714 | #define TCPCB_FLAG_SYN 0x02 | |
715 | #define TCPCB_FLAG_RST 0x04 | |
716 | #define TCPCB_FLAG_PSH 0x08 | |
717 | #define TCPCB_FLAG_ACK 0x10 | |
718 | #define TCPCB_FLAG_URG 0x20 | |
719 | #define TCPCB_FLAG_ECE 0x40 | |
720 | #define TCPCB_FLAG_CWR 0x80 | |
721 | ||
722 | __u8 sacked; /* State flags for SACK/FACK. */ | |
723 | #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */ | |
724 | #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */ | |
725 | #define TCPCB_LOST 0x04 /* SKB is lost */ | |
726 | #define TCPCB_TAGBITS 0x07 /* All tag bits */ | |
727 | ||
728 | #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */ | |
729 | #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS) | |
730 | ||
731 | #define TCPCB_URG 0x20 /* Urgent pointer advenced here */ | |
732 | ||
733 | #define TCPCB_AT_TAIL (TCPCB_URG) | |
734 | ||
735 | __u16 urg_ptr; /* Valid w/URG flags is set. */ | |
736 | __u32 ack_seq; /* Sequence number ACK'd */ | |
737 | }; | |
738 | ||
739 | #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0])) | |
740 | ||
741 | #include <net/tcp_ecn.h> | |
742 | ||
743 | /* Due to TSO, an SKB can be composed of multiple actual | |
744 | * packets. To keep these tracked properly, we use this. | |
745 | */ | |
746 | static inline int tcp_skb_pcount(const struct sk_buff *skb) | |
747 | { | |
748 | return skb_shinfo(skb)->tso_segs; | |
749 | } | |
750 | ||
751 | /* This is valid iff tcp_skb_pcount() > 1. */ | |
752 | static inline int tcp_skb_mss(const struct sk_buff *skb) | |
753 | { | |
754 | return skb_shinfo(skb)->tso_size; | |
755 | } | |
756 | ||
757 | static inline void tcp_dec_pcount_approx(__u32 *count, | |
758 | const struct sk_buff *skb) | |
759 | { | |
760 | if (*count) { | |
761 | *count -= tcp_skb_pcount(skb); | |
762 | if ((int)*count < 0) | |
763 | *count = 0; | |
764 | } | |
765 | } | |
766 | ||
767 | static inline void tcp_packets_out_inc(struct sock *sk, | |
768 | struct tcp_sock *tp, | |
769 | const struct sk_buff *skb) | |
770 | { | |
771 | int orig = tp->packets_out; | |
772 | ||
773 | tp->packets_out += tcp_skb_pcount(skb); | |
774 | if (!orig) | |
775 | tcp_reset_xmit_timer(sk, TCP_TIME_RETRANS, tp->rto); | |
776 | } | |
777 | ||
778 | static inline void tcp_packets_out_dec(struct tcp_sock *tp, | |
779 | const struct sk_buff *skb) | |
780 | { | |
781 | tp->packets_out -= tcp_skb_pcount(skb); | |
782 | } | |
783 | ||
317a76f9 SH |
784 | /* Events passed to congestion control interface */ |
785 | enum tcp_ca_event { | |
786 | CA_EVENT_TX_START, /* first transmit when no packets in flight */ | |
787 | CA_EVENT_CWND_RESTART, /* congestion window restart */ | |
788 | CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */ | |
789 | CA_EVENT_FRTO, /* fast recovery timeout */ | |
790 | CA_EVENT_LOSS, /* loss timeout */ | |
791 | CA_EVENT_FAST_ACK, /* in sequence ack */ | |
792 | CA_EVENT_SLOW_ACK, /* other ack */ | |
793 | }; | |
794 | ||
795 | /* | |
796 | * Interface for adding new TCP congestion control handlers | |
797 | */ | |
798 | #define TCP_CA_NAME_MAX 16 | |
799 | struct tcp_congestion_ops { | |
800 | struct list_head list; | |
801 | ||
802 | /* initialize private data (optional) */ | |
803 | void (*init)(struct tcp_sock *tp); | |
804 | /* cleanup private data (optional) */ | |
805 | void (*release)(struct tcp_sock *tp); | |
806 | ||
807 | /* return slow start threshold (required) */ | |
808 | u32 (*ssthresh)(struct tcp_sock *tp); | |
809 | /* lower bound for congestion window (optional) */ | |
810 | u32 (*min_cwnd)(struct tcp_sock *tp); | |
811 | /* do new cwnd calculation (required) */ | |
812 | void (*cong_avoid)(struct tcp_sock *tp, u32 ack, | |
813 | u32 rtt, u32 in_flight, int good_ack); | |
814 | /* round trip time sample per acked packet (optional) */ | |
815 | void (*rtt_sample)(struct tcp_sock *tp, u32 usrtt); | |
816 | /* call before changing ca_state (optional) */ | |
817 | void (*set_state)(struct tcp_sock *tp, u8 new_state); | |
818 | /* call when cwnd event occurs (optional) */ | |
819 | void (*cwnd_event)(struct tcp_sock *tp, enum tcp_ca_event ev); | |
820 | /* new value of cwnd after loss (optional) */ | |
821 | u32 (*undo_cwnd)(struct tcp_sock *tp); | |
822 | /* hook for packet ack accounting (optional) */ | |
823 | void (*pkts_acked)(struct tcp_sock *tp, u32 num_acked); | |
824 | /* get info for tcp_diag (optional) */ | |
825 | void (*get_info)(struct tcp_sock *tp, u32 ext, struct sk_buff *skb); | |
826 | ||
827 | char name[TCP_CA_NAME_MAX]; | |
828 | struct module *owner; | |
829 | }; | |
830 | ||
831 | extern int tcp_register_congestion_control(struct tcp_congestion_ops *type); | |
832 | extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type); | |
833 | ||
834 | extern void tcp_init_congestion_control(struct tcp_sock *tp); | |
835 | extern void tcp_cleanup_congestion_control(struct tcp_sock *tp); | |
836 | extern int tcp_set_default_congestion_control(const char *name); | |
837 | extern void tcp_get_default_congestion_control(char *name); | |
5f8ef48d | 838 | extern int tcp_set_congestion_control(struct tcp_sock *tp, const char *name); |
317a76f9 | 839 | |
5f8ef48d | 840 | extern struct tcp_congestion_ops tcp_init_congestion_ops; |
317a76f9 SH |
841 | extern u32 tcp_reno_ssthresh(struct tcp_sock *tp); |
842 | extern void tcp_reno_cong_avoid(struct tcp_sock *tp, u32 ack, | |
843 | u32 rtt, u32 in_flight, int flag); | |
844 | extern u32 tcp_reno_min_cwnd(struct tcp_sock *tp); | |
a8acfbac | 845 | extern struct tcp_congestion_ops tcp_reno; |
317a76f9 SH |
846 | |
847 | static inline void tcp_set_ca_state(struct tcp_sock *tp, u8 ca_state) | |
848 | { | |
849 | if (tp->ca_ops->set_state) | |
850 | tp->ca_ops->set_state(tp, ca_state); | |
851 | tp->ca_state = ca_state; | |
852 | } | |
853 | ||
854 | static inline void tcp_ca_event(struct tcp_sock *tp, enum tcp_ca_event event) | |
855 | { | |
856 | if (tp->ca_ops->cwnd_event) | |
857 | tp->ca_ops->cwnd_event(tp, event); | |
858 | } | |
859 | ||
1da177e4 LT |
860 | /* This determines how many packets are "in the network" to the best |
861 | * of our knowledge. In many cases it is conservative, but where | |
862 | * detailed information is available from the receiver (via SACK | |
863 | * blocks etc.) we can make more aggressive calculations. | |
864 | * | |
865 | * Use this for decisions involving congestion control, use just | |
866 | * tp->packets_out to determine if the send queue is empty or not. | |
867 | * | |
868 | * Read this equation as: | |
869 | * | |
870 | * "Packets sent once on transmission queue" MINUS | |
871 | * "Packets left network, but not honestly ACKed yet" PLUS | |
872 | * "Packets fast retransmitted" | |
873 | */ | |
874 | static __inline__ unsigned int tcp_packets_in_flight(const struct tcp_sock *tp) | |
875 | { | |
876 | return (tp->packets_out - tp->left_out + tp->retrans_out); | |
877 | } | |
878 | ||
1da177e4 LT |
879 | /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd. |
880 | * The exception is rate halving phase, when cwnd is decreasing towards | |
881 | * ssthresh. | |
882 | */ | |
883 | static inline __u32 tcp_current_ssthresh(struct tcp_sock *tp) | |
884 | { | |
885 | if ((1<<tp->ca_state)&(TCPF_CA_CWR|TCPF_CA_Recovery)) | |
886 | return tp->snd_ssthresh; | |
887 | else | |
888 | return max(tp->snd_ssthresh, | |
889 | ((tp->snd_cwnd >> 1) + | |
890 | (tp->snd_cwnd >> 2))); | |
891 | } | |
892 | ||
893 | static inline void tcp_sync_left_out(struct tcp_sock *tp) | |
894 | { | |
895 | if (tp->rx_opt.sack_ok && | |
896 | (tp->sacked_out >= tp->packets_out - tp->lost_out)) | |
897 | tp->sacked_out = tp->packets_out - tp->lost_out; | |
898 | tp->left_out = tp->sacked_out + tp->lost_out; | |
899 | } | |
900 | ||
d1b04c08 | 901 | /* Set slow start threshold and cwnd not falling to slow start */ |
1da177e4 LT |
902 | static inline void __tcp_enter_cwr(struct tcp_sock *tp) |
903 | { | |
904 | tp->undo_marker = 0; | |
317a76f9 | 905 | tp->snd_ssthresh = tp->ca_ops->ssthresh(tp); |
1da177e4 LT |
906 | tp->snd_cwnd = min(tp->snd_cwnd, |
907 | tcp_packets_in_flight(tp) + 1U); | |
908 | tp->snd_cwnd_cnt = 0; | |
909 | tp->high_seq = tp->snd_nxt; | |
910 | tp->snd_cwnd_stamp = tcp_time_stamp; | |
911 | TCP_ECN_queue_cwr(tp); | |
912 | } | |
913 | ||
914 | static inline void tcp_enter_cwr(struct tcp_sock *tp) | |
915 | { | |
916 | tp->prior_ssthresh = 0; | |
917 | if (tp->ca_state < TCP_CA_CWR) { | |
918 | __tcp_enter_cwr(tp); | |
919 | tcp_set_ca_state(tp, TCP_CA_CWR); | |
920 | } | |
921 | } | |
922 | ||
923 | extern __u32 tcp_init_cwnd(struct tcp_sock *tp, struct dst_entry *dst); | |
924 | ||
925 | /* Slow start with delack produces 3 packets of burst, so that | |
926 | * it is safe "de facto". | |
927 | */ | |
928 | static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp) | |
929 | { | |
930 | return 3; | |
931 | } | |
932 | ||
1da177e4 LT |
933 | static __inline__ void tcp_minshall_update(struct tcp_sock *tp, int mss, |
934 | const struct sk_buff *skb) | |
935 | { | |
936 | if (skb->len < mss) | |
937 | tp->snd_sml = TCP_SKB_CB(skb)->end_seq; | |
938 | } | |
939 | ||
1da177e4 LT |
940 | static __inline__ void tcp_check_probe_timer(struct sock *sk, struct tcp_sock *tp) |
941 | { | |
942 | if (!tp->packets_out && !tp->pending) | |
943 | tcp_reset_xmit_timer(sk, TCP_TIME_PROBE0, tp->rto); | |
944 | } | |
945 | ||
1da177e4 LT |
946 | static __inline__ void tcp_push_pending_frames(struct sock *sk, |
947 | struct tcp_sock *tp) | |
948 | { | |
949 | __tcp_push_pending_frames(sk, tp, tcp_current_mss(sk, 1), tp->nonagle); | |
950 | } | |
951 | ||
1da177e4 LT |
952 | static __inline__ void tcp_init_wl(struct tcp_sock *tp, u32 ack, u32 seq) |
953 | { | |
954 | tp->snd_wl1 = seq; | |
955 | } | |
956 | ||
957 | static __inline__ void tcp_update_wl(struct tcp_sock *tp, u32 ack, u32 seq) | |
958 | { | |
959 | tp->snd_wl1 = seq; | |
960 | } | |
961 | ||
962 | extern void tcp_destroy_sock(struct sock *sk); | |
963 | ||
964 | ||
965 | /* | |
966 | * Calculate(/check) TCP checksum | |
967 | */ | |
968 | static __inline__ u16 tcp_v4_check(struct tcphdr *th, int len, | |
969 | unsigned long saddr, unsigned long daddr, | |
970 | unsigned long base) | |
971 | { | |
972 | return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base); | |
973 | } | |
974 | ||
975 | static __inline__ int __tcp_checksum_complete(struct sk_buff *skb) | |
976 | { | |
977 | return (unsigned short)csum_fold(skb_checksum(skb, 0, skb->len, skb->csum)); | |
978 | } | |
979 | ||
980 | static __inline__ int tcp_checksum_complete(struct sk_buff *skb) | |
981 | { | |
982 | return skb->ip_summed != CHECKSUM_UNNECESSARY && | |
983 | __tcp_checksum_complete(skb); | |
984 | } | |
985 | ||
986 | /* Prequeue for VJ style copy to user, combined with checksumming. */ | |
987 | ||
988 | static __inline__ void tcp_prequeue_init(struct tcp_sock *tp) | |
989 | { | |
990 | tp->ucopy.task = NULL; | |
991 | tp->ucopy.len = 0; | |
992 | tp->ucopy.memory = 0; | |
993 | skb_queue_head_init(&tp->ucopy.prequeue); | |
994 | } | |
995 | ||
996 | /* Packet is added to VJ-style prequeue for processing in process | |
997 | * context, if a reader task is waiting. Apparently, this exciting | |
998 | * idea (VJ's mail "Re: query about TCP header on tcp-ip" of 07 Sep 93) | |
999 | * failed somewhere. Latency? Burstiness? Well, at least now we will | |
1000 | * see, why it failed. 8)8) --ANK | |
1001 | * | |
1002 | * NOTE: is this not too big to inline? | |
1003 | */ | |
1004 | static __inline__ int tcp_prequeue(struct sock *sk, struct sk_buff *skb) | |
1005 | { | |
1006 | struct tcp_sock *tp = tcp_sk(sk); | |
1007 | ||
1008 | if (!sysctl_tcp_low_latency && tp->ucopy.task) { | |
1009 | __skb_queue_tail(&tp->ucopy.prequeue, skb); | |
1010 | tp->ucopy.memory += skb->truesize; | |
1011 | if (tp->ucopy.memory > sk->sk_rcvbuf) { | |
1012 | struct sk_buff *skb1; | |
1013 | ||
1014 | BUG_ON(sock_owned_by_user(sk)); | |
1015 | ||
1016 | while ((skb1 = __skb_dequeue(&tp->ucopy.prequeue)) != NULL) { | |
1017 | sk->sk_backlog_rcv(sk, skb1); | |
1018 | NET_INC_STATS_BH(LINUX_MIB_TCPPREQUEUEDROPPED); | |
1019 | } | |
1020 | ||
1021 | tp->ucopy.memory = 0; | |
1022 | } else if (skb_queue_len(&tp->ucopy.prequeue) == 1) { | |
1023 | wake_up_interruptible(sk->sk_sleep); | |
1024 | if (!tcp_ack_scheduled(tp)) | |
1025 | tcp_reset_xmit_timer(sk, TCP_TIME_DACK, (3*TCP_RTO_MIN)/4); | |
1026 | } | |
1027 | return 1; | |
1028 | } | |
1029 | return 0; | |
1030 | } | |
1031 | ||
1032 | ||
1033 | #undef STATE_TRACE | |
1034 | ||
1035 | #ifdef STATE_TRACE | |
1036 | static const char *statename[]={ | |
1037 | "Unused","Established","Syn Sent","Syn Recv", | |
1038 | "Fin Wait 1","Fin Wait 2","Time Wait", "Close", | |
1039 | "Close Wait","Last ACK","Listen","Closing" | |
1040 | }; | |
1041 | #endif | |
1042 | ||
1043 | static __inline__ void tcp_set_state(struct sock *sk, int state) | |
1044 | { | |
1045 | int oldstate = sk->sk_state; | |
1046 | ||
1047 | switch (state) { | |
1048 | case TCP_ESTABLISHED: | |
1049 | if (oldstate != TCP_ESTABLISHED) | |
1050 | TCP_INC_STATS(TCP_MIB_CURRESTAB); | |
1051 | break; | |
1052 | ||
1053 | case TCP_CLOSE: | |
1054 | if (oldstate == TCP_CLOSE_WAIT || oldstate == TCP_ESTABLISHED) | |
1055 | TCP_INC_STATS(TCP_MIB_ESTABRESETS); | |
1056 | ||
1057 | sk->sk_prot->unhash(sk); | |
a55ebcc4 | 1058 | if (inet_sk(sk)->bind_hash && |
1da177e4 | 1059 | !(sk->sk_userlocks & SOCK_BINDPORT_LOCK)) |
2d8c4ce5 | 1060 | inet_put_port(&tcp_hashinfo, sk); |
1da177e4 LT |
1061 | /* fall through */ |
1062 | default: | |
1063 | if (oldstate==TCP_ESTABLISHED) | |
1064 | TCP_DEC_STATS(TCP_MIB_CURRESTAB); | |
1065 | } | |
1066 | ||
1067 | /* Change state AFTER socket is unhashed to avoid closed | |
1068 | * socket sitting in hash tables. | |
1069 | */ | |
1070 | sk->sk_state = state; | |
1071 | ||
1072 | #ifdef STATE_TRACE | |
1073 | SOCK_DEBUG(sk, "TCP sk=%p, State %s -> %s\n",sk, statename[oldstate],statename[state]); | |
1074 | #endif | |
1075 | } | |
1076 | ||
1077 | static __inline__ void tcp_done(struct sock *sk) | |
1078 | { | |
1079 | tcp_set_state(sk, TCP_CLOSE); | |
1080 | tcp_clear_xmit_timers(sk); | |
1081 | ||
1082 | sk->sk_shutdown = SHUTDOWN_MASK; | |
1083 | ||
1084 | if (!sock_flag(sk, SOCK_DEAD)) | |
1085 | sk->sk_state_change(sk); | |
1086 | else | |
1087 | tcp_destroy_sock(sk); | |
1088 | } | |
1089 | ||
1090 | static __inline__ void tcp_sack_reset(struct tcp_options_received *rx_opt) | |
1091 | { | |
1092 | rx_opt->dsack = 0; | |
1093 | rx_opt->eff_sacks = 0; | |
1094 | rx_opt->num_sacks = 0; | |
1095 | } | |
1096 | ||
1097 | static __inline__ void tcp_build_and_update_options(__u32 *ptr, struct tcp_sock *tp, __u32 tstamp) | |
1098 | { | |
1099 | if (tp->rx_opt.tstamp_ok) { | |
1100 | *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | | |
1101 | (TCPOPT_NOP << 16) | | |
1102 | (TCPOPT_TIMESTAMP << 8) | | |
1103 | TCPOLEN_TIMESTAMP); | |
1104 | *ptr++ = htonl(tstamp); | |
1105 | *ptr++ = htonl(tp->rx_opt.ts_recent); | |
1106 | } | |
1107 | if (tp->rx_opt.eff_sacks) { | |
1108 | struct tcp_sack_block *sp = tp->rx_opt.dsack ? tp->duplicate_sack : tp->selective_acks; | |
1109 | int this_sack; | |
1110 | ||
1111 | *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | | |
1112 | (TCPOPT_NOP << 16) | | |
1113 | (TCPOPT_SACK << 8) | | |
1114 | (TCPOLEN_SACK_BASE + | |
1115 | (tp->rx_opt.eff_sacks * TCPOLEN_SACK_PERBLOCK))); | |
1116 | for(this_sack = 0; this_sack < tp->rx_opt.eff_sacks; this_sack++) { | |
1117 | *ptr++ = htonl(sp[this_sack].start_seq); | |
1118 | *ptr++ = htonl(sp[this_sack].end_seq); | |
1119 | } | |
1120 | if (tp->rx_opt.dsack) { | |
1121 | tp->rx_opt.dsack = 0; | |
1122 | tp->rx_opt.eff_sacks--; | |
1123 | } | |
1124 | } | |
1125 | } | |
1126 | ||
1127 | /* Construct a tcp options header for a SYN or SYN_ACK packet. | |
1128 | * If this is every changed make sure to change the definition of | |
1129 | * MAX_SYN_SIZE to match the new maximum number of options that you | |
1130 | * can generate. | |
1131 | */ | |
1132 | static inline void tcp_syn_build_options(__u32 *ptr, int mss, int ts, int sack, | |
1133 | int offer_wscale, int wscale, __u32 tstamp, __u32 ts_recent) | |
1134 | { | |
1135 | /* We always get an MSS option. | |
1136 | * The option bytes which will be seen in normal data | |
1137 | * packets should timestamps be used, must be in the MSS | |
1138 | * advertised. But we subtract them from tp->mss_cache so | |
1139 | * that calculations in tcp_sendmsg are simpler etc. | |
1140 | * So account for this fact here if necessary. If we | |
1141 | * don't do this correctly, as a receiver we won't | |
1142 | * recognize data packets as being full sized when we | |
1143 | * should, and thus we won't abide by the delayed ACK | |
1144 | * rules correctly. | |
1145 | * SACKs don't matter, we never delay an ACK when we | |
1146 | * have any of those going out. | |
1147 | */ | |
1148 | *ptr++ = htonl((TCPOPT_MSS << 24) | (TCPOLEN_MSS << 16) | mss); | |
1149 | if (ts) { | |
1150 | if(sack) | |
1151 | *ptr++ = __constant_htonl((TCPOPT_SACK_PERM << 24) | (TCPOLEN_SACK_PERM << 16) | | |
1152 | (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP); | |
1153 | else | |
1154 | *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | | |
1155 | (TCPOPT_TIMESTAMP << 8) | TCPOLEN_TIMESTAMP); | |
1156 | *ptr++ = htonl(tstamp); /* TSVAL */ | |
1157 | *ptr++ = htonl(ts_recent); /* TSECR */ | |
1158 | } else if(sack) | |
1159 | *ptr++ = __constant_htonl((TCPOPT_NOP << 24) | (TCPOPT_NOP << 16) | | |
1160 | (TCPOPT_SACK_PERM << 8) | TCPOLEN_SACK_PERM); | |
1161 | if (offer_wscale) | |
1162 | *ptr++ = htonl((TCPOPT_NOP << 24) | (TCPOPT_WINDOW << 16) | (TCPOLEN_WINDOW << 8) | (wscale)); | |
1163 | } | |
1164 | ||
1165 | /* Determine a window scaling and initial window to offer. */ | |
1166 | extern void tcp_select_initial_window(int __space, __u32 mss, | |
1167 | __u32 *rcv_wnd, __u32 *window_clamp, | |
1168 | int wscale_ok, __u8 *rcv_wscale); | |
1169 | ||
1170 | static inline int tcp_win_from_space(int space) | |
1171 | { | |
1172 | return sysctl_tcp_adv_win_scale<=0 ? | |
1173 | (space>>(-sysctl_tcp_adv_win_scale)) : | |
1174 | space - (space>>sysctl_tcp_adv_win_scale); | |
1175 | } | |
1176 | ||
1177 | /* Note: caller must be prepared to deal with negative returns */ | |
1178 | static inline int tcp_space(const struct sock *sk) | |
1179 | { | |
1180 | return tcp_win_from_space(sk->sk_rcvbuf - | |
1181 | atomic_read(&sk->sk_rmem_alloc)); | |
1182 | } | |
1183 | ||
1184 | static inline int tcp_full_space(const struct sock *sk) | |
1185 | { | |
1186 | return tcp_win_from_space(sk->sk_rcvbuf); | |
1187 | } | |
1188 | ||
60236fdd | 1189 | static inline void tcp_acceptq_queue(struct sock *sk, struct request_sock *req, |
1da177e4 LT |
1190 | struct sock *child) |
1191 | { | |
0e87506f | 1192 | reqsk_queue_add(&tcp_sk(sk)->accept_queue, req, sk, child); |
1da177e4 LT |
1193 | } |
1194 | ||
1da177e4 | 1195 | static inline void |
60236fdd | 1196 | tcp_synq_removed(struct sock *sk, struct request_sock *req) |
1da177e4 | 1197 | { |
0e87506f | 1198 | if (reqsk_queue_removed(&tcp_sk(sk)->accept_queue, req) == 0) |
1da177e4 | 1199 | tcp_delete_keepalive_timer(sk); |
1da177e4 LT |
1200 | } |
1201 | ||
1202 | static inline void tcp_synq_added(struct sock *sk) | |
1203 | { | |
0e87506f | 1204 | if (reqsk_queue_added(&tcp_sk(sk)->accept_queue) == 0) |
1da177e4 | 1205 | tcp_reset_keepalive_timer(sk, TCP_TIMEOUT_INIT); |
1da177e4 LT |
1206 | } |
1207 | ||
1208 | static inline int tcp_synq_len(struct sock *sk) | |
1209 | { | |
0e87506f | 1210 | return reqsk_queue_len(&tcp_sk(sk)->accept_queue); |
1da177e4 LT |
1211 | } |
1212 | ||
1213 | static inline int tcp_synq_young(struct sock *sk) | |
1214 | { | |
0e87506f | 1215 | return reqsk_queue_len_young(&tcp_sk(sk)->accept_queue); |
1da177e4 LT |
1216 | } |
1217 | ||
1218 | static inline int tcp_synq_is_full(struct sock *sk) | |
1219 | { | |
0e87506f | 1220 | return reqsk_queue_is_full(&tcp_sk(sk)->accept_queue); |
1da177e4 LT |
1221 | } |
1222 | ||
60236fdd | 1223 | static inline void tcp_synq_unlink(struct tcp_sock *tp, struct request_sock *req, |
0e87506f | 1224 | struct request_sock **prev) |
1da177e4 | 1225 | { |
0e87506f | 1226 | reqsk_queue_unlink(&tp->accept_queue, req, prev); |
1da177e4 LT |
1227 | } |
1228 | ||
60236fdd ACM |
1229 | static inline void tcp_synq_drop(struct sock *sk, struct request_sock *req, |
1230 | struct request_sock **prev) | |
1da177e4 LT |
1231 | { |
1232 | tcp_synq_unlink(tcp_sk(sk), req, prev); | |
1233 | tcp_synq_removed(sk, req); | |
60236fdd | 1234 | reqsk_free(req); |
1da177e4 LT |
1235 | } |
1236 | ||
60236fdd | 1237 | static __inline__ void tcp_openreq_init(struct request_sock *req, |
1da177e4 LT |
1238 | struct tcp_options_received *rx_opt, |
1239 | struct sk_buff *skb) | |
1240 | { | |
2e6599cb ACM |
1241 | struct inet_request_sock *ireq = inet_rsk(req); |
1242 | ||
1da177e4 | 1243 | req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */ |
2e6599cb | 1244 | tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq; |
1da177e4 LT |
1245 | req->mss = rx_opt->mss_clamp; |
1246 | req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0; | |
2e6599cb ACM |
1247 | ireq->tstamp_ok = rx_opt->tstamp_ok; |
1248 | ireq->sack_ok = rx_opt->sack_ok; | |
1249 | ireq->snd_wscale = rx_opt->snd_wscale; | |
1250 | ireq->wscale_ok = rx_opt->wscale_ok; | |
1251 | ireq->acked = 0; | |
1252 | ireq->ecn_ok = 0; | |
1253 | ireq->rmt_port = skb->h.th->source; | |
1da177e4 LT |
1254 | } |
1255 | ||
1256 | extern void tcp_enter_memory_pressure(void); | |
1257 | ||
1da177e4 LT |
1258 | static inline int keepalive_intvl_when(const struct tcp_sock *tp) |
1259 | { | |
1260 | return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl; | |
1261 | } | |
1262 | ||
1263 | static inline int keepalive_time_when(const struct tcp_sock *tp) | |
1264 | { | |
1265 | return tp->keepalive_time ? : sysctl_tcp_keepalive_time; | |
1266 | } | |
1267 | ||
1268 | static inline int tcp_fin_time(const struct tcp_sock *tp) | |
1269 | { | |
1270 | int fin_timeout = tp->linger2 ? : sysctl_tcp_fin_timeout; | |
1271 | ||
1272 | if (fin_timeout < (tp->rto<<2) - (tp->rto>>1)) | |
1273 | fin_timeout = (tp->rto<<2) - (tp->rto>>1); | |
1274 | ||
1275 | return fin_timeout; | |
1276 | } | |
1277 | ||
1278 | static inline int tcp_paws_check(const struct tcp_options_received *rx_opt, int rst) | |
1279 | { | |
1280 | if ((s32)(rx_opt->rcv_tsval - rx_opt->ts_recent) >= 0) | |
1281 | return 0; | |
1282 | if (xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS) | |
1283 | return 0; | |
1284 | ||
1285 | /* RST segments are not recommended to carry timestamp, | |
1286 | and, if they do, it is recommended to ignore PAWS because | |
1287 | "their cleanup function should take precedence over timestamps." | |
1288 | Certainly, it is mistake. It is necessary to understand the reasons | |
1289 | of this constraint to relax it: if peer reboots, clock may go | |
1290 | out-of-sync and half-open connections will not be reset. | |
1291 | Actually, the problem would be not existing if all | |
1292 | the implementations followed draft about maintaining clock | |
1293 | via reboots. Linux-2.2 DOES NOT! | |
1294 | ||
1295 | However, we can relax time bounds for RST segments to MSL. | |
1296 | */ | |
1297 | if (rst && xtime.tv_sec >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL) | |
1298 | return 0; | |
1299 | return 1; | |
1300 | } | |
1301 | ||
1da177e4 LT |
1302 | #define TCP_CHECK_TIMER(sk) do { } while (0) |
1303 | ||
1304 | static inline int tcp_use_frto(const struct sock *sk) | |
1305 | { | |
1306 | const struct tcp_sock *tp = tcp_sk(sk); | |
1307 | ||
1308 | /* F-RTO must be activated in sysctl and there must be some | |
1309 | * unsent new data, and the advertised window should allow | |
1310 | * sending it. | |
1311 | */ | |
1312 | return (sysctl_tcp_frto && sk->sk_send_head && | |
1313 | !after(TCP_SKB_CB(sk->sk_send_head)->end_seq, | |
1314 | tp->snd_una + tp->snd_wnd)); | |
1315 | } | |
1316 | ||
1317 | static inline void tcp_mib_init(void) | |
1318 | { | |
1319 | /* See RFC 2012 */ | |
1320 | TCP_ADD_STATS_USER(TCP_MIB_RTOALGORITHM, 1); | |
1321 | TCP_ADD_STATS_USER(TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ); | |
1322 | TCP_ADD_STATS_USER(TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ); | |
1323 | TCP_ADD_STATS_USER(TCP_MIB_MAXCONN, -1); | |
1324 | } | |
1325 | ||
1326 | /* /proc */ | |
1327 | enum tcp_seq_states { | |
1328 | TCP_SEQ_STATE_LISTENING, | |
1329 | TCP_SEQ_STATE_OPENREQ, | |
1330 | TCP_SEQ_STATE_ESTABLISHED, | |
1331 | TCP_SEQ_STATE_TIME_WAIT, | |
1332 | }; | |
1333 | ||
1334 | struct tcp_seq_afinfo { | |
1335 | struct module *owner; | |
1336 | char *name; | |
1337 | sa_family_t family; | |
1338 | int (*seq_show) (struct seq_file *m, void *v); | |
1339 | struct file_operations *seq_fops; | |
1340 | }; | |
1341 | ||
1342 | struct tcp_iter_state { | |
1343 | sa_family_t family; | |
1344 | enum tcp_seq_states state; | |
1345 | struct sock *syn_wait_sk; | |
1346 | int bucket, sbucket, num, uid; | |
1347 | struct seq_operations seq_ops; | |
1348 | }; | |
1349 | ||
1350 | extern int tcp_proc_register(struct tcp_seq_afinfo *afinfo); | |
1351 | extern void tcp_proc_unregister(struct tcp_seq_afinfo *afinfo); | |
1352 | ||
1da177e4 | 1353 | #endif /* _TCP_H */ |