?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 3.12.39
[linux/fpc-iii.git] / include / net / tcp.h
blobda22d3a23a325e81a54db5ec0260ce1e37fda7c8
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 FASTRETRANS_DEBUG 1
22
23 #include <linux/list.h>
24 #include <linux/tcp.h>
25 #include <linux/bug.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 #include <linux/memcontrol.h>
49
50 extern struct inet_hashinfo tcp_hashinfo;
51
52 extern struct percpu_counter tcp_orphan_count;
53 extern void tcp_time_wait(struct sock *sk, int state, int timeo);
54
55 #define MAX_TCP_HEADER (128 + MAX_HEADER)
56 #define MAX_TCP_OPTION_SPACE 40
57
58 /*
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
61 */
62 #define MAX_TCP_WINDOW 32767U
63
64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
65 #define TCP_MIN_MSS 88U
66
67 /* The least MTU to use for probing */
68 #define TCP_BASE_MSS 512
69
70 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
71 #define TCP_FASTRETRANS_THRESH 3
72
73 /* Maximal reordering. */
74 #define TCP_MAX_REORDERING 127
75
76 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
77 #define TCP_MAX_QUICKACKS 16U
78
79 /* urg_data states */
80 #define TCP_URG_VALID 0x0100
81 #define TCP_URG_NOTYET 0x0200
82 #define TCP_URG_READ 0x0400
83
84 #define TCP_RETR1 3 /*
85 * This is how many retries it does before it
86 * tries to figure out if the gateway is
87 * down. Minimal RFC value is 3; it corresponds
88 * to ~3sec-8min depending on RTO.
89 */
90
91 #define TCP_RETR2 15 /*
92 * This should take at least
93 * 90 minutes to time out.
94 * RFC1122 says that the limit is 100 sec.
95 * 15 is ~13-30min depending on RTO.
96 */
97
98 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
99 * when active opening a connection.
100 * RFC1122 says the minimum retry MUST
101 * be at least 180secs. Nevertheless
102 * this value is corresponding to
103 * 63secs of retransmission with the
104 * current initial RTO.
105 */
106
107 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
108 * when passive opening a connection.
109 * This is corresponding to 31secs of
110 * retransmission with the current
111 * initial RTO.
112 */
113
114 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
115 * state, about 60 seconds */
116 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
117 /* BSD style FIN_WAIT2 deadlock breaker.
118 * It used to be 3min, new value is 60sec,
119 * to combine FIN-WAIT-2 timeout with
120 * TIME-WAIT timer.
121 */
122
123 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
124 #if HZ >= 100
125 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
126 #define TCP_ATO_MIN ((unsigned)(HZ/25))
127 #else
128 #define TCP_DELACK_MIN 4U
129 #define TCP_ATO_MIN 4U
130 #endif
131 #define TCP_RTO_MAX ((unsigned)(120*HZ))
132 #define TCP_RTO_MIN ((unsigned)(HZ/5))
133 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
134 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
135 * used as a fallback RTO for the
136 * initial data transmission if no
137 * valid RTT sample has been acquired,
138 * most likely due to retrans in 3WHS.
139 */
140
141 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
142 * for local resources.
143 */
144
145 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
146 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
147 #define TCP_KEEPALIVE_INTVL (75*HZ)
148
149 #define MAX_TCP_KEEPIDLE 32767
150 #define MAX_TCP_KEEPINTVL 32767
151 #define MAX_TCP_KEEPCNT 127
152 #define MAX_TCP_SYNCNT 127
153
154 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
155
156 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
157 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
158 * after this time. It should be equal
159 * (or greater than) TCP_TIMEWAIT_LEN
160 * to provide reliability equal to one
161 * provided by timewait state.
162 */
163 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
164 * timestamps. It must be less than
165 * minimal timewait lifetime.
166 */
167 /*
168 * TCP option
169 */
170
171 #define TCPOPT_NOP 1 /* Padding */
172 #define TCPOPT_EOL 0 /* End of options */
173 #define TCPOPT_MSS 2 /* Segment size negotiating */
174 #define TCPOPT_WINDOW 3 /* Window scaling */
175 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
176 #define TCPOPT_SACK 5 /* SACK Block */
177 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
178 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
179 #define TCPOPT_EXP 254 /* Experimental */
180 /* Magic number to be after the option value for sharing TCP
181 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
182 */
183 #define TCPOPT_FASTOPEN_MAGIC 0xF989
184
185 /*
186 * TCP option lengths
187 */
188
189 #define TCPOLEN_MSS 4
190 #define TCPOLEN_WINDOW 3
191 #define TCPOLEN_SACK_PERM 2
192 #define TCPOLEN_TIMESTAMP 10
193 #define TCPOLEN_MD5SIG 18
194 #define TCPOLEN_EXP_FASTOPEN_BASE 4
195
196 /* But this is what stacks really send out. */
197 #define TCPOLEN_TSTAMP_ALIGNED 12
198 #define TCPOLEN_WSCALE_ALIGNED 4
199 #define TCPOLEN_SACKPERM_ALIGNED 4
200 #define TCPOLEN_SACK_BASE 2
201 #define TCPOLEN_SACK_BASE_ALIGNED 4
202 #define TCPOLEN_SACK_PERBLOCK 8
203 #define TCPOLEN_MD5SIG_ALIGNED 20
204 #define TCPOLEN_MSS_ALIGNED 4
205
206 /* Flags in tp->nonagle */
207 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
208 #define TCP_NAGLE_CORK 2 /* Socket is corked */
209 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
210
211 /* TCP thin-stream limits */
212 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
213
214 /* TCP initial congestion window as per draft-hkchu-tcpm-initcwnd-01 */
215 #define TCP_INIT_CWND 10
216
217 /* Bit Flags for sysctl_tcp_fastopen */
218 #define TFO_CLIENT_ENABLE 1
219 #define TFO_SERVER_ENABLE 2
220 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
221
222 /* Process SYN data but skip cookie validation */
223 #define TFO_SERVER_COOKIE_NOT_CHKED 0x100
224 /* Accept SYN data w/o any cookie option */
225 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
226
227 /* Force enable TFO on all listeners, i.e., not requiring the
228 * TCP_FASTOPEN socket option. SOCKOPT1/2 determine how to set max_qlen.
229 */
230 #define TFO_SERVER_WO_SOCKOPT1 0x400
231 #define TFO_SERVER_WO_SOCKOPT2 0x800
232 /* Always create TFO child sockets on a TFO listener even when
233 * cookie/data not present. (For testing purpose!)
234 */
235 #define TFO_SERVER_ALWAYS 0x1000
236
237 extern struct inet_timewait_death_row tcp_death_row;
238
239 /* sysctl variables for tcp */
240 extern int sysctl_tcp_timestamps;
241 extern int sysctl_tcp_window_scaling;
242 extern int sysctl_tcp_sack;
243 extern int sysctl_tcp_fin_timeout;
244 extern int sysctl_tcp_keepalive_time;
245 extern int sysctl_tcp_keepalive_probes;
246 extern int sysctl_tcp_keepalive_intvl;
247 extern int sysctl_tcp_syn_retries;
248 extern int sysctl_tcp_synack_retries;
249 extern int sysctl_tcp_retries1;
250 extern int sysctl_tcp_retries2;
251 extern int sysctl_tcp_orphan_retries;
252 extern int sysctl_tcp_syncookies;
253 extern int sysctl_tcp_fastopen;
254 extern int sysctl_tcp_retrans_collapse;
255 extern int sysctl_tcp_stdurg;
256 extern int sysctl_tcp_rfc1337;
257 extern int sysctl_tcp_abort_on_overflow;
258 extern int sysctl_tcp_max_orphans;
259 extern int sysctl_tcp_fack;
260 extern int sysctl_tcp_reordering;
261 extern int sysctl_tcp_dsack;
262 extern int sysctl_tcp_wmem[3];
263 extern int sysctl_tcp_rmem[3];
264 extern int sysctl_tcp_app_win;
265 extern int sysctl_tcp_adv_win_scale;
266 extern int sysctl_tcp_tw_reuse;
267 extern int sysctl_tcp_frto;
268 extern int sysctl_tcp_low_latency;
269 extern int sysctl_tcp_dma_copybreak;
270 extern int sysctl_tcp_nometrics_save;
271 extern int sysctl_tcp_moderate_rcvbuf;
272 extern int sysctl_tcp_tso_win_divisor;
273 extern int sysctl_tcp_mtu_probing;
274 extern int sysctl_tcp_base_mss;
275 extern int sysctl_tcp_workaround_signed_windows;
276 extern int sysctl_tcp_slow_start_after_idle;
277 extern int sysctl_tcp_max_ssthresh;
278 extern int sysctl_tcp_thin_linear_timeouts;
279 extern int sysctl_tcp_thin_dupack;
280 extern int sysctl_tcp_early_retrans;
281 extern int sysctl_tcp_limit_output_bytes;
282 extern int sysctl_tcp_challenge_ack_limit;
283 extern unsigned int sysctl_tcp_notsent_lowat;
284 extern int sysctl_tcp_min_tso_segs;
285
286 extern atomic_long_t tcp_memory_allocated;
287 extern struct percpu_counter tcp_sockets_allocated;
288 extern int tcp_memory_pressure;
289
290 /*
291 * The next routines deal with comparing 32 bit unsigned ints
292 * and worry about wraparound (automatic with unsigned arithmetic).
293 */
294
295 static inline bool before(__u32 seq1, __u32 seq2)
296 {
297 return (__s32)(seq1-seq2) < 0;
298 }
299 #define after(seq2, seq1) before(seq1, seq2)
300
301 /* is s2<=s1<=s3 ? */
302 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
303 {
304 return seq3 - seq2 >= seq1 - seq2;
305 }
306
307 static inline bool tcp_out_of_memory(struct sock *sk)
308 {
309 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
310 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
311 return true;
312 return false;
313 }
314
315 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
316 {
317 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
318 int orphans = percpu_counter_read_positive(ocp);
319
320 if (orphans << shift > sysctl_tcp_max_orphans) {
321 orphans = percpu_counter_sum_positive(ocp);
322 if (orphans << shift > sysctl_tcp_max_orphans)
323 return true;
324 }
325 return false;
326 }
327
328 extern bool tcp_check_oom(struct sock *sk, int shift);
329
330 /* syncookies: remember time of last synqueue overflow */
331 static inline void tcp_synq_overflow(struct sock *sk)
332 {
333 tcp_sk(sk)->rx_opt.ts_recent_stamp = jiffies;
334 }
335
336 /* syncookies: no recent synqueue overflow on this listening socket? */
337 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
338 {
339 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
340 return time_after(jiffies, last_overflow + TCP_TIMEOUT_FALLBACK);
341 }
342
343 extern struct proto tcp_prot;
344
345 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
346 #define TCP_INC_STATS_BH(net, field) SNMP_INC_STATS_BH((net)->mib.tcp_statistics, field)
347 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
348 #define TCP_ADD_STATS_USER(net, field, val) SNMP_ADD_STATS_USER((net)->mib.tcp_statistics, field, val)
349 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
350
351 extern void tcp_init_mem(struct net *net);
352
353 extern void tcp_tasklet_init(void);
354
355 extern void tcp_v4_err(struct sk_buff *skb, u32);
356
357 extern void tcp_shutdown (struct sock *sk, int how);
358
359 extern void tcp_v4_early_demux(struct sk_buff *skb);
360 extern int tcp_v4_rcv(struct sk_buff *skb);
361
362 extern int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
363 extern int tcp_sendmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
364 size_t size);
365 extern int tcp_sendpage(struct sock *sk, struct page *page, int offset,
366 size_t size, int flags);
367 extern void tcp_release_cb(struct sock *sk);
368 extern void tcp_wfree(struct sk_buff *skb);
369 extern void tcp_write_timer_handler(struct sock *sk);
370 extern void tcp_delack_timer_handler(struct sock *sk);
371 extern int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
372 extern int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb,
373 const struct tcphdr *th, unsigned int len);
374 extern void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
375 const struct tcphdr *th, unsigned int len);
376 extern void tcp_rcv_space_adjust(struct sock *sk);
377 extern void tcp_cleanup_rbuf(struct sock *sk, int copied);
378 extern int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
379 extern void tcp_twsk_destructor(struct sock *sk);
380 extern ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
381 struct pipe_inode_info *pipe, size_t len,
382 unsigned int flags);
383
384 static inline void tcp_dec_quickack_mode(struct sock *sk,
385 const unsigned int pkts)
386 {
387 struct inet_connection_sock *icsk = inet_csk(sk);
388
389 if (icsk->icsk_ack.quick) {
390 if (pkts >= icsk->icsk_ack.quick) {
391 icsk->icsk_ack.quick = 0;
392 /* Leaving quickack mode we deflate ATO. */
393 icsk->icsk_ack.ato = TCP_ATO_MIN;
394 } else
395 icsk->icsk_ack.quick -= pkts;
396 }
397 }
398
399 #define TCP_ECN_OK 1
400 #define TCP_ECN_QUEUE_CWR 2
401 #define TCP_ECN_DEMAND_CWR 4
402 #define TCP_ECN_SEEN 8
403
404 enum tcp_tw_status {
405 TCP_TW_SUCCESS = 0,
406 TCP_TW_RST = 1,
407 TCP_TW_ACK = 2,
408 TCP_TW_SYN = 3
409 };
410
411
412 extern enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
413 struct sk_buff *skb,
414 const struct tcphdr *th);
415 extern struct sock * tcp_check_req(struct sock *sk,struct sk_buff *skb,
416 struct request_sock *req,
417 struct request_sock **prev,
418 bool fastopen);
419 extern int tcp_child_process(struct sock *parent, struct sock *child,
420 struct sk_buff *skb);
421 extern void tcp_enter_loss(struct sock *sk, int how);
422 extern void tcp_clear_retrans(struct tcp_sock *tp);
423 extern void tcp_update_metrics(struct sock *sk);
424 extern void tcp_init_metrics(struct sock *sk);
425 extern void tcp_metrics_init(void);
426 extern bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst, bool paws_check);
427 extern bool tcp_remember_stamp(struct sock *sk);
428 extern bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
429 extern void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
430 extern void tcp_disable_fack(struct tcp_sock *tp);
431 extern void tcp_close(struct sock *sk, long timeout);
432 extern void tcp_init_sock(struct sock *sk);
433 extern unsigned int tcp_poll(struct file * file, struct socket *sock,
434 struct poll_table_struct *wait);
435 extern int tcp_getsockopt(struct sock *sk, int level, int optname,
436 char __user *optval, int __user *optlen);
437 extern int tcp_setsockopt(struct sock *sk, int level, int optname,
438 char __user *optval, unsigned int optlen);
439 extern int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
440 char __user *optval, int __user *optlen);
441 extern int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
442 char __user *optval, unsigned int optlen);
443 extern void tcp_set_keepalive(struct sock *sk, int val);
444 extern void tcp_syn_ack_timeout(struct sock *sk, struct request_sock *req);
445 extern int tcp_recvmsg(struct kiocb *iocb, struct sock *sk, struct msghdr *msg,
446 size_t len, int nonblock, int flags, int *addr_len);
447 extern void tcp_parse_options(const struct sk_buff *skb,
448 struct tcp_options_received *opt_rx,
449 int estab, struct tcp_fastopen_cookie *foc);
450 extern const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
451
452 /*
453 * TCP v4 functions exported for the inet6 API
454 */
455
456 extern void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
457 void tcp_v4_mtu_reduced(struct sock *sk);
458 extern int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
459 extern struct sock * tcp_create_openreq_child(struct sock *sk,
460 struct request_sock *req,
461 struct sk_buff *skb);
462 extern struct sock * tcp_v4_syn_recv_sock(struct sock *sk, struct sk_buff *skb,
463 struct request_sock *req,
464 struct dst_entry *dst);
465 extern int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
466 extern int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr,
467 int addr_len);
468 extern int tcp_connect(struct sock *sk);
469 extern struct sk_buff * tcp_make_synack(struct sock *sk, struct dst_entry *dst,
470 struct request_sock *req,
471 struct tcp_fastopen_cookie *foc);
472 extern int tcp_disconnect(struct sock *sk, int flags);
473
474 void tcp_connect_init(struct sock *sk);
475 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
476 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
477 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
478
479 /* From syncookies.c */
480 extern __u32 syncookie_secret[2][16-4+SHA_DIGEST_WORDS];
481 extern int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
482 u32 cookie);
483 extern struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb,
484 struct ip_options *opt);
485 #ifdef CONFIG_SYN_COOKIES
486 #include <linux/ktime.h>
487
488 /* Syncookies use a monotonic timer which increments every 60 seconds.
489 * This counter is used both as a hash input and partially encoded into
490 * the cookie value. A cookie is only validated further if the delta
491 * between the current counter value and the encoded one is less than this,
492 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
493 * the counter advances immediately after a cookie is generated).
494 */
495 #define MAX_SYNCOOKIE_AGE 2
496
497 static inline u32 tcp_cookie_time(void)
498 {
499 u64 val = get_jiffies_64();
500
501 do_div(val, 60 * HZ);
502 return val;
503 }
504
505 extern u32 __cookie_v4_init_sequence(const struct iphdr *iph,
506 const struct tcphdr *th, u16 *mssp);
507 extern __u32 cookie_v4_init_sequence(struct sock *sk, struct sk_buff *skb,
508 __u16 *mss);
509 #else
510 static inline __u32 cookie_v4_init_sequence(struct sock *sk,
511 struct sk_buff *skb,
512 __u16 *mss)
513 {
514 return 0;
515 }
516 #endif
517
518 extern __u32 cookie_init_timestamp(struct request_sock *req);
519 extern bool cookie_check_timestamp(struct tcp_options_received *opt,
520 struct net *net, bool *ecn_ok);
521
522 /* From net/ipv6/syncookies.c */
523 extern int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
524 u32 cookie);
525 extern struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
526 #ifdef CONFIG_SYN_COOKIES
527 extern u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
528 const struct tcphdr *th, u16 *mssp);
529 extern __u32 cookie_v6_init_sequence(struct sock *sk, const struct sk_buff *skb,
530 __u16 *mss);
531 #else
532 static inline __u32 cookie_v6_init_sequence(struct sock *sk,
533 struct sk_buff *skb,
534 __u16 *mss)
535 {
536 return 0;
537 }
538 #endif
539 /* tcp_output.c */
540
541 extern void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
542 int nonagle);
543 extern bool tcp_may_send_now(struct sock *sk);
544 extern int __tcp_retransmit_skb(struct sock *, struct sk_buff *);
545 extern int tcp_retransmit_skb(struct sock *, struct sk_buff *);
546 extern void tcp_retransmit_timer(struct sock *sk);
547 extern void tcp_xmit_retransmit_queue(struct sock *);
548 extern void tcp_simple_retransmit(struct sock *);
549 extern int tcp_trim_head(struct sock *, struct sk_buff *, u32);
550 extern int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int);
551
552 extern void tcp_send_probe0(struct sock *);
553 extern void tcp_send_partial(struct sock *);
554 extern int tcp_write_wakeup(struct sock *);
555 extern void tcp_send_fin(struct sock *sk);
556 extern void tcp_send_active_reset(struct sock *sk, gfp_t priority);
557 extern int tcp_send_synack(struct sock *);
558 extern bool tcp_syn_flood_action(struct sock *sk,
559 const struct sk_buff *skb,
560 const char *proto);
561 extern void tcp_push_one(struct sock *, unsigned int mss_now);
562 extern void tcp_send_ack(struct sock *sk);
563 extern void tcp_send_delayed_ack(struct sock *sk);
564 extern void tcp_send_loss_probe(struct sock *sk);
565 extern bool tcp_schedule_loss_probe(struct sock *sk);
566
567 /* tcp_input.c */
568 extern void tcp_cwnd_application_limited(struct sock *sk);
569 extern void tcp_resume_early_retransmit(struct sock *sk);
570 extern void tcp_rearm_rto(struct sock *sk);
571 extern void tcp_reset(struct sock *sk);
572
573 /* tcp_timer.c */
574 extern void tcp_init_xmit_timers(struct sock *);
575 static inline void tcp_clear_xmit_timers(struct sock *sk)
576 {
577 inet_csk_clear_xmit_timers(sk);
578 }
579
580 extern unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
581 extern unsigned int tcp_current_mss(struct sock *sk);
582
583 /* Bound MSS / TSO packet size with the half of the window */
584 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
585 {
586 int cutoff;
587
588 /* When peer uses tiny windows, there is no use in packetizing
589 * to sub-MSS pieces for the sake of SWS or making sure there
590 * are enough packets in the pipe for fast recovery.
591 *
592 * On the other hand, for extremely large MSS devices, handling
593 * smaller than MSS windows in this way does make sense.
594 */
595 if (tp->max_window >= 512)
596 cutoff = (tp->max_window >> 1);
597 else
598 cutoff = tp->max_window;
599
600 if (cutoff && pktsize > cutoff)
601 return max_t(int, cutoff, 68U - tp->tcp_header_len);
602 else
603 return pktsize;
604 }
605
606 /* tcp.c */
607 extern void tcp_get_info(const struct sock *, struct tcp_info *);
608
609 /* Read 'sendfile()'-style from a TCP socket */
610 typedef int (*sk_read_actor_t)(read_descriptor_t *, struct sk_buff *,
611 unsigned int, size_t);
612 extern int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
613 sk_read_actor_t recv_actor);
614
615 extern void tcp_initialize_rcv_mss(struct sock *sk);
616
617 extern int tcp_mtu_to_mss(struct sock *sk, int pmtu);
618 extern int tcp_mss_to_mtu(struct sock *sk, int mss);
619 extern void tcp_mtup_init(struct sock *sk);
620 extern void tcp_init_buffer_space(struct sock *sk);
621
622 static inline void tcp_bound_rto(const struct sock *sk)
623 {
624 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
625 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
626 }
627
628 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
629 {
630 return (tp->srtt >> 3) + tp->rttvar;
631 }
632
633 extern void tcp_set_rto(struct sock *sk);
634
635 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
636 {
637 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
638 ntohl(TCP_FLAG_ACK) |
639 snd_wnd);
640 }
641
642 static inline void tcp_fast_path_on(struct tcp_sock *tp)
643 {
644 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
645 }
646
647 static inline void tcp_fast_path_check(struct sock *sk)
648 {
649 struct tcp_sock *tp = tcp_sk(sk);
650
651 if (skb_queue_empty(&tp->out_of_order_queue) &&
652 tp->rcv_wnd &&
653 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
654 !tp->urg_data)
655 tcp_fast_path_on(tp);
656 }
657
658 /* Compute the actual rto_min value */
659 static inline u32 tcp_rto_min(struct sock *sk)
660 {
661 const struct dst_entry *dst = __sk_dst_get(sk);
662 u32 rto_min = TCP_RTO_MIN;
663
664 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
665 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
666 return rto_min;
667 }
668
669 /* Compute the actual receive window we are currently advertising.
670 * Rcv_nxt can be after the window if our peer push more data
671 * than the offered window.
672 */
673 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
674 {
675 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
676
677 if (win < 0)
678 win = 0;
679 return (u32) win;
680 }
681
682 /* Choose a new window, without checks for shrinking, and without
683 * scaling applied to the result. The caller does these things
684 * if necessary. This is a "raw" window selection.
685 */
686 extern u32 __tcp_select_window(struct sock *sk);
687
688 void tcp_send_window_probe(struct sock *sk);
689
690 /* TCP timestamps are only 32-bits, this causes a slight
691 * complication on 64-bit systems since we store a snapshot
692 * of jiffies in the buffer control blocks below. We decided
693 * to use only the low 32-bits of jiffies and hide the ugly
694 * casts with the following macro.
695 */
696 #define tcp_time_stamp ((__u32)(jiffies))
697
698 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
699
700 #define TCPHDR_FIN 0x01
701 #define TCPHDR_SYN 0x02
702 #define TCPHDR_RST 0x04
703 #define TCPHDR_PSH 0x08
704 #define TCPHDR_ACK 0x10
705 #define TCPHDR_URG 0x20
706 #define TCPHDR_ECE 0x40
707 #define TCPHDR_CWR 0x80
708
709 /* This is what the send packet queuing engine uses to pass
710 * TCP per-packet control information to the transmission code.
711 * We also store the host-order sequence numbers in here too.
712 * This is 44 bytes if IPV6 is enabled.
713 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
714 */
715 struct tcp_skb_cb {
716 union {
717 struct inet_skb_parm h4;
718 #if IS_ENABLED(CONFIG_IPV6)
719 struct inet6_skb_parm h6;
720 #endif
721 } header; /* For incoming frames */
722 __u32 seq; /* Starting sequence number */
723 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
724 __u32 when; /* used to compute rtt's */
725 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
726
727 __u8 sacked; /* State flags for SACK/FACK. */
728 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
729 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
730 #define TCPCB_LOST 0x04 /* SKB is lost */
731 #define TCPCB_TAGBITS 0x07 /* All tag bits */
732 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
733 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS)
734
735 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
736 /* 1 byte hole */
737 __u32 ack_seq; /* Sequence number ACK'd */
738 };
739
740 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
741
742 /* RFC3168 : 6.1.1 SYN packets must not have ECT/ECN bits set
743 *
744 * If we receive a SYN packet with these bits set, it means a network is
745 * playing bad games with TOS bits. In order to avoid possible false congestion
746 * notifications, we disable TCP ECN negociation.
747 */
748 static inline void
749 TCP_ECN_create_request(struct request_sock *req, const struct sk_buff *skb,
750 struct net *net)
751 {
752 const struct tcphdr *th = tcp_hdr(skb);
753
754 if (net->ipv4.sysctl_tcp_ecn && th->ece && th->cwr &&
755 INET_ECN_is_not_ect(TCP_SKB_CB(skb)->ip_dsfield))
756 inet_rsk(req)->ecn_ok = 1;
757 }
758
759 /* Due to TSO, an SKB can be composed of multiple actual
760 * packets. To keep these tracked properly, we use this.
761 */
762 static inline int tcp_skb_pcount(const struct sk_buff *skb)
763 {
764 return skb_shinfo(skb)->gso_segs;
765 }
766
767 /* This is valid iff tcp_skb_pcount() > 1. */
768 static inline int tcp_skb_mss(const struct sk_buff *skb)
769 {
770 return skb_shinfo(skb)->gso_size;
771 }
772
773 /* Events passed to congestion control interface */
774 enum tcp_ca_event {
775 CA_EVENT_TX_START, /* first transmit when no packets in flight */
776 CA_EVENT_CWND_RESTART, /* congestion window restart */
777 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
778 CA_EVENT_LOSS, /* loss timeout */
779 CA_EVENT_FAST_ACK, /* in sequence ack */
780 CA_EVENT_SLOW_ACK, /* other ack */
781 };
782
783 /*
784 * Interface for adding new TCP congestion control handlers
785 */
786 #define TCP_CA_NAME_MAX 16
787 #define TCP_CA_MAX 128
788 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
789
790 #define TCP_CONG_NON_RESTRICTED 0x1
791 #define TCP_CONG_RTT_STAMP 0x2
792
793 struct tcp_congestion_ops {
794 struct list_head list;
795 unsigned long flags;
796
797 /* initialize private data (optional) */
798 void (*init)(struct sock *sk);
799 /* cleanup private data (optional) */
800 void (*release)(struct sock *sk);
801
802 /* return slow start threshold (required) */
803 u32 (*ssthresh)(struct sock *sk);
804 /* lower bound for congestion window (optional) */
805 u32 (*min_cwnd)(const struct sock *sk);
806 /* do new cwnd calculation (required) */
807 void (*cong_avoid)(struct sock *sk, u32 ack, u32 in_flight);
808 /* call before changing ca_state (optional) */
809 void (*set_state)(struct sock *sk, u8 new_state);
810 /* call when cwnd event occurs (optional) */
811 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
812 /* new value of cwnd after loss (optional) */
813 u32 (*undo_cwnd)(struct sock *sk);
814 /* hook for packet ack accounting (optional) */
815 void (*pkts_acked)(struct sock *sk, u32 num_acked, s32 rtt_us);
816 /* get info for inet_diag (optional) */
817 void (*get_info)(struct sock *sk, u32 ext, struct sk_buff *skb);
818
819 char name[TCP_CA_NAME_MAX];
820 struct module *owner;
821 };
822
823 extern int tcp_register_congestion_control(struct tcp_congestion_ops *type);
824 extern void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
825
826 extern void tcp_init_congestion_control(struct sock *sk);
827 extern void tcp_cleanup_congestion_control(struct sock *sk);
828 extern int tcp_set_default_congestion_control(const char *name);
829 extern void tcp_get_default_congestion_control(char *name);
830 extern void tcp_get_available_congestion_control(char *buf, size_t len);
831 extern void tcp_get_allowed_congestion_control(char *buf, size_t len);
832 extern int tcp_set_allowed_congestion_control(char *allowed);
833 extern int tcp_set_congestion_control(struct sock *sk, const char *name);
834 extern void tcp_slow_start(struct tcp_sock *tp);
835 extern void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w);
836
837 extern struct tcp_congestion_ops tcp_init_congestion_ops;
838 extern u32 tcp_reno_ssthresh(struct sock *sk);
839 extern void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 in_flight);
840 extern u32 tcp_reno_min_cwnd(const struct sock *sk);
841 extern struct tcp_congestion_ops tcp_reno;
842
843 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
844 {
845 struct inet_connection_sock *icsk = inet_csk(sk);
846
847 if (icsk->icsk_ca_ops->set_state)
848 icsk->icsk_ca_ops->set_state(sk, ca_state);
849 icsk->icsk_ca_state = ca_state;
850 }
851
852 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
853 {
854 const struct inet_connection_sock *icsk = inet_csk(sk);
855
856 if (icsk->icsk_ca_ops->cwnd_event)
857 icsk->icsk_ca_ops->cwnd_event(sk, event);
858 }
859
860 /* These functions determine how the current flow behaves in respect of SACK
861 * handling. SACK is negotiated with the peer, and therefore it can vary
862 * between different flows.
863 *
864 * tcp_is_sack - SACK enabled
865 * tcp_is_reno - No SACK
866 * tcp_is_fack - FACK enabled, implies SACK enabled
867 */
868 static inline int tcp_is_sack(const struct tcp_sock *tp)
869 {
870 return tp->rx_opt.sack_ok;
871 }
872
873 static inline bool tcp_is_reno(const struct tcp_sock *tp)
874 {
875 return !tcp_is_sack(tp);
876 }
877
878 static inline bool tcp_is_fack(const struct tcp_sock *tp)
879 {
880 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
881 }
882
883 static inline void tcp_enable_fack(struct tcp_sock *tp)
884 {
885 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
886 }
887
888 /* TCP early-retransmit (ER) is similar to but more conservative than
889 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
890 */
891 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
892 {
893 tp->do_early_retrans = sysctl_tcp_early_retrans &&
894 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
895 sysctl_tcp_reordering == 3;
896 }
897
898 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
899 {
900 tp->do_early_retrans = 0;
901 }
902
903 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
904 {
905 return tp->sacked_out + tp->lost_out;
906 }
907
908 /* This determines how many packets are "in the network" to the best
909 * of our knowledge. In many cases it is conservative, but where
910 * detailed information is available from the receiver (via SACK
911 * blocks etc.) we can make more aggressive calculations.
912 *
913 * Use this for decisions involving congestion control, use just
914 * tp->packets_out to determine if the send queue is empty or not.
915 *
916 * Read this equation as:
917 *
918 * "Packets sent once on transmission queue" MINUS
919 * "Packets left network, but not honestly ACKed yet" PLUS
920 * "Packets fast retransmitted"
921 */
922 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
923 {
924 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
925 }
926
927 #define TCP_INFINITE_SSTHRESH 0x7fffffff
928
929 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
930 {
931 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
932 }
933
934 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
935 {
936 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
937 (1 << inet_csk(sk)->icsk_ca_state);
938 }
939
940 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
941 * The exception is cwnd reduction phase, when cwnd is decreasing towards
942 * ssthresh.
943 */
944 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
945 {
946 const struct tcp_sock *tp = tcp_sk(sk);
947
948 if (tcp_in_cwnd_reduction(sk))
949 return tp->snd_ssthresh;
950 else
951 return max(tp->snd_ssthresh,
952 ((tp->snd_cwnd >> 1) +
953 (tp->snd_cwnd >> 2)));
954 }
955
956 /* Use define here intentionally to get WARN_ON location shown at the caller */
957 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
958
959 extern void tcp_enter_cwr(struct sock *sk, const int set_ssthresh);
960 extern __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
961
962 /* The maximum number of MSS of available cwnd for which TSO defers
963 * sending if not using sysctl_tcp_tso_win_divisor.
964 */
965 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
966 {
967 return 3;
968 }
969
970 /* Slow start with delack produces 3 packets of burst, so that
971 * it is safe "de facto". This will be the default - same as
972 * the default reordering threshold - but if reordering increases,
973 * we must be able to allow cwnd to burst at least this much in order
974 * to not pull it back when holes are filled.
975 */
976 static __inline__ __u32 tcp_max_burst(const struct tcp_sock *tp)
977 {
978 return tp->reordering;
979 }
980
981 /* Returns end sequence number of the receiver's advertised window */
982 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
983 {
984 return tp->snd_una + tp->snd_wnd;
985 }
986 extern bool tcp_is_cwnd_limited(const struct sock *sk, u32 in_flight);
987
988 static inline void tcp_minshall_update(struct tcp_sock *tp, unsigned int mss,
989 const struct sk_buff *skb)
990 {
991 if (skb->len < mss)
992 tp->snd_sml = TCP_SKB_CB(skb)->end_seq;
993 }
994
995 static inline void tcp_check_probe_timer(struct sock *sk)
996 {
997 const struct tcp_sock *tp = tcp_sk(sk);
998 const struct inet_connection_sock *icsk = inet_csk(sk);
999
1000 if (!tp->packets_out && !icsk->icsk_pending)
1001 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1002 icsk->icsk_rto, TCP_RTO_MAX);
1003 }
1004
1005 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1006 {
1007 tp->snd_wl1 = seq;
1008 }
1009
1010 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1011 {
1012 tp->snd_wl1 = seq;
1013 }
1014
1015 /*
1016 * Calculate(/check) TCP checksum
1017 */
1018 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1019 __be32 daddr, __wsum base)
1020 {
1021 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1022 }
1023
1024 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1025 {
1026 return __skb_checksum_complete(skb);
1027 }
1028
1029 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1030 {
1031 return !skb_csum_unnecessary(skb) &&
1032 __tcp_checksum_complete(skb);
1033 }
1034
1035 /* Prequeue for VJ style copy to user, combined with checksumming. */
1036
1037 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1038 {
1039 tp->ucopy.task = NULL;
1040 tp->ucopy.len = 0;
1041 tp->ucopy.memory = 0;
1042 skb_queue_head_init(&tp->ucopy.prequeue);
1043 #ifdef CONFIG_NET_DMA
1044 tp->ucopy.dma_chan = NULL;
1045 tp->ucopy.wakeup = 0;
1046 tp->ucopy.pinned_list = NULL;
1047 tp->ucopy.dma_cookie = 0;
1048 #endif
1049 }
1050
1051 extern bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1052
1053 #undef STATE_TRACE
1054
1055 #ifdef STATE_TRACE
1056 static const char *statename[]={
1057 "Unused","Established","Syn Sent","Syn Recv",
1058 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1059 "Close Wait","Last ACK","Listen","Closing"
1060 };
1061 #endif
1062 extern void tcp_set_state(struct sock *sk, int state);
1063
1064 extern void tcp_done(struct sock *sk);
1065
1066 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1067 {
1068 rx_opt->dsack = 0;
1069 rx_opt->num_sacks = 0;
1070 }
1071
1072 extern u32 tcp_default_init_rwnd(u32 mss);
1073
1074 /* Determine a window scaling and initial window to offer. */
1075 extern void tcp_select_initial_window(int __space, __u32 mss,
1076 __u32 *rcv_wnd, __u32 *window_clamp,
1077 int wscale_ok, __u8 *rcv_wscale,
1078 __u32 init_rcv_wnd);
1079
1080 static inline int tcp_win_from_space(int space)
1081 {
1082 return sysctl_tcp_adv_win_scale<=0 ?
1083 (space>>(-sysctl_tcp_adv_win_scale)) :
1084 space - (space>>sysctl_tcp_adv_win_scale);
1085 }
1086
1087 /* Note: caller must be prepared to deal with negative returns */
1088 static inline int tcp_space(const struct sock *sk)
1089 {
1090 return tcp_win_from_space(sk->sk_rcvbuf -
1091 atomic_read(&sk->sk_rmem_alloc));
1092 }
1093
1094 static inline int tcp_full_space(const struct sock *sk)
1095 {
1096 return tcp_win_from_space(sk->sk_rcvbuf);
1097 }
1098
1099 static inline void tcp_openreq_init(struct request_sock *req,
1100 struct tcp_options_received *rx_opt,
1101 struct sk_buff *skb)
1102 {
1103 struct inet_request_sock *ireq = inet_rsk(req);
1104
1105 req->rcv_wnd = 0; /* So that tcp_send_synack() knows! */
1106 req->cookie_ts = 0;
1107 tcp_rsk(req)->rcv_isn = TCP_SKB_CB(skb)->seq;
1108 tcp_rsk(req)->rcv_nxt = TCP_SKB_CB(skb)->seq + 1;
1109 tcp_rsk(req)->snt_synack = 0;
1110 req->mss = rx_opt->mss_clamp;
1111 req->ts_recent = rx_opt->saw_tstamp ? rx_opt->rcv_tsval : 0;
1112 ireq->tstamp_ok = rx_opt->tstamp_ok;
1113 ireq->sack_ok = rx_opt->sack_ok;
1114 ireq->snd_wscale = rx_opt->snd_wscale;
1115 ireq->wscale_ok = rx_opt->wscale_ok;
1116 ireq->acked = 0;
1117 ireq->ecn_ok = 0;
1118 ireq->rmt_port = tcp_hdr(skb)->source;
1119 ireq->loc_port = tcp_hdr(skb)->dest;
1120 }
1121
1122 extern void tcp_enter_memory_pressure(struct sock *sk);
1123
1124 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1125 {
1126 return tp->keepalive_intvl ? : sysctl_tcp_keepalive_intvl;
1127 }
1128
1129 static inline int keepalive_time_when(const struct tcp_sock *tp)
1130 {
1131 return tp->keepalive_time ? : sysctl_tcp_keepalive_time;
1132 }
1133
1134 static inline int keepalive_probes(const struct tcp_sock *tp)
1135 {
1136 return tp->keepalive_probes ? : sysctl_tcp_keepalive_probes;
1137 }
1138
1139 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1140 {
1141 const struct inet_connection_sock *icsk = &tp->inet_conn;
1142
1143 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1144 tcp_time_stamp - tp->rcv_tstamp);
1145 }
1146
1147 static inline int tcp_fin_time(const struct sock *sk)
1148 {
1149 int fin_timeout = tcp_sk(sk)->linger2 ? : sysctl_tcp_fin_timeout;
1150 const int rto = inet_csk(sk)->icsk_rto;
1151
1152 if (fin_timeout < (rto << 2) - (rto >> 1))
1153 fin_timeout = (rto << 2) - (rto >> 1);
1154
1155 return fin_timeout;
1156 }
1157
1158 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1159 int paws_win)
1160 {
1161 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1162 return true;
1163 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1164 return true;
1165 /*
1166 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1167 * then following tcp messages have valid values. Ignore 0 value,
1168 * or else 'negative' tsval might forbid us to accept their packets.
1169 */
1170 if (!rx_opt->ts_recent)
1171 return true;
1172 return false;
1173 }
1174
1175 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1176 int rst)
1177 {
1178 if (tcp_paws_check(rx_opt, 0))
1179 return false;
1180
1181 /* RST segments are not recommended to carry timestamp,
1182 and, if they do, it is recommended to ignore PAWS because
1183 "their cleanup function should take precedence over timestamps."
1184 Certainly, it is mistake. It is necessary to understand the reasons
1185 of this constraint to relax it: if peer reboots, clock may go
1186 out-of-sync and half-open connections will not be reset.
1187 Actually, the problem would be not existing if all
1188 the implementations followed draft about maintaining clock
1189 via reboots. Linux-2.2 DOES NOT!
1190
1191 However, we can relax time bounds for RST segments to MSL.
1192 */
1193 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1194 return false;
1195 return true;
1196 }
1197
1198 static inline void tcp_mib_init(struct net *net)
1199 {
1200 /* See RFC 2012 */
1201 TCP_ADD_STATS_USER(net, TCP_MIB_RTOALGORITHM, 1);
1202 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1203 TCP_ADD_STATS_USER(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1204 TCP_ADD_STATS_USER(net, TCP_MIB_MAXCONN, -1);
1205 }
1206
1207 /* from STCP */
1208 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1209 {
1210 tp->lost_skb_hint = NULL;
1211 }
1212
1213 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1214 {
1215 tcp_clear_retrans_hints_partial(tp);
1216 tp->retransmit_skb_hint = NULL;
1217 }
1218
1219 /* MD5 Signature */
1220 struct crypto_hash;
1221
1222 union tcp_md5_addr {
1223 struct in_addr a4;
1224 #if IS_ENABLED(CONFIG_IPV6)
1225 struct in6_addr a6;
1226 #endif
1227 };
1228
1229 /* - key database */
1230 struct tcp_md5sig_key {
1231 struct hlist_node node;
1232 u8 keylen;
1233 u8 family; /* AF_INET or AF_INET6 */
1234 union tcp_md5_addr addr;
1235 u8 key[TCP_MD5SIG_MAXKEYLEN];
1236 struct rcu_head rcu;
1237 };
1238
1239 /* - sock block */
1240 struct tcp_md5sig_info {
1241 struct hlist_head head;
1242 struct rcu_head rcu;
1243 };
1244
1245 /* - pseudo header */
1246 struct tcp4_pseudohdr {
1247 __be32 saddr;
1248 __be32 daddr;
1249 __u8 pad;
1250 __u8 protocol;
1251 __be16 len;
1252 };
1253
1254 struct tcp6_pseudohdr {
1255 struct in6_addr saddr;
1256 struct in6_addr daddr;
1257 __be32 len;
1258 __be32 protocol; /* including padding */
1259 };
1260
1261 union tcp_md5sum_block {
1262 struct tcp4_pseudohdr ip4;
1263 #if IS_ENABLED(CONFIG_IPV6)
1264 struct tcp6_pseudohdr ip6;
1265 #endif
1266 };
1267
1268 /* - pool: digest algorithm, hash description and scratch buffer */
1269 struct tcp_md5sig_pool {
1270 struct hash_desc md5_desc;
1271 union tcp_md5sum_block md5_blk;
1272 };
1273
1274 /* - functions */
1275 extern int tcp_v4_md5_hash_skb(char *md5_hash, struct tcp_md5sig_key *key,
1276 const struct sock *sk,
1277 const struct request_sock *req,
1278 const struct sk_buff *skb);
1279 extern int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1280 int family, const u8 *newkey,
1281 u8 newkeylen, gfp_t gfp);
1282 extern int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1283 int family);
1284 extern struct tcp_md5sig_key *tcp_v4_md5_lookup(struct sock *sk,
1285 struct sock *addr_sk);
1286
1287 #ifdef CONFIG_TCP_MD5SIG
1288 extern struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1289 const union tcp_md5_addr *addr, int family);
1290 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1291 #else
1292 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(struct sock *sk,
1293 const union tcp_md5_addr *addr,
1294 int family)
1295 {
1296 return NULL;
1297 }
1298 #define tcp_twsk_md5_key(twsk) NULL
1299 #endif
1300
1301 extern bool tcp_alloc_md5sig_pool(void);
1302
1303 extern struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1304 static inline void tcp_put_md5sig_pool(void)
1305 {
1306 local_bh_enable();
1307 }
1308
1309 extern int tcp_md5_hash_header(struct tcp_md5sig_pool *, const struct tcphdr *);
1310 extern int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1311 unsigned int header_len);
1312 extern int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1313 const struct tcp_md5sig_key *key);
1314
1315 /* From tcp_fastopen.c */
1316 extern void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1317 struct tcp_fastopen_cookie *cookie,
1318 int *syn_loss, unsigned long *last_syn_loss);
1319 extern void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1320 struct tcp_fastopen_cookie *cookie,
1321 bool syn_lost);
1322 struct tcp_fastopen_request {
1323 /* Fast Open cookie. Size 0 means a cookie request */
1324 struct tcp_fastopen_cookie cookie;
1325 struct msghdr *data; /* data in MSG_FASTOPEN */
1326 size_t size;
1327 int copied; /* queued in tcp_connect() */
1328 };
1329 void tcp_free_fastopen_req(struct tcp_sock *tp);
1330
1331 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1332 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1333 extern void tcp_fastopen_cookie_gen(__be32 src, __be32 dst,
1334 struct tcp_fastopen_cookie *foc);
1335
1336 #define TCP_FASTOPEN_KEY_LENGTH 16
1337
1338 /* Fastopen key context */
1339 struct tcp_fastopen_context {
1340 struct crypto_cipher *tfm;
1341 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1342 struct rcu_head rcu;
1343 };
1344
1345 /* write queue abstraction */
1346 static inline void tcp_write_queue_purge(struct sock *sk)
1347 {
1348 struct sk_buff *skb;
1349
1350 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1351 sk_wmem_free_skb(sk, skb);
1352 sk_mem_reclaim(sk);
1353 tcp_clear_all_retrans_hints(tcp_sk(sk));
1354 }
1355
1356 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1357 {
1358 return skb_peek(&sk->sk_write_queue);
1359 }
1360
1361 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1362 {
1363 return skb_peek_tail(&sk->sk_write_queue);
1364 }
1365
1366 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1367 const struct sk_buff *skb)
1368 {
1369 return skb_queue_next(&sk->sk_write_queue, skb);
1370 }
1371
1372 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1373 const struct sk_buff *skb)
1374 {
1375 return skb_queue_prev(&sk->sk_write_queue, skb);
1376 }
1377
1378 #define tcp_for_write_queue(skb, sk) \
1379 skb_queue_walk(&(sk)->sk_write_queue, skb)
1380
1381 #define tcp_for_write_queue_from(skb, sk) \
1382 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1383
1384 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1385 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1386
1387 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1388 {
1389 return sk->sk_send_head;
1390 }
1391
1392 static inline bool tcp_skb_is_last(const struct sock *sk,
1393 const struct sk_buff *skb)
1394 {
1395 return skb_queue_is_last(&sk->sk_write_queue, skb);
1396 }
1397
1398 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1399 {
1400 if (tcp_skb_is_last(sk, skb))
1401 sk->sk_send_head = NULL;
1402 else
1403 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1404 }
1405
1406 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1407 {
1408 if (sk->sk_send_head == skb_unlinked)
1409 sk->sk_send_head = NULL;
1410 }
1411
1412 static inline void tcp_init_send_head(struct sock *sk)
1413 {
1414 sk->sk_send_head = NULL;
1415 }
1416
1417 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1418 {
1419 __skb_queue_tail(&sk->sk_write_queue, skb);
1420 }
1421
1422 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1423 {
1424 __tcp_add_write_queue_tail(sk, skb);
1425
1426 /* Queue it, remembering where we must start sending. */
1427 if (sk->sk_send_head == NULL) {
1428 sk->sk_send_head = skb;
1429
1430 if (tcp_sk(sk)->highest_sack == NULL)
1431 tcp_sk(sk)->highest_sack = skb;
1432 }
1433 }
1434
1435 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1436 {
1437 __skb_queue_head(&sk->sk_write_queue, skb);
1438 }
1439
1440 /* Insert buff after skb on the write queue of sk. */
1441 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1442 struct sk_buff *buff,
1443 struct sock *sk)
1444 {
1445 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1446 }
1447
1448 /* Insert new before skb on the write queue of sk. */
1449 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1450 struct sk_buff *skb,
1451 struct sock *sk)
1452 {
1453 __skb_queue_before(&sk->sk_write_queue, skb, new);
1454
1455 if (sk->sk_send_head == skb)
1456 sk->sk_send_head = new;
1457 }
1458
1459 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1460 {
1461 __skb_unlink(skb, &sk->sk_write_queue);
1462 }
1463
1464 static inline bool tcp_write_queue_empty(struct sock *sk)
1465 {
1466 return skb_queue_empty(&sk->sk_write_queue);
1467 }
1468
1469 static inline void tcp_push_pending_frames(struct sock *sk)
1470 {
1471 if (tcp_send_head(sk)) {
1472 struct tcp_sock *tp = tcp_sk(sk);
1473
1474 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1475 }
1476 }
1477
1478 /* Start sequence of the skb just after the highest skb with SACKed
1479 * bit, valid only if sacked_out > 0 or when the caller has ensured
1480 * validity by itself.
1481 */
1482 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1483 {
1484 if (!tp->sacked_out)
1485 return tp->snd_una;
1486
1487 if (tp->highest_sack == NULL)
1488 return tp->snd_nxt;
1489
1490 return TCP_SKB_CB(tp->highest_sack)->seq;
1491 }
1492
1493 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1494 {
1495 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1496 tcp_write_queue_next(sk, skb);
1497 }
1498
1499 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1500 {
1501 return tcp_sk(sk)->highest_sack;
1502 }
1503
1504 static inline void tcp_highest_sack_reset(struct sock *sk)
1505 {
1506 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1507 }
1508
1509 /* Called when old skb is about to be deleted (to be combined with new skb) */
1510 static inline void tcp_highest_sack_combine(struct sock *sk,
1511 struct sk_buff *old,
1512 struct sk_buff *new)
1513 {
1514 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1515 tcp_sk(sk)->highest_sack = new;
1516 }
1517
1518 /* Determines whether this is a thin stream (which may suffer from
1519 * increased latency). Used to trigger latency-reducing mechanisms.
1520 */
1521 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1522 {
1523 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1524 }
1525
1526 /* /proc */
1527 enum tcp_seq_states {
1528 TCP_SEQ_STATE_LISTENING,
1529 TCP_SEQ_STATE_OPENREQ,
1530 TCP_SEQ_STATE_ESTABLISHED,
1531 TCP_SEQ_STATE_TIME_WAIT,
1532 };
1533
1534 int tcp_seq_open(struct inode *inode, struct file *file);
1535
1536 struct tcp_seq_afinfo {
1537 char *name;
1538 sa_family_t family;
1539 const struct file_operations *seq_fops;
1540 struct seq_operations seq_ops;
1541 };
1542
1543 struct tcp_iter_state {
1544 struct seq_net_private p;
1545 sa_family_t family;
1546 enum tcp_seq_states state;
1547 struct sock *syn_wait_sk;
1548 int bucket, offset, sbucket, num;
1549 kuid_t uid;
1550 loff_t last_pos;
1551 };
1552
1553 extern int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1554 extern void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1555
1556 extern struct request_sock_ops tcp_request_sock_ops;
1557 extern struct request_sock_ops tcp6_request_sock_ops;
1558
1559 extern void tcp_v4_destroy_sock(struct sock *sk);
1560
1561 extern struct sk_buff *tcp_tso_segment(struct sk_buff *skb,
1562 netdev_features_t features);
1563 extern struct sk_buff **tcp_gro_receive(struct sk_buff **head,
1564 struct sk_buff *skb);
1565 extern int tcp_gro_complete(struct sk_buff *skb);
1566
1567 extern void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr,
1568 __be32 daddr);
1569
1570 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1571 {
1572 return tp->notsent_lowat ?: sysctl_tcp_notsent_lowat;
1573 }
1574
1575 static inline bool tcp_stream_memory_free(const struct sock *sk)
1576 {
1577 const struct tcp_sock *tp = tcp_sk(sk);
1578 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1579
1580 return notsent_bytes < tcp_notsent_lowat(tp);
1581 }
1582
1583 #ifdef CONFIG_PROC_FS
1584 extern int tcp4_proc_init(void);
1585 extern void tcp4_proc_exit(void);
1586 #endif
1587
1588 /* TCP af-specific functions */
1589 struct tcp_sock_af_ops {
1590 #ifdef CONFIG_TCP_MD5SIG
1591 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1592 struct sock *addr_sk);
1593 int (*calc_md5_hash) (char *location,
1594 struct tcp_md5sig_key *md5,
1595 const struct sock *sk,
1596 const struct request_sock *req,
1597 const struct sk_buff *skb);
1598 int (*md5_parse) (struct sock *sk,
1599 char __user *optval,
1600 int optlen);
1601 #endif
1602 };
1603
1604 struct tcp_request_sock_ops {
1605 #ifdef CONFIG_TCP_MD5SIG
1606 struct tcp_md5sig_key *(*md5_lookup) (struct sock *sk,
1607 struct request_sock *req);
1608 int (*calc_md5_hash) (char *location,
1609 struct tcp_md5sig_key *md5,
1610 const struct sock *sk,
1611 const struct request_sock *req,
1612 const struct sk_buff *skb);
1613 #endif
1614 };
1615
1616 extern int tcpv4_offload_init(void);
1617
1618 extern void tcp_v4_init(void);
1619 extern void tcp_init(void);
1620
1621 #endif /* _TCP_H */
Linux with added FPC-III support