]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - include/net/tcp.h
ASoC: max98504: Add missing MAX98504 on SND_SOC_ALL_CODECS
[mirror_ubuntu-bionic-kernel.git] / include / net / tcp.h
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/cryptohash.h>
31 #include <linux/kref.h>
32 #include <linux/ktime.h>
33
34 #include <net/inet_connection_sock.h>
35 #include <net/inet_timewait_sock.h>
36 #include <net/inet_hashtables.h>
37 #include <net/checksum.h>
38 #include <net/request_sock.h>
39 #include <net/sock.h>
40 #include <net/snmp.h>
41 #include <net/ip.h>
42 #include <net/tcp_states.h>
43 #include <net/inet_ecn.h>
44 #include <net/dst.h>
45
46 #include <linux/seq_file.h>
47 #include <linux/memcontrol.h>
48
49 extern struct inet_hashinfo tcp_hashinfo;
50
51 extern struct percpu_counter tcp_orphan_count;
52 void tcp_time_wait(struct sock *sk, int state, int timeo);
53
54 #define MAX_TCP_HEADER (128 + MAX_HEADER)
55 #define MAX_TCP_OPTION_SPACE 40
56
57 /*
58 * Never offer a window over 32767 without using window scaling. Some
59 * poor stacks do signed 16bit maths!
60 */
61 #define MAX_TCP_WINDOW 32767U
62
63 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
64 #define TCP_MIN_MSS 88U
65
66 /* The least MTU to use for probing */
67 #define TCP_BASE_MSS 1024
68
69 /* probing interval, default to 10 minutes as per RFC4821 */
70 #define TCP_PROBE_INTERVAL 600
71
72 /* Specify interval when tcp mtu probing will stop */
73 #define TCP_PROBE_THRESHOLD 8
74
75 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
76 #define TCP_FASTRETRANS_THRESH 3
77
78 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
79 #define TCP_MAX_QUICKACKS 16U
80
81 /* urg_data states */
82 #define TCP_URG_VALID 0x0100
83 #define TCP_URG_NOTYET 0x0200
84 #define TCP_URG_READ 0x0400
85
86 #define TCP_RETR1 3 /*
87 * This is how many retries it does before it
88 * tries to figure out if the gateway is
89 * down. Minimal RFC value is 3; it corresponds
90 * to ~3sec-8min depending on RTO.
91 */
92
93 #define TCP_RETR2 15 /*
94 * This should take at least
95 * 90 minutes to time out.
96 * RFC1122 says that the limit is 100 sec.
97 * 15 is ~13-30min depending on RTO.
98 */
99
100 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
101 * when active opening a connection.
102 * RFC1122 says the minimum retry MUST
103 * be at least 180secs. Nevertheless
104 * this value is corresponding to
105 * 63secs of retransmission with the
106 * current initial RTO.
107 */
108
109 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
110 * when passive opening a connection.
111 * This is corresponding to 31secs of
112 * retransmission with the current
113 * initial RTO.
114 */
115
116 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
117 * state, about 60 seconds */
118 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
119 /* BSD style FIN_WAIT2 deadlock breaker.
120 * It used to be 3min, new value is 60sec,
121 * to combine FIN-WAIT-2 timeout with
122 * TIME-WAIT timer.
123 */
124
125 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
126 #if HZ >= 100
127 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
128 #define TCP_ATO_MIN ((unsigned)(HZ/25))
129 #else
130 #define TCP_DELACK_MIN 4U
131 #define TCP_ATO_MIN 4U
132 #endif
133 #define TCP_RTO_MAX ((unsigned)(120*HZ))
134 #define TCP_RTO_MIN ((unsigned)(HZ/5))
135 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
136 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
137 * used as a fallback RTO for the
138 * initial data transmission if no
139 * valid RTT sample has been acquired,
140 * most likely due to retrans in 3WHS.
141 */
142
143 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
144 * for local resources.
145 */
146
147 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
148 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
149 #define TCP_KEEPALIVE_INTVL (75*HZ)
150
151 #define MAX_TCP_KEEPIDLE 32767
152 #define MAX_TCP_KEEPINTVL 32767
153 #define MAX_TCP_KEEPCNT 127
154 #define MAX_TCP_SYNCNT 127
155
156 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
157
158 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
159 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
160 * after this time. It should be equal
161 * (or greater than) TCP_TIMEWAIT_LEN
162 * to provide reliability equal to one
163 * provided by timewait state.
164 */
165 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
166 * timestamps. It must be less than
167 * minimal timewait lifetime.
168 */
169 /*
170 * TCP option
171 */
172
173 #define TCPOPT_NOP 1 /* Padding */
174 #define TCPOPT_EOL 0 /* End of options */
175 #define TCPOPT_MSS 2 /* Segment size negotiating */
176 #define TCPOPT_WINDOW 3 /* Window scaling */
177 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
178 #define TCPOPT_SACK 5 /* SACK Block */
179 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
180 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
181 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
182 #define TCPOPT_EXP 254 /* Experimental */
183 /* Magic number to be after the option value for sharing TCP
184 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
185 */
186 #define TCPOPT_FASTOPEN_MAGIC 0xF989
187
188 /*
189 * TCP option lengths
190 */
191
192 #define TCPOLEN_MSS 4
193 #define TCPOLEN_WINDOW 3
194 #define TCPOLEN_SACK_PERM 2
195 #define TCPOLEN_TIMESTAMP 10
196 #define TCPOLEN_MD5SIG 18
197 #define TCPOLEN_FASTOPEN_BASE 2
198 #define TCPOLEN_EXP_FASTOPEN_BASE 4
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 #define TCPOLEN_MD5SIG_ALIGNED 20
208 #define TCPOLEN_MSS_ALIGNED 4
209
210 /* Flags in tp->nonagle */
211 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
212 #define TCP_NAGLE_CORK 2 /* Socket is corked */
213 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
214
215 /* TCP thin-stream limits */
216 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
217
218 /* TCP initial congestion window as per rfc6928 */
219 #define TCP_INIT_CWND 10
220
221 /* Bit Flags for sysctl_tcp_fastopen */
222 #define TFO_CLIENT_ENABLE 1
223 #define TFO_SERVER_ENABLE 2
224 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
225
226 /* Accept SYN data w/o any cookie option */
227 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
228
229 /* Force enable TFO on all listeners, i.e., not requiring the
230 * TCP_FASTOPEN socket option.
231 */
232 #define TFO_SERVER_WO_SOCKOPT1 0x400
233
234 extern struct inet_timewait_death_row tcp_death_row;
235
236 /* sysctl variables for tcp */
237 extern int sysctl_tcp_timestamps;
238 extern int sysctl_tcp_window_scaling;
239 extern int sysctl_tcp_sack;
240 extern int sysctl_tcp_fastopen;
241 extern int sysctl_tcp_retrans_collapse;
242 extern int sysctl_tcp_stdurg;
243 extern int sysctl_tcp_rfc1337;
244 extern int sysctl_tcp_abort_on_overflow;
245 extern int sysctl_tcp_max_orphans;
246 extern int sysctl_tcp_fack;
247 extern int sysctl_tcp_reordering;
248 extern int sysctl_tcp_max_reordering;
249 extern int sysctl_tcp_dsack;
250 extern long sysctl_tcp_mem[3];
251 extern int sysctl_tcp_wmem[3];
252 extern int sysctl_tcp_rmem[3];
253 extern int sysctl_tcp_app_win;
254 extern int sysctl_tcp_adv_win_scale;
255 extern int sysctl_tcp_tw_reuse;
256 extern int sysctl_tcp_frto;
257 extern int sysctl_tcp_low_latency;
258 extern int sysctl_tcp_nometrics_save;
259 extern int sysctl_tcp_moderate_rcvbuf;
260 extern int sysctl_tcp_tso_win_divisor;
261 extern int sysctl_tcp_workaround_signed_windows;
262 extern int sysctl_tcp_slow_start_after_idle;
263 extern int sysctl_tcp_thin_linear_timeouts;
264 extern int sysctl_tcp_thin_dupack;
265 extern int sysctl_tcp_early_retrans;
266 extern int sysctl_tcp_limit_output_bytes;
267 extern int sysctl_tcp_challenge_ack_limit;
268 extern int sysctl_tcp_min_tso_segs;
269 extern int sysctl_tcp_min_rtt_wlen;
270 extern int sysctl_tcp_autocorking;
271 extern int sysctl_tcp_invalid_ratelimit;
272 extern int sysctl_tcp_pacing_ss_ratio;
273 extern int sysctl_tcp_pacing_ca_ratio;
274
275 extern atomic_long_t tcp_memory_allocated;
276 extern struct percpu_counter tcp_sockets_allocated;
277 extern int tcp_memory_pressure;
278
279 /* optimized version of sk_under_memory_pressure() for TCP sockets */
280 static inline bool tcp_under_memory_pressure(const struct sock *sk)
281 {
282 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
283 mem_cgroup_under_socket_pressure(sk->sk_memcg))
284 return true;
285
286 return tcp_memory_pressure;
287 }
288 /*
289 * The next routines deal with comparing 32 bit unsigned ints
290 * and worry about wraparound (automatic with unsigned arithmetic).
291 */
292
293 static inline bool before(__u32 seq1, __u32 seq2)
294 {
295 return (__s32)(seq1-seq2) < 0;
296 }
297 #define after(seq2, seq1) before(seq1, seq2)
298
299 /* is s2<=s1<=s3 ? */
300 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
301 {
302 return seq3 - seq2 >= seq1 - seq2;
303 }
304
305 static inline bool tcp_out_of_memory(struct sock *sk)
306 {
307 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
308 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
309 return true;
310 return false;
311 }
312
313 void sk_forced_mem_schedule(struct sock *sk, int size);
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 bool tcp_check_oom(struct sock *sk, int shift);
329
330
331 extern struct proto tcp_prot;
332
333 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
334 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
335 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
336 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
337
338 void tcp_tasklet_init(void);
339
340 void tcp_v4_err(struct sk_buff *skb, u32);
341
342 void tcp_shutdown(struct sock *sk, int how);
343
344 void tcp_v4_early_demux(struct sk_buff *skb);
345 int tcp_v4_rcv(struct sk_buff *skb);
346
347 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
348 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
349 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
350 int flags);
351 void tcp_release_cb(struct sock *sk);
352 void tcp_wfree(struct sk_buff *skb);
353 void tcp_write_timer_handler(struct sock *sk);
354 void tcp_delack_timer_handler(struct sock *sk);
355 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
356 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
357 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
358 const struct tcphdr *th, unsigned int len);
359 void tcp_rcv_space_adjust(struct sock *sk);
360 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
361 void tcp_twsk_destructor(struct sock *sk);
362 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
363 struct pipe_inode_info *pipe, size_t len,
364 unsigned int flags);
365
366 static inline void tcp_dec_quickack_mode(struct sock *sk,
367 const unsigned int pkts)
368 {
369 struct inet_connection_sock *icsk = inet_csk(sk);
370
371 if (icsk->icsk_ack.quick) {
372 if (pkts >= icsk->icsk_ack.quick) {
373 icsk->icsk_ack.quick = 0;
374 /* Leaving quickack mode we deflate ATO. */
375 icsk->icsk_ack.ato = TCP_ATO_MIN;
376 } else
377 icsk->icsk_ack.quick -= pkts;
378 }
379 }
380
381 #define TCP_ECN_OK 1
382 #define TCP_ECN_QUEUE_CWR 2
383 #define TCP_ECN_DEMAND_CWR 4
384 #define TCP_ECN_SEEN 8
385
386 enum tcp_tw_status {
387 TCP_TW_SUCCESS = 0,
388 TCP_TW_RST = 1,
389 TCP_TW_ACK = 2,
390 TCP_TW_SYN = 3
391 };
392
393
394 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
395 struct sk_buff *skb,
396 const struct tcphdr *th);
397 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
398 struct request_sock *req, bool fastopen);
399 int tcp_child_process(struct sock *parent, struct sock *child,
400 struct sk_buff *skb);
401 void tcp_enter_loss(struct sock *sk);
402 void tcp_clear_retrans(struct tcp_sock *tp);
403 void tcp_update_metrics(struct sock *sk);
404 void tcp_init_metrics(struct sock *sk);
405 void tcp_metrics_init(void);
406 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst,
407 bool paws_check, bool timestamps);
408 bool tcp_remember_stamp(struct sock *sk);
409 bool tcp_tw_remember_stamp(struct inet_timewait_sock *tw);
410 void tcp_fetch_timewait_stamp(struct sock *sk, struct dst_entry *dst);
411 void tcp_disable_fack(struct tcp_sock *tp);
412 void tcp_close(struct sock *sk, long timeout);
413 void tcp_init_sock(struct sock *sk);
414 unsigned int tcp_poll(struct file *file, struct socket *sock,
415 struct poll_table_struct *wait);
416 int tcp_getsockopt(struct sock *sk, int level, int optname,
417 char __user *optval, int __user *optlen);
418 int tcp_setsockopt(struct sock *sk, int level, int optname,
419 char __user *optval, unsigned int optlen);
420 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
421 char __user *optval, int __user *optlen);
422 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
423 char __user *optval, unsigned int optlen);
424 void tcp_set_keepalive(struct sock *sk, int val);
425 void tcp_syn_ack_timeout(const struct request_sock *req);
426 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
427 int flags, int *addr_len);
428 void tcp_parse_options(const struct sk_buff *skb,
429 struct tcp_options_received *opt_rx,
430 int estab, struct tcp_fastopen_cookie *foc);
431 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
432
433 /*
434 * TCP v4 functions exported for the inet6 API
435 */
436
437 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
438 void tcp_v4_mtu_reduced(struct sock *sk);
439 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
440 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
441 struct sock *tcp_create_openreq_child(const struct sock *sk,
442 struct request_sock *req,
443 struct sk_buff *skb);
444 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
445 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
446 struct request_sock *req,
447 struct dst_entry *dst,
448 struct request_sock *req_unhash,
449 bool *own_req);
450 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
451 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
452 int tcp_connect(struct sock *sk);
453 enum tcp_synack_type {
454 TCP_SYNACK_NORMAL,
455 TCP_SYNACK_FASTOPEN,
456 TCP_SYNACK_COOKIE,
457 };
458 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
459 struct request_sock *req,
460 struct tcp_fastopen_cookie *foc,
461 enum tcp_synack_type synack_type);
462 int tcp_disconnect(struct sock *sk, int flags);
463
464 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
465 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
466 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
467
468 /* From syncookies.c */
469 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
470 struct request_sock *req,
471 struct dst_entry *dst);
472 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
473 u32 cookie);
474 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
475 #ifdef CONFIG_SYN_COOKIES
476
477 /* Syncookies use a monotonic timer which increments every 60 seconds.
478 * This counter is used both as a hash input and partially encoded into
479 * the cookie value. A cookie is only validated further if the delta
480 * between the current counter value and the encoded one is less than this,
481 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
482 * the counter advances immediately after a cookie is generated).
483 */
484 #define MAX_SYNCOOKIE_AGE 2
485 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
486 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
487
488 /* syncookies: remember time of last synqueue overflow
489 * But do not dirty this field too often (once per second is enough)
490 * It is racy as we do not hold a lock, but race is very minor.
491 */
492 static inline void tcp_synq_overflow(const struct sock *sk)
493 {
494 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
495 unsigned long now = jiffies;
496
497 if (time_after(now, last_overflow + HZ))
498 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
499 }
500
501 /* syncookies: no recent synqueue overflow on this listening socket? */
502 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
503 {
504 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
505
506 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
507 }
508
509 static inline u32 tcp_cookie_time(void)
510 {
511 u64 val = get_jiffies_64();
512
513 do_div(val, TCP_SYNCOOKIE_PERIOD);
514 return val;
515 }
516
517 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
518 u16 *mssp);
519 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
520 __u32 cookie_init_timestamp(struct request_sock *req);
521 bool cookie_timestamp_decode(struct tcp_options_received *opt);
522 bool cookie_ecn_ok(const struct tcp_options_received *opt,
523 const struct net *net, const struct dst_entry *dst);
524
525 /* From net/ipv6/syncookies.c */
526 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
527 u32 cookie);
528 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
529
530 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
531 const struct tcphdr *th, u16 *mssp);
532 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
533 #endif
534 /* tcp_output.c */
535
536 u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
537 int min_tso_segs);
538 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
539 int nonagle);
540 bool tcp_may_send_now(struct sock *sk);
541 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
542 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
543 void tcp_retransmit_timer(struct sock *sk);
544 void tcp_xmit_retransmit_queue(struct sock *);
545 void tcp_simple_retransmit(struct sock *);
546 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
547 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
548
549 void tcp_send_probe0(struct sock *);
550 void tcp_send_partial(struct sock *);
551 int tcp_write_wakeup(struct sock *, int mib);
552 void tcp_send_fin(struct sock *sk);
553 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
554 int tcp_send_synack(struct sock *);
555 void tcp_push_one(struct sock *, unsigned int mss_now);
556 void tcp_send_ack(struct sock *sk);
557 void tcp_send_delayed_ack(struct sock *sk);
558 void tcp_send_loss_probe(struct sock *sk);
559 bool tcp_schedule_loss_probe(struct sock *sk);
560 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
561 const struct sk_buff *next_skb);
562
563 /* tcp_input.c */
564 void tcp_resume_early_retransmit(struct sock *sk);
565 void tcp_rearm_rto(struct sock *sk);
566 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
567 void tcp_reset(struct sock *sk);
568 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
569 void tcp_fin(struct sock *sk);
570
571 /* tcp_timer.c */
572 void tcp_init_xmit_timers(struct sock *);
573 static inline void tcp_clear_xmit_timers(struct sock *sk)
574 {
575 inet_csk_clear_xmit_timers(sk);
576 }
577
578 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
579 unsigned int tcp_current_mss(struct sock *sk);
580
581 /* Bound MSS / TSO packet size with the half of the window */
582 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
583 {
584 int cutoff;
585
586 /* When peer uses tiny windows, there is no use in packetizing
587 * to sub-MSS pieces for the sake of SWS or making sure there
588 * are enough packets in the pipe for fast recovery.
589 *
590 * On the other hand, for extremely large MSS devices, handling
591 * smaller than MSS windows in this way does make sense.
592 */
593 if (tp->max_window > TCP_MSS_DEFAULT)
594 cutoff = (tp->max_window >> 1);
595 else
596 cutoff = tp->max_window;
597
598 if (cutoff && pktsize > cutoff)
599 return max_t(int, cutoff, 68U - tp->tcp_header_len);
600 else
601 return pktsize;
602 }
603
604 /* tcp.c */
605 void tcp_get_info(struct sock *, struct tcp_info *);
606
607 /* Read 'sendfile()'-style from a TCP socket */
608 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
609 sk_read_actor_t recv_actor);
610
611 void tcp_initialize_rcv_mss(struct sock *sk);
612
613 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
614 int tcp_mss_to_mtu(struct sock *sk, int mss);
615 void tcp_mtup_init(struct sock *sk);
616 void tcp_init_buffer_space(struct sock *sk);
617
618 static inline void tcp_bound_rto(const struct sock *sk)
619 {
620 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
621 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
622 }
623
624 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
625 {
626 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
627 }
628
629 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
630 {
631 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
632 ntohl(TCP_FLAG_ACK) |
633 snd_wnd);
634 }
635
636 static inline void tcp_fast_path_on(struct tcp_sock *tp)
637 {
638 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
639 }
640
641 static inline void tcp_fast_path_check(struct sock *sk)
642 {
643 struct tcp_sock *tp = tcp_sk(sk);
644
645 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
646 tp->rcv_wnd &&
647 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
648 !tp->urg_data)
649 tcp_fast_path_on(tp);
650 }
651
652 /* Compute the actual rto_min value */
653 static inline u32 tcp_rto_min(struct sock *sk)
654 {
655 const struct dst_entry *dst = __sk_dst_get(sk);
656 u32 rto_min = TCP_RTO_MIN;
657
658 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
659 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
660 return rto_min;
661 }
662
663 static inline u32 tcp_rto_min_us(struct sock *sk)
664 {
665 return jiffies_to_usecs(tcp_rto_min(sk));
666 }
667
668 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
669 {
670 return dst_metric_locked(dst, RTAX_CC_ALGO);
671 }
672
673 /* Minimum RTT in usec. ~0 means not available. */
674 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
675 {
676 return minmax_get(&tp->rtt_min);
677 }
678
679 /* Compute the actual receive window we are currently advertising.
680 * Rcv_nxt can be after the window if our peer push more data
681 * than the offered window.
682 */
683 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
684 {
685 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
686
687 if (win < 0)
688 win = 0;
689 return (u32) win;
690 }
691
692 /* Choose a new window, without checks for shrinking, and without
693 * scaling applied to the result. The caller does these things
694 * if necessary. This is a "raw" window selection.
695 */
696 u32 __tcp_select_window(struct sock *sk);
697
698 void tcp_send_window_probe(struct sock *sk);
699
700 /* TCP timestamps are only 32-bits, this causes a slight
701 * complication on 64-bit systems since we store a snapshot
702 * of jiffies in the buffer control blocks below. We decided
703 * to use only the low 32-bits of jiffies and hide the ugly
704 * casts with the following macro.
705 */
706 #define tcp_time_stamp ((__u32)(jiffies))
707
708 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
709 {
710 return skb->skb_mstamp.stamp_jiffies;
711 }
712
713
714 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
715
716 #define TCPHDR_FIN 0x01
717 #define TCPHDR_SYN 0x02
718 #define TCPHDR_RST 0x04
719 #define TCPHDR_PSH 0x08
720 #define TCPHDR_ACK 0x10
721 #define TCPHDR_URG 0x20
722 #define TCPHDR_ECE 0x40
723 #define TCPHDR_CWR 0x80
724
725 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
726
727 /* This is what the send packet queuing engine uses to pass
728 * TCP per-packet control information to the transmission code.
729 * We also store the host-order sequence numbers in here too.
730 * This is 44 bytes if IPV6 is enabled.
731 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
732 */
733 struct tcp_skb_cb {
734 __u32 seq; /* Starting sequence number */
735 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
736 union {
737 /* Note : tcp_tw_isn is used in input path only
738 * (isn chosen by tcp_timewait_state_process())
739 *
740 * tcp_gso_segs/size are used in write queue only,
741 * cf tcp_skb_pcount()/tcp_skb_mss()
742 */
743 __u32 tcp_tw_isn;
744 struct {
745 u16 tcp_gso_segs;
746 u16 tcp_gso_size;
747 };
748 };
749 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
750
751 __u8 sacked; /* State flags for SACK/FACK. */
752 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
753 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
754 #define TCPCB_LOST 0x04 /* SKB is lost */
755 #define TCPCB_TAGBITS 0x07 /* All tag bits */
756 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
757 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
758 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
759 TCPCB_REPAIRED)
760
761 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
762 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
763 eor:1, /* Is skb MSG_EOR marked? */
764 unused:6;
765 __u32 ack_seq; /* Sequence number ACK'd */
766 union {
767 struct {
768 /* There is space for up to 24 bytes */
769 __u32 in_flight:30,/* Bytes in flight at transmit */
770 is_app_limited:1, /* cwnd not fully used? */
771 unused:1;
772 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
773 __u32 delivered;
774 /* start of send pipeline phase */
775 struct skb_mstamp first_tx_mstamp;
776 /* when we reached the "delivered" count */
777 struct skb_mstamp delivered_mstamp;
778 } tx; /* only used for outgoing skbs */
779 union {
780 struct inet_skb_parm h4;
781 #if IS_ENABLED(CONFIG_IPV6)
782 struct inet6_skb_parm h6;
783 #endif
784 } header; /* For incoming skbs */
785 };
786 };
787
788 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
789
790
791 #if IS_ENABLED(CONFIG_IPV6)
792 /* This is the variant of inet6_iif() that must be used by TCP,
793 * as TCP moves IP6CB into a different location in skb->cb[]
794 */
795 static inline int tcp_v6_iif(const struct sk_buff *skb)
796 {
797 bool l3_slave = skb_l3mdev_slave(TCP_SKB_CB(skb)->header.h6.flags);
798
799 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
800 }
801 #endif
802
803 /* Due to TSO, an SKB can be composed of multiple actual
804 * packets. To keep these tracked properly, we use this.
805 */
806 static inline int tcp_skb_pcount(const struct sk_buff *skb)
807 {
808 return TCP_SKB_CB(skb)->tcp_gso_segs;
809 }
810
811 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
812 {
813 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
814 }
815
816 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
817 {
818 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
819 }
820
821 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
822 static inline int tcp_skb_mss(const struct sk_buff *skb)
823 {
824 return TCP_SKB_CB(skb)->tcp_gso_size;
825 }
826
827 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
828 {
829 return likely(!TCP_SKB_CB(skb)->eor);
830 }
831
832 /* Events passed to congestion control interface */
833 enum tcp_ca_event {
834 CA_EVENT_TX_START, /* first transmit when no packets in flight */
835 CA_EVENT_CWND_RESTART, /* congestion window restart */
836 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
837 CA_EVENT_LOSS, /* loss timeout */
838 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
839 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
840 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
841 CA_EVENT_NON_DELAYED_ACK,
842 };
843
844 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
845 enum tcp_ca_ack_event_flags {
846 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
847 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
848 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
849 };
850
851 /*
852 * Interface for adding new TCP congestion control handlers
853 */
854 #define TCP_CA_NAME_MAX 16
855 #define TCP_CA_MAX 128
856 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
857
858 #define TCP_CA_UNSPEC 0
859
860 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
861 #define TCP_CONG_NON_RESTRICTED 0x1
862 /* Requires ECN/ECT set on all packets */
863 #define TCP_CONG_NEEDS_ECN 0x2
864
865 union tcp_cc_info;
866
867 struct ack_sample {
868 u32 pkts_acked;
869 s32 rtt_us;
870 u32 in_flight;
871 };
872
873 /* A rate sample measures the number of (original/retransmitted) data
874 * packets delivered "delivered" over an interval of time "interval_us".
875 * The tcp_rate.c code fills in the rate sample, and congestion
876 * control modules that define a cong_control function to run at the end
877 * of ACK processing can optionally chose to consult this sample when
878 * setting cwnd and pacing rate.
879 * A sample is invalid if "delivered" or "interval_us" is negative.
880 */
881 struct rate_sample {
882 struct skb_mstamp prior_mstamp; /* starting timestamp for interval */
883 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
884 s32 delivered; /* number of packets delivered over interval */
885 long interval_us; /* time for tp->delivered to incr "delivered" */
886 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
887 int losses; /* number of packets marked lost upon ACK */
888 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
889 u32 prior_in_flight; /* in flight before this ACK */
890 bool is_app_limited; /* is sample from packet with bubble in pipe? */
891 bool is_retrans; /* is sample from retransmission? */
892 };
893
894 struct tcp_congestion_ops {
895 struct list_head list;
896 u32 key;
897 u32 flags;
898
899 /* initialize private data (optional) */
900 void (*init)(struct sock *sk);
901 /* cleanup private data (optional) */
902 void (*release)(struct sock *sk);
903
904 /* return slow start threshold (required) */
905 u32 (*ssthresh)(struct sock *sk);
906 /* do new cwnd calculation (required) */
907 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
908 /* call before changing ca_state (optional) */
909 void (*set_state)(struct sock *sk, u8 new_state);
910 /* call when cwnd event occurs (optional) */
911 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
912 /* call when ack arrives (optional) */
913 void (*in_ack_event)(struct sock *sk, u32 flags);
914 /* new value of cwnd after loss (optional) */
915 u32 (*undo_cwnd)(struct sock *sk);
916 /* hook for packet ack accounting (optional) */
917 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
918 /* suggest number of segments for each skb to transmit (optional) */
919 u32 (*tso_segs_goal)(struct sock *sk);
920 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
921 u32 (*sndbuf_expand)(struct sock *sk);
922 /* call when packets are delivered to update cwnd and pacing rate,
923 * after all the ca_state processing. (optional)
924 */
925 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
926 /* get info for inet_diag (optional) */
927 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
928 union tcp_cc_info *info);
929
930 char name[TCP_CA_NAME_MAX];
931 struct module *owner;
932 };
933
934 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
935 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
936
937 void tcp_assign_congestion_control(struct sock *sk);
938 void tcp_init_congestion_control(struct sock *sk);
939 void tcp_cleanup_congestion_control(struct sock *sk);
940 int tcp_set_default_congestion_control(const char *name);
941 void tcp_get_default_congestion_control(char *name);
942 void tcp_get_available_congestion_control(char *buf, size_t len);
943 void tcp_get_allowed_congestion_control(char *buf, size_t len);
944 int tcp_set_allowed_congestion_control(char *allowed);
945 int tcp_set_congestion_control(struct sock *sk, const char *name);
946 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
947 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
948
949 u32 tcp_reno_ssthresh(struct sock *sk);
950 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
951 extern struct tcp_congestion_ops tcp_reno;
952
953 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
954 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
955 #ifdef CONFIG_INET
956 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
957 #else
958 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
959 {
960 return NULL;
961 }
962 #endif
963
964 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
965 {
966 const struct inet_connection_sock *icsk = inet_csk(sk);
967
968 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
969 }
970
971 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
972 {
973 struct inet_connection_sock *icsk = inet_csk(sk);
974
975 if (icsk->icsk_ca_ops->set_state)
976 icsk->icsk_ca_ops->set_state(sk, ca_state);
977 icsk->icsk_ca_state = ca_state;
978 }
979
980 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
981 {
982 const struct inet_connection_sock *icsk = inet_csk(sk);
983
984 if (icsk->icsk_ca_ops->cwnd_event)
985 icsk->icsk_ca_ops->cwnd_event(sk, event);
986 }
987
988 /* From tcp_rate.c */
989 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
990 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
991 struct rate_sample *rs);
992 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
993 struct skb_mstamp *now, struct rate_sample *rs);
994 void tcp_rate_check_app_limited(struct sock *sk);
995
996 /* These functions determine how the current flow behaves in respect of SACK
997 * handling. SACK is negotiated with the peer, and therefore it can vary
998 * between different flows.
999 *
1000 * tcp_is_sack - SACK enabled
1001 * tcp_is_reno - No SACK
1002 * tcp_is_fack - FACK enabled, implies SACK enabled
1003 */
1004 static inline int tcp_is_sack(const struct tcp_sock *tp)
1005 {
1006 return tp->rx_opt.sack_ok;
1007 }
1008
1009 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1010 {
1011 return !tcp_is_sack(tp);
1012 }
1013
1014 static inline bool tcp_is_fack(const struct tcp_sock *tp)
1015 {
1016 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
1017 }
1018
1019 static inline void tcp_enable_fack(struct tcp_sock *tp)
1020 {
1021 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
1022 }
1023
1024 /* TCP early-retransmit (ER) is similar to but more conservative than
1025 * the thin-dupack feature. Enable ER only if thin-dupack is disabled.
1026 */
1027 static inline void tcp_enable_early_retrans(struct tcp_sock *tp)
1028 {
1029 struct net *net = sock_net((struct sock *)tp);
1030
1031 tp->do_early_retrans = sysctl_tcp_early_retrans &&
1032 sysctl_tcp_early_retrans < 4 && !sysctl_tcp_thin_dupack &&
1033 net->ipv4.sysctl_tcp_reordering == 3;
1034 }
1035
1036 static inline void tcp_disable_early_retrans(struct tcp_sock *tp)
1037 {
1038 tp->do_early_retrans = 0;
1039 }
1040
1041 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1042 {
1043 return tp->sacked_out + tp->lost_out;
1044 }
1045
1046 /* This determines how many packets are "in the network" to the best
1047 * of our knowledge. In many cases it is conservative, but where
1048 * detailed information is available from the receiver (via SACK
1049 * blocks etc.) we can make more aggressive calculations.
1050 *
1051 * Use this for decisions involving congestion control, use just
1052 * tp->packets_out to determine if the send queue is empty or not.
1053 *
1054 * Read this equation as:
1055 *
1056 * "Packets sent once on transmission queue" MINUS
1057 * "Packets left network, but not honestly ACKed yet" PLUS
1058 * "Packets fast retransmitted"
1059 */
1060 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1061 {
1062 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1063 }
1064
1065 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1066
1067 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1068 {
1069 return tp->snd_cwnd < tp->snd_ssthresh;
1070 }
1071
1072 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1073 {
1074 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1075 }
1076
1077 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1078 {
1079 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1080 (1 << inet_csk(sk)->icsk_ca_state);
1081 }
1082
1083 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1084 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1085 * ssthresh.
1086 */
1087 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1088 {
1089 const struct tcp_sock *tp = tcp_sk(sk);
1090
1091 if (tcp_in_cwnd_reduction(sk))
1092 return tp->snd_ssthresh;
1093 else
1094 return max(tp->snd_ssthresh,
1095 ((tp->snd_cwnd >> 1) +
1096 (tp->snd_cwnd >> 2)));
1097 }
1098
1099 /* Use define here intentionally to get WARN_ON location shown at the caller */
1100 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1101
1102 void tcp_enter_cwr(struct sock *sk);
1103 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1104
1105 /* The maximum number of MSS of available cwnd for which TSO defers
1106 * sending if not using sysctl_tcp_tso_win_divisor.
1107 */
1108 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1109 {
1110 return 3;
1111 }
1112
1113 /* Returns end sequence number of the receiver's advertised window */
1114 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1115 {
1116 return tp->snd_una + tp->snd_wnd;
1117 }
1118
1119 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1120 * flexible approach. The RFC suggests cwnd should not be raised unless
1121 * it was fully used previously. And that's exactly what we do in
1122 * congestion avoidance mode. But in slow start we allow cwnd to grow
1123 * as long as the application has used half the cwnd.
1124 * Example :
1125 * cwnd is 10 (IW10), but application sends 9 frames.
1126 * We allow cwnd to reach 18 when all frames are ACKed.
1127 * This check is safe because it's as aggressive as slow start which already
1128 * risks 100% overshoot. The advantage is that we discourage application to
1129 * either send more filler packets or data to artificially blow up the cwnd
1130 * usage, and allow application-limited process to probe bw more aggressively.
1131 */
1132 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1133 {
1134 const struct tcp_sock *tp = tcp_sk(sk);
1135
1136 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1137 if (tcp_in_slow_start(tp))
1138 return tp->snd_cwnd < 2 * tp->max_packets_out;
1139
1140 return tp->is_cwnd_limited;
1141 }
1142
1143 /* Something is really bad, we could not queue an additional packet,
1144 * because qdisc is full or receiver sent a 0 window.
1145 * We do not want to add fuel to the fire, or abort too early,
1146 * so make sure the timer we arm now is at least 200ms in the future,
1147 * regardless of current icsk_rto value (as it could be ~2ms)
1148 */
1149 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1150 {
1151 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1152 }
1153
1154 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1155 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1156 unsigned long max_when)
1157 {
1158 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1159
1160 return (unsigned long)min_t(u64, when, max_when);
1161 }
1162
1163 static inline void tcp_check_probe_timer(struct sock *sk)
1164 {
1165 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1166 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1167 tcp_probe0_base(sk), TCP_RTO_MAX);
1168 }
1169
1170 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1171 {
1172 tp->snd_wl1 = seq;
1173 }
1174
1175 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1176 {
1177 tp->snd_wl1 = seq;
1178 }
1179
1180 /*
1181 * Calculate(/check) TCP checksum
1182 */
1183 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1184 __be32 daddr, __wsum base)
1185 {
1186 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1187 }
1188
1189 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1190 {
1191 return __skb_checksum_complete(skb);
1192 }
1193
1194 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1195 {
1196 return !skb_csum_unnecessary(skb) &&
1197 __tcp_checksum_complete(skb);
1198 }
1199
1200 /* Prequeue for VJ style copy to user, combined with checksumming. */
1201
1202 static inline void tcp_prequeue_init(struct tcp_sock *tp)
1203 {
1204 tp->ucopy.task = NULL;
1205 tp->ucopy.len = 0;
1206 tp->ucopy.memory = 0;
1207 skb_queue_head_init(&tp->ucopy.prequeue);
1208 }
1209
1210 bool tcp_prequeue(struct sock *sk, struct sk_buff *skb);
1211 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1212
1213 #undef STATE_TRACE
1214
1215 #ifdef STATE_TRACE
1216 static const char *statename[]={
1217 "Unused","Established","Syn Sent","Syn Recv",
1218 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1219 "Close Wait","Last ACK","Listen","Closing"
1220 };
1221 #endif
1222 void tcp_set_state(struct sock *sk, int state);
1223
1224 void tcp_done(struct sock *sk);
1225
1226 int tcp_abort(struct sock *sk, int err);
1227
1228 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1229 {
1230 rx_opt->dsack = 0;
1231 rx_opt->num_sacks = 0;
1232 }
1233
1234 u32 tcp_default_init_rwnd(u32 mss);
1235 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1236
1237 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1238 {
1239 struct tcp_sock *tp = tcp_sk(sk);
1240 s32 delta;
1241
1242 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out)
1243 return;
1244 delta = tcp_time_stamp - tp->lsndtime;
1245 if (delta > inet_csk(sk)->icsk_rto)
1246 tcp_cwnd_restart(sk, delta);
1247 }
1248
1249 /* Determine a window scaling and initial window to offer. */
1250 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1251 __u32 *window_clamp, int wscale_ok,
1252 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1253
1254 static inline int tcp_win_from_space(int space)
1255 {
1256 return sysctl_tcp_adv_win_scale<=0 ?
1257 (space>>(-sysctl_tcp_adv_win_scale)) :
1258 space - (space>>sysctl_tcp_adv_win_scale);
1259 }
1260
1261 /* Note: caller must be prepared to deal with negative returns */
1262 static inline int tcp_space(const struct sock *sk)
1263 {
1264 return tcp_win_from_space(sk->sk_rcvbuf -
1265 atomic_read(&sk->sk_rmem_alloc));
1266 }
1267
1268 static inline int tcp_full_space(const struct sock *sk)
1269 {
1270 return tcp_win_from_space(sk->sk_rcvbuf);
1271 }
1272
1273 extern void tcp_openreq_init_rwin(struct request_sock *req,
1274 const struct sock *sk_listener,
1275 const struct dst_entry *dst);
1276
1277 void tcp_enter_memory_pressure(struct sock *sk);
1278
1279 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1280 {
1281 struct net *net = sock_net((struct sock *)tp);
1282
1283 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1284 }
1285
1286 static inline int keepalive_time_when(const struct tcp_sock *tp)
1287 {
1288 struct net *net = sock_net((struct sock *)tp);
1289
1290 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1291 }
1292
1293 static inline int keepalive_probes(const struct tcp_sock *tp)
1294 {
1295 struct net *net = sock_net((struct sock *)tp);
1296
1297 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1298 }
1299
1300 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1301 {
1302 const struct inet_connection_sock *icsk = &tp->inet_conn;
1303
1304 return min_t(u32, tcp_time_stamp - icsk->icsk_ack.lrcvtime,
1305 tcp_time_stamp - tp->rcv_tstamp);
1306 }
1307
1308 static inline int tcp_fin_time(const struct sock *sk)
1309 {
1310 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1311 const int rto = inet_csk(sk)->icsk_rto;
1312
1313 if (fin_timeout < (rto << 2) - (rto >> 1))
1314 fin_timeout = (rto << 2) - (rto >> 1);
1315
1316 return fin_timeout;
1317 }
1318
1319 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1320 int paws_win)
1321 {
1322 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1323 return true;
1324 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1325 return true;
1326 /*
1327 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1328 * then following tcp messages have valid values. Ignore 0 value,
1329 * or else 'negative' tsval might forbid us to accept their packets.
1330 */
1331 if (!rx_opt->ts_recent)
1332 return true;
1333 return false;
1334 }
1335
1336 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1337 int rst)
1338 {
1339 if (tcp_paws_check(rx_opt, 0))
1340 return false;
1341
1342 /* RST segments are not recommended to carry timestamp,
1343 and, if they do, it is recommended to ignore PAWS because
1344 "their cleanup function should take precedence over timestamps."
1345 Certainly, it is mistake. It is necessary to understand the reasons
1346 of this constraint to relax it: if peer reboots, clock may go
1347 out-of-sync and half-open connections will not be reset.
1348 Actually, the problem would be not existing if all
1349 the implementations followed draft about maintaining clock
1350 via reboots. Linux-2.2 DOES NOT!
1351
1352 However, we can relax time bounds for RST segments to MSL.
1353 */
1354 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1355 return false;
1356 return true;
1357 }
1358
1359 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1360 int mib_idx, u32 *last_oow_ack_time);
1361
1362 static inline void tcp_mib_init(struct net *net)
1363 {
1364 /* See RFC 2012 */
1365 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1366 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1367 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1368 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1369 }
1370
1371 /* from STCP */
1372 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1373 {
1374 tp->lost_skb_hint = NULL;
1375 }
1376
1377 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1378 {
1379 tcp_clear_retrans_hints_partial(tp);
1380 tp->retransmit_skb_hint = NULL;
1381 }
1382
1383 union tcp_md5_addr {
1384 struct in_addr a4;
1385 #if IS_ENABLED(CONFIG_IPV6)
1386 struct in6_addr a6;
1387 #endif
1388 };
1389
1390 /* - key database */
1391 struct tcp_md5sig_key {
1392 struct hlist_node node;
1393 u8 keylen;
1394 u8 family; /* AF_INET or AF_INET6 */
1395 union tcp_md5_addr addr;
1396 u8 key[TCP_MD5SIG_MAXKEYLEN];
1397 struct rcu_head rcu;
1398 };
1399
1400 /* - sock block */
1401 struct tcp_md5sig_info {
1402 struct hlist_head head;
1403 struct rcu_head rcu;
1404 };
1405
1406 /* - pseudo header */
1407 struct tcp4_pseudohdr {
1408 __be32 saddr;
1409 __be32 daddr;
1410 __u8 pad;
1411 __u8 protocol;
1412 __be16 len;
1413 };
1414
1415 struct tcp6_pseudohdr {
1416 struct in6_addr saddr;
1417 struct in6_addr daddr;
1418 __be32 len;
1419 __be32 protocol; /* including padding */
1420 };
1421
1422 union tcp_md5sum_block {
1423 struct tcp4_pseudohdr ip4;
1424 #if IS_ENABLED(CONFIG_IPV6)
1425 struct tcp6_pseudohdr ip6;
1426 #endif
1427 };
1428
1429 /* - pool: digest algorithm, hash description and scratch buffer */
1430 struct tcp_md5sig_pool {
1431 struct ahash_request *md5_req;
1432 void *scratch;
1433 };
1434
1435 /* - functions */
1436 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1437 const struct sock *sk, const struct sk_buff *skb);
1438 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1439 int family, const u8 *newkey, u8 newkeylen, gfp_t gfp);
1440 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1441 int family);
1442 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1443 const struct sock *addr_sk);
1444
1445 #ifdef CONFIG_TCP_MD5SIG
1446 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1447 const union tcp_md5_addr *addr,
1448 int family);
1449 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1450 #else
1451 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1452 const union tcp_md5_addr *addr,
1453 int family)
1454 {
1455 return NULL;
1456 }
1457 #define tcp_twsk_md5_key(twsk) NULL
1458 #endif
1459
1460 bool tcp_alloc_md5sig_pool(void);
1461
1462 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1463 static inline void tcp_put_md5sig_pool(void)
1464 {
1465 local_bh_enable();
1466 }
1467
1468 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1469 unsigned int header_len);
1470 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1471 const struct tcp_md5sig_key *key);
1472
1473 /* From tcp_fastopen.c */
1474 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1475 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1476 unsigned long *last_syn_loss);
1477 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1478 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1479 u16 try_exp);
1480 struct tcp_fastopen_request {
1481 /* Fast Open cookie. Size 0 means a cookie request */
1482 struct tcp_fastopen_cookie cookie;
1483 struct msghdr *data; /* data in MSG_FASTOPEN */
1484 size_t size;
1485 int copied; /* queued in tcp_connect() */
1486 };
1487 void tcp_free_fastopen_req(struct tcp_sock *tp);
1488
1489 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1490 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1491 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1492 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1493 struct request_sock *req,
1494 struct tcp_fastopen_cookie *foc,
1495 struct dst_entry *dst);
1496 void tcp_fastopen_init_key_once(bool publish);
1497 #define TCP_FASTOPEN_KEY_LENGTH 16
1498
1499 /* Fastopen key context */
1500 struct tcp_fastopen_context {
1501 struct crypto_cipher *tfm;
1502 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1503 struct rcu_head rcu;
1504 };
1505
1506 /* write queue abstraction */
1507 static inline void tcp_write_queue_purge(struct sock *sk)
1508 {
1509 struct sk_buff *skb;
1510
1511 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1512 sk_wmem_free_skb(sk, skb);
1513 sk_mem_reclaim(sk);
1514 tcp_clear_all_retrans_hints(tcp_sk(sk));
1515 }
1516
1517 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1518 {
1519 return skb_peek(&sk->sk_write_queue);
1520 }
1521
1522 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1523 {
1524 return skb_peek_tail(&sk->sk_write_queue);
1525 }
1526
1527 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1528 const struct sk_buff *skb)
1529 {
1530 return skb_queue_next(&sk->sk_write_queue, skb);
1531 }
1532
1533 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1534 const struct sk_buff *skb)
1535 {
1536 return skb_queue_prev(&sk->sk_write_queue, skb);
1537 }
1538
1539 #define tcp_for_write_queue(skb, sk) \
1540 skb_queue_walk(&(sk)->sk_write_queue, skb)
1541
1542 #define tcp_for_write_queue_from(skb, sk) \
1543 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1544
1545 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1546 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1547
1548 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1549 {
1550 return sk->sk_send_head;
1551 }
1552
1553 static inline bool tcp_skb_is_last(const struct sock *sk,
1554 const struct sk_buff *skb)
1555 {
1556 return skb_queue_is_last(&sk->sk_write_queue, skb);
1557 }
1558
1559 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1560 {
1561 if (tcp_skb_is_last(sk, skb))
1562 sk->sk_send_head = NULL;
1563 else
1564 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1565 }
1566
1567 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1568 {
1569 if (sk->sk_send_head == skb_unlinked)
1570 sk->sk_send_head = NULL;
1571 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1572 tcp_sk(sk)->highest_sack = NULL;
1573 }
1574
1575 static inline void tcp_init_send_head(struct sock *sk)
1576 {
1577 sk->sk_send_head = NULL;
1578 }
1579
1580 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1581 {
1582 __skb_queue_tail(&sk->sk_write_queue, skb);
1583 }
1584
1585 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1586 {
1587 __tcp_add_write_queue_tail(sk, skb);
1588
1589 /* Queue it, remembering where we must start sending. */
1590 if (sk->sk_send_head == NULL) {
1591 sk->sk_send_head = skb;
1592
1593 if (tcp_sk(sk)->highest_sack == NULL)
1594 tcp_sk(sk)->highest_sack = skb;
1595 }
1596 }
1597
1598 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1599 {
1600 __skb_queue_head(&sk->sk_write_queue, skb);
1601 }
1602
1603 /* Insert buff after skb on the write queue of sk. */
1604 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1605 struct sk_buff *buff,
1606 struct sock *sk)
1607 {
1608 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1609 }
1610
1611 /* Insert new before skb on the write queue of sk. */
1612 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1613 struct sk_buff *skb,
1614 struct sock *sk)
1615 {
1616 __skb_queue_before(&sk->sk_write_queue, skb, new);
1617
1618 if (sk->sk_send_head == skb)
1619 sk->sk_send_head = new;
1620 }
1621
1622 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1623 {
1624 __skb_unlink(skb, &sk->sk_write_queue);
1625 }
1626
1627 static inline bool tcp_write_queue_empty(struct sock *sk)
1628 {
1629 return skb_queue_empty(&sk->sk_write_queue);
1630 }
1631
1632 static inline void tcp_push_pending_frames(struct sock *sk)
1633 {
1634 if (tcp_send_head(sk)) {
1635 struct tcp_sock *tp = tcp_sk(sk);
1636
1637 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1638 }
1639 }
1640
1641 /* Start sequence of the skb just after the highest skb with SACKed
1642 * bit, valid only if sacked_out > 0 or when the caller has ensured
1643 * validity by itself.
1644 */
1645 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1646 {
1647 if (!tp->sacked_out)
1648 return tp->snd_una;
1649
1650 if (tp->highest_sack == NULL)
1651 return tp->snd_nxt;
1652
1653 return TCP_SKB_CB(tp->highest_sack)->seq;
1654 }
1655
1656 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1657 {
1658 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1659 tcp_write_queue_next(sk, skb);
1660 }
1661
1662 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1663 {
1664 return tcp_sk(sk)->highest_sack;
1665 }
1666
1667 static inline void tcp_highest_sack_reset(struct sock *sk)
1668 {
1669 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1670 }
1671
1672 /* Called when old skb is about to be deleted (to be combined with new skb) */
1673 static inline void tcp_highest_sack_combine(struct sock *sk,
1674 struct sk_buff *old,
1675 struct sk_buff *new)
1676 {
1677 if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1678 tcp_sk(sk)->highest_sack = new;
1679 }
1680
1681 /* This helper checks if socket has IP_TRANSPARENT set */
1682 static inline bool inet_sk_transparent(const struct sock *sk)
1683 {
1684 switch (sk->sk_state) {
1685 case TCP_TIME_WAIT:
1686 return inet_twsk(sk)->tw_transparent;
1687 case TCP_NEW_SYN_RECV:
1688 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1689 }
1690 return inet_sk(sk)->transparent;
1691 }
1692
1693 /* Determines whether this is a thin stream (which may suffer from
1694 * increased latency). Used to trigger latency-reducing mechanisms.
1695 */
1696 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1697 {
1698 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1699 }
1700
1701 /* /proc */
1702 enum tcp_seq_states {
1703 TCP_SEQ_STATE_LISTENING,
1704 TCP_SEQ_STATE_ESTABLISHED,
1705 };
1706
1707 int tcp_seq_open(struct inode *inode, struct file *file);
1708
1709 struct tcp_seq_afinfo {
1710 char *name;
1711 sa_family_t family;
1712 const struct file_operations *seq_fops;
1713 struct seq_operations seq_ops;
1714 };
1715
1716 struct tcp_iter_state {
1717 struct seq_net_private p;
1718 sa_family_t family;
1719 enum tcp_seq_states state;
1720 struct sock *syn_wait_sk;
1721 int bucket, offset, sbucket, num;
1722 loff_t last_pos;
1723 };
1724
1725 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1726 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1727
1728 extern struct request_sock_ops tcp_request_sock_ops;
1729 extern struct request_sock_ops tcp6_request_sock_ops;
1730
1731 void tcp_v4_destroy_sock(struct sock *sk);
1732
1733 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1734 netdev_features_t features);
1735 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1736 int tcp_gro_complete(struct sk_buff *skb);
1737
1738 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1739
1740 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1741 {
1742 struct net *net = sock_net((struct sock *)tp);
1743 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1744 }
1745
1746 static inline bool tcp_stream_memory_free(const struct sock *sk)
1747 {
1748 const struct tcp_sock *tp = tcp_sk(sk);
1749 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1750
1751 return notsent_bytes < tcp_notsent_lowat(tp);
1752 }
1753
1754 #ifdef CONFIG_PROC_FS
1755 int tcp4_proc_init(void);
1756 void tcp4_proc_exit(void);
1757 #endif
1758
1759 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1760 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1761 const struct tcp_request_sock_ops *af_ops,
1762 struct sock *sk, struct sk_buff *skb);
1763
1764 /* TCP af-specific functions */
1765 struct tcp_sock_af_ops {
1766 #ifdef CONFIG_TCP_MD5SIG
1767 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1768 const struct sock *addr_sk);
1769 int (*calc_md5_hash)(char *location,
1770 const struct tcp_md5sig_key *md5,
1771 const struct sock *sk,
1772 const struct sk_buff *skb);
1773 int (*md5_parse)(struct sock *sk,
1774 char __user *optval,
1775 int optlen);
1776 #endif
1777 };
1778
1779 struct tcp_request_sock_ops {
1780 u16 mss_clamp;
1781 #ifdef CONFIG_TCP_MD5SIG
1782 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1783 const struct sock *addr_sk);
1784 int (*calc_md5_hash) (char *location,
1785 const struct tcp_md5sig_key *md5,
1786 const struct sock *sk,
1787 const struct sk_buff *skb);
1788 #endif
1789 void (*init_req)(struct request_sock *req,
1790 const struct sock *sk_listener,
1791 struct sk_buff *skb);
1792 #ifdef CONFIG_SYN_COOKIES
1793 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1794 __u16 *mss);
1795 #endif
1796 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1797 const struct request_sock *req,
1798 bool *strict);
1799 __u32 (*init_seq)(const struct sk_buff *skb);
1800 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1801 struct flowi *fl, struct request_sock *req,
1802 struct tcp_fastopen_cookie *foc,
1803 enum tcp_synack_type synack_type);
1804 };
1805
1806 #ifdef CONFIG_SYN_COOKIES
1807 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1808 const struct sock *sk, struct sk_buff *skb,
1809 __u16 *mss)
1810 {
1811 tcp_synq_overflow(sk);
1812 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1813 return ops->cookie_init_seq(skb, mss);
1814 }
1815 #else
1816 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1817 const struct sock *sk, struct sk_buff *skb,
1818 __u16 *mss)
1819 {
1820 return 0;
1821 }
1822 #endif
1823
1824 int tcpv4_offload_init(void);
1825
1826 void tcp_v4_init(void);
1827 void tcp_init(void);
1828
1829 /* tcp_recovery.c */
1830
1831 /* Flags to enable various loss recovery features. See below */
1832 extern int sysctl_tcp_recovery;
1833
1834 /* Use TCP RACK to detect (some) tail and retransmit losses */
1835 #define TCP_RACK_LOST_RETRANS 0x1
1836
1837 extern int tcp_rack_mark_lost(struct sock *sk);
1838
1839 extern void tcp_rack_advance(struct tcp_sock *tp,
1840 const struct skb_mstamp *xmit_time, u8 sacked);
1841
1842 /*
1843 * Save and compile IPv4 options, return a pointer to it
1844 */
1845 static inline struct ip_options_rcu *tcp_v4_save_options(struct sk_buff *skb)
1846 {
1847 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1848 struct ip_options_rcu *dopt = NULL;
1849
1850 if (opt->optlen) {
1851 int opt_size = sizeof(*dopt) + opt->optlen;
1852
1853 dopt = kmalloc(opt_size, GFP_ATOMIC);
1854 if (dopt && __ip_options_echo(&dopt->opt, skb, opt)) {
1855 kfree(dopt);
1856 dopt = NULL;
1857 }
1858 }
1859 return dopt;
1860 }
1861
1862 /* locally generated TCP pure ACKs have skb->truesize == 2
1863 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1864 * This is much faster than dissecting the packet to find out.
1865 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1866 */
1867 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1868 {
1869 return skb->truesize == 2;
1870 }
1871
1872 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1873 {
1874 skb->truesize = 2;
1875 }
1876
1877 static inline int tcp_inq(struct sock *sk)
1878 {
1879 struct tcp_sock *tp = tcp_sk(sk);
1880 int answ;
1881
1882 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1883 answ = 0;
1884 } else if (sock_flag(sk, SOCK_URGINLINE) ||
1885 !tp->urg_data ||
1886 before(tp->urg_seq, tp->copied_seq) ||
1887 !before(tp->urg_seq, tp->rcv_nxt)) {
1888
1889 answ = tp->rcv_nxt - tp->copied_seq;
1890
1891 /* Subtract 1, if FIN was received */
1892 if (answ && sock_flag(sk, SOCK_DONE))
1893 answ--;
1894 } else {
1895 answ = tp->urg_seq - tp->copied_seq;
1896 }
1897
1898 return answ;
1899 }
1900
1901 int tcp_peek_len(struct socket *sock);
1902
1903 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1904 {
1905 u16 segs_in;
1906
1907 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1908 tp->segs_in += segs_in;
1909 if (skb->len > tcp_hdrlen(skb))
1910 tp->data_segs_in += segs_in;
1911 }
1912
1913 /*
1914 * TCP listen path runs lockless.
1915 * We forced "struct sock" to be const qualified to make sure
1916 * we don't modify one of its field by mistake.
1917 * Here, we increment sk_drops which is an atomic_t, so we can safely
1918 * make sock writable again.
1919 */
1920 static inline void tcp_listendrop(const struct sock *sk)
1921 {
1922 atomic_inc(&((struct sock *)sk)->sk_drops);
1923 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
1924 }
1925
1926 #endif /* _TCP_H */