xref: /openbmc/linux/include/net/tcp.h (revision a20eefae)
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		Definitions for the TCP module.
8  *
9  * Version:	@(#)tcp.h	1.0.5	05/23/93
10  *
11  * Authors:	Ross Biro
12  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13  */
14 #ifndef _TCP_H
15 #define _TCP_H
16 
17 #define FASTRETRANS_DEBUG 1
18 
19 #include <linux/list.h>
20 #include <linux/tcp.h>
21 #include <linux/bug.h>
22 #include <linux/slab.h>
23 #include <linux/cache.h>
24 #include <linux/percpu.h>
25 #include <linux/skbuff.h>
26 #include <linux/cryptohash.h>
27 #include <linux/kref.h>
28 #include <linux/ktime.h>
29 
30 #include <net/inet_connection_sock.h>
31 #include <net/inet_timewait_sock.h>
32 #include <net/inet_hashtables.h>
33 #include <net/checksum.h>
34 #include <net/request_sock.h>
35 #include <net/sock_reuseport.h>
36 #include <net/sock.h>
37 #include <net/snmp.h>
38 #include <net/ip.h>
39 #include <net/tcp_states.h>
40 #include <net/inet_ecn.h>
41 #include <net/dst.h>
42 
43 #include <linux/seq_file.h>
44 #include <linux/memcontrol.h>
45 #include <linux/bpf-cgroup.h>
46 #include <linux/siphash.h>
47 
48 extern struct inet_hashinfo tcp_hashinfo;
49 
50 extern struct percpu_counter tcp_orphan_count;
51 void tcp_time_wait(struct sock *sk, int state, int timeo);
52 
53 #define MAX_TCP_HEADER	(128 + MAX_HEADER)
54 #define MAX_TCP_OPTION_SPACE 40
55 #define TCP_MIN_SND_MSS		48
56 #define TCP_MIN_GSO_SIZE	(TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
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		1024
69 
70 /* probing interval, default to 10 minutes as per RFC4821 */
71 #define TCP_PROBE_INTERVAL	600
72 
73 /* Specify interval when tcp mtu probing will stop */
74 #define TCP_PROBE_THRESHOLD	8
75 
76 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
77 #define TCP_FASTRETRANS_THRESH 3
78 
79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
80 #define TCP_MAX_QUICKACKS	16U
81 
82 /* Maximal number of window scale according to RFC1323 */
83 #define TCP_MAX_WSCALE		14U
84 
85 /* urg_data states */
86 #define TCP_URG_VALID	0x0100
87 #define TCP_URG_NOTYET	0x0200
88 #define TCP_URG_READ	0x0400
89 
90 #define TCP_RETR1	3	/*
91 				 * This is how many retries it does before it
92 				 * tries to figure out if the gateway is
93 				 * down. Minimal RFC value is 3; it corresponds
94 				 * to ~3sec-8min depending on RTO.
95 				 */
96 
97 #define TCP_RETR2	15	/*
98 				 * This should take at least
99 				 * 90 minutes to time out.
100 				 * RFC1122 says that the limit is 100 sec.
101 				 * 15 is ~13-30min depending on RTO.
102 				 */
103 
104 #define TCP_SYN_RETRIES	 6	/* This is how many retries are done
105 				 * when active opening a connection.
106 				 * RFC1122 says the minimum retry MUST
107 				 * be at least 180secs.  Nevertheless
108 				 * this value is corresponding to
109 				 * 63secs of retransmission with the
110 				 * current initial RTO.
111 				 */
112 
113 #define TCP_SYNACK_RETRIES 5	/* This is how may retries are done
114 				 * when passive opening a connection.
115 				 * This is corresponding to 31secs of
116 				 * retransmission with the current
117 				 * initial RTO.
118 				 */
119 
120 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
121 				  * state, about 60 seconds	*/
122 #define TCP_FIN_TIMEOUT	TCP_TIMEWAIT_LEN
123                                  /* BSD style FIN_WAIT2 deadlock breaker.
124 				  * It used to be 3min, new value is 60sec,
125 				  * to combine FIN-WAIT-2 timeout with
126 				  * TIME-WAIT timer.
127 				  */
128 
129 #define TCP_DELACK_MAX	((unsigned)(HZ/5))	/* maximal time to delay before sending an ACK */
130 #if HZ >= 100
131 #define TCP_DELACK_MIN	((unsigned)(HZ/25))	/* minimal time to delay before sending an ACK */
132 #define TCP_ATO_MIN	((unsigned)(HZ/25))
133 #else
134 #define TCP_DELACK_MIN	4U
135 #define TCP_ATO_MIN	4U
136 #endif
137 #define TCP_RTO_MAX	((unsigned)(120*HZ))
138 #define TCP_RTO_MIN	((unsigned)(HZ/5))
139 #define TCP_TIMEOUT_MIN	(2U) /* Min timeout for TCP timers in jiffies */
140 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ))	/* RFC6298 2.1 initial RTO value	*/
141 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ))	/* RFC 1122 initial RTO value, now
142 						 * used as a fallback RTO for the
143 						 * initial data transmission if no
144 						 * valid RTT sample has been acquired,
145 						 * most likely due to retrans in 3WHS.
146 						 */
147 
148 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
149 					                 * for local resources.
150 					                 */
151 #define TCP_KEEPALIVE_TIME	(120*60*HZ)	/* two hours */
152 #define TCP_KEEPALIVE_PROBES	9		/* Max of 9 keepalive probes	*/
153 #define TCP_KEEPALIVE_INTVL	(75*HZ)
154 
155 #define MAX_TCP_KEEPIDLE	32767
156 #define MAX_TCP_KEEPINTVL	32767
157 #define MAX_TCP_KEEPCNT		127
158 #define MAX_TCP_SYNCNT		127
159 
160 #define TCP_SYNQ_INTERVAL	(HZ/5)	/* Period of SYNACK timer */
161 
162 #define TCP_PAWS_24DAYS	(60 * 60 * 24 * 24)
163 #define TCP_PAWS_MSL	60		/* Per-host timestamps are invalidated
164 					 * after this time. It should be equal
165 					 * (or greater than) TCP_TIMEWAIT_LEN
166 					 * to provide reliability equal to one
167 					 * provided by timewait state.
168 					 */
169 #define TCP_PAWS_WINDOW	1		/* Replay window for per-host
170 					 * timestamps. It must be less than
171 					 * minimal timewait lifetime.
172 					 */
173 /*
174  *	TCP option
175  */
176 
177 #define TCPOPT_NOP		1	/* Padding */
178 #define TCPOPT_EOL		0	/* End of options */
179 #define TCPOPT_MSS		2	/* Segment size negotiating */
180 #define TCPOPT_WINDOW		3	/* Window scaling */
181 #define TCPOPT_SACK_PERM        4       /* SACK Permitted */
182 #define TCPOPT_SACK             5       /* SACK Block */
183 #define TCPOPT_TIMESTAMP	8	/* Better RTT estimations/PAWS */
184 #define TCPOPT_MD5SIG		19	/* MD5 Signature (RFC2385) */
185 #define TCPOPT_FASTOPEN		34	/* Fast open (RFC7413) */
186 #define TCPOPT_EXP		254	/* Experimental */
187 /* Magic number to be after the option value for sharing TCP
188  * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
189  */
190 #define TCPOPT_FASTOPEN_MAGIC	0xF989
191 #define TCPOPT_SMC_MAGIC	0xE2D4C3D9
192 
193 /*
194  *     TCP option lengths
195  */
196 
197 #define TCPOLEN_MSS            4
198 #define TCPOLEN_WINDOW         3
199 #define TCPOLEN_SACK_PERM      2
200 #define TCPOLEN_TIMESTAMP      10
201 #define TCPOLEN_MD5SIG         18
202 #define TCPOLEN_FASTOPEN_BASE  2
203 #define TCPOLEN_EXP_FASTOPEN_BASE  4
204 #define TCPOLEN_EXP_SMC_BASE   6
205 
206 /* But this is what stacks really send out. */
207 #define TCPOLEN_TSTAMP_ALIGNED		12
208 #define TCPOLEN_WSCALE_ALIGNED		4
209 #define TCPOLEN_SACKPERM_ALIGNED	4
210 #define TCPOLEN_SACK_BASE		2
211 #define TCPOLEN_SACK_BASE_ALIGNED	4
212 #define TCPOLEN_SACK_PERBLOCK		8
213 #define TCPOLEN_MD5SIG_ALIGNED		20
214 #define TCPOLEN_MSS_ALIGNED		4
215 #define TCPOLEN_EXP_SMC_BASE_ALIGNED	8
216 
217 /* Flags in tp->nonagle */
218 #define TCP_NAGLE_OFF		1	/* Nagle's algo is disabled */
219 #define TCP_NAGLE_CORK		2	/* Socket is corked	    */
220 #define TCP_NAGLE_PUSH		4	/* Cork is overridden for already queued data */
221 
222 /* TCP thin-stream limits */
223 #define TCP_THIN_LINEAR_RETRIES 6       /* After 6 linear retries, do exp. backoff */
224 
225 /* TCP initial congestion window as per rfc6928 */
226 #define TCP_INIT_CWND		10
227 
228 /* Bit Flags for sysctl_tcp_fastopen */
229 #define	TFO_CLIENT_ENABLE	1
230 #define	TFO_SERVER_ENABLE	2
231 #define	TFO_CLIENT_NO_COOKIE	4	/* Data in SYN w/o cookie option */
232 
233 /* Accept SYN data w/o any cookie option */
234 #define	TFO_SERVER_COOKIE_NOT_REQD	0x200
235 
236 /* Force enable TFO on all listeners, i.e., not requiring the
237  * TCP_FASTOPEN socket option.
238  */
239 #define	TFO_SERVER_WO_SOCKOPT1	0x400
240 
241 
242 /* sysctl variables for tcp */
243 extern int sysctl_tcp_max_orphans;
244 extern long sysctl_tcp_mem[3];
245 
246 #define TCP_RACK_LOSS_DETECTION  0x1 /* Use RACK to detect losses */
247 #define TCP_RACK_STATIC_REO_WND  0x2 /* Use static RACK reo wnd */
248 #define TCP_RACK_NO_DUPTHRESH    0x4 /* Do not use DUPACK threshold in RACK */
249 
250 extern atomic_long_t tcp_memory_allocated;
251 extern struct percpu_counter tcp_sockets_allocated;
252 extern unsigned long tcp_memory_pressure;
253 
254 /* optimized version of sk_under_memory_pressure() for TCP sockets */
255 static inline bool tcp_under_memory_pressure(const struct sock *sk)
256 {
257 	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
258 	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
259 		return true;
260 
261 	return tcp_memory_pressure;
262 }
263 /*
264  * The next routines deal with comparing 32 bit unsigned ints
265  * and worry about wraparound (automatic with unsigned arithmetic).
266  */
267 
268 static inline bool before(__u32 seq1, __u32 seq2)
269 {
270         return (__s32)(seq1-seq2) < 0;
271 }
272 #define after(seq2, seq1) 	before(seq1, seq2)
273 
274 /* is s2<=s1<=s3 ? */
275 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
276 {
277 	return seq3 - seq2 >= seq1 - seq2;
278 }
279 
280 static inline bool tcp_out_of_memory(struct sock *sk)
281 {
282 	if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
283 	    sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
284 		return true;
285 	return false;
286 }
287 
288 void sk_forced_mem_schedule(struct sock *sk, int size);
289 
290 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
291 {
292 	struct percpu_counter *ocp = sk->sk_prot->orphan_count;
293 	int orphans = percpu_counter_read_positive(ocp);
294 
295 	if (orphans << shift > sysctl_tcp_max_orphans) {
296 		orphans = percpu_counter_sum_positive(ocp);
297 		if (orphans << shift > sysctl_tcp_max_orphans)
298 			return true;
299 	}
300 	return false;
301 }
302 
303 bool tcp_check_oom(struct sock *sk, int shift);
304 
305 
306 extern struct proto tcp_prot;
307 
308 #define TCP_INC_STATS(net, field)	SNMP_INC_STATS((net)->mib.tcp_statistics, field)
309 #define __TCP_INC_STATS(net, field)	__SNMP_INC_STATS((net)->mib.tcp_statistics, field)
310 #define TCP_DEC_STATS(net, field)	SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
311 #define TCP_ADD_STATS(net, field, val)	SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
312 
313 void tcp_tasklet_init(void);
314 
315 int tcp_v4_err(struct sk_buff *skb, u32);
316 
317 void tcp_shutdown(struct sock *sk, int how);
318 
319 int tcp_v4_early_demux(struct sk_buff *skb);
320 int tcp_v4_rcv(struct sk_buff *skb);
321 
322 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
323 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
324 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
325 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
326 		 int flags);
327 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
328 			size_t size, int flags);
329 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
330 		 size_t size, int flags);
331 void tcp_release_cb(struct sock *sk);
332 void tcp_wfree(struct sk_buff *skb);
333 void tcp_write_timer_handler(struct sock *sk);
334 void tcp_delack_timer_handler(struct sock *sk);
335 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
336 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
337 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
338 void tcp_rcv_space_adjust(struct sock *sk);
339 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
340 void tcp_twsk_destructor(struct sock *sk);
341 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
342 			struct pipe_inode_info *pipe, size_t len,
343 			unsigned int flags);
344 
345 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
346 static inline void tcp_dec_quickack_mode(struct sock *sk,
347 					 const unsigned int pkts)
348 {
349 	struct inet_connection_sock *icsk = inet_csk(sk);
350 
351 	if (icsk->icsk_ack.quick) {
352 		if (pkts >= icsk->icsk_ack.quick) {
353 			icsk->icsk_ack.quick = 0;
354 			/* Leaving quickack mode we deflate ATO. */
355 			icsk->icsk_ack.ato   = TCP_ATO_MIN;
356 		} else
357 			icsk->icsk_ack.quick -= pkts;
358 	}
359 }
360 
361 #define	TCP_ECN_OK		1
362 #define	TCP_ECN_QUEUE_CWR	2
363 #define	TCP_ECN_DEMAND_CWR	4
364 #define	TCP_ECN_SEEN		8
365 
366 enum tcp_tw_status {
367 	TCP_TW_SUCCESS = 0,
368 	TCP_TW_RST = 1,
369 	TCP_TW_ACK = 2,
370 	TCP_TW_SYN = 3
371 };
372 
373 
374 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
375 					      struct sk_buff *skb,
376 					      const struct tcphdr *th);
377 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
378 			   struct request_sock *req, bool fastopen,
379 			   bool *lost_race);
380 int tcp_child_process(struct sock *parent, struct sock *child,
381 		      struct sk_buff *skb);
382 void tcp_enter_loss(struct sock *sk);
383 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
384 void tcp_clear_retrans(struct tcp_sock *tp);
385 void tcp_update_metrics(struct sock *sk);
386 void tcp_init_metrics(struct sock *sk);
387 void tcp_metrics_init(void);
388 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
389 void tcp_close(struct sock *sk, long timeout);
390 void tcp_init_sock(struct sock *sk);
391 void tcp_init_transfer(struct sock *sk, int bpf_op);
392 __poll_t tcp_poll(struct file *file, struct socket *sock,
393 		      struct poll_table_struct *wait);
394 int tcp_getsockopt(struct sock *sk, int level, int optname,
395 		   char __user *optval, int __user *optlen);
396 int tcp_setsockopt(struct sock *sk, int level, int optname,
397 		   char __user *optval, unsigned int optlen);
398 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
399 			  char __user *optval, int __user *optlen);
400 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
401 			  char __user *optval, unsigned int optlen);
402 void tcp_set_keepalive(struct sock *sk, int val);
403 void tcp_syn_ack_timeout(const struct request_sock *req);
404 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
405 		int flags, int *addr_len);
406 int tcp_set_rcvlowat(struct sock *sk, int val);
407 void tcp_data_ready(struct sock *sk);
408 #ifdef CONFIG_MMU
409 int tcp_mmap(struct file *file, struct socket *sock,
410 	     struct vm_area_struct *vma);
411 #endif
412 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
413 		       struct tcp_options_received *opt_rx,
414 		       int estab, struct tcp_fastopen_cookie *foc);
415 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
416 
417 /*
418  *	TCP v4 functions exported for the inet6 API
419  */
420 
421 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
422 void tcp_v4_mtu_reduced(struct sock *sk);
423 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
424 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
425 struct sock *tcp_create_openreq_child(const struct sock *sk,
426 				      struct request_sock *req,
427 				      struct sk_buff *skb);
428 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
429 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
430 				  struct request_sock *req,
431 				  struct dst_entry *dst,
432 				  struct request_sock *req_unhash,
433 				  bool *own_req);
434 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
435 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
436 int tcp_connect(struct sock *sk);
437 enum tcp_synack_type {
438 	TCP_SYNACK_NORMAL,
439 	TCP_SYNACK_FASTOPEN,
440 	TCP_SYNACK_COOKIE,
441 };
442 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
443 				struct request_sock *req,
444 				struct tcp_fastopen_cookie *foc,
445 				enum tcp_synack_type synack_type);
446 int tcp_disconnect(struct sock *sk, int flags);
447 
448 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
449 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
450 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
451 
452 /* From syncookies.c */
453 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
454 				 struct request_sock *req,
455 				 struct dst_entry *dst, u32 tsoff);
456 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
457 		      u32 cookie);
458 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
459 #ifdef CONFIG_SYN_COOKIES
460 
461 /* Syncookies use a monotonic timer which increments every 60 seconds.
462  * This counter is used both as a hash input and partially encoded into
463  * the cookie value.  A cookie is only validated further if the delta
464  * between the current counter value and the encoded one is less than this,
465  * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
466  * the counter advances immediately after a cookie is generated).
467  */
468 #define MAX_SYNCOOKIE_AGE	2
469 #define TCP_SYNCOOKIE_PERIOD	(60 * HZ)
470 #define TCP_SYNCOOKIE_VALID	(MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
471 
472 /* syncookies: remember time of last synqueue overflow
473  * But do not dirty this field too often (once per second is enough)
474  * It is racy as we do not hold a lock, but race is very minor.
475  */
476 static inline void tcp_synq_overflow(const struct sock *sk)
477 {
478 	unsigned int last_overflow;
479 	unsigned int now = jiffies;
480 
481 	if (sk->sk_reuseport) {
482 		struct sock_reuseport *reuse;
483 
484 		reuse = rcu_dereference(sk->sk_reuseport_cb);
485 		if (likely(reuse)) {
486 			last_overflow = READ_ONCE(reuse->synq_overflow_ts);
487 			if (time_after32(now, last_overflow + HZ))
488 				WRITE_ONCE(reuse->synq_overflow_ts, now);
489 			return;
490 		}
491 	}
492 
493 	last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
494 	if (time_after32(now, last_overflow + HZ))
495 		tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
496 }
497 
498 /* syncookies: no recent synqueue overflow on this listening socket? */
499 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
500 {
501 	unsigned int last_overflow;
502 	unsigned int now = jiffies;
503 
504 	if (sk->sk_reuseport) {
505 		struct sock_reuseport *reuse;
506 
507 		reuse = rcu_dereference(sk->sk_reuseport_cb);
508 		if (likely(reuse)) {
509 			last_overflow = READ_ONCE(reuse->synq_overflow_ts);
510 			return time_after32(now, last_overflow +
511 					    TCP_SYNCOOKIE_VALID);
512 		}
513 	}
514 
515 	last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
516 	return time_after32(now, last_overflow + TCP_SYNCOOKIE_VALID);
517 }
518 
519 static inline u32 tcp_cookie_time(void)
520 {
521 	u64 val = get_jiffies_64();
522 
523 	do_div(val, TCP_SYNCOOKIE_PERIOD);
524 	return val;
525 }
526 
527 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
528 			      u16 *mssp);
529 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
530 u64 cookie_init_timestamp(struct request_sock *req);
531 bool cookie_timestamp_decode(const struct net *net,
532 			     struct tcp_options_received *opt);
533 bool cookie_ecn_ok(const struct tcp_options_received *opt,
534 		   const struct net *net, const struct dst_entry *dst);
535 
536 /* From net/ipv6/syncookies.c */
537 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
538 		      u32 cookie);
539 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
540 
541 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
542 			      const struct tcphdr *th, u16 *mssp);
543 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
544 #endif
545 /* tcp_output.c */
546 
547 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
548 			       int nonagle);
549 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
550 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
551 void tcp_retransmit_timer(struct sock *sk);
552 void tcp_xmit_retransmit_queue(struct sock *);
553 void tcp_simple_retransmit(struct sock *);
554 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
555 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
556 enum tcp_queue {
557 	TCP_FRAG_IN_WRITE_QUEUE,
558 	TCP_FRAG_IN_RTX_QUEUE,
559 };
560 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
561 		 struct sk_buff *skb, u32 len,
562 		 unsigned int mss_now, gfp_t gfp);
563 
564 void tcp_send_probe0(struct sock *);
565 void tcp_send_partial(struct sock *);
566 int tcp_write_wakeup(struct sock *, int mib);
567 void tcp_send_fin(struct sock *sk);
568 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
569 int tcp_send_synack(struct sock *);
570 void tcp_push_one(struct sock *, unsigned int mss_now);
571 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
572 void tcp_send_ack(struct sock *sk);
573 void tcp_send_delayed_ack(struct sock *sk);
574 void tcp_send_loss_probe(struct sock *sk);
575 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
576 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
577 			     const struct sk_buff *next_skb);
578 
579 /* tcp_input.c */
580 void tcp_rearm_rto(struct sock *sk);
581 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
582 void tcp_reset(struct sock *sk);
583 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
584 void tcp_fin(struct sock *sk);
585 
586 /* tcp_timer.c */
587 void tcp_init_xmit_timers(struct sock *);
588 static inline void tcp_clear_xmit_timers(struct sock *sk)
589 {
590 	if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
591 		__sock_put(sk);
592 
593 	if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
594 		__sock_put(sk);
595 
596 	inet_csk_clear_xmit_timers(sk);
597 }
598 
599 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
600 unsigned int tcp_current_mss(struct sock *sk);
601 
602 /* Bound MSS / TSO packet size with the half of the window */
603 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
604 {
605 	int cutoff;
606 
607 	/* When peer uses tiny windows, there is no use in packetizing
608 	 * to sub-MSS pieces for the sake of SWS or making sure there
609 	 * are enough packets in the pipe for fast recovery.
610 	 *
611 	 * On the other hand, for extremely large MSS devices, handling
612 	 * smaller than MSS windows in this way does make sense.
613 	 */
614 	if (tp->max_window > TCP_MSS_DEFAULT)
615 		cutoff = (tp->max_window >> 1);
616 	else
617 		cutoff = tp->max_window;
618 
619 	if (cutoff && pktsize > cutoff)
620 		return max_t(int, cutoff, 68U - tp->tcp_header_len);
621 	else
622 		return pktsize;
623 }
624 
625 /* tcp.c */
626 void tcp_get_info(struct sock *, struct tcp_info *);
627 
628 /* Read 'sendfile()'-style from a TCP socket */
629 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
630 		  sk_read_actor_t recv_actor);
631 
632 void tcp_initialize_rcv_mss(struct sock *sk);
633 
634 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
635 int tcp_mss_to_mtu(struct sock *sk, int mss);
636 void tcp_mtup_init(struct sock *sk);
637 void tcp_init_buffer_space(struct sock *sk);
638 
639 static inline void tcp_bound_rto(const struct sock *sk)
640 {
641 	if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
642 		inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
643 }
644 
645 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
646 {
647 	return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
648 }
649 
650 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
651 {
652 	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
653 			       ntohl(TCP_FLAG_ACK) |
654 			       snd_wnd);
655 }
656 
657 static inline void tcp_fast_path_on(struct tcp_sock *tp)
658 {
659 	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
660 }
661 
662 static inline void tcp_fast_path_check(struct sock *sk)
663 {
664 	struct tcp_sock *tp = tcp_sk(sk);
665 
666 	if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
667 	    tp->rcv_wnd &&
668 	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
669 	    !tp->urg_data)
670 		tcp_fast_path_on(tp);
671 }
672 
673 /* Compute the actual rto_min value */
674 static inline u32 tcp_rto_min(struct sock *sk)
675 {
676 	const struct dst_entry *dst = __sk_dst_get(sk);
677 	u32 rto_min = TCP_RTO_MIN;
678 
679 	if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
680 		rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
681 	return rto_min;
682 }
683 
684 static inline u32 tcp_rto_min_us(struct sock *sk)
685 {
686 	return jiffies_to_usecs(tcp_rto_min(sk));
687 }
688 
689 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
690 {
691 	return dst_metric_locked(dst, RTAX_CC_ALGO);
692 }
693 
694 /* Minimum RTT in usec. ~0 means not available. */
695 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
696 {
697 	return minmax_get(&tp->rtt_min);
698 }
699 
700 /* Compute the actual receive window we are currently advertising.
701  * Rcv_nxt can be after the window if our peer push more data
702  * than the offered window.
703  */
704 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
705 {
706 	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
707 
708 	if (win < 0)
709 		win = 0;
710 	return (u32) win;
711 }
712 
713 /* Choose a new window, without checks for shrinking, and without
714  * scaling applied to the result.  The caller does these things
715  * if necessary.  This is a "raw" window selection.
716  */
717 u32 __tcp_select_window(struct sock *sk);
718 
719 void tcp_send_window_probe(struct sock *sk);
720 
721 /* TCP uses 32bit jiffies to save some space.
722  * Note that this is different from tcp_time_stamp, which
723  * historically has been the same until linux-4.13.
724  */
725 #define tcp_jiffies32 ((u32)jiffies)
726 
727 /*
728  * Deliver a 32bit value for TCP timestamp option (RFC 7323)
729  * It is no longer tied to jiffies, but to 1 ms clock.
730  * Note: double check if you want to use tcp_jiffies32 instead of this.
731  */
732 #define TCP_TS_HZ	1000
733 
734 static inline u64 tcp_clock_ns(void)
735 {
736 	return ktime_get_ns();
737 }
738 
739 static inline u64 tcp_clock_us(void)
740 {
741 	return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
742 }
743 
744 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
745 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
746 {
747 	return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
748 }
749 
750 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
751 static inline u32 tcp_time_stamp_raw(void)
752 {
753 	return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ);
754 }
755 
756 void tcp_mstamp_refresh(struct tcp_sock *tp);
757 
758 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
759 {
760 	return max_t(s64, t1 - t0, 0);
761 }
762 
763 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
764 {
765 	return div_u64(skb->skb_mstamp_ns, NSEC_PER_SEC / TCP_TS_HZ);
766 }
767 
768 /* provide the departure time in us unit */
769 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
770 {
771 	return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
772 }
773 
774 
775 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
776 
777 #define TCPHDR_FIN 0x01
778 #define TCPHDR_SYN 0x02
779 #define TCPHDR_RST 0x04
780 #define TCPHDR_PSH 0x08
781 #define TCPHDR_ACK 0x10
782 #define TCPHDR_URG 0x20
783 #define TCPHDR_ECE 0x40
784 #define TCPHDR_CWR 0x80
785 
786 #define TCPHDR_SYN_ECN	(TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
787 
788 /* This is what the send packet queuing engine uses to pass
789  * TCP per-packet control information to the transmission code.
790  * We also store the host-order sequence numbers in here too.
791  * This is 44 bytes if IPV6 is enabled.
792  * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
793  */
794 struct tcp_skb_cb {
795 	__u32		seq;		/* Starting sequence number	*/
796 	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
797 	union {
798 		/* Note : tcp_tw_isn is used in input path only
799 		 *	  (isn chosen by tcp_timewait_state_process())
800 		 *
801 		 * 	  tcp_gso_segs/size are used in write queue only,
802 		 *	  cf tcp_skb_pcount()/tcp_skb_mss()
803 		 */
804 		__u32		tcp_tw_isn;
805 		struct {
806 			u16	tcp_gso_segs;
807 			u16	tcp_gso_size;
808 		};
809 	};
810 	__u8		tcp_flags;	/* TCP header flags. (tcp[13])	*/
811 
812 	__u8		sacked;		/* State flags for SACK.	*/
813 #define TCPCB_SACKED_ACKED	0x01	/* SKB ACK'd by a SACK block	*/
814 #define TCPCB_SACKED_RETRANS	0x02	/* SKB retransmitted		*/
815 #define TCPCB_LOST		0x04	/* SKB is lost			*/
816 #define TCPCB_TAGBITS		0x07	/* All tag bits			*/
817 #define TCPCB_REPAIRED		0x10	/* SKB repaired (no skb_mstamp_ns)	*/
818 #define TCPCB_EVER_RETRANS	0x80	/* Ever retransmitted frame	*/
819 #define TCPCB_RETRANS		(TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
820 				TCPCB_REPAIRED)
821 
822 	__u8		ip_dsfield;	/* IPv4 tos or IPv6 dsfield	*/
823 	__u8		txstamp_ack:1,	/* Record TX timestamp for ack? */
824 			eor:1,		/* Is skb MSG_EOR marked? */
825 			has_rxtstamp:1,	/* SKB has a RX timestamp	*/
826 			unused:5;
827 	__u32		ack_seq;	/* Sequence number ACK'd	*/
828 	union {
829 		struct {
830 			/* There is space for up to 24 bytes */
831 			__u32 in_flight:30,/* Bytes in flight at transmit */
832 			      is_app_limited:1, /* cwnd not fully used? */
833 			      unused:1;
834 			/* pkts S/ACKed so far upon tx of skb, incl retrans: */
835 			__u32 delivered;
836 			/* start of send pipeline phase */
837 			u64 first_tx_mstamp;
838 			/* when we reached the "delivered" count */
839 			u64 delivered_mstamp;
840 		} tx;   /* only used for outgoing skbs */
841 		union {
842 			struct inet_skb_parm	h4;
843 #if IS_ENABLED(CONFIG_IPV6)
844 			struct inet6_skb_parm	h6;
845 #endif
846 		} header;	/* For incoming skbs */
847 		struct {
848 			__u32 flags;
849 			struct sock *sk_redir;
850 			void *data_end;
851 		} bpf;
852 	};
853 };
854 
855 #define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
856 
857 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb)
858 {
859 	TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb);
860 }
861 
862 static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb)
863 {
864 	return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS;
865 }
866 
867 static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb)
868 {
869 	return TCP_SKB_CB(skb)->bpf.sk_redir;
870 }
871 
872 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb)
873 {
874 	TCP_SKB_CB(skb)->bpf.sk_redir = NULL;
875 }
876 
877 #if IS_ENABLED(CONFIG_IPV6)
878 /* This is the variant of inet6_iif() that must be used by TCP,
879  * as TCP moves IP6CB into a different location in skb->cb[]
880  */
881 static inline int tcp_v6_iif(const struct sk_buff *skb)
882 {
883 	return TCP_SKB_CB(skb)->header.h6.iif;
884 }
885 
886 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
887 {
888 	bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
889 
890 	return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
891 }
892 
893 /* TCP_SKB_CB reference means this can not be used from early demux */
894 static inline int tcp_v6_sdif(const struct sk_buff *skb)
895 {
896 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
897 	if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
898 		return TCP_SKB_CB(skb)->header.h6.iif;
899 #endif
900 	return 0;
901 }
902 #endif
903 
904 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
905 {
906 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
907 	if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
908 	    skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
909 		return true;
910 #endif
911 	return false;
912 }
913 
914 /* TCP_SKB_CB reference means this can not be used from early demux */
915 static inline int tcp_v4_sdif(struct sk_buff *skb)
916 {
917 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
918 	if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
919 		return TCP_SKB_CB(skb)->header.h4.iif;
920 #endif
921 	return 0;
922 }
923 
924 /* Due to TSO, an SKB can be composed of multiple actual
925  * packets.  To keep these tracked properly, we use this.
926  */
927 static inline int tcp_skb_pcount(const struct sk_buff *skb)
928 {
929 	return TCP_SKB_CB(skb)->tcp_gso_segs;
930 }
931 
932 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
933 {
934 	TCP_SKB_CB(skb)->tcp_gso_segs = segs;
935 }
936 
937 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
938 {
939 	TCP_SKB_CB(skb)->tcp_gso_segs += segs;
940 }
941 
942 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
943 static inline int tcp_skb_mss(const struct sk_buff *skb)
944 {
945 	return TCP_SKB_CB(skb)->tcp_gso_size;
946 }
947 
948 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
949 {
950 	return likely(!TCP_SKB_CB(skb)->eor);
951 }
952 
953 /* Events passed to congestion control interface */
954 enum tcp_ca_event {
955 	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
956 	CA_EVENT_CWND_RESTART,	/* congestion window restart */
957 	CA_EVENT_COMPLETE_CWR,	/* end of congestion recovery */
958 	CA_EVENT_LOSS,		/* loss timeout */
959 	CA_EVENT_ECN_NO_CE,	/* ECT set, but not CE marked */
960 	CA_EVENT_ECN_IS_CE,	/* received CE marked IP packet */
961 };
962 
963 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
964 enum tcp_ca_ack_event_flags {
965 	CA_ACK_SLOWPATH		= (1 << 0),	/* In slow path processing */
966 	CA_ACK_WIN_UPDATE	= (1 << 1),	/* ACK updated window */
967 	CA_ACK_ECE		= (1 << 2),	/* ECE bit is set on ack */
968 };
969 
970 /*
971  * Interface for adding new TCP congestion control handlers
972  */
973 #define TCP_CA_NAME_MAX	16
974 #define TCP_CA_MAX	128
975 #define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
976 
977 #define TCP_CA_UNSPEC	0
978 
979 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
980 #define TCP_CONG_NON_RESTRICTED 0x1
981 /* Requires ECN/ECT set on all packets */
982 #define TCP_CONG_NEEDS_ECN	0x2
983 
984 union tcp_cc_info;
985 
986 struct ack_sample {
987 	u32 pkts_acked;
988 	s32 rtt_us;
989 	u32 in_flight;
990 };
991 
992 /* A rate sample measures the number of (original/retransmitted) data
993  * packets delivered "delivered" over an interval of time "interval_us".
994  * The tcp_rate.c code fills in the rate sample, and congestion
995  * control modules that define a cong_control function to run at the end
996  * of ACK processing can optionally chose to consult this sample when
997  * setting cwnd and pacing rate.
998  * A sample is invalid if "delivered" or "interval_us" is negative.
999  */
1000 struct rate_sample {
1001 	u64  prior_mstamp; /* starting timestamp for interval */
1002 	u32  prior_delivered;	/* tp->delivered at "prior_mstamp" */
1003 	s32  delivered;		/* number of packets delivered over interval */
1004 	long interval_us;	/* time for tp->delivered to incr "delivered" */
1005 	u32 snd_interval_us;	/* snd interval for delivered packets */
1006 	u32 rcv_interval_us;	/* rcv interval for delivered packets */
1007 	long rtt_us;		/* RTT of last (S)ACKed packet (or -1) */
1008 	int  losses;		/* number of packets marked lost upon ACK */
1009 	u32  acked_sacked;	/* number of packets newly (S)ACKed upon ACK */
1010 	u32  prior_in_flight;	/* in flight before this ACK */
1011 	bool is_app_limited;	/* is sample from packet with bubble in pipe? */
1012 	bool is_retrans;	/* is sample from retransmission? */
1013 	bool is_ack_delayed;	/* is this (likely) a delayed ACK? */
1014 };
1015 
1016 struct tcp_congestion_ops {
1017 	struct list_head	list;
1018 	u32 key;
1019 	u32 flags;
1020 
1021 	/* initialize private data (optional) */
1022 	void (*init)(struct sock *sk);
1023 	/* cleanup private data  (optional) */
1024 	void (*release)(struct sock *sk);
1025 
1026 	/* return slow start threshold (required) */
1027 	u32 (*ssthresh)(struct sock *sk);
1028 	/* do new cwnd calculation (required) */
1029 	void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1030 	/* call before changing ca_state (optional) */
1031 	void (*set_state)(struct sock *sk, u8 new_state);
1032 	/* call when cwnd event occurs (optional) */
1033 	void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1034 	/* call when ack arrives (optional) */
1035 	void (*in_ack_event)(struct sock *sk, u32 flags);
1036 	/* new value of cwnd after loss (required) */
1037 	u32  (*undo_cwnd)(struct sock *sk);
1038 	/* hook for packet ack accounting (optional) */
1039 	void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1040 	/* override sysctl_tcp_min_tso_segs */
1041 	u32 (*min_tso_segs)(struct sock *sk);
1042 	/* returns the multiplier used in tcp_sndbuf_expand (optional) */
1043 	u32 (*sndbuf_expand)(struct sock *sk);
1044 	/* call when packets are delivered to update cwnd and pacing rate,
1045 	 * after all the ca_state processing. (optional)
1046 	 */
1047 	void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1048 	/* get info for inet_diag (optional) */
1049 	size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1050 			   union tcp_cc_info *info);
1051 
1052 	char 		name[TCP_CA_NAME_MAX];
1053 	struct module 	*owner;
1054 };
1055 
1056 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1057 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1058 
1059 void tcp_assign_congestion_control(struct sock *sk);
1060 void tcp_init_congestion_control(struct sock *sk);
1061 void tcp_cleanup_congestion_control(struct sock *sk);
1062 int tcp_set_default_congestion_control(struct net *net, const char *name);
1063 void tcp_get_default_congestion_control(struct net *net, char *name);
1064 void tcp_get_available_congestion_control(char *buf, size_t len);
1065 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1066 int tcp_set_allowed_congestion_control(char *allowed);
1067 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1068 			       bool reinit, bool cap_net_admin);
1069 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1070 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1071 
1072 u32 tcp_reno_ssthresh(struct sock *sk);
1073 u32 tcp_reno_undo_cwnd(struct sock *sk);
1074 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1075 extern struct tcp_congestion_ops tcp_reno;
1076 
1077 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1078 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1079 #ifdef CONFIG_INET
1080 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1081 #else
1082 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1083 {
1084 	return NULL;
1085 }
1086 #endif
1087 
1088 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1089 {
1090 	const struct inet_connection_sock *icsk = inet_csk(sk);
1091 
1092 	return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1093 }
1094 
1095 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1096 {
1097 	struct inet_connection_sock *icsk = inet_csk(sk);
1098 
1099 	if (icsk->icsk_ca_ops->set_state)
1100 		icsk->icsk_ca_ops->set_state(sk, ca_state);
1101 	icsk->icsk_ca_state = ca_state;
1102 }
1103 
1104 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1105 {
1106 	const struct inet_connection_sock *icsk = inet_csk(sk);
1107 
1108 	if (icsk->icsk_ca_ops->cwnd_event)
1109 		icsk->icsk_ca_ops->cwnd_event(sk, event);
1110 }
1111 
1112 /* From tcp_rate.c */
1113 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1114 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1115 			    struct rate_sample *rs);
1116 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1117 		  bool is_sack_reneg, struct rate_sample *rs);
1118 void tcp_rate_check_app_limited(struct sock *sk);
1119 
1120 /* These functions determine how the current flow behaves in respect of SACK
1121  * handling. SACK is negotiated with the peer, and therefore it can vary
1122  * between different flows.
1123  *
1124  * tcp_is_sack - SACK enabled
1125  * tcp_is_reno - No SACK
1126  */
1127 static inline int tcp_is_sack(const struct tcp_sock *tp)
1128 {
1129 	return likely(tp->rx_opt.sack_ok);
1130 }
1131 
1132 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1133 {
1134 	return !tcp_is_sack(tp);
1135 }
1136 
1137 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1138 {
1139 	return tp->sacked_out + tp->lost_out;
1140 }
1141 
1142 /* This determines how many packets are "in the network" to the best
1143  * of our knowledge.  In many cases it is conservative, but where
1144  * detailed information is available from the receiver (via SACK
1145  * blocks etc.) we can make more aggressive calculations.
1146  *
1147  * Use this for decisions involving congestion control, use just
1148  * tp->packets_out to determine if the send queue is empty or not.
1149  *
1150  * Read this equation as:
1151  *
1152  *	"Packets sent once on transmission queue" MINUS
1153  *	"Packets left network, but not honestly ACKed yet" PLUS
1154  *	"Packets fast retransmitted"
1155  */
1156 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1157 {
1158 	return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1159 }
1160 
1161 #define TCP_INFINITE_SSTHRESH	0x7fffffff
1162 
1163 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1164 {
1165 	return tp->snd_cwnd < tp->snd_ssthresh;
1166 }
1167 
1168 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1169 {
1170 	return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1171 }
1172 
1173 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1174 {
1175 	return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1176 	       (1 << inet_csk(sk)->icsk_ca_state);
1177 }
1178 
1179 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1180  * The exception is cwnd reduction phase, when cwnd is decreasing towards
1181  * ssthresh.
1182  */
1183 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1184 {
1185 	const struct tcp_sock *tp = tcp_sk(sk);
1186 
1187 	if (tcp_in_cwnd_reduction(sk))
1188 		return tp->snd_ssthresh;
1189 	else
1190 		return max(tp->snd_ssthresh,
1191 			   ((tp->snd_cwnd >> 1) +
1192 			    (tp->snd_cwnd >> 2)));
1193 }
1194 
1195 /* Use define here intentionally to get WARN_ON location shown at the caller */
1196 #define tcp_verify_left_out(tp)	WARN_ON(tcp_left_out(tp) > tp->packets_out)
1197 
1198 void tcp_enter_cwr(struct sock *sk);
1199 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1200 
1201 /* The maximum number of MSS of available cwnd for which TSO defers
1202  * sending if not using sysctl_tcp_tso_win_divisor.
1203  */
1204 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1205 {
1206 	return 3;
1207 }
1208 
1209 /* Returns end sequence number of the receiver's advertised window */
1210 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1211 {
1212 	return tp->snd_una + tp->snd_wnd;
1213 }
1214 
1215 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1216  * flexible approach. The RFC suggests cwnd should not be raised unless
1217  * it was fully used previously. And that's exactly what we do in
1218  * congestion avoidance mode. But in slow start we allow cwnd to grow
1219  * as long as the application has used half the cwnd.
1220  * Example :
1221  *    cwnd is 10 (IW10), but application sends 9 frames.
1222  *    We allow cwnd to reach 18 when all frames are ACKed.
1223  * This check is safe because it's as aggressive as slow start which already
1224  * risks 100% overshoot. The advantage is that we discourage application to
1225  * either send more filler packets or data to artificially blow up the cwnd
1226  * usage, and allow application-limited process to probe bw more aggressively.
1227  */
1228 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1229 {
1230 	const struct tcp_sock *tp = tcp_sk(sk);
1231 
1232 	/* If in slow start, ensure cwnd grows to twice what was ACKed. */
1233 	if (tcp_in_slow_start(tp))
1234 		return tp->snd_cwnd < 2 * tp->max_packets_out;
1235 
1236 	return tp->is_cwnd_limited;
1237 }
1238 
1239 /* BBR congestion control needs pacing.
1240  * Same remark for SO_MAX_PACING_RATE.
1241  * sch_fq packet scheduler is efficiently handling pacing,
1242  * but is not always installed/used.
1243  * Return true if TCP stack should pace packets itself.
1244  */
1245 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1246 {
1247 	return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1248 }
1249 
1250 /* Return in jiffies the delay before one skb is sent.
1251  * If @skb is NULL, we look at EDT for next packet being sent on the socket.
1252  */
1253 static inline unsigned long tcp_pacing_delay(const struct sock *sk,
1254 					     const struct sk_buff *skb)
1255 {
1256 	s64 pacing_delay = skb ? skb->tstamp : tcp_sk(sk)->tcp_wstamp_ns;
1257 
1258 	pacing_delay -= tcp_sk(sk)->tcp_clock_cache;
1259 
1260 	return pacing_delay > 0 ? nsecs_to_jiffies(pacing_delay) : 0;
1261 }
1262 
1263 static inline void tcp_reset_xmit_timer(struct sock *sk,
1264 					const int what,
1265 					unsigned long when,
1266 					const unsigned long max_when,
1267 					const struct sk_buff *skb)
1268 {
1269 	inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk, skb),
1270 				  max_when);
1271 }
1272 
1273 /* Something is really bad, we could not queue an additional packet,
1274  * because qdisc is full or receiver sent a 0 window, or we are paced.
1275  * We do not want to add fuel to the fire, or abort too early,
1276  * so make sure the timer we arm now is at least 200ms in the future,
1277  * regardless of current icsk_rto value (as it could be ~2ms)
1278  */
1279 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1280 {
1281 	return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1282 }
1283 
1284 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1285 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1286 					    unsigned long max_when)
1287 {
1288 	u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1289 
1290 	return (unsigned long)min_t(u64, when, max_when);
1291 }
1292 
1293 static inline void tcp_check_probe_timer(struct sock *sk)
1294 {
1295 	if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1296 		tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1297 				     tcp_probe0_base(sk), TCP_RTO_MAX,
1298 				     NULL);
1299 }
1300 
1301 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1302 {
1303 	tp->snd_wl1 = seq;
1304 }
1305 
1306 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1307 {
1308 	tp->snd_wl1 = seq;
1309 }
1310 
1311 /*
1312  * Calculate(/check) TCP checksum
1313  */
1314 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1315 				   __be32 daddr, __wsum base)
1316 {
1317 	return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1318 }
1319 
1320 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1321 {
1322 	return !skb_csum_unnecessary(skb) &&
1323 		__skb_checksum_complete(skb);
1324 }
1325 
1326 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1327 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1328 void tcp_set_state(struct sock *sk, int state);
1329 void tcp_done(struct sock *sk);
1330 int tcp_abort(struct sock *sk, int err);
1331 
1332 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1333 {
1334 	rx_opt->dsack = 0;
1335 	rx_opt->num_sacks = 0;
1336 }
1337 
1338 u32 tcp_default_init_rwnd(u32 mss);
1339 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1340 
1341 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1342 {
1343 	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1344 	struct tcp_sock *tp = tcp_sk(sk);
1345 	s32 delta;
1346 
1347 	if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1348 	    ca_ops->cong_control)
1349 		return;
1350 	delta = tcp_jiffies32 - tp->lsndtime;
1351 	if (delta > inet_csk(sk)->icsk_rto)
1352 		tcp_cwnd_restart(sk, delta);
1353 }
1354 
1355 /* Determine a window scaling and initial window to offer. */
1356 void tcp_select_initial_window(const struct sock *sk, int __space,
1357 			       __u32 mss, __u32 *rcv_wnd,
1358 			       __u32 *window_clamp, int wscale_ok,
1359 			       __u8 *rcv_wscale, __u32 init_rcv_wnd);
1360 
1361 static inline int tcp_win_from_space(const struct sock *sk, int space)
1362 {
1363 	int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1364 
1365 	return tcp_adv_win_scale <= 0 ?
1366 		(space>>(-tcp_adv_win_scale)) :
1367 		space - (space>>tcp_adv_win_scale);
1368 }
1369 
1370 /* Note: caller must be prepared to deal with negative returns */
1371 static inline int tcp_space(const struct sock *sk)
1372 {
1373 	return tcp_win_from_space(sk, sk->sk_rcvbuf - sk->sk_backlog.len -
1374 				  atomic_read(&sk->sk_rmem_alloc));
1375 }
1376 
1377 static inline int tcp_full_space(const struct sock *sk)
1378 {
1379 	return tcp_win_from_space(sk, sk->sk_rcvbuf);
1380 }
1381 
1382 extern void tcp_openreq_init_rwin(struct request_sock *req,
1383 				  const struct sock *sk_listener,
1384 				  const struct dst_entry *dst);
1385 
1386 void tcp_enter_memory_pressure(struct sock *sk);
1387 void tcp_leave_memory_pressure(struct sock *sk);
1388 
1389 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1390 {
1391 	struct net *net = sock_net((struct sock *)tp);
1392 
1393 	return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1394 }
1395 
1396 static inline int keepalive_time_when(const struct tcp_sock *tp)
1397 {
1398 	struct net *net = sock_net((struct sock *)tp);
1399 
1400 	return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1401 }
1402 
1403 static inline int keepalive_probes(const struct tcp_sock *tp)
1404 {
1405 	struct net *net = sock_net((struct sock *)tp);
1406 
1407 	return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1408 }
1409 
1410 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1411 {
1412 	const struct inet_connection_sock *icsk = &tp->inet_conn;
1413 
1414 	return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1415 			  tcp_jiffies32 - tp->rcv_tstamp);
1416 }
1417 
1418 static inline int tcp_fin_time(const struct sock *sk)
1419 {
1420 	int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1421 	const int rto = inet_csk(sk)->icsk_rto;
1422 
1423 	if (fin_timeout < (rto << 2) - (rto >> 1))
1424 		fin_timeout = (rto << 2) - (rto >> 1);
1425 
1426 	return fin_timeout;
1427 }
1428 
1429 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1430 				  int paws_win)
1431 {
1432 	if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1433 		return true;
1434 	if (unlikely(!time_before32(ktime_get_seconds(),
1435 				    rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1436 		return true;
1437 	/*
1438 	 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1439 	 * then following tcp messages have valid values. Ignore 0 value,
1440 	 * or else 'negative' tsval might forbid us to accept their packets.
1441 	 */
1442 	if (!rx_opt->ts_recent)
1443 		return true;
1444 	return false;
1445 }
1446 
1447 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1448 				   int rst)
1449 {
1450 	if (tcp_paws_check(rx_opt, 0))
1451 		return false;
1452 
1453 	/* RST segments are not recommended to carry timestamp,
1454 	   and, if they do, it is recommended to ignore PAWS because
1455 	   "their cleanup function should take precedence over timestamps."
1456 	   Certainly, it is mistake. It is necessary to understand the reasons
1457 	   of this constraint to relax it: if peer reboots, clock may go
1458 	   out-of-sync and half-open connections will not be reset.
1459 	   Actually, the problem would be not existing if all
1460 	   the implementations followed draft about maintaining clock
1461 	   via reboots. Linux-2.2 DOES NOT!
1462 
1463 	   However, we can relax time bounds for RST segments to MSL.
1464 	 */
1465 	if (rst && !time_before32(ktime_get_seconds(),
1466 				  rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1467 		return false;
1468 	return true;
1469 }
1470 
1471 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1472 			  int mib_idx, u32 *last_oow_ack_time);
1473 
1474 static inline void tcp_mib_init(struct net *net)
1475 {
1476 	/* See RFC 2012 */
1477 	TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1478 	TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1479 	TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1480 	TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1481 }
1482 
1483 /* from STCP */
1484 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1485 {
1486 	tp->lost_skb_hint = NULL;
1487 }
1488 
1489 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1490 {
1491 	tcp_clear_retrans_hints_partial(tp);
1492 	tp->retransmit_skb_hint = NULL;
1493 }
1494 
1495 union tcp_md5_addr {
1496 	struct in_addr  a4;
1497 #if IS_ENABLED(CONFIG_IPV6)
1498 	struct in6_addr	a6;
1499 #endif
1500 };
1501 
1502 /* - key database */
1503 struct tcp_md5sig_key {
1504 	struct hlist_node	node;
1505 	u8			keylen;
1506 	u8			family; /* AF_INET or AF_INET6 */
1507 	union tcp_md5_addr	addr;
1508 	u8			prefixlen;
1509 	u8			key[TCP_MD5SIG_MAXKEYLEN];
1510 	struct rcu_head		rcu;
1511 };
1512 
1513 /* - sock block */
1514 struct tcp_md5sig_info {
1515 	struct hlist_head	head;
1516 	struct rcu_head		rcu;
1517 };
1518 
1519 /* - pseudo header */
1520 struct tcp4_pseudohdr {
1521 	__be32		saddr;
1522 	__be32		daddr;
1523 	__u8		pad;
1524 	__u8		protocol;
1525 	__be16		len;
1526 };
1527 
1528 struct tcp6_pseudohdr {
1529 	struct in6_addr	saddr;
1530 	struct in6_addr daddr;
1531 	__be32		len;
1532 	__be32		protocol;	/* including padding */
1533 };
1534 
1535 union tcp_md5sum_block {
1536 	struct tcp4_pseudohdr ip4;
1537 #if IS_ENABLED(CONFIG_IPV6)
1538 	struct tcp6_pseudohdr ip6;
1539 #endif
1540 };
1541 
1542 /* - pool: digest algorithm, hash description and scratch buffer */
1543 struct tcp_md5sig_pool {
1544 	struct ahash_request	*md5_req;
1545 	void			*scratch;
1546 };
1547 
1548 /* - functions */
1549 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1550 			const struct sock *sk, const struct sk_buff *skb);
1551 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1552 		   int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1553 		   gfp_t gfp);
1554 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1555 		   int family, u8 prefixlen);
1556 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1557 					 const struct sock *addr_sk);
1558 
1559 #ifdef CONFIG_TCP_MD5SIG
1560 #include <linux/jump_label.h>
1561 extern struct static_key_false tcp_md5_needed;
1562 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk,
1563 					   const union tcp_md5_addr *addr,
1564 					   int family);
1565 static inline struct tcp_md5sig_key *
1566 tcp_md5_do_lookup(const struct sock *sk,
1567 		  const union tcp_md5_addr *addr,
1568 		  int family)
1569 {
1570 	if (!static_branch_unlikely(&tcp_md5_needed))
1571 		return NULL;
1572 	return __tcp_md5_do_lookup(sk, addr, family);
1573 }
1574 
1575 #define tcp_twsk_md5_key(twsk)	((twsk)->tw_md5_key)
1576 #else
1577 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1578 					 const union tcp_md5_addr *addr,
1579 					 int family)
1580 {
1581 	return NULL;
1582 }
1583 #define tcp_twsk_md5_key(twsk)	NULL
1584 #endif
1585 
1586 bool tcp_alloc_md5sig_pool(void);
1587 
1588 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1589 static inline void tcp_put_md5sig_pool(void)
1590 {
1591 	local_bh_enable();
1592 }
1593 
1594 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1595 			  unsigned int header_len);
1596 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1597 		     const struct tcp_md5sig_key *key);
1598 
1599 /* From tcp_fastopen.c */
1600 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1601 			    struct tcp_fastopen_cookie *cookie);
1602 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1603 			    struct tcp_fastopen_cookie *cookie, bool syn_lost,
1604 			    u16 try_exp);
1605 struct tcp_fastopen_request {
1606 	/* Fast Open cookie. Size 0 means a cookie request */
1607 	struct tcp_fastopen_cookie	cookie;
1608 	struct msghdr			*data;  /* data in MSG_FASTOPEN */
1609 	size_t				size;
1610 	int				copied;	/* queued in tcp_connect() */
1611 	struct ubuf_info		*uarg;
1612 };
1613 void tcp_free_fastopen_req(struct tcp_sock *tp);
1614 void tcp_fastopen_destroy_cipher(struct sock *sk);
1615 void tcp_fastopen_ctx_destroy(struct net *net);
1616 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1617 			      void *primary_key, void *backup_key);
1618 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1619 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1620 			      struct request_sock *req,
1621 			      struct tcp_fastopen_cookie *foc,
1622 			      const struct dst_entry *dst);
1623 void tcp_fastopen_init_key_once(struct net *net);
1624 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1625 			     struct tcp_fastopen_cookie *cookie);
1626 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1627 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1628 #define TCP_FASTOPEN_KEY_MAX 2
1629 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1630 	(TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1631 
1632 /* Fastopen key context */
1633 struct tcp_fastopen_context {
1634 	siphash_key_t	key[TCP_FASTOPEN_KEY_MAX];
1635 	int		num;
1636 	struct rcu_head	rcu;
1637 };
1638 
1639 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1640 void tcp_fastopen_active_disable(struct sock *sk);
1641 bool tcp_fastopen_active_should_disable(struct sock *sk);
1642 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1643 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1644 
1645 /* Caller needs to wrap with rcu_read_(un)lock() */
1646 static inline
1647 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1648 {
1649 	struct tcp_fastopen_context *ctx;
1650 
1651 	ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1652 	if (!ctx)
1653 		ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1654 	return ctx;
1655 }
1656 
1657 static inline
1658 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1659 			       const struct tcp_fastopen_cookie *orig)
1660 {
1661 	if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1662 	    orig->len == foc->len &&
1663 	    !memcmp(orig->val, foc->val, foc->len))
1664 		return true;
1665 	return false;
1666 }
1667 
1668 static inline
1669 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1670 {
1671 	return ctx->num;
1672 }
1673 
1674 /* Latencies incurred by various limits for a sender. They are
1675  * chronograph-like stats that are mutually exclusive.
1676  */
1677 enum tcp_chrono {
1678 	TCP_CHRONO_UNSPEC,
1679 	TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1680 	TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1681 	TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1682 	__TCP_CHRONO_MAX,
1683 };
1684 
1685 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1686 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1687 
1688 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1689  * the same memory storage than skb->destructor/_skb_refdst
1690  */
1691 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1692 {
1693 	skb->destructor = NULL;
1694 	skb->_skb_refdst = 0UL;
1695 }
1696 
1697 #define tcp_skb_tsorted_save(skb) {		\
1698 	unsigned long _save = skb->_skb_refdst;	\
1699 	skb->_skb_refdst = 0UL;
1700 
1701 #define tcp_skb_tsorted_restore(skb)		\
1702 	skb->_skb_refdst = _save;		\
1703 }
1704 
1705 void tcp_write_queue_purge(struct sock *sk);
1706 
1707 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1708 {
1709 	return skb_rb_first(&sk->tcp_rtx_queue);
1710 }
1711 
1712 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1713 {
1714 	return skb_peek(&sk->sk_write_queue);
1715 }
1716 
1717 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1718 {
1719 	return skb_peek_tail(&sk->sk_write_queue);
1720 }
1721 
1722 #define tcp_for_write_queue_from_safe(skb, tmp, sk)			\
1723 	skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1724 
1725 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1726 {
1727 	return skb_peek(&sk->sk_write_queue);
1728 }
1729 
1730 static inline bool tcp_skb_is_last(const struct sock *sk,
1731 				   const struct sk_buff *skb)
1732 {
1733 	return skb_queue_is_last(&sk->sk_write_queue, skb);
1734 }
1735 
1736 static inline bool tcp_write_queue_empty(const struct sock *sk)
1737 {
1738 	return skb_queue_empty(&sk->sk_write_queue);
1739 }
1740 
1741 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1742 {
1743 	return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1744 }
1745 
1746 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1747 {
1748 	return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1749 }
1750 
1751 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1752 {
1753 	__skb_queue_tail(&sk->sk_write_queue, skb);
1754 
1755 	/* Queue it, remembering where we must start sending. */
1756 	if (sk->sk_write_queue.next == skb)
1757 		tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1758 }
1759 
1760 /* Insert new before skb on the write queue of sk.  */
1761 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1762 						  struct sk_buff *skb,
1763 						  struct sock *sk)
1764 {
1765 	__skb_queue_before(&sk->sk_write_queue, skb, new);
1766 }
1767 
1768 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1769 {
1770 	tcp_skb_tsorted_anchor_cleanup(skb);
1771 	__skb_unlink(skb, &sk->sk_write_queue);
1772 }
1773 
1774 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1775 
1776 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1777 {
1778 	tcp_skb_tsorted_anchor_cleanup(skb);
1779 	rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1780 }
1781 
1782 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1783 {
1784 	list_del(&skb->tcp_tsorted_anchor);
1785 	tcp_rtx_queue_unlink(skb, sk);
1786 	sk_wmem_free_skb(sk, skb);
1787 }
1788 
1789 static inline void tcp_push_pending_frames(struct sock *sk)
1790 {
1791 	if (tcp_send_head(sk)) {
1792 		struct tcp_sock *tp = tcp_sk(sk);
1793 
1794 		__tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1795 	}
1796 }
1797 
1798 /* Start sequence of the skb just after the highest skb with SACKed
1799  * bit, valid only if sacked_out > 0 or when the caller has ensured
1800  * validity by itself.
1801  */
1802 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1803 {
1804 	if (!tp->sacked_out)
1805 		return tp->snd_una;
1806 
1807 	if (tp->highest_sack == NULL)
1808 		return tp->snd_nxt;
1809 
1810 	return TCP_SKB_CB(tp->highest_sack)->seq;
1811 }
1812 
1813 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1814 {
1815 	tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1816 }
1817 
1818 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1819 {
1820 	return tcp_sk(sk)->highest_sack;
1821 }
1822 
1823 static inline void tcp_highest_sack_reset(struct sock *sk)
1824 {
1825 	tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1826 }
1827 
1828 /* Called when old skb is about to be deleted and replaced by new skb */
1829 static inline void tcp_highest_sack_replace(struct sock *sk,
1830 					    struct sk_buff *old,
1831 					    struct sk_buff *new)
1832 {
1833 	if (old == tcp_highest_sack(sk))
1834 		tcp_sk(sk)->highest_sack = new;
1835 }
1836 
1837 /* This helper checks if socket has IP_TRANSPARENT set */
1838 static inline bool inet_sk_transparent(const struct sock *sk)
1839 {
1840 	switch (sk->sk_state) {
1841 	case TCP_TIME_WAIT:
1842 		return inet_twsk(sk)->tw_transparent;
1843 	case TCP_NEW_SYN_RECV:
1844 		return inet_rsk(inet_reqsk(sk))->no_srccheck;
1845 	}
1846 	return inet_sk(sk)->transparent;
1847 }
1848 
1849 /* Determines whether this is a thin stream (which may suffer from
1850  * increased latency). Used to trigger latency-reducing mechanisms.
1851  */
1852 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1853 {
1854 	return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1855 }
1856 
1857 /* /proc */
1858 enum tcp_seq_states {
1859 	TCP_SEQ_STATE_LISTENING,
1860 	TCP_SEQ_STATE_ESTABLISHED,
1861 };
1862 
1863 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1864 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1865 void tcp_seq_stop(struct seq_file *seq, void *v);
1866 
1867 struct tcp_seq_afinfo {
1868 	sa_family_t			family;
1869 };
1870 
1871 struct tcp_iter_state {
1872 	struct seq_net_private	p;
1873 	enum tcp_seq_states	state;
1874 	struct sock		*syn_wait_sk;
1875 	int			bucket, offset, sbucket, num;
1876 	loff_t			last_pos;
1877 };
1878 
1879 extern struct request_sock_ops tcp_request_sock_ops;
1880 extern struct request_sock_ops tcp6_request_sock_ops;
1881 
1882 void tcp_v4_destroy_sock(struct sock *sk);
1883 
1884 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1885 				netdev_features_t features);
1886 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
1887 int tcp_gro_complete(struct sk_buff *skb);
1888 
1889 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1890 
1891 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1892 {
1893 	struct net *net = sock_net((struct sock *)tp);
1894 	return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1895 }
1896 
1897 /* @wake is one when sk_stream_write_space() calls us.
1898  * This sends EPOLLOUT only if notsent_bytes is half the limit.
1899  * This mimics the strategy used in sock_def_write_space().
1900  */
1901 static inline bool tcp_stream_memory_free(const struct sock *sk, int wake)
1902 {
1903 	const struct tcp_sock *tp = tcp_sk(sk);
1904 	u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1905 
1906 	return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
1907 }
1908 
1909 #ifdef CONFIG_PROC_FS
1910 int tcp4_proc_init(void);
1911 void tcp4_proc_exit(void);
1912 #endif
1913 
1914 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1915 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1916 		     const struct tcp_request_sock_ops *af_ops,
1917 		     struct sock *sk, struct sk_buff *skb);
1918 
1919 /* TCP af-specific functions */
1920 struct tcp_sock_af_ops {
1921 #ifdef CONFIG_TCP_MD5SIG
1922 	struct tcp_md5sig_key	*(*md5_lookup) (const struct sock *sk,
1923 						const struct sock *addr_sk);
1924 	int		(*calc_md5_hash)(char *location,
1925 					 const struct tcp_md5sig_key *md5,
1926 					 const struct sock *sk,
1927 					 const struct sk_buff *skb);
1928 	int		(*md5_parse)(struct sock *sk,
1929 				     int optname,
1930 				     char __user *optval,
1931 				     int optlen);
1932 #endif
1933 };
1934 
1935 struct tcp_request_sock_ops {
1936 	u16 mss_clamp;
1937 #ifdef CONFIG_TCP_MD5SIG
1938 	struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1939 						 const struct sock *addr_sk);
1940 	int		(*calc_md5_hash) (char *location,
1941 					  const struct tcp_md5sig_key *md5,
1942 					  const struct sock *sk,
1943 					  const struct sk_buff *skb);
1944 #endif
1945 	void (*init_req)(struct request_sock *req,
1946 			 const struct sock *sk_listener,
1947 			 struct sk_buff *skb);
1948 #ifdef CONFIG_SYN_COOKIES
1949 	__u32 (*cookie_init_seq)(const struct sk_buff *skb,
1950 				 __u16 *mss);
1951 #endif
1952 	struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1953 				       const struct request_sock *req);
1954 	u32 (*init_seq)(const struct sk_buff *skb);
1955 	u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
1956 	int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1957 			   struct flowi *fl, struct request_sock *req,
1958 			   struct tcp_fastopen_cookie *foc,
1959 			   enum tcp_synack_type synack_type);
1960 };
1961 
1962 #ifdef CONFIG_SYN_COOKIES
1963 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1964 					 const struct sock *sk, struct sk_buff *skb,
1965 					 __u16 *mss)
1966 {
1967 	tcp_synq_overflow(sk);
1968 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1969 	return ops->cookie_init_seq(skb, mss);
1970 }
1971 #else
1972 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1973 					 const struct sock *sk, struct sk_buff *skb,
1974 					 __u16 *mss)
1975 {
1976 	return 0;
1977 }
1978 #endif
1979 
1980 int tcpv4_offload_init(void);
1981 
1982 void tcp_v4_init(void);
1983 void tcp_init(void);
1984 
1985 /* tcp_recovery.c */
1986 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
1987 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
1988 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
1989 				u32 reo_wnd);
1990 extern void tcp_rack_mark_lost(struct sock *sk);
1991 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
1992 			     u64 xmit_time);
1993 extern void tcp_rack_reo_timeout(struct sock *sk);
1994 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
1995 
1996 /* At how many usecs into the future should the RTO fire? */
1997 static inline s64 tcp_rto_delta_us(const struct sock *sk)
1998 {
1999 	const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2000 	u32 rto = inet_csk(sk)->icsk_rto;
2001 	u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2002 
2003 	return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2004 }
2005 
2006 /*
2007  * Save and compile IPv4 options, return a pointer to it
2008  */
2009 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2010 							 struct sk_buff *skb)
2011 {
2012 	const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2013 	struct ip_options_rcu *dopt = NULL;
2014 
2015 	if (opt->optlen) {
2016 		int opt_size = sizeof(*dopt) + opt->optlen;
2017 
2018 		dopt = kmalloc(opt_size, GFP_ATOMIC);
2019 		if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2020 			kfree(dopt);
2021 			dopt = NULL;
2022 		}
2023 	}
2024 	return dopt;
2025 }
2026 
2027 /* locally generated TCP pure ACKs have skb->truesize == 2
2028  * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2029  * This is much faster than dissecting the packet to find out.
2030  * (Think of GRE encapsulations, IPv4, IPv6, ...)
2031  */
2032 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2033 {
2034 	return skb->truesize == 2;
2035 }
2036 
2037 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2038 {
2039 	skb->truesize = 2;
2040 }
2041 
2042 static inline int tcp_inq(struct sock *sk)
2043 {
2044 	struct tcp_sock *tp = tcp_sk(sk);
2045 	int answ;
2046 
2047 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2048 		answ = 0;
2049 	} else if (sock_flag(sk, SOCK_URGINLINE) ||
2050 		   !tp->urg_data ||
2051 		   before(tp->urg_seq, tp->copied_seq) ||
2052 		   !before(tp->urg_seq, tp->rcv_nxt)) {
2053 
2054 		answ = tp->rcv_nxt - tp->copied_seq;
2055 
2056 		/* Subtract 1, if FIN was received */
2057 		if (answ && sock_flag(sk, SOCK_DONE))
2058 			answ--;
2059 	} else {
2060 		answ = tp->urg_seq - tp->copied_seq;
2061 	}
2062 
2063 	return answ;
2064 }
2065 
2066 int tcp_peek_len(struct socket *sock);
2067 
2068 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2069 {
2070 	u16 segs_in;
2071 
2072 	segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2073 	tp->segs_in += segs_in;
2074 	if (skb->len > tcp_hdrlen(skb))
2075 		tp->data_segs_in += segs_in;
2076 }
2077 
2078 /*
2079  * TCP listen path runs lockless.
2080  * We forced "struct sock" to be const qualified to make sure
2081  * we don't modify one of its field by mistake.
2082  * Here, we increment sk_drops which is an atomic_t, so we can safely
2083  * make sock writable again.
2084  */
2085 static inline void tcp_listendrop(const struct sock *sk)
2086 {
2087 	atomic_inc(&((struct sock *)sk)->sk_drops);
2088 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2089 }
2090 
2091 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2092 
2093 /*
2094  * Interface for adding Upper Level Protocols over TCP
2095  */
2096 
2097 #define TCP_ULP_NAME_MAX	16
2098 #define TCP_ULP_MAX		128
2099 #define TCP_ULP_BUF_MAX		(TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2100 
2101 struct tcp_ulp_ops {
2102 	struct list_head	list;
2103 
2104 	/* initialize ulp */
2105 	int (*init)(struct sock *sk);
2106 	/* cleanup ulp */
2107 	void (*release)(struct sock *sk);
2108 
2109 	char		name[TCP_ULP_NAME_MAX];
2110 	struct module	*owner;
2111 };
2112 int tcp_register_ulp(struct tcp_ulp_ops *type);
2113 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2114 int tcp_set_ulp(struct sock *sk, const char *name);
2115 void tcp_get_available_ulp(char *buf, size_t len);
2116 void tcp_cleanup_ulp(struct sock *sk);
2117 
2118 #define MODULE_ALIAS_TCP_ULP(name)				\
2119 	__MODULE_INFO(alias, alias_userspace, name);		\
2120 	__MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2121 
2122 struct sk_msg;
2123 struct sk_psock;
2124 
2125 int tcp_bpf_init(struct sock *sk);
2126 void tcp_bpf_reinit(struct sock *sk);
2127 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes,
2128 			  int flags);
2129 int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
2130 		    int nonblock, int flags, int *addr_len);
2131 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock,
2132 		      struct msghdr *msg, int len, int flags);
2133 
2134 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2135  * is < 0, then the BPF op failed (for example if the loaded BPF
2136  * program does not support the chosen operation or there is no BPF
2137  * program loaded).
2138  */
2139 #ifdef CONFIG_BPF
2140 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2141 {
2142 	struct bpf_sock_ops_kern sock_ops;
2143 	int ret;
2144 
2145 	memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2146 	if (sk_fullsock(sk)) {
2147 		sock_ops.is_fullsock = 1;
2148 		sock_owned_by_me(sk);
2149 	}
2150 
2151 	sock_ops.sk = sk;
2152 	sock_ops.op = op;
2153 	if (nargs > 0)
2154 		memcpy(sock_ops.args, args, nargs * sizeof(*args));
2155 
2156 	ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2157 	if (ret == 0)
2158 		ret = sock_ops.reply;
2159 	else
2160 		ret = -1;
2161 	return ret;
2162 }
2163 
2164 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2165 {
2166 	u32 args[2] = {arg1, arg2};
2167 
2168 	return tcp_call_bpf(sk, op, 2, args);
2169 }
2170 
2171 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2172 				    u32 arg3)
2173 {
2174 	u32 args[3] = {arg1, arg2, arg3};
2175 
2176 	return tcp_call_bpf(sk, op, 3, args);
2177 }
2178 
2179 #else
2180 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2181 {
2182 	return -EPERM;
2183 }
2184 
2185 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2186 {
2187 	return -EPERM;
2188 }
2189 
2190 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2191 				    u32 arg3)
2192 {
2193 	return -EPERM;
2194 }
2195 
2196 #endif
2197 
2198 static inline u32 tcp_timeout_init(struct sock *sk)
2199 {
2200 	int timeout;
2201 
2202 	timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2203 
2204 	if (timeout <= 0)
2205 		timeout = TCP_TIMEOUT_INIT;
2206 	return timeout;
2207 }
2208 
2209 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2210 {
2211 	int rwnd;
2212 
2213 	rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2214 
2215 	if (rwnd < 0)
2216 		rwnd = 0;
2217 	return rwnd;
2218 }
2219 
2220 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2221 {
2222 	return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2223 }
2224 
2225 static inline void tcp_bpf_rtt(struct sock *sk)
2226 {
2227 	if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2228 		tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2229 }
2230 
2231 #if IS_ENABLED(CONFIG_SMC)
2232 extern struct static_key_false tcp_have_smc;
2233 #endif
2234 
2235 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2236 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2237 			     void (*cad)(struct sock *sk, u32 ack_seq));
2238 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2239 void clean_acked_data_flush(void);
2240 #endif
2241 
2242 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2243 static inline void tcp_add_tx_delay(struct sk_buff *skb,
2244 				    const struct tcp_sock *tp)
2245 {
2246 	if (static_branch_unlikely(&tcp_tx_delay_enabled))
2247 		skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2248 }
2249 
2250 /* Compute Earliest Departure Time for some control packets
2251  * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2252  */
2253 static inline u64 tcp_transmit_time(const struct sock *sk)
2254 {
2255 	if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2256 		u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2257 			tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2258 
2259 		return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
2260 	}
2261 	return 0;
2262 }
2263 
2264 #endif	/* _TCP_H */
2265