xref: /linux/include/net/tcp.h (revision 995231c820e3bd3633cb38bf4ea6f2541e1da331)
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 #include <linux/bpf-cgroup.h>
49 
50 extern struct inet_hashinfo tcp_hashinfo;
51 
52 extern struct percpu_counter tcp_orphan_count;
53 void tcp_time_wait(struct sock *sk, int state, int timeo);
54 
55 #define MAX_TCP_HEADER	(128 + MAX_HEADER)
56 #define MAX_TCP_OPTION_SPACE 40
57 
58 /*
59  * Never offer a window over 32767 without using window scaling. Some
60  * poor stacks do signed 16bit maths!
61  */
62 #define MAX_TCP_WINDOW		32767U
63 
64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
65 #define TCP_MIN_MSS		88U
66 
67 /* The least MTU to use for probing */
68 #define TCP_BASE_MSS		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 extern int sysctl_tcp_wmem[3];
246 extern int sysctl_tcp_rmem[3];
247 
248 #define TCP_RACK_LOSS_DETECTION  0x1 /* Use RACK to detect losses */
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 void 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 			 const struct tcphdr *th);
339 void tcp_rcv_space_adjust(struct sock *sk);
340 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
341 void tcp_twsk_destructor(struct sock *sk);
342 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
343 			struct pipe_inode_info *pipe, size_t len,
344 			unsigned int flags);
345 
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 int tcp_child_process(struct sock *parent, struct sock *child,
380 		      struct sk_buff *skb);
381 void tcp_enter_loss(struct sock *sk);
382 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
383 void tcp_clear_retrans(struct tcp_sock *tp);
384 void tcp_update_metrics(struct sock *sk);
385 void tcp_init_metrics(struct sock *sk);
386 void tcp_metrics_init(void);
387 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
388 void tcp_disable_fack(struct tcp_sock *tp);
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 unsigned int 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 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
407 		       struct tcp_options_received *opt_rx,
408 		       int estab, struct tcp_fastopen_cookie *foc);
409 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
410 
411 /*
412  *	TCP v4 functions exported for the inet6 API
413  */
414 
415 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
416 void tcp_v4_mtu_reduced(struct sock *sk);
417 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
418 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
419 struct sock *tcp_create_openreq_child(const struct sock *sk,
420 				      struct request_sock *req,
421 				      struct sk_buff *skb);
422 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
423 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
424 				  struct request_sock *req,
425 				  struct dst_entry *dst,
426 				  struct request_sock *req_unhash,
427 				  bool *own_req);
428 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
429 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
430 int tcp_connect(struct sock *sk);
431 enum tcp_synack_type {
432 	TCP_SYNACK_NORMAL,
433 	TCP_SYNACK_FASTOPEN,
434 	TCP_SYNACK_COOKIE,
435 };
436 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
437 				struct request_sock *req,
438 				struct tcp_fastopen_cookie *foc,
439 				enum tcp_synack_type synack_type);
440 int tcp_disconnect(struct sock *sk, int flags);
441 
442 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
443 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
444 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
445 
446 /* From syncookies.c */
447 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
448 				 struct request_sock *req,
449 				 struct dst_entry *dst, u32 tsoff);
450 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
451 		      u32 cookie);
452 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
453 #ifdef CONFIG_SYN_COOKIES
454 
455 /* Syncookies use a monotonic timer which increments every 60 seconds.
456  * This counter is used both as a hash input and partially encoded into
457  * the cookie value.  A cookie is only validated further if the delta
458  * between the current counter value and the encoded one is less than this,
459  * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
460  * the counter advances immediately after a cookie is generated).
461  */
462 #define MAX_SYNCOOKIE_AGE	2
463 #define TCP_SYNCOOKIE_PERIOD	(60 * HZ)
464 #define TCP_SYNCOOKIE_VALID	(MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
465 
466 /* syncookies: remember time of last synqueue overflow
467  * But do not dirty this field too often (once per second is enough)
468  * It is racy as we do not hold a lock, but race is very minor.
469  */
470 static inline void tcp_synq_overflow(const struct sock *sk)
471 {
472 	unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
473 	unsigned long now = jiffies;
474 
475 	if (time_after(now, last_overflow + HZ))
476 		tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
477 }
478 
479 /* syncookies: no recent synqueue overflow on this listening socket? */
480 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
481 {
482 	unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
483 
484 	return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
485 }
486 
487 static inline u32 tcp_cookie_time(void)
488 {
489 	u64 val = get_jiffies_64();
490 
491 	do_div(val, TCP_SYNCOOKIE_PERIOD);
492 	return val;
493 }
494 
495 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
496 			      u16 *mssp);
497 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
498 u64 cookie_init_timestamp(struct request_sock *req);
499 bool cookie_timestamp_decode(const struct net *net,
500 			     struct tcp_options_received *opt);
501 bool cookie_ecn_ok(const struct tcp_options_received *opt,
502 		   const struct net *net, const struct dst_entry *dst);
503 
504 /* From net/ipv6/syncookies.c */
505 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
506 		      u32 cookie);
507 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
508 
509 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
510 			      const struct tcphdr *th, u16 *mssp);
511 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
512 #endif
513 /* tcp_output.c */
514 
515 u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
516 		     int min_tso_segs);
517 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
518 			       int nonagle);
519 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
520 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
521 void tcp_retransmit_timer(struct sock *sk);
522 void tcp_xmit_retransmit_queue(struct sock *);
523 void tcp_simple_retransmit(struct sock *);
524 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
525 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
526 enum tcp_queue {
527 	TCP_FRAG_IN_WRITE_QUEUE,
528 	TCP_FRAG_IN_RTX_QUEUE,
529 };
530 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
531 		 struct sk_buff *skb, u32 len,
532 		 unsigned int mss_now, gfp_t gfp);
533 
534 void tcp_send_probe0(struct sock *);
535 void tcp_send_partial(struct sock *);
536 int tcp_write_wakeup(struct sock *, int mib);
537 void tcp_send_fin(struct sock *sk);
538 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
539 int tcp_send_synack(struct sock *);
540 void tcp_push_one(struct sock *, unsigned int mss_now);
541 void tcp_send_ack(struct sock *sk);
542 void tcp_send_delayed_ack(struct sock *sk);
543 void tcp_send_loss_probe(struct sock *sk);
544 bool tcp_schedule_loss_probe(struct sock *sk);
545 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
546 			     const struct sk_buff *next_skb);
547 
548 /* tcp_input.c */
549 void tcp_rearm_rto(struct sock *sk);
550 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
551 void tcp_reset(struct sock *sk);
552 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
553 void tcp_fin(struct sock *sk);
554 
555 /* tcp_timer.c */
556 void tcp_init_xmit_timers(struct sock *);
557 static inline void tcp_clear_xmit_timers(struct sock *sk)
558 {
559 	hrtimer_cancel(&tcp_sk(sk)->pacing_timer);
560 	inet_csk_clear_xmit_timers(sk);
561 }
562 
563 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
564 unsigned int tcp_current_mss(struct sock *sk);
565 
566 /* Bound MSS / TSO packet size with the half of the window */
567 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
568 {
569 	int cutoff;
570 
571 	/* When peer uses tiny windows, there is no use in packetizing
572 	 * to sub-MSS pieces for the sake of SWS or making sure there
573 	 * are enough packets in the pipe for fast recovery.
574 	 *
575 	 * On the other hand, for extremely large MSS devices, handling
576 	 * smaller than MSS windows in this way does make sense.
577 	 */
578 	if (tp->max_window > TCP_MSS_DEFAULT)
579 		cutoff = (tp->max_window >> 1);
580 	else
581 		cutoff = tp->max_window;
582 
583 	if (cutoff && pktsize > cutoff)
584 		return max_t(int, cutoff, 68U - tp->tcp_header_len);
585 	else
586 		return pktsize;
587 }
588 
589 /* tcp.c */
590 void tcp_get_info(struct sock *, struct tcp_info *);
591 
592 /* Read 'sendfile()'-style from a TCP socket */
593 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
594 		  sk_read_actor_t recv_actor);
595 
596 void tcp_initialize_rcv_mss(struct sock *sk);
597 
598 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
599 int tcp_mss_to_mtu(struct sock *sk, int mss);
600 void tcp_mtup_init(struct sock *sk);
601 void tcp_init_buffer_space(struct sock *sk);
602 
603 static inline void tcp_bound_rto(const struct sock *sk)
604 {
605 	if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
606 		inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
607 }
608 
609 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
610 {
611 	return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
612 }
613 
614 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
615 {
616 	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
617 			       ntohl(TCP_FLAG_ACK) |
618 			       snd_wnd);
619 }
620 
621 static inline void tcp_fast_path_on(struct tcp_sock *tp)
622 {
623 	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
624 }
625 
626 static inline void tcp_fast_path_check(struct sock *sk)
627 {
628 	struct tcp_sock *tp = tcp_sk(sk);
629 
630 	if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
631 	    tp->rcv_wnd &&
632 	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
633 	    !tp->urg_data)
634 		tcp_fast_path_on(tp);
635 }
636 
637 /* Compute the actual rto_min value */
638 static inline u32 tcp_rto_min(struct sock *sk)
639 {
640 	const struct dst_entry *dst = __sk_dst_get(sk);
641 	u32 rto_min = TCP_RTO_MIN;
642 
643 	if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
644 		rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
645 	return rto_min;
646 }
647 
648 static inline u32 tcp_rto_min_us(struct sock *sk)
649 {
650 	return jiffies_to_usecs(tcp_rto_min(sk));
651 }
652 
653 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
654 {
655 	return dst_metric_locked(dst, RTAX_CC_ALGO);
656 }
657 
658 /* Minimum RTT in usec. ~0 means not available. */
659 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
660 {
661 	return minmax_get(&tp->rtt_min);
662 }
663 
664 /* Compute the actual receive window we are currently advertising.
665  * Rcv_nxt can be after the window if our peer push more data
666  * than the offered window.
667  */
668 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
669 {
670 	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
671 
672 	if (win < 0)
673 		win = 0;
674 	return (u32) win;
675 }
676 
677 /* Choose a new window, without checks for shrinking, and without
678  * scaling applied to the result.  The caller does these things
679  * if necessary.  This is a "raw" window selection.
680  */
681 u32 __tcp_select_window(struct sock *sk);
682 
683 void tcp_send_window_probe(struct sock *sk);
684 
685 /* TCP uses 32bit jiffies to save some space.
686  * Note that this is different from tcp_time_stamp, which
687  * historically has been the same until linux-4.13.
688  */
689 #define tcp_jiffies32 ((u32)jiffies)
690 
691 /*
692  * Deliver a 32bit value for TCP timestamp option (RFC 7323)
693  * It is no longer tied to jiffies, but to 1 ms clock.
694  * Note: double check if you want to use tcp_jiffies32 instead of this.
695  */
696 #define TCP_TS_HZ	1000
697 
698 static inline u64 tcp_clock_ns(void)
699 {
700 	return local_clock();
701 }
702 
703 static inline u64 tcp_clock_us(void)
704 {
705 	return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
706 }
707 
708 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
709 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
710 {
711 	return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
712 }
713 
714 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
715 static inline u32 tcp_time_stamp_raw(void)
716 {
717 	return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ);
718 }
719 
720 
721 /* Refresh 1us clock of a TCP socket,
722  * ensuring monotically increasing values.
723  */
724 static inline void tcp_mstamp_refresh(struct tcp_sock *tp)
725 {
726 	u64 val = tcp_clock_us();
727 
728 	if (val > tp->tcp_mstamp)
729 		tp->tcp_mstamp = val;
730 }
731 
732 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
733 {
734 	return max_t(s64, t1 - t0, 0);
735 }
736 
737 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
738 {
739 	return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ);
740 }
741 
742 
743 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
744 
745 #define TCPHDR_FIN 0x01
746 #define TCPHDR_SYN 0x02
747 #define TCPHDR_RST 0x04
748 #define TCPHDR_PSH 0x08
749 #define TCPHDR_ACK 0x10
750 #define TCPHDR_URG 0x20
751 #define TCPHDR_ECE 0x40
752 #define TCPHDR_CWR 0x80
753 
754 #define TCPHDR_SYN_ECN	(TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
755 
756 /* This is what the send packet queuing engine uses to pass
757  * TCP per-packet control information to the transmission code.
758  * We also store the host-order sequence numbers in here too.
759  * This is 44 bytes if IPV6 is enabled.
760  * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
761  */
762 struct tcp_skb_cb {
763 	__u32		seq;		/* Starting sequence number	*/
764 	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
765 	union {
766 		/* Note : tcp_tw_isn is used in input path only
767 		 *	  (isn chosen by tcp_timewait_state_process())
768 		 *
769 		 * 	  tcp_gso_segs/size are used in write queue only,
770 		 *	  cf tcp_skb_pcount()/tcp_skb_mss()
771 		 */
772 		__u32		tcp_tw_isn;
773 		struct {
774 			u16	tcp_gso_segs;
775 			u16	tcp_gso_size;
776 		};
777 	};
778 	__u8		tcp_flags;	/* TCP header flags. (tcp[13])	*/
779 
780 	__u8		sacked;		/* State flags for SACK/FACK.	*/
781 #define TCPCB_SACKED_ACKED	0x01	/* SKB ACK'd by a SACK block	*/
782 #define TCPCB_SACKED_RETRANS	0x02	/* SKB retransmitted		*/
783 #define TCPCB_LOST		0x04	/* SKB is lost			*/
784 #define TCPCB_TAGBITS		0x07	/* All tag bits			*/
785 #define TCPCB_REPAIRED		0x10	/* SKB repaired (no skb_mstamp)	*/
786 #define TCPCB_EVER_RETRANS	0x80	/* Ever retransmitted frame	*/
787 #define TCPCB_RETRANS		(TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
788 				TCPCB_REPAIRED)
789 
790 	__u8		ip_dsfield;	/* IPv4 tos or IPv6 dsfield	*/
791 	__u8		txstamp_ack:1,	/* Record TX timestamp for ack? */
792 			eor:1,		/* Is skb MSG_EOR marked? */
793 			has_rxtstamp:1,	/* SKB has a RX timestamp	*/
794 			unused:5;
795 	__u32		ack_seq;	/* Sequence number ACK'd	*/
796 	union {
797 		struct {
798 			/* There is space for up to 24 bytes */
799 			__u32 in_flight:30,/* Bytes in flight at transmit */
800 			      is_app_limited:1, /* cwnd not fully used? */
801 			      unused:1;
802 			/* pkts S/ACKed so far upon tx of skb, incl retrans: */
803 			__u32 delivered;
804 			/* start of send pipeline phase */
805 			u64 first_tx_mstamp;
806 			/* when we reached the "delivered" count */
807 			u64 delivered_mstamp;
808 		} tx;   /* only used for outgoing skbs */
809 		union {
810 			struct inet_skb_parm	h4;
811 #if IS_ENABLED(CONFIG_IPV6)
812 			struct inet6_skb_parm	h6;
813 #endif
814 		} header;	/* For incoming skbs */
815 		struct {
816 			__u32 key;
817 			__u32 flags;
818 			struct bpf_map *map;
819 		} bpf;
820 	};
821 };
822 
823 #define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
824 
825 
826 #if IS_ENABLED(CONFIG_IPV6)
827 /* This is the variant of inet6_iif() that must be used by TCP,
828  * as TCP moves IP6CB into a different location in skb->cb[]
829  */
830 static inline int tcp_v6_iif(const struct sk_buff *skb)
831 {
832 	bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
833 
834 	return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
835 }
836 
837 /* TCP_SKB_CB reference means this can not be used from early demux */
838 static inline int tcp_v6_sdif(const struct sk_buff *skb)
839 {
840 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
841 	if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
842 		return TCP_SKB_CB(skb)->header.h6.iif;
843 #endif
844 	return 0;
845 }
846 #endif
847 
848 /* TCP_SKB_CB reference means this can not be used from early demux */
849 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
850 {
851 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
852 	if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
853 	    skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
854 		return true;
855 #endif
856 	return false;
857 }
858 
859 /* TCP_SKB_CB reference means this can not be used from early demux */
860 static inline int tcp_v4_sdif(struct sk_buff *skb)
861 {
862 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
863 	if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
864 		return TCP_SKB_CB(skb)->header.h4.iif;
865 #endif
866 	return 0;
867 }
868 
869 /* Due to TSO, an SKB can be composed of multiple actual
870  * packets.  To keep these tracked properly, we use this.
871  */
872 static inline int tcp_skb_pcount(const struct sk_buff *skb)
873 {
874 	return TCP_SKB_CB(skb)->tcp_gso_segs;
875 }
876 
877 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
878 {
879 	TCP_SKB_CB(skb)->tcp_gso_segs = segs;
880 }
881 
882 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
883 {
884 	TCP_SKB_CB(skb)->tcp_gso_segs += segs;
885 }
886 
887 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
888 static inline int tcp_skb_mss(const struct sk_buff *skb)
889 {
890 	return TCP_SKB_CB(skb)->tcp_gso_size;
891 }
892 
893 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
894 {
895 	return likely(!TCP_SKB_CB(skb)->eor);
896 }
897 
898 /* Events passed to congestion control interface */
899 enum tcp_ca_event {
900 	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
901 	CA_EVENT_CWND_RESTART,	/* congestion window restart */
902 	CA_EVENT_COMPLETE_CWR,	/* end of congestion recovery */
903 	CA_EVENT_LOSS,		/* loss timeout */
904 	CA_EVENT_ECN_NO_CE,	/* ECT set, but not CE marked */
905 	CA_EVENT_ECN_IS_CE,	/* received CE marked IP packet */
906 	CA_EVENT_DELAYED_ACK,	/* Delayed ack is sent */
907 	CA_EVENT_NON_DELAYED_ACK,
908 };
909 
910 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
911 enum tcp_ca_ack_event_flags {
912 	CA_ACK_SLOWPATH		= (1 << 0),	/* In slow path processing */
913 	CA_ACK_WIN_UPDATE	= (1 << 1),	/* ACK updated window */
914 	CA_ACK_ECE		= (1 << 2),	/* ECE bit is set on ack */
915 };
916 
917 /*
918  * Interface for adding new TCP congestion control handlers
919  */
920 #define TCP_CA_NAME_MAX	16
921 #define TCP_CA_MAX	128
922 #define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
923 
924 #define TCP_CA_UNSPEC	0
925 
926 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
927 #define TCP_CONG_NON_RESTRICTED 0x1
928 /* Requires ECN/ECT set on all packets */
929 #define TCP_CONG_NEEDS_ECN	0x2
930 
931 union tcp_cc_info;
932 
933 struct ack_sample {
934 	u32 pkts_acked;
935 	s32 rtt_us;
936 	u32 in_flight;
937 };
938 
939 /* A rate sample measures the number of (original/retransmitted) data
940  * packets delivered "delivered" over an interval of time "interval_us".
941  * The tcp_rate.c code fills in the rate sample, and congestion
942  * control modules that define a cong_control function to run at the end
943  * of ACK processing can optionally chose to consult this sample when
944  * setting cwnd and pacing rate.
945  * A sample is invalid if "delivered" or "interval_us" is negative.
946  */
947 struct rate_sample {
948 	u64  prior_mstamp; /* starting timestamp for interval */
949 	u32  prior_delivered;	/* tp->delivered at "prior_mstamp" */
950 	s32  delivered;		/* number of packets delivered over interval */
951 	long interval_us;	/* time for tp->delivered to incr "delivered" */
952 	long rtt_us;		/* RTT of last (S)ACKed packet (or -1) */
953 	int  losses;		/* number of packets marked lost upon ACK */
954 	u32  acked_sacked;	/* number of packets newly (S)ACKed upon ACK */
955 	u32  prior_in_flight;	/* in flight before this ACK */
956 	bool is_app_limited;	/* is sample from packet with bubble in pipe? */
957 	bool is_retrans;	/* is sample from retransmission? */
958 };
959 
960 struct tcp_congestion_ops {
961 	struct list_head	list;
962 	u32 key;
963 	u32 flags;
964 
965 	/* initialize private data (optional) */
966 	void (*init)(struct sock *sk);
967 	/* cleanup private data  (optional) */
968 	void (*release)(struct sock *sk);
969 
970 	/* return slow start threshold (required) */
971 	u32 (*ssthresh)(struct sock *sk);
972 	/* do new cwnd calculation (required) */
973 	void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
974 	/* call before changing ca_state (optional) */
975 	void (*set_state)(struct sock *sk, u8 new_state);
976 	/* call when cwnd event occurs (optional) */
977 	void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
978 	/* call when ack arrives (optional) */
979 	void (*in_ack_event)(struct sock *sk, u32 flags);
980 	/* new value of cwnd after loss (required) */
981 	u32  (*undo_cwnd)(struct sock *sk);
982 	/* hook for packet ack accounting (optional) */
983 	void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
984 	/* suggest number of segments for each skb to transmit (optional) */
985 	u32 (*tso_segs_goal)(struct sock *sk);
986 	/* returns the multiplier used in tcp_sndbuf_expand (optional) */
987 	u32 (*sndbuf_expand)(struct sock *sk);
988 	/* call when packets are delivered to update cwnd and pacing rate,
989 	 * after all the ca_state processing. (optional)
990 	 */
991 	void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
992 	/* get info for inet_diag (optional) */
993 	size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
994 			   union tcp_cc_info *info);
995 
996 	char 		name[TCP_CA_NAME_MAX];
997 	struct module 	*owner;
998 };
999 
1000 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1001 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1002 
1003 void tcp_assign_congestion_control(struct sock *sk);
1004 void tcp_init_congestion_control(struct sock *sk);
1005 void tcp_cleanup_congestion_control(struct sock *sk);
1006 int tcp_set_default_congestion_control(const char *name);
1007 void tcp_get_default_congestion_control(char *name);
1008 void tcp_get_available_congestion_control(char *buf, size_t len);
1009 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1010 int tcp_set_allowed_congestion_control(char *allowed);
1011 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit);
1012 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1013 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1014 
1015 u32 tcp_reno_ssthresh(struct sock *sk);
1016 u32 tcp_reno_undo_cwnd(struct sock *sk);
1017 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1018 extern struct tcp_congestion_ops tcp_reno;
1019 
1020 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1021 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
1022 #ifdef CONFIG_INET
1023 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1024 #else
1025 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1026 {
1027 	return NULL;
1028 }
1029 #endif
1030 
1031 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1032 {
1033 	const struct inet_connection_sock *icsk = inet_csk(sk);
1034 
1035 	return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1036 }
1037 
1038 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1039 {
1040 	struct inet_connection_sock *icsk = inet_csk(sk);
1041 
1042 	if (icsk->icsk_ca_ops->set_state)
1043 		icsk->icsk_ca_ops->set_state(sk, ca_state);
1044 	icsk->icsk_ca_state = ca_state;
1045 }
1046 
1047 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1048 {
1049 	const struct inet_connection_sock *icsk = inet_csk(sk);
1050 
1051 	if (icsk->icsk_ca_ops->cwnd_event)
1052 		icsk->icsk_ca_ops->cwnd_event(sk, event);
1053 }
1054 
1055 /* From tcp_rate.c */
1056 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1057 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1058 			    struct rate_sample *rs);
1059 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1060 		  struct rate_sample *rs);
1061 void tcp_rate_check_app_limited(struct sock *sk);
1062 
1063 /* These functions determine how the current flow behaves in respect of SACK
1064  * handling. SACK is negotiated with the peer, and therefore it can vary
1065  * between different flows.
1066  *
1067  * tcp_is_sack - SACK enabled
1068  * tcp_is_reno - No SACK
1069  * tcp_is_fack - FACK enabled, implies SACK enabled
1070  */
1071 static inline int tcp_is_sack(const struct tcp_sock *tp)
1072 {
1073 	return tp->rx_opt.sack_ok;
1074 }
1075 
1076 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1077 {
1078 	return !tcp_is_sack(tp);
1079 }
1080 
1081 static inline bool tcp_is_fack(const struct tcp_sock *tp)
1082 {
1083 	return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
1084 }
1085 
1086 static inline void tcp_enable_fack(struct tcp_sock *tp)
1087 {
1088 	tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
1089 }
1090 
1091 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1092 {
1093 	return tp->sacked_out + tp->lost_out;
1094 }
1095 
1096 /* This determines how many packets are "in the network" to the best
1097  * of our knowledge.  In many cases it is conservative, but where
1098  * detailed information is available from the receiver (via SACK
1099  * blocks etc.) we can make more aggressive calculations.
1100  *
1101  * Use this for decisions involving congestion control, use just
1102  * tp->packets_out to determine if the send queue is empty or not.
1103  *
1104  * Read this equation as:
1105  *
1106  *	"Packets sent once on transmission queue" MINUS
1107  *	"Packets left network, but not honestly ACKed yet" PLUS
1108  *	"Packets fast retransmitted"
1109  */
1110 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1111 {
1112 	return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1113 }
1114 
1115 #define TCP_INFINITE_SSTHRESH	0x7fffffff
1116 
1117 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1118 {
1119 	return tp->snd_cwnd < tp->snd_ssthresh;
1120 }
1121 
1122 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1123 {
1124 	return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1125 }
1126 
1127 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1128 {
1129 	return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1130 	       (1 << inet_csk(sk)->icsk_ca_state);
1131 }
1132 
1133 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1134  * The exception is cwnd reduction phase, when cwnd is decreasing towards
1135  * ssthresh.
1136  */
1137 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1138 {
1139 	const struct tcp_sock *tp = tcp_sk(sk);
1140 
1141 	if (tcp_in_cwnd_reduction(sk))
1142 		return tp->snd_ssthresh;
1143 	else
1144 		return max(tp->snd_ssthresh,
1145 			   ((tp->snd_cwnd >> 1) +
1146 			    (tp->snd_cwnd >> 2)));
1147 }
1148 
1149 /* Use define here intentionally to get WARN_ON location shown at the caller */
1150 #define tcp_verify_left_out(tp)	WARN_ON(tcp_left_out(tp) > tp->packets_out)
1151 
1152 void tcp_enter_cwr(struct sock *sk);
1153 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1154 
1155 /* The maximum number of MSS of available cwnd for which TSO defers
1156  * sending if not using sysctl_tcp_tso_win_divisor.
1157  */
1158 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1159 {
1160 	return 3;
1161 }
1162 
1163 /* Returns end sequence number of the receiver's advertised window */
1164 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1165 {
1166 	return tp->snd_una + tp->snd_wnd;
1167 }
1168 
1169 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1170  * flexible approach. The RFC suggests cwnd should not be raised unless
1171  * it was fully used previously. And that's exactly what we do in
1172  * congestion avoidance mode. But in slow start we allow cwnd to grow
1173  * as long as the application has used half the cwnd.
1174  * Example :
1175  *    cwnd is 10 (IW10), but application sends 9 frames.
1176  *    We allow cwnd to reach 18 when all frames are ACKed.
1177  * This check is safe because it's as aggressive as slow start which already
1178  * risks 100% overshoot. The advantage is that we discourage application to
1179  * either send more filler packets or data to artificially blow up the cwnd
1180  * usage, and allow application-limited process to probe bw more aggressively.
1181  */
1182 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1183 {
1184 	const struct tcp_sock *tp = tcp_sk(sk);
1185 
1186 	/* If in slow start, ensure cwnd grows to twice what was ACKed. */
1187 	if (tcp_in_slow_start(tp))
1188 		return tp->snd_cwnd < 2 * tp->max_packets_out;
1189 
1190 	return tp->is_cwnd_limited;
1191 }
1192 
1193 /* Something is really bad, we could not queue an additional packet,
1194  * because qdisc is full or receiver sent a 0 window.
1195  * We do not want to add fuel to the fire, or abort too early,
1196  * so make sure the timer we arm now is at least 200ms in the future,
1197  * regardless of current icsk_rto value (as it could be ~2ms)
1198  */
1199 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1200 {
1201 	return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1202 }
1203 
1204 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1205 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1206 					    unsigned long max_when)
1207 {
1208 	u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1209 
1210 	return (unsigned long)min_t(u64, when, max_when);
1211 }
1212 
1213 static inline void tcp_check_probe_timer(struct sock *sk)
1214 {
1215 	if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1216 		inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1217 					  tcp_probe0_base(sk), TCP_RTO_MAX);
1218 }
1219 
1220 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1221 {
1222 	tp->snd_wl1 = seq;
1223 }
1224 
1225 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1226 {
1227 	tp->snd_wl1 = seq;
1228 }
1229 
1230 /*
1231  * Calculate(/check) TCP checksum
1232  */
1233 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1234 				   __be32 daddr, __wsum base)
1235 {
1236 	return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1237 }
1238 
1239 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1240 {
1241 	return __skb_checksum_complete(skb);
1242 }
1243 
1244 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1245 {
1246 	return !skb_csum_unnecessary(skb) &&
1247 		__tcp_checksum_complete(skb);
1248 }
1249 
1250 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1251 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1252 
1253 #undef STATE_TRACE
1254 
1255 #ifdef STATE_TRACE
1256 static const char *statename[]={
1257 	"Unused","Established","Syn Sent","Syn Recv",
1258 	"Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1259 	"Close Wait","Last ACK","Listen","Closing"
1260 };
1261 #endif
1262 void tcp_set_state(struct sock *sk, int state);
1263 
1264 void tcp_done(struct sock *sk);
1265 
1266 int tcp_abort(struct sock *sk, int err);
1267 
1268 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1269 {
1270 	rx_opt->dsack = 0;
1271 	rx_opt->num_sacks = 0;
1272 }
1273 
1274 u32 tcp_default_init_rwnd(u32 mss);
1275 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1276 
1277 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1278 {
1279 	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1280 	struct tcp_sock *tp = tcp_sk(sk);
1281 	s32 delta;
1282 
1283 	if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1284 	    ca_ops->cong_control)
1285 		return;
1286 	delta = tcp_jiffies32 - tp->lsndtime;
1287 	if (delta > inet_csk(sk)->icsk_rto)
1288 		tcp_cwnd_restart(sk, delta);
1289 }
1290 
1291 /* Determine a window scaling and initial window to offer. */
1292 void tcp_select_initial_window(const struct sock *sk, int __space,
1293 			       __u32 mss, __u32 *rcv_wnd,
1294 			       __u32 *window_clamp, int wscale_ok,
1295 			       __u8 *rcv_wscale, __u32 init_rcv_wnd);
1296 
1297 static inline int tcp_win_from_space(const struct sock *sk, int space)
1298 {
1299 	int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1300 
1301 	return tcp_adv_win_scale <= 0 ?
1302 		(space>>(-tcp_adv_win_scale)) :
1303 		space - (space>>tcp_adv_win_scale);
1304 }
1305 
1306 /* Note: caller must be prepared to deal with negative returns */
1307 static inline int tcp_space(const struct sock *sk)
1308 {
1309 	return tcp_win_from_space(sk, sk->sk_rcvbuf -
1310 				  atomic_read(&sk->sk_rmem_alloc));
1311 }
1312 
1313 static inline int tcp_full_space(const struct sock *sk)
1314 {
1315 	return tcp_win_from_space(sk, sk->sk_rcvbuf);
1316 }
1317 
1318 extern void tcp_openreq_init_rwin(struct request_sock *req,
1319 				  const struct sock *sk_listener,
1320 				  const struct dst_entry *dst);
1321 
1322 void tcp_enter_memory_pressure(struct sock *sk);
1323 void tcp_leave_memory_pressure(struct sock *sk);
1324 
1325 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1326 {
1327 	struct net *net = sock_net((struct sock *)tp);
1328 
1329 	return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1330 }
1331 
1332 static inline int keepalive_time_when(const struct tcp_sock *tp)
1333 {
1334 	struct net *net = sock_net((struct sock *)tp);
1335 
1336 	return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1337 }
1338 
1339 static inline int keepalive_probes(const struct tcp_sock *tp)
1340 {
1341 	struct net *net = sock_net((struct sock *)tp);
1342 
1343 	return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1344 }
1345 
1346 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1347 {
1348 	const struct inet_connection_sock *icsk = &tp->inet_conn;
1349 
1350 	return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1351 			  tcp_jiffies32 - tp->rcv_tstamp);
1352 }
1353 
1354 static inline int tcp_fin_time(const struct sock *sk)
1355 {
1356 	int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1357 	const int rto = inet_csk(sk)->icsk_rto;
1358 
1359 	if (fin_timeout < (rto << 2) - (rto >> 1))
1360 		fin_timeout = (rto << 2) - (rto >> 1);
1361 
1362 	return fin_timeout;
1363 }
1364 
1365 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1366 				  int paws_win)
1367 {
1368 	if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1369 		return true;
1370 	if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1371 		return true;
1372 	/*
1373 	 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1374 	 * then following tcp messages have valid values. Ignore 0 value,
1375 	 * or else 'negative' tsval might forbid us to accept their packets.
1376 	 */
1377 	if (!rx_opt->ts_recent)
1378 		return true;
1379 	return false;
1380 }
1381 
1382 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1383 				   int rst)
1384 {
1385 	if (tcp_paws_check(rx_opt, 0))
1386 		return false;
1387 
1388 	/* RST segments are not recommended to carry timestamp,
1389 	   and, if they do, it is recommended to ignore PAWS because
1390 	   "their cleanup function should take precedence over timestamps."
1391 	   Certainly, it is mistake. It is necessary to understand the reasons
1392 	   of this constraint to relax it: if peer reboots, clock may go
1393 	   out-of-sync and half-open connections will not be reset.
1394 	   Actually, the problem would be not existing if all
1395 	   the implementations followed draft about maintaining clock
1396 	   via reboots. Linux-2.2 DOES NOT!
1397 
1398 	   However, we can relax time bounds for RST segments to MSL.
1399 	 */
1400 	if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1401 		return false;
1402 	return true;
1403 }
1404 
1405 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1406 			  int mib_idx, u32 *last_oow_ack_time);
1407 
1408 static inline void tcp_mib_init(struct net *net)
1409 {
1410 	/* See RFC 2012 */
1411 	TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1412 	TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1413 	TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1414 	TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1415 }
1416 
1417 /* from STCP */
1418 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1419 {
1420 	tp->lost_skb_hint = NULL;
1421 }
1422 
1423 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1424 {
1425 	tcp_clear_retrans_hints_partial(tp);
1426 	tp->retransmit_skb_hint = NULL;
1427 }
1428 
1429 union tcp_md5_addr {
1430 	struct in_addr  a4;
1431 #if IS_ENABLED(CONFIG_IPV6)
1432 	struct in6_addr	a6;
1433 #endif
1434 };
1435 
1436 /* - key database */
1437 struct tcp_md5sig_key {
1438 	struct hlist_node	node;
1439 	u8			keylen;
1440 	u8			family; /* AF_INET or AF_INET6 */
1441 	union tcp_md5_addr	addr;
1442 	u8			prefixlen;
1443 	u8			key[TCP_MD5SIG_MAXKEYLEN];
1444 	struct rcu_head		rcu;
1445 };
1446 
1447 /* - sock block */
1448 struct tcp_md5sig_info {
1449 	struct hlist_head	head;
1450 	struct rcu_head		rcu;
1451 };
1452 
1453 /* - pseudo header */
1454 struct tcp4_pseudohdr {
1455 	__be32		saddr;
1456 	__be32		daddr;
1457 	__u8		pad;
1458 	__u8		protocol;
1459 	__be16		len;
1460 };
1461 
1462 struct tcp6_pseudohdr {
1463 	struct in6_addr	saddr;
1464 	struct in6_addr daddr;
1465 	__be32		len;
1466 	__be32		protocol;	/* including padding */
1467 };
1468 
1469 union tcp_md5sum_block {
1470 	struct tcp4_pseudohdr ip4;
1471 #if IS_ENABLED(CONFIG_IPV6)
1472 	struct tcp6_pseudohdr ip6;
1473 #endif
1474 };
1475 
1476 /* - pool: digest algorithm, hash description and scratch buffer */
1477 struct tcp_md5sig_pool {
1478 	struct ahash_request	*md5_req;
1479 	void			*scratch;
1480 };
1481 
1482 /* - functions */
1483 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1484 			const struct sock *sk, const struct sk_buff *skb);
1485 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1486 		   int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1487 		   gfp_t gfp);
1488 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1489 		   int family, u8 prefixlen);
1490 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1491 					 const struct sock *addr_sk);
1492 
1493 #ifdef CONFIG_TCP_MD5SIG
1494 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1495 					 const union tcp_md5_addr *addr,
1496 					 int family);
1497 #define tcp_twsk_md5_key(twsk)	((twsk)->tw_md5_key)
1498 #else
1499 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1500 					 const union tcp_md5_addr *addr,
1501 					 int family)
1502 {
1503 	return NULL;
1504 }
1505 #define tcp_twsk_md5_key(twsk)	NULL
1506 #endif
1507 
1508 bool tcp_alloc_md5sig_pool(void);
1509 
1510 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1511 static inline void tcp_put_md5sig_pool(void)
1512 {
1513 	local_bh_enable();
1514 }
1515 
1516 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1517 			  unsigned int header_len);
1518 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1519 		     const struct tcp_md5sig_key *key);
1520 
1521 /* From tcp_fastopen.c */
1522 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1523 			    struct tcp_fastopen_cookie *cookie, int *syn_loss,
1524 			    unsigned long *last_syn_loss);
1525 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1526 			    struct tcp_fastopen_cookie *cookie, bool syn_lost,
1527 			    u16 try_exp);
1528 struct tcp_fastopen_request {
1529 	/* Fast Open cookie. Size 0 means a cookie request */
1530 	struct tcp_fastopen_cookie	cookie;
1531 	struct msghdr			*data;  /* data in MSG_FASTOPEN */
1532 	size_t				size;
1533 	int				copied;	/* queued in tcp_connect() */
1534 };
1535 void tcp_free_fastopen_req(struct tcp_sock *tp);
1536 void tcp_fastopen_destroy_cipher(struct sock *sk);
1537 void tcp_fastopen_ctx_destroy(struct net *net);
1538 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1539 			      void *key, unsigned int len);
1540 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1541 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1542 			      struct request_sock *req,
1543 			      struct tcp_fastopen_cookie *foc,
1544 			      const struct dst_entry *dst);
1545 void tcp_fastopen_init_key_once(struct net *net);
1546 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1547 			     struct tcp_fastopen_cookie *cookie);
1548 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1549 #define TCP_FASTOPEN_KEY_LENGTH 16
1550 
1551 /* Fastopen key context */
1552 struct tcp_fastopen_context {
1553 	struct crypto_cipher	*tfm;
1554 	__u8			key[TCP_FASTOPEN_KEY_LENGTH];
1555 	struct rcu_head		rcu;
1556 };
1557 
1558 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1559 void tcp_fastopen_active_disable(struct sock *sk);
1560 bool tcp_fastopen_active_should_disable(struct sock *sk);
1561 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1562 void tcp_fastopen_active_timeout_reset(void);
1563 
1564 /* Latencies incurred by various limits for a sender. They are
1565  * chronograph-like stats that are mutually exclusive.
1566  */
1567 enum tcp_chrono {
1568 	TCP_CHRONO_UNSPEC,
1569 	TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1570 	TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1571 	TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1572 	__TCP_CHRONO_MAX,
1573 };
1574 
1575 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1576 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1577 
1578 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1579  * the same memory storage than skb->destructor/_skb_refdst
1580  */
1581 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1582 {
1583 	skb->destructor = NULL;
1584 	skb->_skb_refdst = 0UL;
1585 }
1586 
1587 #define tcp_skb_tsorted_save(skb) {		\
1588 	unsigned long _save = skb->_skb_refdst;	\
1589 	skb->_skb_refdst = 0UL;
1590 
1591 #define tcp_skb_tsorted_restore(skb)		\
1592 	skb->_skb_refdst = _save;		\
1593 }
1594 
1595 void tcp_write_queue_purge(struct sock *sk);
1596 
1597 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1598 {
1599 	return skb_rb_first(&sk->tcp_rtx_queue);
1600 }
1601 
1602 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1603 {
1604 	return skb_peek(&sk->sk_write_queue);
1605 }
1606 
1607 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1608 {
1609 	return skb_peek_tail(&sk->sk_write_queue);
1610 }
1611 
1612 #define tcp_for_write_queue_from_safe(skb, tmp, sk)			\
1613 	skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1614 
1615 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1616 {
1617 	return skb_peek(&sk->sk_write_queue);
1618 }
1619 
1620 static inline bool tcp_skb_is_last(const struct sock *sk,
1621 				   const struct sk_buff *skb)
1622 {
1623 	return skb_queue_is_last(&sk->sk_write_queue, skb);
1624 }
1625 
1626 static inline bool tcp_write_queue_empty(const struct sock *sk)
1627 {
1628 	return skb_queue_empty(&sk->sk_write_queue);
1629 }
1630 
1631 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1632 {
1633 	return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1634 }
1635 
1636 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1637 {
1638 	return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1639 }
1640 
1641 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1642 {
1643 	if (tcp_write_queue_empty(sk))
1644 		tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1645 
1646 	if (tcp_sk(sk)->highest_sack == skb_unlinked)
1647 		tcp_sk(sk)->highest_sack = NULL;
1648 }
1649 
1650 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1651 {
1652 	__skb_queue_tail(&sk->sk_write_queue, skb);
1653 }
1654 
1655 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1656 {
1657 	__tcp_add_write_queue_tail(sk, skb);
1658 
1659 	/* Queue it, remembering where we must start sending. */
1660 	if (sk->sk_write_queue.next == skb) {
1661 		tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1662 
1663 		if (tcp_sk(sk)->highest_sack == NULL)
1664 			tcp_sk(sk)->highest_sack = skb;
1665 	}
1666 }
1667 
1668 /* Insert new before skb on the write queue of sk.  */
1669 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1670 						  struct sk_buff *skb,
1671 						  struct sock *sk)
1672 {
1673 	__skb_queue_before(&sk->sk_write_queue, skb, new);
1674 }
1675 
1676 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1677 {
1678 	tcp_skb_tsorted_anchor_cleanup(skb);
1679 	__skb_unlink(skb, &sk->sk_write_queue);
1680 }
1681 
1682 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1683 
1684 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1685 {
1686 	tcp_skb_tsorted_anchor_cleanup(skb);
1687 	rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1688 }
1689 
1690 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1691 {
1692 	list_del(&skb->tcp_tsorted_anchor);
1693 	tcp_rtx_queue_unlink(skb, sk);
1694 	sk_wmem_free_skb(sk, skb);
1695 }
1696 
1697 static inline void tcp_push_pending_frames(struct sock *sk)
1698 {
1699 	if (tcp_send_head(sk)) {
1700 		struct tcp_sock *tp = tcp_sk(sk);
1701 
1702 		__tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1703 	}
1704 }
1705 
1706 /* Start sequence of the skb just after the highest skb with SACKed
1707  * bit, valid only if sacked_out > 0 or when the caller has ensured
1708  * validity by itself.
1709  */
1710 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1711 {
1712 	if (!tp->sacked_out)
1713 		return tp->snd_una;
1714 
1715 	if (tp->highest_sack == NULL)
1716 		return tp->snd_nxt;
1717 
1718 	return TCP_SKB_CB(tp->highest_sack)->seq;
1719 }
1720 
1721 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1722 {
1723 	struct sk_buff *next = skb_rb_next(skb);
1724 
1725 	tcp_sk(sk)->highest_sack = next ?: tcp_send_head(sk);
1726 }
1727 
1728 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1729 {
1730 	return tcp_sk(sk)->highest_sack;
1731 }
1732 
1733 static inline void tcp_highest_sack_reset(struct sock *sk)
1734 {
1735 	struct sk_buff *skb = tcp_rtx_queue_head(sk);
1736 
1737 	tcp_sk(sk)->highest_sack = skb ?: tcp_send_head(sk);
1738 }
1739 
1740 /* Called when old skb is about to be deleted (to be combined with new skb) */
1741 static inline void tcp_highest_sack_combine(struct sock *sk,
1742 					    struct sk_buff *old,
1743 					    struct sk_buff *new)
1744 {
1745 	if (tcp_sk(sk)->sacked_out && (old == tcp_sk(sk)->highest_sack))
1746 		tcp_sk(sk)->highest_sack = new;
1747 }
1748 
1749 /* This helper checks if socket has IP_TRANSPARENT set */
1750 static inline bool inet_sk_transparent(const struct sock *sk)
1751 {
1752 	switch (sk->sk_state) {
1753 	case TCP_TIME_WAIT:
1754 		return inet_twsk(sk)->tw_transparent;
1755 	case TCP_NEW_SYN_RECV:
1756 		return inet_rsk(inet_reqsk(sk))->no_srccheck;
1757 	}
1758 	return inet_sk(sk)->transparent;
1759 }
1760 
1761 /* Determines whether this is a thin stream (which may suffer from
1762  * increased latency). Used to trigger latency-reducing mechanisms.
1763  */
1764 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1765 {
1766 	return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1767 }
1768 
1769 /* /proc */
1770 enum tcp_seq_states {
1771 	TCP_SEQ_STATE_LISTENING,
1772 	TCP_SEQ_STATE_ESTABLISHED,
1773 };
1774 
1775 int tcp_seq_open(struct inode *inode, struct file *file);
1776 
1777 struct tcp_seq_afinfo {
1778 	char				*name;
1779 	sa_family_t			family;
1780 	const struct file_operations	*seq_fops;
1781 	struct seq_operations		seq_ops;
1782 };
1783 
1784 struct tcp_iter_state {
1785 	struct seq_net_private	p;
1786 	sa_family_t		family;
1787 	enum tcp_seq_states	state;
1788 	struct sock		*syn_wait_sk;
1789 	int			bucket, offset, sbucket, num;
1790 	loff_t			last_pos;
1791 };
1792 
1793 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1794 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1795 
1796 extern struct request_sock_ops tcp_request_sock_ops;
1797 extern struct request_sock_ops tcp6_request_sock_ops;
1798 
1799 void tcp_v4_destroy_sock(struct sock *sk);
1800 
1801 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1802 				netdev_features_t features);
1803 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1804 int tcp_gro_complete(struct sk_buff *skb);
1805 
1806 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1807 
1808 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1809 {
1810 	struct net *net = sock_net((struct sock *)tp);
1811 	return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1812 }
1813 
1814 static inline bool tcp_stream_memory_free(const struct sock *sk)
1815 {
1816 	const struct tcp_sock *tp = tcp_sk(sk);
1817 	u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1818 
1819 	return notsent_bytes < tcp_notsent_lowat(tp);
1820 }
1821 
1822 #ifdef CONFIG_PROC_FS
1823 int tcp4_proc_init(void);
1824 void tcp4_proc_exit(void);
1825 #endif
1826 
1827 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1828 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1829 		     const struct tcp_request_sock_ops *af_ops,
1830 		     struct sock *sk, struct sk_buff *skb);
1831 
1832 /* TCP af-specific functions */
1833 struct tcp_sock_af_ops {
1834 #ifdef CONFIG_TCP_MD5SIG
1835 	struct tcp_md5sig_key	*(*md5_lookup) (const struct sock *sk,
1836 						const struct sock *addr_sk);
1837 	int		(*calc_md5_hash)(char *location,
1838 					 const struct tcp_md5sig_key *md5,
1839 					 const struct sock *sk,
1840 					 const struct sk_buff *skb);
1841 	int		(*md5_parse)(struct sock *sk,
1842 				     int optname,
1843 				     char __user *optval,
1844 				     int optlen);
1845 #endif
1846 };
1847 
1848 struct tcp_request_sock_ops {
1849 	u16 mss_clamp;
1850 #ifdef CONFIG_TCP_MD5SIG
1851 	struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1852 						 const struct sock *addr_sk);
1853 	int		(*calc_md5_hash) (char *location,
1854 					  const struct tcp_md5sig_key *md5,
1855 					  const struct sock *sk,
1856 					  const struct sk_buff *skb);
1857 #endif
1858 	void (*init_req)(struct request_sock *req,
1859 			 const struct sock *sk_listener,
1860 			 struct sk_buff *skb);
1861 #ifdef CONFIG_SYN_COOKIES
1862 	__u32 (*cookie_init_seq)(const struct sk_buff *skb,
1863 				 __u16 *mss);
1864 #endif
1865 	struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1866 				       const struct request_sock *req);
1867 	u32 (*init_seq)(const struct sk_buff *skb);
1868 	u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
1869 	int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1870 			   struct flowi *fl, struct request_sock *req,
1871 			   struct tcp_fastopen_cookie *foc,
1872 			   enum tcp_synack_type synack_type);
1873 };
1874 
1875 #ifdef CONFIG_SYN_COOKIES
1876 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1877 					 const struct sock *sk, struct sk_buff *skb,
1878 					 __u16 *mss)
1879 {
1880 	tcp_synq_overflow(sk);
1881 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1882 	return ops->cookie_init_seq(skb, mss);
1883 }
1884 #else
1885 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1886 					 const struct sock *sk, struct sk_buff *skb,
1887 					 __u16 *mss)
1888 {
1889 	return 0;
1890 }
1891 #endif
1892 
1893 int tcpv4_offload_init(void);
1894 
1895 void tcp_v4_init(void);
1896 void tcp_init(void);
1897 
1898 /* tcp_recovery.c */
1899 extern void tcp_rack_mark_lost(struct sock *sk);
1900 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
1901 			     u64 xmit_time);
1902 extern void tcp_rack_reo_timeout(struct sock *sk);
1903 
1904 /* At how many usecs into the future should the RTO fire? */
1905 static inline s64 tcp_rto_delta_us(const struct sock *sk)
1906 {
1907 	const struct sk_buff *skb = tcp_rtx_queue_head(sk);
1908 	u32 rto = inet_csk(sk)->icsk_rto;
1909 	u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto);
1910 
1911 	return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
1912 }
1913 
1914 /*
1915  * Save and compile IPv4 options, return a pointer to it
1916  */
1917 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
1918 							 struct sk_buff *skb)
1919 {
1920 	const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1921 	struct ip_options_rcu *dopt = NULL;
1922 
1923 	if (opt->optlen) {
1924 		int opt_size = sizeof(*dopt) + opt->optlen;
1925 
1926 		dopt = kmalloc(opt_size, GFP_ATOMIC);
1927 		if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
1928 			kfree(dopt);
1929 			dopt = NULL;
1930 		}
1931 	}
1932 	return dopt;
1933 }
1934 
1935 /* locally generated TCP pure ACKs have skb->truesize == 2
1936  * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1937  * This is much faster than dissecting the packet to find out.
1938  * (Think of GRE encapsulations, IPv4, IPv6, ...)
1939  */
1940 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1941 {
1942 	return skb->truesize == 2;
1943 }
1944 
1945 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1946 {
1947 	skb->truesize = 2;
1948 }
1949 
1950 static inline int tcp_inq(struct sock *sk)
1951 {
1952 	struct tcp_sock *tp = tcp_sk(sk);
1953 	int answ;
1954 
1955 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1956 		answ = 0;
1957 	} else if (sock_flag(sk, SOCK_URGINLINE) ||
1958 		   !tp->urg_data ||
1959 		   before(tp->urg_seq, tp->copied_seq) ||
1960 		   !before(tp->urg_seq, tp->rcv_nxt)) {
1961 
1962 		answ = tp->rcv_nxt - tp->copied_seq;
1963 
1964 		/* Subtract 1, if FIN was received */
1965 		if (answ && sock_flag(sk, SOCK_DONE))
1966 			answ--;
1967 	} else {
1968 		answ = tp->urg_seq - tp->copied_seq;
1969 	}
1970 
1971 	return answ;
1972 }
1973 
1974 int tcp_peek_len(struct socket *sock);
1975 
1976 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1977 {
1978 	u16 segs_in;
1979 
1980 	segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1981 	tp->segs_in += segs_in;
1982 	if (skb->len > tcp_hdrlen(skb))
1983 		tp->data_segs_in += segs_in;
1984 }
1985 
1986 /*
1987  * TCP listen path runs lockless.
1988  * We forced "struct sock" to be const qualified to make sure
1989  * we don't modify one of its field by mistake.
1990  * Here, we increment sk_drops which is an atomic_t, so we can safely
1991  * make sock writable again.
1992  */
1993 static inline void tcp_listendrop(const struct sock *sk)
1994 {
1995 	atomic_inc(&((struct sock *)sk)->sk_drops);
1996 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
1997 }
1998 
1999 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2000 
2001 /*
2002  * Interface for adding Upper Level Protocols over TCP
2003  */
2004 
2005 #define TCP_ULP_NAME_MAX	16
2006 #define TCP_ULP_MAX		128
2007 #define TCP_ULP_BUF_MAX		(TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2008 
2009 struct tcp_ulp_ops {
2010 	struct list_head	list;
2011 
2012 	/* initialize ulp */
2013 	int (*init)(struct sock *sk);
2014 	/* cleanup ulp */
2015 	void (*release)(struct sock *sk);
2016 
2017 	char		name[TCP_ULP_NAME_MAX];
2018 	struct module	*owner;
2019 };
2020 int tcp_register_ulp(struct tcp_ulp_ops *type);
2021 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2022 int tcp_set_ulp(struct sock *sk, const char *name);
2023 void tcp_get_available_ulp(char *buf, size_t len);
2024 void tcp_cleanup_ulp(struct sock *sk);
2025 
2026 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2027  * is < 0, then the BPF op failed (for example if the loaded BPF
2028  * program does not support the chosen operation or there is no BPF
2029  * program loaded).
2030  */
2031 #ifdef CONFIG_BPF
2032 static inline int tcp_call_bpf(struct sock *sk, int op)
2033 {
2034 	struct bpf_sock_ops_kern sock_ops;
2035 	int ret;
2036 
2037 	if (sk_fullsock(sk))
2038 		sock_owned_by_me(sk);
2039 
2040 	memset(&sock_ops, 0, sizeof(sock_ops));
2041 	sock_ops.sk = sk;
2042 	sock_ops.op = op;
2043 
2044 	ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2045 	if (ret == 0)
2046 		ret = sock_ops.reply;
2047 	else
2048 		ret = -1;
2049 	return ret;
2050 }
2051 #else
2052 static inline int tcp_call_bpf(struct sock *sk, int op)
2053 {
2054 	return -EPERM;
2055 }
2056 #endif
2057 
2058 static inline u32 tcp_timeout_init(struct sock *sk)
2059 {
2060 	int timeout;
2061 
2062 	timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT);
2063 
2064 	if (timeout <= 0)
2065 		timeout = TCP_TIMEOUT_INIT;
2066 	return timeout;
2067 }
2068 
2069 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2070 {
2071 	int rwnd;
2072 
2073 	rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT);
2074 
2075 	if (rwnd < 0)
2076 		rwnd = 0;
2077 	return rwnd;
2078 }
2079 
2080 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2081 {
2082 	return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN) == 1);
2083 }
2084 
2085 #if IS_ENABLED(CONFIG_SMC)
2086 extern struct static_key_false tcp_have_smc;
2087 #endif
2088 #endif	/* _TCP_H */
2089