xref: /linux/include/net/tcp.h (revision 9406b485dea5e25bed7c81cd822747d494cc8bde)
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 #include <net/mptcp.h>
43 
44 #include <linux/seq_file.h>
45 #include <linux/memcontrol.h>
46 #include <linux/bpf-cgroup.h>
47 #include <linux/siphash.h>
48 
49 extern struct inet_hashinfo tcp_hashinfo;
50 
51 extern struct percpu_counter tcp_orphan_count;
52 void tcp_time_wait(struct sock *sk, int state, int timeo);
53 
54 #define MAX_TCP_HEADER	L1_CACHE_ALIGN(128 + MAX_HEADER)
55 #define MAX_TCP_OPTION_SPACE 40
56 #define TCP_MIN_SND_MSS		48
57 #define TCP_MIN_GSO_SIZE	(TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
58 
59 /*
60  * Never offer a window over 32767 without using window scaling. Some
61  * poor stacks do signed 16bit maths!
62  */
63 #define MAX_TCP_WINDOW		32767U
64 
65 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
66 #define TCP_MIN_MSS		88U
67 
68 /* The initial MTU to use for probing */
69 #define TCP_BASE_MSS		1024
70 
71 /* probing interval, default to 10 minutes as per RFC4821 */
72 #define TCP_PROBE_INTERVAL	600
73 
74 /* Specify interval when tcp mtu probing will stop */
75 #define TCP_PROBE_THRESHOLD	8
76 
77 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
78 #define TCP_FASTRETRANS_THRESH 3
79 
80 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
81 #define TCP_MAX_QUICKACKS	16U
82 
83 /* Maximal number of window scale according to RFC1323 */
84 #define TCP_MAX_WSCALE		14U
85 
86 /* urg_data states */
87 #define TCP_URG_VALID	0x0100
88 #define TCP_URG_NOTYET	0x0200
89 #define TCP_URG_READ	0x0400
90 
91 #define TCP_RETR1	3	/*
92 				 * This is how many retries it does before it
93 				 * tries to figure out if the gateway is
94 				 * down. Minimal RFC value is 3; it corresponds
95 				 * to ~3sec-8min depending on RTO.
96 				 */
97 
98 #define TCP_RETR2	15	/*
99 				 * This should take at least
100 				 * 90 minutes to time out.
101 				 * RFC1122 says that the limit is 100 sec.
102 				 * 15 is ~13-30min depending on RTO.
103 				 */
104 
105 #define TCP_SYN_RETRIES	 6	/* This is how many retries are done
106 				 * when active opening a connection.
107 				 * RFC1122 says the minimum retry MUST
108 				 * be at least 180secs.  Nevertheless
109 				 * this value is corresponding to
110 				 * 63secs of retransmission with the
111 				 * current initial RTO.
112 				 */
113 
114 #define TCP_SYNACK_RETRIES 5	/* This is how may retries are done
115 				 * when passive opening a connection.
116 				 * This is corresponding to 31secs of
117 				 * retransmission with the current
118 				 * initial RTO.
119 				 */
120 
121 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
122 				  * state, about 60 seconds	*/
123 #define TCP_FIN_TIMEOUT	TCP_TIMEWAIT_LEN
124                                  /* BSD style FIN_WAIT2 deadlock breaker.
125 				  * It used to be 3min, new value is 60sec,
126 				  * to combine FIN-WAIT-2 timeout with
127 				  * TIME-WAIT timer.
128 				  */
129 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */
130 
131 #define TCP_DELACK_MAX	((unsigned)(HZ/5))	/* maximal time to delay before sending an ACK */
132 #if HZ >= 100
133 #define TCP_DELACK_MIN	((unsigned)(HZ/25))	/* minimal time to delay before sending an ACK */
134 #define TCP_ATO_MIN	((unsigned)(HZ/25))
135 #else
136 #define TCP_DELACK_MIN	4U
137 #define TCP_ATO_MIN	4U
138 #endif
139 #define TCP_RTO_MAX	((unsigned)(120*HZ))
140 #define TCP_RTO_MIN	((unsigned)(HZ/5))
141 #define TCP_TIMEOUT_MIN	(2U) /* Min timeout for TCP timers in jiffies */
142 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ))	/* RFC6298 2.1 initial RTO value	*/
143 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ))	/* RFC 1122 initial RTO value, now
144 						 * used as a fallback RTO for the
145 						 * initial data transmission if no
146 						 * valid RTT sample has been acquired,
147 						 * most likely due to retrans in 3WHS.
148 						 */
149 
150 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
151 					                 * for local resources.
152 					                 */
153 #define TCP_KEEPALIVE_TIME	(120*60*HZ)	/* two hours */
154 #define TCP_KEEPALIVE_PROBES	9		/* Max of 9 keepalive probes	*/
155 #define TCP_KEEPALIVE_INTVL	(75*HZ)
156 
157 #define MAX_TCP_KEEPIDLE	32767
158 #define MAX_TCP_KEEPINTVL	32767
159 #define MAX_TCP_KEEPCNT		127
160 #define MAX_TCP_SYNCNT		127
161 
162 #define TCP_SYNQ_INTERVAL	(HZ/5)	/* Period of SYNACK timer */
163 
164 #define TCP_PAWS_24DAYS	(60 * 60 * 24 * 24)
165 #define TCP_PAWS_MSL	60		/* Per-host timestamps are invalidated
166 					 * after this time. It should be equal
167 					 * (or greater than) TCP_TIMEWAIT_LEN
168 					 * to provide reliability equal to one
169 					 * provided by timewait state.
170 					 */
171 #define TCP_PAWS_WINDOW	1		/* Replay window for per-host
172 					 * timestamps. It must be less than
173 					 * minimal timewait lifetime.
174 					 */
175 /*
176  *	TCP option
177  */
178 
179 #define TCPOPT_NOP		1	/* Padding */
180 #define TCPOPT_EOL		0	/* End of options */
181 #define TCPOPT_MSS		2	/* Segment size negotiating */
182 #define TCPOPT_WINDOW		3	/* Window scaling */
183 #define TCPOPT_SACK_PERM        4       /* SACK Permitted */
184 #define TCPOPT_SACK             5       /* SACK Block */
185 #define TCPOPT_TIMESTAMP	8	/* Better RTT estimations/PAWS */
186 #define TCPOPT_MD5SIG		19	/* MD5 Signature (RFC2385) */
187 #define TCPOPT_MPTCP		30	/* Multipath TCP (RFC6824) */
188 #define TCPOPT_FASTOPEN		34	/* Fast open (RFC7413) */
189 #define TCPOPT_EXP		254	/* Experimental */
190 /* Magic number to be after the option value for sharing TCP
191  * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
192  */
193 #define TCPOPT_FASTOPEN_MAGIC	0xF989
194 #define TCPOPT_SMC_MAGIC	0xE2D4C3D9
195 
196 /*
197  *     TCP option lengths
198  */
199 
200 #define TCPOLEN_MSS            4
201 #define TCPOLEN_WINDOW         3
202 #define TCPOLEN_SACK_PERM      2
203 #define TCPOLEN_TIMESTAMP      10
204 #define TCPOLEN_MD5SIG         18
205 #define TCPOLEN_FASTOPEN_BASE  2
206 #define TCPOLEN_EXP_FASTOPEN_BASE  4
207 #define TCPOLEN_EXP_SMC_BASE   6
208 
209 /* But this is what stacks really send out. */
210 #define TCPOLEN_TSTAMP_ALIGNED		12
211 #define TCPOLEN_WSCALE_ALIGNED		4
212 #define TCPOLEN_SACKPERM_ALIGNED	4
213 #define TCPOLEN_SACK_BASE		2
214 #define TCPOLEN_SACK_BASE_ALIGNED	4
215 #define TCPOLEN_SACK_PERBLOCK		8
216 #define TCPOLEN_MD5SIG_ALIGNED		20
217 #define TCPOLEN_MSS_ALIGNED		4
218 #define TCPOLEN_EXP_SMC_BASE_ALIGNED	8
219 
220 /* Flags in tp->nonagle */
221 #define TCP_NAGLE_OFF		1	/* Nagle's algo is disabled */
222 #define TCP_NAGLE_CORK		2	/* Socket is corked	    */
223 #define TCP_NAGLE_PUSH		4	/* Cork is overridden for already queued data */
224 
225 /* TCP thin-stream limits */
226 #define TCP_THIN_LINEAR_RETRIES 6       /* After 6 linear retries, do exp. backoff */
227 
228 /* TCP initial congestion window as per rfc6928 */
229 #define TCP_INIT_CWND		10
230 
231 /* Bit Flags for sysctl_tcp_fastopen */
232 #define	TFO_CLIENT_ENABLE	1
233 #define	TFO_SERVER_ENABLE	2
234 #define	TFO_CLIENT_NO_COOKIE	4	/* Data in SYN w/o cookie option */
235 
236 /* Accept SYN data w/o any cookie option */
237 #define	TFO_SERVER_COOKIE_NOT_REQD	0x200
238 
239 /* Force enable TFO on all listeners, i.e., not requiring the
240  * TCP_FASTOPEN socket option.
241  */
242 #define	TFO_SERVER_WO_SOCKOPT1	0x400
243 
244 
245 /* sysctl variables for tcp */
246 extern int sysctl_tcp_max_orphans;
247 extern long sysctl_tcp_mem[3];
248 
249 #define TCP_RACK_LOSS_DETECTION  0x1 /* Use RACK to detect losses */
250 #define TCP_RACK_STATIC_REO_WND  0x2 /* Use static RACK reo wnd */
251 #define TCP_RACK_NO_DUPTHRESH    0x4 /* Do not use DUPACK threshold in RACK */
252 
253 extern atomic_long_t tcp_memory_allocated;
254 extern struct percpu_counter tcp_sockets_allocated;
255 extern unsigned long tcp_memory_pressure;
256 
257 /* optimized version of sk_under_memory_pressure() for TCP sockets */
258 static inline bool tcp_under_memory_pressure(const struct sock *sk)
259 {
260 	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
261 	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
262 		return true;
263 
264 	return READ_ONCE(tcp_memory_pressure);
265 }
266 /*
267  * The next routines deal with comparing 32 bit unsigned ints
268  * and worry about wraparound (automatic with unsigned arithmetic).
269  */
270 
271 static inline bool before(__u32 seq1, __u32 seq2)
272 {
273         return (__s32)(seq1-seq2) < 0;
274 }
275 #define after(seq2, seq1) 	before(seq1, seq2)
276 
277 /* is s2<=s1<=s3 ? */
278 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
279 {
280 	return seq3 - seq2 >= seq1 - seq2;
281 }
282 
283 static inline bool tcp_out_of_memory(struct sock *sk)
284 {
285 	if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
286 	    sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
287 		return true;
288 	return false;
289 }
290 
291 void sk_forced_mem_schedule(struct sock *sk, int size);
292 
293 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
294 {
295 	struct percpu_counter *ocp = sk->sk_prot->orphan_count;
296 	int orphans = percpu_counter_read_positive(ocp);
297 
298 	if (orphans << shift > sysctl_tcp_max_orphans) {
299 		orphans = percpu_counter_sum_positive(ocp);
300 		if (orphans << shift > sysctl_tcp_max_orphans)
301 			return true;
302 	}
303 	return false;
304 }
305 
306 bool tcp_check_oom(struct sock *sk, int shift);
307 
308 
309 extern struct proto tcp_prot;
310 
311 #define TCP_INC_STATS(net, field)	SNMP_INC_STATS((net)->mib.tcp_statistics, field)
312 #define __TCP_INC_STATS(net, field)	__SNMP_INC_STATS((net)->mib.tcp_statistics, field)
313 #define TCP_DEC_STATS(net, field)	SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
314 #define TCP_ADD_STATS(net, field, val)	SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
315 
316 void tcp_tasklet_init(void);
317 
318 int tcp_v4_err(struct sk_buff *skb, u32);
319 
320 void tcp_shutdown(struct sock *sk, int how);
321 
322 int tcp_v4_early_demux(struct sk_buff *skb);
323 int tcp_v4_rcv(struct sk_buff *skb);
324 
325 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
326 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
327 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
328 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
329 		 int flags);
330 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
331 			size_t size, int flags);
332 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
333 		 size_t size, int flags);
334 int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
335 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
336 	      int size_goal);
337 void tcp_release_cb(struct sock *sk);
338 void tcp_wfree(struct sk_buff *skb);
339 void tcp_write_timer_handler(struct sock *sk);
340 void tcp_delack_timer_handler(struct sock *sk);
341 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
342 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
343 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
344 void tcp_rcv_space_adjust(struct sock *sk);
345 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
346 void tcp_twsk_destructor(struct sock *sk);
347 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
348 			struct pipe_inode_info *pipe, size_t len,
349 			unsigned int flags);
350 
351 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
352 static inline void tcp_dec_quickack_mode(struct sock *sk,
353 					 const unsigned int pkts)
354 {
355 	struct inet_connection_sock *icsk = inet_csk(sk);
356 
357 	if (icsk->icsk_ack.quick) {
358 		if (pkts >= icsk->icsk_ack.quick) {
359 			icsk->icsk_ack.quick = 0;
360 			/* Leaving quickack mode we deflate ATO. */
361 			icsk->icsk_ack.ato   = TCP_ATO_MIN;
362 		} else
363 			icsk->icsk_ack.quick -= pkts;
364 	}
365 }
366 
367 #define	TCP_ECN_OK		1
368 #define	TCP_ECN_QUEUE_CWR	2
369 #define	TCP_ECN_DEMAND_CWR	4
370 #define	TCP_ECN_SEEN		8
371 
372 enum tcp_tw_status {
373 	TCP_TW_SUCCESS = 0,
374 	TCP_TW_RST = 1,
375 	TCP_TW_ACK = 2,
376 	TCP_TW_SYN = 3
377 };
378 
379 
380 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
381 					      struct sk_buff *skb,
382 					      const struct tcphdr *th);
383 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
384 			   struct request_sock *req, bool fastopen,
385 			   bool *lost_race);
386 int tcp_child_process(struct sock *parent, struct sock *child,
387 		      struct sk_buff *skb);
388 void tcp_enter_loss(struct sock *sk);
389 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
390 void tcp_clear_retrans(struct tcp_sock *tp);
391 void tcp_update_metrics(struct sock *sk);
392 void tcp_init_metrics(struct sock *sk);
393 void tcp_metrics_init(void);
394 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
395 void tcp_close(struct sock *sk, long timeout);
396 void tcp_init_sock(struct sock *sk);
397 void tcp_init_transfer(struct sock *sk, int bpf_op);
398 __poll_t tcp_poll(struct file *file, struct socket *sock,
399 		      struct poll_table_struct *wait);
400 int tcp_getsockopt(struct sock *sk, int level, int optname,
401 		   char __user *optval, int __user *optlen);
402 int tcp_setsockopt(struct sock *sk, int level, int optname,
403 		   char __user *optval, unsigned int optlen);
404 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
405 			  char __user *optval, int __user *optlen);
406 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
407 			  char __user *optval, unsigned int optlen);
408 void tcp_set_keepalive(struct sock *sk, int val);
409 void tcp_syn_ack_timeout(const struct request_sock *req);
410 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
411 		int flags, int *addr_len);
412 int tcp_set_rcvlowat(struct sock *sk, int val);
413 void tcp_data_ready(struct sock *sk);
414 #ifdef CONFIG_MMU
415 int tcp_mmap(struct file *file, struct socket *sock,
416 	     struct vm_area_struct *vma);
417 #endif
418 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
419 		       struct tcp_options_received *opt_rx,
420 		       int estab, struct tcp_fastopen_cookie *foc);
421 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
422 
423 /*
424  *	BPF SKB-less helpers
425  */
426 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
427 			 struct tcphdr *th, u32 *cookie);
428 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
429 			 struct tcphdr *th, u32 *cookie);
430 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
431 			  const struct tcp_request_sock_ops *af_ops,
432 			  struct sock *sk, struct tcphdr *th);
433 /*
434  *	TCP v4 functions exported for the inet6 API
435  */
436 
437 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
438 void tcp_v4_mtu_reduced(struct sock *sk);
439 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
440 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
441 struct sock *tcp_create_openreq_child(const struct sock *sk,
442 				      struct request_sock *req,
443 				      struct sk_buff *skb);
444 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
445 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
446 				  struct request_sock *req,
447 				  struct dst_entry *dst,
448 				  struct request_sock *req_unhash,
449 				  bool *own_req);
450 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
451 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
452 int tcp_connect(struct sock *sk);
453 enum tcp_synack_type {
454 	TCP_SYNACK_NORMAL,
455 	TCP_SYNACK_FASTOPEN,
456 	TCP_SYNACK_COOKIE,
457 };
458 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
459 				struct request_sock *req,
460 				struct tcp_fastopen_cookie *foc,
461 				enum tcp_synack_type synack_type);
462 int tcp_disconnect(struct sock *sk, int flags);
463 
464 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
465 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
466 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
467 
468 /* From syncookies.c */
469 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
470 				 struct request_sock *req,
471 				 struct dst_entry *dst, u32 tsoff);
472 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
473 		      u32 cookie);
474 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
475 #ifdef CONFIG_SYN_COOKIES
476 
477 /* Syncookies use a monotonic timer which increments every 60 seconds.
478  * This counter is used both as a hash input and partially encoded into
479  * the cookie value.  A cookie is only validated further if the delta
480  * between the current counter value and the encoded one is less than this,
481  * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
482  * the counter advances immediately after a cookie is generated).
483  */
484 #define MAX_SYNCOOKIE_AGE	2
485 #define TCP_SYNCOOKIE_PERIOD	(60 * HZ)
486 #define TCP_SYNCOOKIE_VALID	(MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
487 
488 /* syncookies: remember time of last synqueue overflow
489  * But do not dirty this field too often (once per second is enough)
490  * It is racy as we do not hold a lock, but race is very minor.
491  */
492 static inline void tcp_synq_overflow(const struct sock *sk)
493 {
494 	unsigned int last_overflow;
495 	unsigned int now = jiffies;
496 
497 	if (sk->sk_reuseport) {
498 		struct sock_reuseport *reuse;
499 
500 		reuse = rcu_dereference(sk->sk_reuseport_cb);
501 		if (likely(reuse)) {
502 			last_overflow = READ_ONCE(reuse->synq_overflow_ts);
503 			if (!time_between32(now, last_overflow,
504 					    last_overflow + HZ))
505 				WRITE_ONCE(reuse->synq_overflow_ts, now);
506 			return;
507 		}
508 	}
509 
510 	last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
511 	if (!time_between32(now, last_overflow, last_overflow + HZ))
512 		WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
513 }
514 
515 /* syncookies: no recent synqueue overflow on this listening socket? */
516 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
517 {
518 	unsigned int last_overflow;
519 	unsigned int now = jiffies;
520 
521 	if (sk->sk_reuseport) {
522 		struct sock_reuseport *reuse;
523 
524 		reuse = rcu_dereference(sk->sk_reuseport_cb);
525 		if (likely(reuse)) {
526 			last_overflow = READ_ONCE(reuse->synq_overflow_ts);
527 			return !time_between32(now, last_overflow - HZ,
528 					       last_overflow +
529 					       TCP_SYNCOOKIE_VALID);
530 		}
531 	}
532 
533 	last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
534 
535 	/* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
536 	 * then we're under synflood. However, we have to use
537 	 * 'last_overflow - HZ' as lower bound. That's because a concurrent
538 	 * tcp_synq_overflow() could update .ts_recent_stamp after we read
539 	 * jiffies but before we store .ts_recent_stamp into last_overflow,
540 	 * which could lead to rejecting a valid syncookie.
541 	 */
542 	return !time_between32(now, last_overflow - HZ,
543 			       last_overflow + TCP_SYNCOOKIE_VALID);
544 }
545 
546 static inline u32 tcp_cookie_time(void)
547 {
548 	u64 val = get_jiffies_64();
549 
550 	do_div(val, TCP_SYNCOOKIE_PERIOD);
551 	return val;
552 }
553 
554 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
555 			      u16 *mssp);
556 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
557 u64 cookie_init_timestamp(struct request_sock *req, u64 now);
558 bool cookie_timestamp_decode(const struct net *net,
559 			     struct tcp_options_received *opt);
560 bool cookie_ecn_ok(const struct tcp_options_received *opt,
561 		   const struct net *net, const struct dst_entry *dst);
562 
563 /* From net/ipv6/syncookies.c */
564 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
565 		      u32 cookie);
566 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
567 
568 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
569 			      const struct tcphdr *th, u16 *mssp);
570 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
571 #endif
572 /* tcp_output.c */
573 
574 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
575 			       int nonagle);
576 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
577 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
578 void tcp_retransmit_timer(struct sock *sk);
579 void tcp_xmit_retransmit_queue(struct sock *);
580 void tcp_simple_retransmit(struct sock *);
581 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
582 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
583 enum tcp_queue {
584 	TCP_FRAG_IN_WRITE_QUEUE,
585 	TCP_FRAG_IN_RTX_QUEUE,
586 };
587 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
588 		 struct sk_buff *skb, u32 len,
589 		 unsigned int mss_now, gfp_t gfp);
590 
591 void tcp_send_probe0(struct sock *);
592 void tcp_send_partial(struct sock *);
593 int tcp_write_wakeup(struct sock *, int mib);
594 void tcp_send_fin(struct sock *sk);
595 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
596 int tcp_send_synack(struct sock *);
597 void tcp_push_one(struct sock *, unsigned int mss_now);
598 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
599 void tcp_send_ack(struct sock *sk);
600 void tcp_send_delayed_ack(struct sock *sk);
601 void tcp_send_loss_probe(struct sock *sk);
602 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
603 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
604 			     const struct sk_buff *next_skb);
605 
606 /* tcp_input.c */
607 void tcp_rearm_rto(struct sock *sk);
608 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
609 void tcp_reset(struct sock *sk);
610 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
611 void tcp_fin(struct sock *sk);
612 
613 /* tcp_timer.c */
614 void tcp_init_xmit_timers(struct sock *);
615 static inline void tcp_clear_xmit_timers(struct sock *sk)
616 {
617 	if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
618 		__sock_put(sk);
619 
620 	if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
621 		__sock_put(sk);
622 
623 	inet_csk_clear_xmit_timers(sk);
624 }
625 
626 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
627 unsigned int tcp_current_mss(struct sock *sk);
628 
629 /* Bound MSS / TSO packet size with the half of the window */
630 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
631 {
632 	int cutoff;
633 
634 	/* When peer uses tiny windows, there is no use in packetizing
635 	 * to sub-MSS pieces for the sake of SWS or making sure there
636 	 * are enough packets in the pipe for fast recovery.
637 	 *
638 	 * On the other hand, for extremely large MSS devices, handling
639 	 * smaller than MSS windows in this way does make sense.
640 	 */
641 	if (tp->max_window > TCP_MSS_DEFAULT)
642 		cutoff = (tp->max_window >> 1);
643 	else
644 		cutoff = tp->max_window;
645 
646 	if (cutoff && pktsize > cutoff)
647 		return max_t(int, cutoff, 68U - tp->tcp_header_len);
648 	else
649 		return pktsize;
650 }
651 
652 /* tcp.c */
653 void tcp_get_info(struct sock *, struct tcp_info *);
654 
655 /* Read 'sendfile()'-style from a TCP socket */
656 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
657 		  sk_read_actor_t recv_actor);
658 
659 void tcp_initialize_rcv_mss(struct sock *sk);
660 
661 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
662 int tcp_mss_to_mtu(struct sock *sk, int mss);
663 void tcp_mtup_init(struct sock *sk);
664 void tcp_init_buffer_space(struct sock *sk);
665 
666 static inline void tcp_bound_rto(const struct sock *sk)
667 {
668 	if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
669 		inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
670 }
671 
672 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
673 {
674 	return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
675 }
676 
677 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
678 {
679 	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
680 			       ntohl(TCP_FLAG_ACK) |
681 			       snd_wnd);
682 }
683 
684 static inline void tcp_fast_path_on(struct tcp_sock *tp)
685 {
686 	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
687 }
688 
689 static inline void tcp_fast_path_check(struct sock *sk)
690 {
691 	struct tcp_sock *tp = tcp_sk(sk);
692 
693 	if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
694 	    tp->rcv_wnd &&
695 	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
696 	    !tp->urg_data)
697 		tcp_fast_path_on(tp);
698 }
699 
700 /* Compute the actual rto_min value */
701 static inline u32 tcp_rto_min(struct sock *sk)
702 {
703 	const struct dst_entry *dst = __sk_dst_get(sk);
704 	u32 rto_min = TCP_RTO_MIN;
705 
706 	if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
707 		rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
708 	return rto_min;
709 }
710 
711 static inline u32 tcp_rto_min_us(struct sock *sk)
712 {
713 	return jiffies_to_usecs(tcp_rto_min(sk));
714 }
715 
716 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
717 {
718 	return dst_metric_locked(dst, RTAX_CC_ALGO);
719 }
720 
721 /* Minimum RTT in usec. ~0 means not available. */
722 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
723 {
724 	return minmax_get(&tp->rtt_min);
725 }
726 
727 /* Compute the actual receive window we are currently advertising.
728  * Rcv_nxt can be after the window if our peer push more data
729  * than the offered window.
730  */
731 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
732 {
733 	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
734 
735 	if (win < 0)
736 		win = 0;
737 	return (u32) win;
738 }
739 
740 /* Choose a new window, without checks for shrinking, and without
741  * scaling applied to the result.  The caller does these things
742  * if necessary.  This is a "raw" window selection.
743  */
744 u32 __tcp_select_window(struct sock *sk);
745 
746 void tcp_send_window_probe(struct sock *sk);
747 
748 /* TCP uses 32bit jiffies to save some space.
749  * Note that this is different from tcp_time_stamp, which
750  * historically has been the same until linux-4.13.
751  */
752 #define tcp_jiffies32 ((u32)jiffies)
753 
754 /*
755  * Deliver a 32bit value for TCP timestamp option (RFC 7323)
756  * It is no longer tied to jiffies, but to 1 ms clock.
757  * Note: double check if you want to use tcp_jiffies32 instead of this.
758  */
759 #define TCP_TS_HZ	1000
760 
761 static inline u64 tcp_clock_ns(void)
762 {
763 	return ktime_get_ns();
764 }
765 
766 static inline u64 tcp_clock_us(void)
767 {
768 	return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
769 }
770 
771 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
772 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
773 {
774 	return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
775 }
776 
777 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
778 static inline u32 tcp_ns_to_ts(u64 ns)
779 {
780 	return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
781 }
782 
783 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
784 static inline u32 tcp_time_stamp_raw(void)
785 {
786 	return tcp_ns_to_ts(tcp_clock_ns());
787 }
788 
789 void tcp_mstamp_refresh(struct tcp_sock *tp);
790 
791 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
792 {
793 	return max_t(s64, t1 - t0, 0);
794 }
795 
796 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
797 {
798 	return tcp_ns_to_ts(skb->skb_mstamp_ns);
799 }
800 
801 /* provide the departure time in us unit */
802 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
803 {
804 	return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
805 }
806 
807 
808 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
809 
810 #define TCPHDR_FIN 0x01
811 #define TCPHDR_SYN 0x02
812 #define TCPHDR_RST 0x04
813 #define TCPHDR_PSH 0x08
814 #define TCPHDR_ACK 0x10
815 #define TCPHDR_URG 0x20
816 #define TCPHDR_ECE 0x40
817 #define TCPHDR_CWR 0x80
818 
819 #define TCPHDR_SYN_ECN	(TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
820 
821 /* This is what the send packet queuing engine uses to pass
822  * TCP per-packet control information to the transmission code.
823  * We also store the host-order sequence numbers in here too.
824  * This is 44 bytes if IPV6 is enabled.
825  * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
826  */
827 struct tcp_skb_cb {
828 	__u32		seq;		/* Starting sequence number	*/
829 	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
830 	union {
831 		/* Note : tcp_tw_isn is used in input path only
832 		 *	  (isn chosen by tcp_timewait_state_process())
833 		 *
834 		 * 	  tcp_gso_segs/size are used in write queue only,
835 		 *	  cf tcp_skb_pcount()/tcp_skb_mss()
836 		 */
837 		__u32		tcp_tw_isn;
838 		struct {
839 			u16	tcp_gso_segs;
840 			u16	tcp_gso_size;
841 		};
842 	};
843 	__u8		tcp_flags;	/* TCP header flags. (tcp[13])	*/
844 
845 	__u8		sacked;		/* State flags for SACK.	*/
846 #define TCPCB_SACKED_ACKED	0x01	/* SKB ACK'd by a SACK block	*/
847 #define TCPCB_SACKED_RETRANS	0x02	/* SKB retransmitted		*/
848 #define TCPCB_LOST		0x04	/* SKB is lost			*/
849 #define TCPCB_TAGBITS		0x07	/* All tag bits			*/
850 #define TCPCB_REPAIRED		0x10	/* SKB repaired (no skb_mstamp_ns)	*/
851 #define TCPCB_EVER_RETRANS	0x80	/* Ever retransmitted frame	*/
852 #define TCPCB_RETRANS		(TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
853 				TCPCB_REPAIRED)
854 
855 	__u8		ip_dsfield;	/* IPv4 tos or IPv6 dsfield	*/
856 	__u8		txstamp_ack:1,	/* Record TX timestamp for ack? */
857 			eor:1,		/* Is skb MSG_EOR marked? */
858 			has_rxtstamp:1,	/* SKB has a RX timestamp	*/
859 			unused:5;
860 	__u32		ack_seq;	/* Sequence number ACK'd	*/
861 	union {
862 		struct {
863 			/* There is space for up to 24 bytes */
864 			__u32 in_flight:30,/* Bytes in flight at transmit */
865 			      is_app_limited:1, /* cwnd not fully used? */
866 			      unused:1;
867 			/* pkts S/ACKed so far upon tx of skb, incl retrans: */
868 			__u32 delivered;
869 			/* start of send pipeline phase */
870 			u64 first_tx_mstamp;
871 			/* when we reached the "delivered" count */
872 			u64 delivered_mstamp;
873 		} tx;   /* only used for outgoing skbs */
874 		union {
875 			struct inet_skb_parm	h4;
876 #if IS_ENABLED(CONFIG_IPV6)
877 			struct inet6_skb_parm	h6;
878 #endif
879 		} header;	/* For incoming skbs */
880 		struct {
881 			__u32 flags;
882 			struct sock *sk_redir;
883 			void *data_end;
884 		} bpf;
885 	};
886 };
887 
888 #define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
889 
890 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb)
891 {
892 	TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb);
893 }
894 
895 static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb)
896 {
897 	return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS;
898 }
899 
900 static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb)
901 {
902 	return TCP_SKB_CB(skb)->bpf.sk_redir;
903 }
904 
905 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb)
906 {
907 	TCP_SKB_CB(skb)->bpf.sk_redir = NULL;
908 }
909 
910 #if IS_ENABLED(CONFIG_IPV6)
911 /* This is the variant of inet6_iif() that must be used by TCP,
912  * as TCP moves IP6CB into a different location in skb->cb[]
913  */
914 static inline int tcp_v6_iif(const struct sk_buff *skb)
915 {
916 	return TCP_SKB_CB(skb)->header.h6.iif;
917 }
918 
919 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
920 {
921 	bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
922 
923 	return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
924 }
925 
926 /* TCP_SKB_CB reference means this can not be used from early demux */
927 static inline int tcp_v6_sdif(const struct sk_buff *skb)
928 {
929 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
930 	if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
931 		return TCP_SKB_CB(skb)->header.h6.iif;
932 #endif
933 	return 0;
934 }
935 #endif
936 
937 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
938 {
939 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
940 	if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
941 	    skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
942 		return true;
943 #endif
944 	return false;
945 }
946 
947 /* TCP_SKB_CB reference means this can not be used from early demux */
948 static inline int tcp_v4_sdif(struct sk_buff *skb)
949 {
950 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
951 	if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
952 		return TCP_SKB_CB(skb)->header.h4.iif;
953 #endif
954 	return 0;
955 }
956 
957 /* Due to TSO, an SKB can be composed of multiple actual
958  * packets.  To keep these tracked properly, we use this.
959  */
960 static inline int tcp_skb_pcount(const struct sk_buff *skb)
961 {
962 	return TCP_SKB_CB(skb)->tcp_gso_segs;
963 }
964 
965 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
966 {
967 	TCP_SKB_CB(skb)->tcp_gso_segs = segs;
968 }
969 
970 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
971 {
972 	TCP_SKB_CB(skb)->tcp_gso_segs += segs;
973 }
974 
975 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
976 static inline int tcp_skb_mss(const struct sk_buff *skb)
977 {
978 	return TCP_SKB_CB(skb)->tcp_gso_size;
979 }
980 
981 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
982 {
983 	return likely(!TCP_SKB_CB(skb)->eor);
984 }
985 
986 static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
987 					const struct sk_buff *from)
988 {
989 	return likely(tcp_skb_can_collapse_to(to) &&
990 		      mptcp_skb_can_collapse(to, from));
991 }
992 
993 /* Events passed to congestion control interface */
994 enum tcp_ca_event {
995 	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
996 	CA_EVENT_CWND_RESTART,	/* congestion window restart */
997 	CA_EVENT_COMPLETE_CWR,	/* end of congestion recovery */
998 	CA_EVENT_LOSS,		/* loss timeout */
999 	CA_EVENT_ECN_NO_CE,	/* ECT set, but not CE marked */
1000 	CA_EVENT_ECN_IS_CE,	/* received CE marked IP packet */
1001 };
1002 
1003 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
1004 enum tcp_ca_ack_event_flags {
1005 	CA_ACK_SLOWPATH		= (1 << 0),	/* In slow path processing */
1006 	CA_ACK_WIN_UPDATE	= (1 << 1),	/* ACK updated window */
1007 	CA_ACK_ECE		= (1 << 2),	/* ECE bit is set on ack */
1008 };
1009 
1010 /*
1011  * Interface for adding new TCP congestion control handlers
1012  */
1013 #define TCP_CA_NAME_MAX	16
1014 #define TCP_CA_MAX	128
1015 #define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
1016 
1017 #define TCP_CA_UNSPEC	0
1018 
1019 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1020 #define TCP_CONG_NON_RESTRICTED 0x1
1021 /* Requires ECN/ECT set on all packets */
1022 #define TCP_CONG_NEEDS_ECN	0x2
1023 #define TCP_CONG_MASK	(TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1024 
1025 union tcp_cc_info;
1026 
1027 struct ack_sample {
1028 	u32 pkts_acked;
1029 	s32 rtt_us;
1030 	u32 in_flight;
1031 };
1032 
1033 /* A rate sample measures the number of (original/retransmitted) data
1034  * packets delivered "delivered" over an interval of time "interval_us".
1035  * The tcp_rate.c code fills in the rate sample, and congestion
1036  * control modules that define a cong_control function to run at the end
1037  * of ACK processing can optionally chose to consult this sample when
1038  * setting cwnd and pacing rate.
1039  * A sample is invalid if "delivered" or "interval_us" is negative.
1040  */
1041 struct rate_sample {
1042 	u64  prior_mstamp; /* starting timestamp for interval */
1043 	u32  prior_delivered;	/* tp->delivered at "prior_mstamp" */
1044 	s32  delivered;		/* number of packets delivered over interval */
1045 	long interval_us;	/* time for tp->delivered to incr "delivered" */
1046 	u32 snd_interval_us;	/* snd interval for delivered packets */
1047 	u32 rcv_interval_us;	/* rcv interval for delivered packets */
1048 	long rtt_us;		/* RTT of last (S)ACKed packet (or -1) */
1049 	int  losses;		/* number of packets marked lost upon ACK */
1050 	u32  acked_sacked;	/* number of packets newly (S)ACKed upon ACK */
1051 	u32  prior_in_flight;	/* in flight before this ACK */
1052 	bool is_app_limited;	/* is sample from packet with bubble in pipe? */
1053 	bool is_retrans;	/* is sample from retransmission? */
1054 	bool is_ack_delayed;	/* is this (likely) a delayed ACK? */
1055 };
1056 
1057 struct tcp_congestion_ops {
1058 	struct list_head	list;
1059 	u32 key;
1060 	u32 flags;
1061 
1062 	/* initialize private data (optional) */
1063 	void (*init)(struct sock *sk);
1064 	/* cleanup private data  (optional) */
1065 	void (*release)(struct sock *sk);
1066 
1067 	/* return slow start threshold (required) */
1068 	u32 (*ssthresh)(struct sock *sk);
1069 	/* do new cwnd calculation (required) */
1070 	void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1071 	/* call before changing ca_state (optional) */
1072 	void (*set_state)(struct sock *sk, u8 new_state);
1073 	/* call when cwnd event occurs (optional) */
1074 	void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1075 	/* call when ack arrives (optional) */
1076 	void (*in_ack_event)(struct sock *sk, u32 flags);
1077 	/* new value of cwnd after loss (required) */
1078 	u32  (*undo_cwnd)(struct sock *sk);
1079 	/* hook for packet ack accounting (optional) */
1080 	void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1081 	/* override sysctl_tcp_min_tso_segs */
1082 	u32 (*min_tso_segs)(struct sock *sk);
1083 	/* returns the multiplier used in tcp_sndbuf_expand (optional) */
1084 	u32 (*sndbuf_expand)(struct sock *sk);
1085 	/* call when packets are delivered to update cwnd and pacing rate,
1086 	 * after all the ca_state processing. (optional)
1087 	 */
1088 	void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1089 	/* get info for inet_diag (optional) */
1090 	size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1091 			   union tcp_cc_info *info);
1092 
1093 	char 		name[TCP_CA_NAME_MAX];
1094 	struct module 	*owner;
1095 };
1096 
1097 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1098 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1099 
1100 void tcp_assign_congestion_control(struct sock *sk);
1101 void tcp_init_congestion_control(struct sock *sk);
1102 void tcp_cleanup_congestion_control(struct sock *sk);
1103 int tcp_set_default_congestion_control(struct net *net, const char *name);
1104 void tcp_get_default_congestion_control(struct net *net, char *name);
1105 void tcp_get_available_congestion_control(char *buf, size_t len);
1106 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1107 int tcp_set_allowed_congestion_control(char *allowed);
1108 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1109 			       bool reinit, bool cap_net_admin);
1110 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1111 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1112 
1113 u32 tcp_reno_ssthresh(struct sock *sk);
1114 u32 tcp_reno_undo_cwnd(struct sock *sk);
1115 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1116 extern struct tcp_congestion_ops tcp_reno;
1117 
1118 struct tcp_congestion_ops *tcp_ca_find(const char *name);
1119 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1120 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1121 #ifdef CONFIG_INET
1122 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1123 #else
1124 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1125 {
1126 	return NULL;
1127 }
1128 #endif
1129 
1130 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1131 {
1132 	const struct inet_connection_sock *icsk = inet_csk(sk);
1133 
1134 	return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1135 }
1136 
1137 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1138 {
1139 	struct inet_connection_sock *icsk = inet_csk(sk);
1140 
1141 	if (icsk->icsk_ca_ops->set_state)
1142 		icsk->icsk_ca_ops->set_state(sk, ca_state);
1143 	icsk->icsk_ca_state = ca_state;
1144 }
1145 
1146 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1147 {
1148 	const struct inet_connection_sock *icsk = inet_csk(sk);
1149 
1150 	if (icsk->icsk_ca_ops->cwnd_event)
1151 		icsk->icsk_ca_ops->cwnd_event(sk, event);
1152 }
1153 
1154 /* From tcp_rate.c */
1155 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1156 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1157 			    struct rate_sample *rs);
1158 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1159 		  bool is_sack_reneg, struct rate_sample *rs);
1160 void tcp_rate_check_app_limited(struct sock *sk);
1161 
1162 /* These functions determine how the current flow behaves in respect of SACK
1163  * handling. SACK is negotiated with the peer, and therefore it can vary
1164  * between different flows.
1165  *
1166  * tcp_is_sack - SACK enabled
1167  * tcp_is_reno - No SACK
1168  */
1169 static inline int tcp_is_sack(const struct tcp_sock *tp)
1170 {
1171 	return likely(tp->rx_opt.sack_ok);
1172 }
1173 
1174 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1175 {
1176 	return !tcp_is_sack(tp);
1177 }
1178 
1179 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1180 {
1181 	return tp->sacked_out + tp->lost_out;
1182 }
1183 
1184 /* This determines how many packets are "in the network" to the best
1185  * of our knowledge.  In many cases it is conservative, but where
1186  * detailed information is available from the receiver (via SACK
1187  * blocks etc.) we can make more aggressive calculations.
1188  *
1189  * Use this for decisions involving congestion control, use just
1190  * tp->packets_out to determine if the send queue is empty or not.
1191  *
1192  * Read this equation as:
1193  *
1194  *	"Packets sent once on transmission queue" MINUS
1195  *	"Packets left network, but not honestly ACKed yet" PLUS
1196  *	"Packets fast retransmitted"
1197  */
1198 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1199 {
1200 	return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1201 }
1202 
1203 #define TCP_INFINITE_SSTHRESH	0x7fffffff
1204 
1205 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1206 {
1207 	return tp->snd_cwnd < tp->snd_ssthresh;
1208 }
1209 
1210 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1211 {
1212 	return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1213 }
1214 
1215 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1216 {
1217 	return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1218 	       (1 << inet_csk(sk)->icsk_ca_state);
1219 }
1220 
1221 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1222  * The exception is cwnd reduction phase, when cwnd is decreasing towards
1223  * ssthresh.
1224  */
1225 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1226 {
1227 	const struct tcp_sock *tp = tcp_sk(sk);
1228 
1229 	if (tcp_in_cwnd_reduction(sk))
1230 		return tp->snd_ssthresh;
1231 	else
1232 		return max(tp->snd_ssthresh,
1233 			   ((tp->snd_cwnd >> 1) +
1234 			    (tp->snd_cwnd >> 2)));
1235 }
1236 
1237 /* Use define here intentionally to get WARN_ON location shown at the caller */
1238 #define tcp_verify_left_out(tp)	WARN_ON(tcp_left_out(tp) > tp->packets_out)
1239 
1240 void tcp_enter_cwr(struct sock *sk);
1241 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1242 
1243 /* The maximum number of MSS of available cwnd for which TSO defers
1244  * sending if not using sysctl_tcp_tso_win_divisor.
1245  */
1246 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1247 {
1248 	return 3;
1249 }
1250 
1251 /* Returns end sequence number of the receiver's advertised window */
1252 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1253 {
1254 	return tp->snd_una + tp->snd_wnd;
1255 }
1256 
1257 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1258  * flexible approach. The RFC suggests cwnd should not be raised unless
1259  * it was fully used previously. And that's exactly what we do in
1260  * congestion avoidance mode. But in slow start we allow cwnd to grow
1261  * as long as the application has used half the cwnd.
1262  * Example :
1263  *    cwnd is 10 (IW10), but application sends 9 frames.
1264  *    We allow cwnd to reach 18 when all frames are ACKed.
1265  * This check is safe because it's as aggressive as slow start which already
1266  * risks 100% overshoot. The advantage is that we discourage application to
1267  * either send more filler packets or data to artificially blow up the cwnd
1268  * usage, and allow application-limited process to probe bw more aggressively.
1269  */
1270 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1271 {
1272 	const struct tcp_sock *tp = tcp_sk(sk);
1273 
1274 	/* If in slow start, ensure cwnd grows to twice what was ACKed. */
1275 	if (tcp_in_slow_start(tp))
1276 		return tp->snd_cwnd < 2 * tp->max_packets_out;
1277 
1278 	return tp->is_cwnd_limited;
1279 }
1280 
1281 /* BBR congestion control needs pacing.
1282  * Same remark for SO_MAX_PACING_RATE.
1283  * sch_fq packet scheduler is efficiently handling pacing,
1284  * but is not always installed/used.
1285  * Return true if TCP stack should pace packets itself.
1286  */
1287 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1288 {
1289 	return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1290 }
1291 
1292 /* Return in jiffies the delay before one skb is sent.
1293  * If @skb is NULL, we look at EDT for next packet being sent on the socket.
1294  */
1295 static inline unsigned long tcp_pacing_delay(const struct sock *sk,
1296 					     const struct sk_buff *skb)
1297 {
1298 	s64 pacing_delay = skb ? skb->tstamp : tcp_sk(sk)->tcp_wstamp_ns;
1299 
1300 	pacing_delay -= tcp_sk(sk)->tcp_clock_cache;
1301 
1302 	return pacing_delay > 0 ? nsecs_to_jiffies(pacing_delay) : 0;
1303 }
1304 
1305 static inline void tcp_reset_xmit_timer(struct sock *sk,
1306 					const int what,
1307 					unsigned long when,
1308 					const unsigned long max_when,
1309 					const struct sk_buff *skb)
1310 {
1311 	inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk, skb),
1312 				  max_when);
1313 }
1314 
1315 /* Something is really bad, we could not queue an additional packet,
1316  * because qdisc is full or receiver sent a 0 window, or we are paced.
1317  * We do not want to add fuel to the fire, or abort too early,
1318  * so make sure the timer we arm now is at least 200ms in the future,
1319  * regardless of current icsk_rto value (as it could be ~2ms)
1320  */
1321 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1322 {
1323 	return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1324 }
1325 
1326 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1327 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1328 					    unsigned long max_when)
1329 {
1330 	u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1331 
1332 	return (unsigned long)min_t(u64, when, max_when);
1333 }
1334 
1335 static inline void tcp_check_probe_timer(struct sock *sk)
1336 {
1337 	if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1338 		tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1339 				     tcp_probe0_base(sk), TCP_RTO_MAX,
1340 				     NULL);
1341 }
1342 
1343 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1344 {
1345 	tp->snd_wl1 = seq;
1346 }
1347 
1348 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1349 {
1350 	tp->snd_wl1 = seq;
1351 }
1352 
1353 /*
1354  * Calculate(/check) TCP checksum
1355  */
1356 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1357 				   __be32 daddr, __wsum base)
1358 {
1359 	return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1360 }
1361 
1362 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1363 {
1364 	return !skb_csum_unnecessary(skb) &&
1365 		__skb_checksum_complete(skb);
1366 }
1367 
1368 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1369 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1370 void tcp_set_state(struct sock *sk, int state);
1371 void tcp_done(struct sock *sk);
1372 int tcp_abort(struct sock *sk, int err);
1373 
1374 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1375 {
1376 	rx_opt->dsack = 0;
1377 	rx_opt->num_sacks = 0;
1378 }
1379 
1380 u32 tcp_default_init_rwnd(u32 mss);
1381 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1382 
1383 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1384 {
1385 	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1386 	struct tcp_sock *tp = tcp_sk(sk);
1387 	s32 delta;
1388 
1389 	if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1390 	    ca_ops->cong_control)
1391 		return;
1392 	delta = tcp_jiffies32 - tp->lsndtime;
1393 	if (delta > inet_csk(sk)->icsk_rto)
1394 		tcp_cwnd_restart(sk, delta);
1395 }
1396 
1397 /* Determine a window scaling and initial window to offer. */
1398 void tcp_select_initial_window(const struct sock *sk, int __space,
1399 			       __u32 mss, __u32 *rcv_wnd,
1400 			       __u32 *window_clamp, int wscale_ok,
1401 			       __u8 *rcv_wscale, __u32 init_rcv_wnd);
1402 
1403 static inline int tcp_win_from_space(const struct sock *sk, int space)
1404 {
1405 	int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1406 
1407 	return tcp_adv_win_scale <= 0 ?
1408 		(space>>(-tcp_adv_win_scale)) :
1409 		space - (space>>tcp_adv_win_scale);
1410 }
1411 
1412 /* Note: caller must be prepared to deal with negative returns */
1413 static inline int tcp_space(const struct sock *sk)
1414 {
1415 	return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1416 				  READ_ONCE(sk->sk_backlog.len) -
1417 				  atomic_read(&sk->sk_rmem_alloc));
1418 }
1419 
1420 static inline int tcp_full_space(const struct sock *sk)
1421 {
1422 	return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1423 }
1424 
1425 extern void tcp_openreq_init_rwin(struct request_sock *req,
1426 				  const struct sock *sk_listener,
1427 				  const struct dst_entry *dst);
1428 
1429 void tcp_enter_memory_pressure(struct sock *sk);
1430 void tcp_leave_memory_pressure(struct sock *sk);
1431 
1432 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1433 {
1434 	struct net *net = sock_net((struct sock *)tp);
1435 
1436 	return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1437 }
1438 
1439 static inline int keepalive_time_when(const struct tcp_sock *tp)
1440 {
1441 	struct net *net = sock_net((struct sock *)tp);
1442 
1443 	return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1444 }
1445 
1446 static inline int keepalive_probes(const struct tcp_sock *tp)
1447 {
1448 	struct net *net = sock_net((struct sock *)tp);
1449 
1450 	return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1451 }
1452 
1453 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1454 {
1455 	const struct inet_connection_sock *icsk = &tp->inet_conn;
1456 
1457 	return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1458 			  tcp_jiffies32 - tp->rcv_tstamp);
1459 }
1460 
1461 static inline int tcp_fin_time(const struct sock *sk)
1462 {
1463 	int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1464 	const int rto = inet_csk(sk)->icsk_rto;
1465 
1466 	if (fin_timeout < (rto << 2) - (rto >> 1))
1467 		fin_timeout = (rto << 2) - (rto >> 1);
1468 
1469 	return fin_timeout;
1470 }
1471 
1472 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1473 				  int paws_win)
1474 {
1475 	if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1476 		return true;
1477 	if (unlikely(!time_before32(ktime_get_seconds(),
1478 				    rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1479 		return true;
1480 	/*
1481 	 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1482 	 * then following tcp messages have valid values. Ignore 0 value,
1483 	 * or else 'negative' tsval might forbid us to accept their packets.
1484 	 */
1485 	if (!rx_opt->ts_recent)
1486 		return true;
1487 	return false;
1488 }
1489 
1490 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1491 				   int rst)
1492 {
1493 	if (tcp_paws_check(rx_opt, 0))
1494 		return false;
1495 
1496 	/* RST segments are not recommended to carry timestamp,
1497 	   and, if they do, it is recommended to ignore PAWS because
1498 	   "their cleanup function should take precedence over timestamps."
1499 	   Certainly, it is mistake. It is necessary to understand the reasons
1500 	   of this constraint to relax it: if peer reboots, clock may go
1501 	   out-of-sync and half-open connections will not be reset.
1502 	   Actually, the problem would be not existing if all
1503 	   the implementations followed draft about maintaining clock
1504 	   via reboots. Linux-2.2 DOES NOT!
1505 
1506 	   However, we can relax time bounds for RST segments to MSL.
1507 	 */
1508 	if (rst && !time_before32(ktime_get_seconds(),
1509 				  rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1510 		return false;
1511 	return true;
1512 }
1513 
1514 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1515 			  int mib_idx, u32 *last_oow_ack_time);
1516 
1517 static inline void tcp_mib_init(struct net *net)
1518 {
1519 	/* See RFC 2012 */
1520 	TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1521 	TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1522 	TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1523 	TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1524 }
1525 
1526 /* from STCP */
1527 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1528 {
1529 	tp->lost_skb_hint = NULL;
1530 }
1531 
1532 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1533 {
1534 	tcp_clear_retrans_hints_partial(tp);
1535 	tp->retransmit_skb_hint = NULL;
1536 }
1537 
1538 union tcp_md5_addr {
1539 	struct in_addr  a4;
1540 #if IS_ENABLED(CONFIG_IPV6)
1541 	struct in6_addr	a6;
1542 #endif
1543 };
1544 
1545 /* - key database */
1546 struct tcp_md5sig_key {
1547 	struct hlist_node	node;
1548 	u8			keylen;
1549 	u8			family; /* AF_INET or AF_INET6 */
1550 	u8			prefixlen;
1551 	union tcp_md5_addr	addr;
1552 	int			l3index; /* set if key added with L3 scope */
1553 	u8			key[TCP_MD5SIG_MAXKEYLEN];
1554 	struct rcu_head		rcu;
1555 };
1556 
1557 /* - sock block */
1558 struct tcp_md5sig_info {
1559 	struct hlist_head	head;
1560 	struct rcu_head		rcu;
1561 };
1562 
1563 /* - pseudo header */
1564 struct tcp4_pseudohdr {
1565 	__be32		saddr;
1566 	__be32		daddr;
1567 	__u8		pad;
1568 	__u8		protocol;
1569 	__be16		len;
1570 };
1571 
1572 struct tcp6_pseudohdr {
1573 	struct in6_addr	saddr;
1574 	struct in6_addr daddr;
1575 	__be32		len;
1576 	__be32		protocol;	/* including padding */
1577 };
1578 
1579 union tcp_md5sum_block {
1580 	struct tcp4_pseudohdr ip4;
1581 #if IS_ENABLED(CONFIG_IPV6)
1582 	struct tcp6_pseudohdr ip6;
1583 #endif
1584 };
1585 
1586 /* - pool: digest algorithm, hash description and scratch buffer */
1587 struct tcp_md5sig_pool {
1588 	struct ahash_request	*md5_req;
1589 	void			*scratch;
1590 };
1591 
1592 /* - functions */
1593 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1594 			const struct sock *sk, const struct sk_buff *skb);
1595 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1596 		   int family, u8 prefixlen, int l3index,
1597 		   const u8 *newkey, u8 newkeylen, gfp_t gfp);
1598 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1599 		   int family, u8 prefixlen, int l3index);
1600 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1601 					 const struct sock *addr_sk);
1602 
1603 #ifdef CONFIG_TCP_MD5SIG
1604 #include <linux/jump_label.h>
1605 extern struct static_key_false tcp_md5_needed;
1606 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1607 					   const union tcp_md5_addr *addr,
1608 					   int family);
1609 static inline struct tcp_md5sig_key *
1610 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1611 		  const union tcp_md5_addr *addr, int family)
1612 {
1613 	if (!static_branch_unlikely(&tcp_md5_needed))
1614 		return NULL;
1615 	return __tcp_md5_do_lookup(sk, l3index, addr, family);
1616 }
1617 
1618 #define tcp_twsk_md5_key(twsk)	((twsk)->tw_md5_key)
1619 #else
1620 static inline struct tcp_md5sig_key *
1621 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1622 		  const union tcp_md5_addr *addr, int family)
1623 {
1624 	return NULL;
1625 }
1626 #define tcp_twsk_md5_key(twsk)	NULL
1627 #endif
1628 
1629 bool tcp_alloc_md5sig_pool(void);
1630 
1631 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1632 static inline void tcp_put_md5sig_pool(void)
1633 {
1634 	local_bh_enable();
1635 }
1636 
1637 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1638 			  unsigned int header_len);
1639 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1640 		     const struct tcp_md5sig_key *key);
1641 
1642 /* From tcp_fastopen.c */
1643 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1644 			    struct tcp_fastopen_cookie *cookie);
1645 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1646 			    struct tcp_fastopen_cookie *cookie, bool syn_lost,
1647 			    u16 try_exp);
1648 struct tcp_fastopen_request {
1649 	/* Fast Open cookie. Size 0 means a cookie request */
1650 	struct tcp_fastopen_cookie	cookie;
1651 	struct msghdr			*data;  /* data in MSG_FASTOPEN */
1652 	size_t				size;
1653 	int				copied;	/* queued in tcp_connect() */
1654 	struct ubuf_info		*uarg;
1655 };
1656 void tcp_free_fastopen_req(struct tcp_sock *tp);
1657 void tcp_fastopen_destroy_cipher(struct sock *sk);
1658 void tcp_fastopen_ctx_destroy(struct net *net);
1659 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1660 			      void *primary_key, void *backup_key);
1661 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1662 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1663 			      struct request_sock *req,
1664 			      struct tcp_fastopen_cookie *foc,
1665 			      const struct dst_entry *dst);
1666 void tcp_fastopen_init_key_once(struct net *net);
1667 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1668 			     struct tcp_fastopen_cookie *cookie);
1669 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1670 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1671 #define TCP_FASTOPEN_KEY_MAX 2
1672 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1673 	(TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1674 
1675 /* Fastopen key context */
1676 struct tcp_fastopen_context {
1677 	siphash_key_t	key[TCP_FASTOPEN_KEY_MAX];
1678 	int		num;
1679 	struct rcu_head	rcu;
1680 };
1681 
1682 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1683 void tcp_fastopen_active_disable(struct sock *sk);
1684 bool tcp_fastopen_active_should_disable(struct sock *sk);
1685 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1686 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1687 
1688 /* Caller needs to wrap with rcu_read_(un)lock() */
1689 static inline
1690 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1691 {
1692 	struct tcp_fastopen_context *ctx;
1693 
1694 	ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1695 	if (!ctx)
1696 		ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1697 	return ctx;
1698 }
1699 
1700 static inline
1701 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1702 			       const struct tcp_fastopen_cookie *orig)
1703 {
1704 	if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1705 	    orig->len == foc->len &&
1706 	    !memcmp(orig->val, foc->val, foc->len))
1707 		return true;
1708 	return false;
1709 }
1710 
1711 static inline
1712 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1713 {
1714 	return ctx->num;
1715 }
1716 
1717 /* Latencies incurred by various limits for a sender. They are
1718  * chronograph-like stats that are mutually exclusive.
1719  */
1720 enum tcp_chrono {
1721 	TCP_CHRONO_UNSPEC,
1722 	TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1723 	TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1724 	TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1725 	__TCP_CHRONO_MAX,
1726 };
1727 
1728 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1729 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1730 
1731 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1732  * the same memory storage than skb->destructor/_skb_refdst
1733  */
1734 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1735 {
1736 	skb->destructor = NULL;
1737 	skb->_skb_refdst = 0UL;
1738 }
1739 
1740 #define tcp_skb_tsorted_save(skb) {		\
1741 	unsigned long _save = skb->_skb_refdst;	\
1742 	skb->_skb_refdst = 0UL;
1743 
1744 #define tcp_skb_tsorted_restore(skb)		\
1745 	skb->_skb_refdst = _save;		\
1746 }
1747 
1748 void tcp_write_queue_purge(struct sock *sk);
1749 
1750 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1751 {
1752 	return skb_rb_first(&sk->tcp_rtx_queue);
1753 }
1754 
1755 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1756 {
1757 	return skb_rb_last(&sk->tcp_rtx_queue);
1758 }
1759 
1760 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1761 {
1762 	return skb_peek(&sk->sk_write_queue);
1763 }
1764 
1765 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1766 {
1767 	return skb_peek_tail(&sk->sk_write_queue);
1768 }
1769 
1770 #define tcp_for_write_queue_from_safe(skb, tmp, sk)			\
1771 	skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1772 
1773 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1774 {
1775 	return skb_peek(&sk->sk_write_queue);
1776 }
1777 
1778 static inline bool tcp_skb_is_last(const struct sock *sk,
1779 				   const struct sk_buff *skb)
1780 {
1781 	return skb_queue_is_last(&sk->sk_write_queue, skb);
1782 }
1783 
1784 /**
1785  * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
1786  * @sk: socket
1787  *
1788  * Since the write queue can have a temporary empty skb in it,
1789  * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
1790  */
1791 static inline bool tcp_write_queue_empty(const struct sock *sk)
1792 {
1793 	const struct tcp_sock *tp = tcp_sk(sk);
1794 
1795 	return tp->write_seq == tp->snd_nxt;
1796 }
1797 
1798 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1799 {
1800 	return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1801 }
1802 
1803 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1804 {
1805 	return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1806 }
1807 
1808 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1809 {
1810 	__skb_queue_tail(&sk->sk_write_queue, skb);
1811 
1812 	/* Queue it, remembering where we must start sending. */
1813 	if (sk->sk_write_queue.next == skb)
1814 		tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1815 }
1816 
1817 /* Insert new before skb on the write queue of sk.  */
1818 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1819 						  struct sk_buff *skb,
1820 						  struct sock *sk)
1821 {
1822 	__skb_queue_before(&sk->sk_write_queue, skb, new);
1823 }
1824 
1825 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1826 {
1827 	tcp_skb_tsorted_anchor_cleanup(skb);
1828 	__skb_unlink(skb, &sk->sk_write_queue);
1829 }
1830 
1831 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1832 
1833 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1834 {
1835 	tcp_skb_tsorted_anchor_cleanup(skb);
1836 	rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1837 }
1838 
1839 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1840 {
1841 	list_del(&skb->tcp_tsorted_anchor);
1842 	tcp_rtx_queue_unlink(skb, sk);
1843 	sk_wmem_free_skb(sk, skb);
1844 }
1845 
1846 static inline void tcp_push_pending_frames(struct sock *sk)
1847 {
1848 	if (tcp_send_head(sk)) {
1849 		struct tcp_sock *tp = tcp_sk(sk);
1850 
1851 		__tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1852 	}
1853 }
1854 
1855 /* Start sequence of the skb just after the highest skb with SACKed
1856  * bit, valid only if sacked_out > 0 or when the caller has ensured
1857  * validity by itself.
1858  */
1859 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1860 {
1861 	if (!tp->sacked_out)
1862 		return tp->snd_una;
1863 
1864 	if (tp->highest_sack == NULL)
1865 		return tp->snd_nxt;
1866 
1867 	return TCP_SKB_CB(tp->highest_sack)->seq;
1868 }
1869 
1870 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1871 {
1872 	tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1873 }
1874 
1875 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1876 {
1877 	return tcp_sk(sk)->highest_sack;
1878 }
1879 
1880 static inline void tcp_highest_sack_reset(struct sock *sk)
1881 {
1882 	tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1883 }
1884 
1885 /* Called when old skb is about to be deleted and replaced by new skb */
1886 static inline void tcp_highest_sack_replace(struct sock *sk,
1887 					    struct sk_buff *old,
1888 					    struct sk_buff *new)
1889 {
1890 	if (old == tcp_highest_sack(sk))
1891 		tcp_sk(sk)->highest_sack = new;
1892 }
1893 
1894 /* This helper checks if socket has IP_TRANSPARENT set */
1895 static inline bool inet_sk_transparent(const struct sock *sk)
1896 {
1897 	switch (sk->sk_state) {
1898 	case TCP_TIME_WAIT:
1899 		return inet_twsk(sk)->tw_transparent;
1900 	case TCP_NEW_SYN_RECV:
1901 		return inet_rsk(inet_reqsk(sk))->no_srccheck;
1902 	}
1903 	return inet_sk(sk)->transparent;
1904 }
1905 
1906 /* Determines whether this is a thin stream (which may suffer from
1907  * increased latency). Used to trigger latency-reducing mechanisms.
1908  */
1909 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1910 {
1911 	return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1912 }
1913 
1914 /* /proc */
1915 enum tcp_seq_states {
1916 	TCP_SEQ_STATE_LISTENING,
1917 	TCP_SEQ_STATE_ESTABLISHED,
1918 };
1919 
1920 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1921 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1922 void tcp_seq_stop(struct seq_file *seq, void *v);
1923 
1924 struct tcp_seq_afinfo {
1925 	sa_family_t			family;
1926 };
1927 
1928 struct tcp_iter_state {
1929 	struct seq_net_private	p;
1930 	enum tcp_seq_states	state;
1931 	struct sock		*syn_wait_sk;
1932 	int			bucket, offset, sbucket, num;
1933 	loff_t			last_pos;
1934 };
1935 
1936 extern struct request_sock_ops tcp_request_sock_ops;
1937 extern struct request_sock_ops tcp6_request_sock_ops;
1938 
1939 void tcp_v4_destroy_sock(struct sock *sk);
1940 
1941 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1942 				netdev_features_t features);
1943 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
1944 int tcp_gro_complete(struct sk_buff *skb);
1945 
1946 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1947 
1948 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1949 {
1950 	struct net *net = sock_net((struct sock *)tp);
1951 	return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1952 }
1953 
1954 /* @wake is one when sk_stream_write_space() calls us.
1955  * This sends EPOLLOUT only if notsent_bytes is half the limit.
1956  * This mimics the strategy used in sock_def_write_space().
1957  */
1958 static inline bool tcp_stream_memory_free(const struct sock *sk, int wake)
1959 {
1960 	const struct tcp_sock *tp = tcp_sk(sk);
1961 	u32 notsent_bytes = READ_ONCE(tp->write_seq) -
1962 			    READ_ONCE(tp->snd_nxt);
1963 
1964 	return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
1965 }
1966 
1967 #ifdef CONFIG_PROC_FS
1968 int tcp4_proc_init(void);
1969 void tcp4_proc_exit(void);
1970 #endif
1971 
1972 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1973 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1974 		     const struct tcp_request_sock_ops *af_ops,
1975 		     struct sock *sk, struct sk_buff *skb);
1976 
1977 /* TCP af-specific functions */
1978 struct tcp_sock_af_ops {
1979 #ifdef CONFIG_TCP_MD5SIG
1980 	struct tcp_md5sig_key	*(*md5_lookup) (const struct sock *sk,
1981 						const struct sock *addr_sk);
1982 	int		(*calc_md5_hash)(char *location,
1983 					 const struct tcp_md5sig_key *md5,
1984 					 const struct sock *sk,
1985 					 const struct sk_buff *skb);
1986 	int		(*md5_parse)(struct sock *sk,
1987 				     int optname,
1988 				     char __user *optval,
1989 				     int optlen);
1990 #endif
1991 };
1992 
1993 struct tcp_request_sock_ops {
1994 	u16 mss_clamp;
1995 #ifdef CONFIG_TCP_MD5SIG
1996 	struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1997 						 const struct sock *addr_sk);
1998 	int		(*calc_md5_hash) (char *location,
1999 					  const struct tcp_md5sig_key *md5,
2000 					  const struct sock *sk,
2001 					  const struct sk_buff *skb);
2002 #endif
2003 	void (*init_req)(struct request_sock *req,
2004 			 const struct sock *sk_listener,
2005 			 struct sk_buff *skb);
2006 #ifdef CONFIG_SYN_COOKIES
2007 	__u32 (*cookie_init_seq)(const struct sk_buff *skb,
2008 				 __u16 *mss);
2009 #endif
2010 	struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
2011 				       const struct request_sock *req);
2012 	u32 (*init_seq)(const struct sk_buff *skb);
2013 	u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
2014 	int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2015 			   struct flowi *fl, struct request_sock *req,
2016 			   struct tcp_fastopen_cookie *foc,
2017 			   enum tcp_synack_type synack_type);
2018 };
2019 
2020 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2021 #if IS_ENABLED(CONFIG_IPV6)
2022 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2023 #endif
2024 
2025 #ifdef CONFIG_SYN_COOKIES
2026 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2027 					 const struct sock *sk, struct sk_buff *skb,
2028 					 __u16 *mss)
2029 {
2030 	tcp_synq_overflow(sk);
2031 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2032 	return ops->cookie_init_seq(skb, mss);
2033 }
2034 #else
2035 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2036 					 const struct sock *sk, struct sk_buff *skb,
2037 					 __u16 *mss)
2038 {
2039 	return 0;
2040 }
2041 #endif
2042 
2043 int tcpv4_offload_init(void);
2044 
2045 void tcp_v4_init(void);
2046 void tcp_init(void);
2047 
2048 /* tcp_recovery.c */
2049 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2050 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2051 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2052 				u32 reo_wnd);
2053 extern void tcp_rack_mark_lost(struct sock *sk);
2054 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2055 			     u64 xmit_time);
2056 extern void tcp_rack_reo_timeout(struct sock *sk);
2057 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2058 
2059 /* At how many usecs into the future should the RTO fire? */
2060 static inline s64 tcp_rto_delta_us(const struct sock *sk)
2061 {
2062 	const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2063 	u32 rto = inet_csk(sk)->icsk_rto;
2064 	u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2065 
2066 	return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2067 }
2068 
2069 /*
2070  * Save and compile IPv4 options, return a pointer to it
2071  */
2072 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2073 							 struct sk_buff *skb)
2074 {
2075 	const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2076 	struct ip_options_rcu *dopt = NULL;
2077 
2078 	if (opt->optlen) {
2079 		int opt_size = sizeof(*dopt) + opt->optlen;
2080 
2081 		dopt = kmalloc(opt_size, GFP_ATOMIC);
2082 		if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2083 			kfree(dopt);
2084 			dopt = NULL;
2085 		}
2086 	}
2087 	return dopt;
2088 }
2089 
2090 /* locally generated TCP pure ACKs have skb->truesize == 2
2091  * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2092  * This is much faster than dissecting the packet to find out.
2093  * (Think of GRE encapsulations, IPv4, IPv6, ...)
2094  */
2095 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2096 {
2097 	return skb->truesize == 2;
2098 }
2099 
2100 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2101 {
2102 	skb->truesize = 2;
2103 }
2104 
2105 static inline int tcp_inq(struct sock *sk)
2106 {
2107 	struct tcp_sock *tp = tcp_sk(sk);
2108 	int answ;
2109 
2110 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2111 		answ = 0;
2112 	} else if (sock_flag(sk, SOCK_URGINLINE) ||
2113 		   !tp->urg_data ||
2114 		   before(tp->urg_seq, tp->copied_seq) ||
2115 		   !before(tp->urg_seq, tp->rcv_nxt)) {
2116 
2117 		answ = tp->rcv_nxt - tp->copied_seq;
2118 
2119 		/* Subtract 1, if FIN was received */
2120 		if (answ && sock_flag(sk, SOCK_DONE))
2121 			answ--;
2122 	} else {
2123 		answ = tp->urg_seq - tp->copied_seq;
2124 	}
2125 
2126 	return answ;
2127 }
2128 
2129 int tcp_peek_len(struct socket *sock);
2130 
2131 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2132 {
2133 	u16 segs_in;
2134 
2135 	segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2136 	tp->segs_in += segs_in;
2137 	if (skb->len > tcp_hdrlen(skb))
2138 		tp->data_segs_in += segs_in;
2139 }
2140 
2141 /*
2142  * TCP listen path runs lockless.
2143  * We forced "struct sock" to be const qualified to make sure
2144  * we don't modify one of its field by mistake.
2145  * Here, we increment sk_drops which is an atomic_t, so we can safely
2146  * make sock writable again.
2147  */
2148 static inline void tcp_listendrop(const struct sock *sk)
2149 {
2150 	atomic_inc(&((struct sock *)sk)->sk_drops);
2151 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2152 }
2153 
2154 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2155 
2156 /*
2157  * Interface for adding Upper Level Protocols over TCP
2158  */
2159 
2160 #define TCP_ULP_NAME_MAX	16
2161 #define TCP_ULP_MAX		128
2162 #define TCP_ULP_BUF_MAX		(TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2163 
2164 struct tcp_ulp_ops {
2165 	struct list_head	list;
2166 
2167 	/* initialize ulp */
2168 	int (*init)(struct sock *sk);
2169 	/* update ulp */
2170 	void (*update)(struct sock *sk, struct proto *p,
2171 		       void (*write_space)(struct sock *sk));
2172 	/* cleanup ulp */
2173 	void (*release)(struct sock *sk);
2174 	/* diagnostic */
2175 	int (*get_info)(const struct sock *sk, struct sk_buff *skb);
2176 	size_t (*get_info_size)(const struct sock *sk);
2177 	/* clone ulp */
2178 	void (*clone)(const struct request_sock *req, struct sock *newsk,
2179 		      const gfp_t priority);
2180 
2181 	char		name[TCP_ULP_NAME_MAX];
2182 	struct module	*owner;
2183 };
2184 int tcp_register_ulp(struct tcp_ulp_ops *type);
2185 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2186 int tcp_set_ulp(struct sock *sk, const char *name);
2187 void tcp_get_available_ulp(char *buf, size_t len);
2188 void tcp_cleanup_ulp(struct sock *sk);
2189 void tcp_update_ulp(struct sock *sk, struct proto *p,
2190 		    void (*write_space)(struct sock *sk));
2191 
2192 #define MODULE_ALIAS_TCP_ULP(name)				\
2193 	__MODULE_INFO(alias, alias_userspace, name);		\
2194 	__MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2195 
2196 struct sk_msg;
2197 struct sk_psock;
2198 
2199 #ifdef CONFIG_BPF_STREAM_PARSER
2200 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
2201 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2202 #else
2203 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2204 {
2205 }
2206 #endif /* CONFIG_BPF_STREAM_PARSER */
2207 
2208 #ifdef CONFIG_NET_SOCK_MSG
2209 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes,
2210 			  int flags);
2211 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock,
2212 		      struct msghdr *msg, int len, int flags);
2213 #endif /* CONFIG_NET_SOCK_MSG */
2214 
2215 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2216  * is < 0, then the BPF op failed (for example if the loaded BPF
2217  * program does not support the chosen operation or there is no BPF
2218  * program loaded).
2219  */
2220 #ifdef CONFIG_BPF
2221 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2222 {
2223 	struct bpf_sock_ops_kern sock_ops;
2224 	int ret;
2225 
2226 	memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2227 	if (sk_fullsock(sk)) {
2228 		sock_ops.is_fullsock = 1;
2229 		sock_owned_by_me(sk);
2230 	}
2231 
2232 	sock_ops.sk = sk;
2233 	sock_ops.op = op;
2234 	if (nargs > 0)
2235 		memcpy(sock_ops.args, args, nargs * sizeof(*args));
2236 
2237 	ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2238 	if (ret == 0)
2239 		ret = sock_ops.reply;
2240 	else
2241 		ret = -1;
2242 	return ret;
2243 }
2244 
2245 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2246 {
2247 	u32 args[2] = {arg1, arg2};
2248 
2249 	return tcp_call_bpf(sk, op, 2, args);
2250 }
2251 
2252 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2253 				    u32 arg3)
2254 {
2255 	u32 args[3] = {arg1, arg2, arg3};
2256 
2257 	return tcp_call_bpf(sk, op, 3, args);
2258 }
2259 
2260 #else
2261 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2262 {
2263 	return -EPERM;
2264 }
2265 
2266 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2267 {
2268 	return -EPERM;
2269 }
2270 
2271 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2272 				    u32 arg3)
2273 {
2274 	return -EPERM;
2275 }
2276 
2277 #endif
2278 
2279 static inline u32 tcp_timeout_init(struct sock *sk)
2280 {
2281 	int timeout;
2282 
2283 	timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2284 
2285 	if (timeout <= 0)
2286 		timeout = TCP_TIMEOUT_INIT;
2287 	return timeout;
2288 }
2289 
2290 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2291 {
2292 	int rwnd;
2293 
2294 	rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2295 
2296 	if (rwnd < 0)
2297 		rwnd = 0;
2298 	return rwnd;
2299 }
2300 
2301 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2302 {
2303 	return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2304 }
2305 
2306 static inline void tcp_bpf_rtt(struct sock *sk)
2307 {
2308 	if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2309 		tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2310 }
2311 
2312 #if IS_ENABLED(CONFIG_SMC)
2313 extern struct static_key_false tcp_have_smc;
2314 #endif
2315 
2316 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2317 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2318 			     void (*cad)(struct sock *sk, u32 ack_seq));
2319 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2320 void clean_acked_data_flush(void);
2321 #endif
2322 
2323 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2324 static inline void tcp_add_tx_delay(struct sk_buff *skb,
2325 				    const struct tcp_sock *tp)
2326 {
2327 	if (static_branch_unlikely(&tcp_tx_delay_enabled))
2328 		skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2329 }
2330 
2331 /* Compute Earliest Departure Time for some control packets
2332  * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2333  */
2334 static inline u64 tcp_transmit_time(const struct sock *sk)
2335 {
2336 	if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2337 		u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2338 			tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2339 
2340 		return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
2341 	}
2342 	return 0;
2343 }
2344 
2345 #endif	/* _TCP_H */
2346