xref: /openbmc/linux/include/net/tcp.h (revision 36acd5e2)
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/kref.h>
27 #include <linux/ktime.h>
28 #include <linux/indirect_call_wrapper.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 void tcp_remove_empty_skb(struct sock *sk, struct sk_buff *skb);
326 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
327 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
328 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
329 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
330 		 int flags);
331 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
332 			size_t size, int flags);
333 struct sk_buff *tcp_build_frag(struct sock *sk, int size_goal, int flags,
334 			       struct page *page, int offset, size_t *size);
335 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
336 		 size_t size, int flags);
337 int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
338 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
339 	      int size_goal);
340 void tcp_release_cb(struct sock *sk);
341 void tcp_wfree(struct sk_buff *skb);
342 void tcp_write_timer_handler(struct sock *sk);
343 void tcp_delack_timer_handler(struct sock *sk);
344 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
345 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
346 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
347 void tcp_rcv_space_adjust(struct sock *sk);
348 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
349 void tcp_twsk_destructor(struct sock *sk);
350 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
351 			struct pipe_inode_info *pipe, size_t len,
352 			unsigned int flags);
353 
354 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
355 static inline void tcp_dec_quickack_mode(struct sock *sk,
356 					 const unsigned int pkts)
357 {
358 	struct inet_connection_sock *icsk = inet_csk(sk);
359 
360 	if (icsk->icsk_ack.quick) {
361 		if (pkts >= icsk->icsk_ack.quick) {
362 			icsk->icsk_ack.quick = 0;
363 			/* Leaving quickack mode we deflate ATO. */
364 			icsk->icsk_ack.ato   = TCP_ATO_MIN;
365 		} else
366 			icsk->icsk_ack.quick -= pkts;
367 	}
368 }
369 
370 #define	TCP_ECN_OK		1
371 #define	TCP_ECN_QUEUE_CWR	2
372 #define	TCP_ECN_DEMAND_CWR	4
373 #define	TCP_ECN_SEEN		8
374 
375 enum tcp_tw_status {
376 	TCP_TW_SUCCESS = 0,
377 	TCP_TW_RST = 1,
378 	TCP_TW_ACK = 2,
379 	TCP_TW_SYN = 3
380 };
381 
382 
383 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
384 					      struct sk_buff *skb,
385 					      const struct tcphdr *th);
386 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
387 			   struct request_sock *req, bool fastopen,
388 			   bool *lost_race);
389 int tcp_child_process(struct sock *parent, struct sock *child,
390 		      struct sk_buff *skb);
391 void tcp_enter_loss(struct sock *sk);
392 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag);
393 void tcp_clear_retrans(struct tcp_sock *tp);
394 void tcp_update_metrics(struct sock *sk);
395 void tcp_init_metrics(struct sock *sk);
396 void tcp_metrics_init(void);
397 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
398 void __tcp_close(struct sock *sk, long timeout);
399 void tcp_close(struct sock *sk, long timeout);
400 void tcp_init_sock(struct sock *sk);
401 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb);
402 __poll_t tcp_poll(struct file *file, struct socket *sock,
403 		      struct poll_table_struct *wait);
404 int tcp_getsockopt(struct sock *sk, int level, int optname,
405 		   char __user *optval, int __user *optlen);
406 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
407 		   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 int tcp_set_window_clamp(struct sock *sk, int val);
414 void tcp_data_ready(struct sock *sk);
415 #ifdef CONFIG_MMU
416 int tcp_mmap(struct file *file, struct socket *sock,
417 	     struct vm_area_struct *vma);
418 #endif
419 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
420 		       struct tcp_options_received *opt_rx,
421 		       int estab, struct tcp_fastopen_cookie *foc);
422 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
423 
424 /*
425  *	BPF SKB-less helpers
426  */
427 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
428 			 struct tcphdr *th, u32 *cookie);
429 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
430 			 struct tcphdr *th, u32 *cookie);
431 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
432 			  const struct tcp_request_sock_ops *af_ops,
433 			  struct sock *sk, struct tcphdr *th);
434 /*
435  *	TCP v4 functions exported for the inet6 API
436  */
437 
438 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
439 void tcp_v4_mtu_reduced(struct sock *sk);
440 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
441 void tcp_ld_RTO_revert(struct sock *sk, u32 seq);
442 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
443 struct sock *tcp_create_openreq_child(const struct sock *sk,
444 				      struct request_sock *req,
445 				      struct sk_buff *skb);
446 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
447 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
448 				  struct request_sock *req,
449 				  struct dst_entry *dst,
450 				  struct request_sock *req_unhash,
451 				  bool *own_req);
452 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
453 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
454 int tcp_connect(struct sock *sk);
455 enum tcp_synack_type {
456 	TCP_SYNACK_NORMAL,
457 	TCP_SYNACK_FASTOPEN,
458 	TCP_SYNACK_COOKIE,
459 };
460 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
461 				struct request_sock *req,
462 				struct tcp_fastopen_cookie *foc,
463 				enum tcp_synack_type synack_type,
464 				struct sk_buff *syn_skb);
465 int tcp_disconnect(struct sock *sk, int flags);
466 
467 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
468 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
469 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
470 
471 /* From syncookies.c */
472 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
473 				 struct request_sock *req,
474 				 struct dst_entry *dst, u32 tsoff);
475 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
476 		      u32 cookie);
477 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
478 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
479 					    struct sock *sk, struct sk_buff *skb);
480 #ifdef CONFIG_SYN_COOKIES
481 
482 /* Syncookies use a monotonic timer which increments every 60 seconds.
483  * This counter is used both as a hash input and partially encoded into
484  * the cookie value.  A cookie is only validated further if the delta
485  * between the current counter value and the encoded one is less than this,
486  * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
487  * the counter advances immediately after a cookie is generated).
488  */
489 #define MAX_SYNCOOKIE_AGE	2
490 #define TCP_SYNCOOKIE_PERIOD	(60 * HZ)
491 #define TCP_SYNCOOKIE_VALID	(MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
492 
493 /* syncookies: remember time of last synqueue overflow
494  * But do not dirty this field too often (once per second is enough)
495  * It is racy as we do not hold a lock, but race is very minor.
496  */
497 static inline void tcp_synq_overflow(const struct sock *sk)
498 {
499 	unsigned int last_overflow;
500 	unsigned int now = jiffies;
501 
502 	if (sk->sk_reuseport) {
503 		struct sock_reuseport *reuse;
504 
505 		reuse = rcu_dereference(sk->sk_reuseport_cb);
506 		if (likely(reuse)) {
507 			last_overflow = READ_ONCE(reuse->synq_overflow_ts);
508 			if (!time_between32(now, last_overflow,
509 					    last_overflow + HZ))
510 				WRITE_ONCE(reuse->synq_overflow_ts, now);
511 			return;
512 		}
513 	}
514 
515 	last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
516 	if (!time_between32(now, last_overflow, last_overflow + HZ))
517 		WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
518 }
519 
520 /* syncookies: no recent synqueue overflow on this listening socket? */
521 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
522 {
523 	unsigned int last_overflow;
524 	unsigned int now = jiffies;
525 
526 	if (sk->sk_reuseport) {
527 		struct sock_reuseport *reuse;
528 
529 		reuse = rcu_dereference(sk->sk_reuseport_cb);
530 		if (likely(reuse)) {
531 			last_overflow = READ_ONCE(reuse->synq_overflow_ts);
532 			return !time_between32(now, last_overflow - HZ,
533 					       last_overflow +
534 					       TCP_SYNCOOKIE_VALID);
535 		}
536 	}
537 
538 	last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
539 
540 	/* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
541 	 * then we're under synflood. However, we have to use
542 	 * 'last_overflow - HZ' as lower bound. That's because a concurrent
543 	 * tcp_synq_overflow() could update .ts_recent_stamp after we read
544 	 * jiffies but before we store .ts_recent_stamp into last_overflow,
545 	 * which could lead to rejecting a valid syncookie.
546 	 */
547 	return !time_between32(now, last_overflow - HZ,
548 			       last_overflow + TCP_SYNCOOKIE_VALID);
549 }
550 
551 static inline u32 tcp_cookie_time(void)
552 {
553 	u64 val = get_jiffies_64();
554 
555 	do_div(val, TCP_SYNCOOKIE_PERIOD);
556 	return val;
557 }
558 
559 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
560 			      u16 *mssp);
561 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
562 u64 cookie_init_timestamp(struct request_sock *req, u64 now);
563 bool cookie_timestamp_decode(const struct net *net,
564 			     struct tcp_options_received *opt);
565 bool cookie_ecn_ok(const struct tcp_options_received *opt,
566 		   const struct net *net, const struct dst_entry *dst);
567 
568 /* From net/ipv6/syncookies.c */
569 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
570 		      u32 cookie);
571 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
572 
573 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
574 			      const struct tcphdr *th, u16 *mssp);
575 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
576 #endif
577 /* tcp_output.c */
578 
579 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
580 			       int nonagle);
581 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
582 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
583 void tcp_retransmit_timer(struct sock *sk);
584 void tcp_xmit_retransmit_queue(struct sock *);
585 void tcp_simple_retransmit(struct sock *);
586 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
587 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
588 enum tcp_queue {
589 	TCP_FRAG_IN_WRITE_QUEUE,
590 	TCP_FRAG_IN_RTX_QUEUE,
591 };
592 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
593 		 struct sk_buff *skb, u32 len,
594 		 unsigned int mss_now, gfp_t gfp);
595 
596 void tcp_send_probe0(struct sock *);
597 void tcp_send_partial(struct sock *);
598 int tcp_write_wakeup(struct sock *, int mib);
599 void tcp_send_fin(struct sock *sk);
600 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
601 int tcp_send_synack(struct sock *);
602 void tcp_push_one(struct sock *, unsigned int mss_now);
603 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
604 void tcp_send_ack(struct sock *sk);
605 void tcp_send_delayed_ack(struct sock *sk);
606 void tcp_send_loss_probe(struct sock *sk);
607 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
608 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
609 			     const struct sk_buff *next_skb);
610 
611 /* tcp_input.c */
612 void tcp_rearm_rto(struct sock *sk);
613 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
614 void tcp_reset(struct sock *sk, struct sk_buff *skb);
615 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
616 void tcp_fin(struct sock *sk);
617 
618 /* tcp_timer.c */
619 void tcp_init_xmit_timers(struct sock *);
620 static inline void tcp_clear_xmit_timers(struct sock *sk)
621 {
622 	if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
623 		__sock_put(sk);
624 
625 	if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
626 		__sock_put(sk);
627 
628 	inet_csk_clear_xmit_timers(sk);
629 }
630 
631 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
632 unsigned int tcp_current_mss(struct sock *sk);
633 
634 /* Bound MSS / TSO packet size with the half of the window */
635 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
636 {
637 	int cutoff;
638 
639 	/* When peer uses tiny windows, there is no use in packetizing
640 	 * to sub-MSS pieces for the sake of SWS or making sure there
641 	 * are enough packets in the pipe for fast recovery.
642 	 *
643 	 * On the other hand, for extremely large MSS devices, handling
644 	 * smaller than MSS windows in this way does make sense.
645 	 */
646 	if (tp->max_window > TCP_MSS_DEFAULT)
647 		cutoff = (tp->max_window >> 1);
648 	else
649 		cutoff = tp->max_window;
650 
651 	if (cutoff && pktsize > cutoff)
652 		return max_t(int, cutoff, 68U - tp->tcp_header_len);
653 	else
654 		return pktsize;
655 }
656 
657 /* tcp.c */
658 void tcp_get_info(struct sock *, struct tcp_info *);
659 
660 /* Read 'sendfile()'-style from a TCP socket */
661 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
662 		  sk_read_actor_t recv_actor);
663 
664 void tcp_initialize_rcv_mss(struct sock *sk);
665 
666 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
667 int tcp_mss_to_mtu(struct sock *sk, int mss);
668 void tcp_mtup_init(struct sock *sk);
669 
670 static inline void tcp_bound_rto(const struct sock *sk)
671 {
672 	if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
673 		inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
674 }
675 
676 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
677 {
678 	return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
679 }
680 
681 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
682 {
683 	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
684 			       ntohl(TCP_FLAG_ACK) |
685 			       snd_wnd);
686 }
687 
688 static inline void tcp_fast_path_on(struct tcp_sock *tp)
689 {
690 	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
691 }
692 
693 static inline void tcp_fast_path_check(struct sock *sk)
694 {
695 	struct tcp_sock *tp = tcp_sk(sk);
696 
697 	if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
698 	    tp->rcv_wnd &&
699 	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
700 	    !tp->urg_data)
701 		tcp_fast_path_on(tp);
702 }
703 
704 /* Compute the actual rto_min value */
705 static inline u32 tcp_rto_min(struct sock *sk)
706 {
707 	const struct dst_entry *dst = __sk_dst_get(sk);
708 	u32 rto_min = inet_csk(sk)->icsk_rto_min;
709 
710 	if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
711 		rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
712 	return rto_min;
713 }
714 
715 static inline u32 tcp_rto_min_us(struct sock *sk)
716 {
717 	return jiffies_to_usecs(tcp_rto_min(sk));
718 }
719 
720 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
721 {
722 	return dst_metric_locked(dst, RTAX_CC_ALGO);
723 }
724 
725 /* Minimum RTT in usec. ~0 means not available. */
726 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
727 {
728 	return minmax_get(&tp->rtt_min);
729 }
730 
731 /* Compute the actual receive window we are currently advertising.
732  * Rcv_nxt can be after the window if our peer push more data
733  * than the offered window.
734  */
735 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
736 {
737 	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
738 
739 	if (win < 0)
740 		win = 0;
741 	return (u32) win;
742 }
743 
744 /* Choose a new window, without checks for shrinking, and without
745  * scaling applied to the result.  The caller does these things
746  * if necessary.  This is a "raw" window selection.
747  */
748 u32 __tcp_select_window(struct sock *sk);
749 
750 void tcp_send_window_probe(struct sock *sk);
751 
752 /* TCP uses 32bit jiffies to save some space.
753  * Note that this is different from tcp_time_stamp, which
754  * historically has been the same until linux-4.13.
755  */
756 #define tcp_jiffies32 ((u32)jiffies)
757 
758 /*
759  * Deliver a 32bit value for TCP timestamp option (RFC 7323)
760  * It is no longer tied to jiffies, but to 1 ms clock.
761  * Note: double check if you want to use tcp_jiffies32 instead of this.
762  */
763 #define TCP_TS_HZ	1000
764 
765 static inline u64 tcp_clock_ns(void)
766 {
767 	return ktime_get_ns();
768 }
769 
770 static inline u64 tcp_clock_us(void)
771 {
772 	return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
773 }
774 
775 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
776 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
777 {
778 	return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
779 }
780 
781 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
782 static inline u32 tcp_ns_to_ts(u64 ns)
783 {
784 	return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
785 }
786 
787 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
788 static inline u32 tcp_time_stamp_raw(void)
789 {
790 	return tcp_ns_to_ts(tcp_clock_ns());
791 }
792 
793 void tcp_mstamp_refresh(struct tcp_sock *tp);
794 
795 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
796 {
797 	return max_t(s64, t1 - t0, 0);
798 }
799 
800 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
801 {
802 	return tcp_ns_to_ts(skb->skb_mstamp_ns);
803 }
804 
805 /* provide the departure time in us unit */
806 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
807 {
808 	return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
809 }
810 
811 
812 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
813 
814 #define TCPHDR_FIN 0x01
815 #define TCPHDR_SYN 0x02
816 #define TCPHDR_RST 0x04
817 #define TCPHDR_PSH 0x08
818 #define TCPHDR_ACK 0x10
819 #define TCPHDR_URG 0x20
820 #define TCPHDR_ECE 0x40
821 #define TCPHDR_CWR 0x80
822 
823 #define TCPHDR_SYN_ECN	(TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
824 
825 /* This is what the send packet queuing engine uses to pass
826  * TCP per-packet control information to the transmission code.
827  * We also store the host-order sequence numbers in here too.
828  * This is 44 bytes if IPV6 is enabled.
829  * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
830  */
831 struct tcp_skb_cb {
832 	__u32		seq;		/* Starting sequence number	*/
833 	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
834 	union {
835 		/* Note : tcp_tw_isn is used in input path only
836 		 *	  (isn chosen by tcp_timewait_state_process())
837 		 *
838 		 * 	  tcp_gso_segs/size are used in write queue only,
839 		 *	  cf tcp_skb_pcount()/tcp_skb_mss()
840 		 */
841 		__u32		tcp_tw_isn;
842 		struct {
843 			u16	tcp_gso_segs;
844 			u16	tcp_gso_size;
845 		};
846 	};
847 	__u8		tcp_flags;	/* TCP header flags. (tcp[13])	*/
848 
849 	__u8		sacked;		/* State flags for SACK.	*/
850 #define TCPCB_SACKED_ACKED	0x01	/* SKB ACK'd by a SACK block	*/
851 #define TCPCB_SACKED_RETRANS	0x02	/* SKB retransmitted		*/
852 #define TCPCB_LOST		0x04	/* SKB is lost			*/
853 #define TCPCB_TAGBITS		0x07	/* All tag bits			*/
854 #define TCPCB_REPAIRED		0x10	/* SKB repaired (no skb_mstamp_ns)	*/
855 #define TCPCB_EVER_RETRANS	0x80	/* Ever retransmitted frame	*/
856 #define TCPCB_RETRANS		(TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
857 				TCPCB_REPAIRED)
858 
859 	__u8		ip_dsfield;	/* IPv4 tos or IPv6 dsfield	*/
860 	__u8		txstamp_ack:1,	/* Record TX timestamp for ack? */
861 			eor:1,		/* Is skb MSG_EOR marked? */
862 			has_rxtstamp:1,	/* SKB has a RX timestamp	*/
863 			unused:5;
864 	__u32		ack_seq;	/* Sequence number ACK'd	*/
865 	union {
866 		struct {
867 			/* There is space for up to 24 bytes */
868 			__u32 in_flight:30,/* Bytes in flight at transmit */
869 			      is_app_limited:1, /* cwnd not fully used? */
870 			      unused:1;
871 			/* pkts S/ACKed so far upon tx of skb, incl retrans: */
872 			__u32 delivered;
873 			/* start of send pipeline phase */
874 			u64 first_tx_mstamp;
875 			/* when we reached the "delivered" count */
876 			u64 delivered_mstamp;
877 		} tx;   /* only used for outgoing skbs */
878 		union {
879 			struct inet_skb_parm	h4;
880 #if IS_ENABLED(CONFIG_IPV6)
881 			struct inet6_skb_parm	h6;
882 #endif
883 		} header;	/* For incoming skbs */
884 		struct {
885 			__u32 flags;
886 			struct sock *sk_redir;
887 			void *data_end;
888 		} bpf;
889 	};
890 };
891 
892 #define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
893 
894 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb)
895 {
896 	TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb);
897 }
898 
899 static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb)
900 {
901 	return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS;
902 }
903 
904 static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb)
905 {
906 	return TCP_SKB_CB(skb)->bpf.sk_redir;
907 }
908 
909 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb)
910 {
911 	TCP_SKB_CB(skb)->bpf.sk_redir = NULL;
912 }
913 
914 extern const struct inet_connection_sock_af_ops ipv4_specific;
915 
916 #if IS_ENABLED(CONFIG_IPV6)
917 /* This is the variant of inet6_iif() that must be used by TCP,
918  * as TCP moves IP6CB into a different location in skb->cb[]
919  */
920 static inline int tcp_v6_iif(const struct sk_buff *skb)
921 {
922 	return TCP_SKB_CB(skb)->header.h6.iif;
923 }
924 
925 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
926 {
927 	bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
928 
929 	return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
930 }
931 
932 /* TCP_SKB_CB reference means this can not be used from early demux */
933 static inline int tcp_v6_sdif(const struct sk_buff *skb)
934 {
935 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
936 	if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
937 		return TCP_SKB_CB(skb)->header.h6.iif;
938 #endif
939 	return 0;
940 }
941 
942 extern const struct inet_connection_sock_af_ops ipv6_specific;
943 
944 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
945 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
946 INDIRECT_CALLABLE_DECLARE(void tcp_v6_early_demux(struct sk_buff *skb));
947 
948 #endif
949 
950 /* TCP_SKB_CB reference means this can not be used from early demux */
951 static inline int tcp_v4_sdif(struct sk_buff *skb)
952 {
953 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
954 	if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
955 		return TCP_SKB_CB(skb)->header.h4.iif;
956 #endif
957 	return 0;
958 }
959 
960 /* Due to TSO, an SKB can be composed of multiple actual
961  * packets.  To keep these tracked properly, we use this.
962  */
963 static inline int tcp_skb_pcount(const struct sk_buff *skb)
964 {
965 	return TCP_SKB_CB(skb)->tcp_gso_segs;
966 }
967 
968 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
969 {
970 	TCP_SKB_CB(skb)->tcp_gso_segs = segs;
971 }
972 
973 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
974 {
975 	TCP_SKB_CB(skb)->tcp_gso_segs += segs;
976 }
977 
978 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
979 static inline int tcp_skb_mss(const struct sk_buff *skb)
980 {
981 	return TCP_SKB_CB(skb)->tcp_gso_size;
982 }
983 
984 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
985 {
986 	return likely(!TCP_SKB_CB(skb)->eor);
987 }
988 
989 static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
990 					const struct sk_buff *from)
991 {
992 	return likely(tcp_skb_can_collapse_to(to) &&
993 		      mptcp_skb_can_collapse(to, from));
994 }
995 
996 /* Events passed to congestion control interface */
997 enum tcp_ca_event {
998 	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
999 	CA_EVENT_CWND_RESTART,	/* congestion window restart */
1000 	CA_EVENT_COMPLETE_CWR,	/* end of congestion recovery */
1001 	CA_EVENT_LOSS,		/* loss timeout */
1002 	CA_EVENT_ECN_NO_CE,	/* ECT set, but not CE marked */
1003 	CA_EVENT_ECN_IS_CE,	/* received CE marked IP packet */
1004 };
1005 
1006 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
1007 enum tcp_ca_ack_event_flags {
1008 	CA_ACK_SLOWPATH		= (1 << 0),	/* In slow path processing */
1009 	CA_ACK_WIN_UPDATE	= (1 << 1),	/* ACK updated window */
1010 	CA_ACK_ECE		= (1 << 2),	/* ECE bit is set on ack */
1011 };
1012 
1013 /*
1014  * Interface for adding new TCP congestion control handlers
1015  */
1016 #define TCP_CA_NAME_MAX	16
1017 #define TCP_CA_MAX	128
1018 #define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
1019 
1020 #define TCP_CA_UNSPEC	0
1021 
1022 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1023 #define TCP_CONG_NON_RESTRICTED 0x1
1024 /* Requires ECN/ECT set on all packets */
1025 #define TCP_CONG_NEEDS_ECN	0x2
1026 #define TCP_CONG_MASK	(TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1027 
1028 union tcp_cc_info;
1029 
1030 struct ack_sample {
1031 	u32 pkts_acked;
1032 	s32 rtt_us;
1033 	u32 in_flight;
1034 };
1035 
1036 /* A rate sample measures the number of (original/retransmitted) data
1037  * packets delivered "delivered" over an interval of time "interval_us".
1038  * The tcp_rate.c code fills in the rate sample, and congestion
1039  * control modules that define a cong_control function to run at the end
1040  * of ACK processing can optionally chose to consult this sample when
1041  * setting cwnd and pacing rate.
1042  * A sample is invalid if "delivered" or "interval_us" is negative.
1043  */
1044 struct rate_sample {
1045 	u64  prior_mstamp; /* starting timestamp for interval */
1046 	u32  prior_delivered;	/* tp->delivered at "prior_mstamp" */
1047 	s32  delivered;		/* number of packets delivered over interval */
1048 	long interval_us;	/* time for tp->delivered to incr "delivered" */
1049 	u32 snd_interval_us;	/* snd interval for delivered packets */
1050 	u32 rcv_interval_us;	/* rcv interval for delivered packets */
1051 	long rtt_us;		/* RTT of last (S)ACKed packet (or -1) */
1052 	int  losses;		/* number of packets marked lost upon ACK */
1053 	u32  acked_sacked;	/* number of packets newly (S)ACKed upon ACK */
1054 	u32  prior_in_flight;	/* in flight before this ACK */
1055 	bool is_app_limited;	/* is sample from packet with bubble in pipe? */
1056 	bool is_retrans;	/* is sample from retransmission? */
1057 	bool is_ack_delayed;	/* is this (likely) a delayed ACK? */
1058 };
1059 
1060 struct tcp_congestion_ops {
1061 	struct list_head	list;
1062 	u32 key;
1063 	u32 flags;
1064 
1065 	/* initialize private data (optional) */
1066 	void (*init)(struct sock *sk);
1067 	/* cleanup private data  (optional) */
1068 	void (*release)(struct sock *sk);
1069 
1070 	/* return slow start threshold (required) */
1071 	u32 (*ssthresh)(struct sock *sk);
1072 	/* do new cwnd calculation (required) */
1073 	void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1074 	/* call before changing ca_state (optional) */
1075 	void (*set_state)(struct sock *sk, u8 new_state);
1076 	/* call when cwnd event occurs (optional) */
1077 	void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1078 	/* call when ack arrives (optional) */
1079 	void (*in_ack_event)(struct sock *sk, u32 flags);
1080 	/* new value of cwnd after loss (required) */
1081 	u32  (*undo_cwnd)(struct sock *sk);
1082 	/* hook for packet ack accounting (optional) */
1083 	void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1084 	/* override sysctl_tcp_min_tso_segs */
1085 	u32 (*min_tso_segs)(struct sock *sk);
1086 	/* returns the multiplier used in tcp_sndbuf_expand (optional) */
1087 	u32 (*sndbuf_expand)(struct sock *sk);
1088 	/* call when packets are delivered to update cwnd and pacing rate,
1089 	 * after all the ca_state processing. (optional)
1090 	 */
1091 	void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1092 	/* get info for inet_diag (optional) */
1093 	size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1094 			   union tcp_cc_info *info);
1095 
1096 	char 		name[TCP_CA_NAME_MAX];
1097 	struct module 	*owner;
1098 };
1099 
1100 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1101 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1102 
1103 void tcp_assign_congestion_control(struct sock *sk);
1104 void tcp_init_congestion_control(struct sock *sk);
1105 void tcp_cleanup_congestion_control(struct sock *sk);
1106 int tcp_set_default_congestion_control(struct net *net, const char *name);
1107 void tcp_get_default_congestion_control(struct net *net, char *name);
1108 void tcp_get_available_congestion_control(char *buf, size_t len);
1109 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1110 int tcp_set_allowed_congestion_control(char *allowed);
1111 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1112 			       bool cap_net_admin);
1113 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1114 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1115 
1116 u32 tcp_reno_ssthresh(struct sock *sk);
1117 u32 tcp_reno_undo_cwnd(struct sock *sk);
1118 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1119 extern struct tcp_congestion_ops tcp_reno;
1120 
1121 struct tcp_congestion_ops *tcp_ca_find(const char *name);
1122 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1123 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1124 #ifdef CONFIG_INET
1125 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1126 #else
1127 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1128 {
1129 	return NULL;
1130 }
1131 #endif
1132 
1133 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1134 {
1135 	const struct inet_connection_sock *icsk = inet_csk(sk);
1136 
1137 	return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1138 }
1139 
1140 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1141 {
1142 	struct inet_connection_sock *icsk = inet_csk(sk);
1143 
1144 	if (icsk->icsk_ca_ops->set_state)
1145 		icsk->icsk_ca_ops->set_state(sk, ca_state);
1146 	icsk->icsk_ca_state = ca_state;
1147 }
1148 
1149 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1150 {
1151 	const struct inet_connection_sock *icsk = inet_csk(sk);
1152 
1153 	if (icsk->icsk_ca_ops->cwnd_event)
1154 		icsk->icsk_ca_ops->cwnd_event(sk, event);
1155 }
1156 
1157 /* From tcp_rate.c */
1158 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1159 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1160 			    struct rate_sample *rs);
1161 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1162 		  bool is_sack_reneg, struct rate_sample *rs);
1163 void tcp_rate_check_app_limited(struct sock *sk);
1164 
1165 /* These functions determine how the current flow behaves in respect of SACK
1166  * handling. SACK is negotiated with the peer, and therefore it can vary
1167  * between different flows.
1168  *
1169  * tcp_is_sack - SACK enabled
1170  * tcp_is_reno - No SACK
1171  */
1172 static inline int tcp_is_sack(const struct tcp_sock *tp)
1173 {
1174 	return likely(tp->rx_opt.sack_ok);
1175 }
1176 
1177 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1178 {
1179 	return !tcp_is_sack(tp);
1180 }
1181 
1182 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1183 {
1184 	return tp->sacked_out + tp->lost_out;
1185 }
1186 
1187 /* This determines how many packets are "in the network" to the best
1188  * of our knowledge.  In many cases it is conservative, but where
1189  * detailed information is available from the receiver (via SACK
1190  * blocks etc.) we can make more aggressive calculations.
1191  *
1192  * Use this for decisions involving congestion control, use just
1193  * tp->packets_out to determine if the send queue is empty or not.
1194  *
1195  * Read this equation as:
1196  *
1197  *	"Packets sent once on transmission queue" MINUS
1198  *	"Packets left network, but not honestly ACKed yet" PLUS
1199  *	"Packets fast retransmitted"
1200  */
1201 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1202 {
1203 	return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1204 }
1205 
1206 #define TCP_INFINITE_SSTHRESH	0x7fffffff
1207 
1208 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1209 {
1210 	return tp->snd_cwnd < tp->snd_ssthresh;
1211 }
1212 
1213 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1214 {
1215 	return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1216 }
1217 
1218 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1219 {
1220 	return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1221 	       (1 << inet_csk(sk)->icsk_ca_state);
1222 }
1223 
1224 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1225  * The exception is cwnd reduction phase, when cwnd is decreasing towards
1226  * ssthresh.
1227  */
1228 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1229 {
1230 	const struct tcp_sock *tp = tcp_sk(sk);
1231 
1232 	if (tcp_in_cwnd_reduction(sk))
1233 		return tp->snd_ssthresh;
1234 	else
1235 		return max(tp->snd_ssthresh,
1236 			   ((tp->snd_cwnd >> 1) +
1237 			    (tp->snd_cwnd >> 2)));
1238 }
1239 
1240 /* Use define here intentionally to get WARN_ON location shown at the caller */
1241 #define tcp_verify_left_out(tp)	WARN_ON(tcp_left_out(tp) > tp->packets_out)
1242 
1243 void tcp_enter_cwr(struct sock *sk);
1244 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1245 
1246 /* The maximum number of MSS of available cwnd for which TSO defers
1247  * sending if not using sysctl_tcp_tso_win_divisor.
1248  */
1249 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1250 {
1251 	return 3;
1252 }
1253 
1254 /* Returns end sequence number of the receiver's advertised window */
1255 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1256 {
1257 	return tp->snd_una + tp->snd_wnd;
1258 }
1259 
1260 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1261  * flexible approach. The RFC suggests cwnd should not be raised unless
1262  * it was fully used previously. And that's exactly what we do in
1263  * congestion avoidance mode. But in slow start we allow cwnd to grow
1264  * as long as the application has used half the cwnd.
1265  * Example :
1266  *    cwnd is 10 (IW10), but application sends 9 frames.
1267  *    We allow cwnd to reach 18 when all frames are ACKed.
1268  * This check is safe because it's as aggressive as slow start which already
1269  * risks 100% overshoot. The advantage is that we discourage application to
1270  * either send more filler packets or data to artificially blow up the cwnd
1271  * usage, and allow application-limited process to probe bw more aggressively.
1272  */
1273 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1274 {
1275 	const struct tcp_sock *tp = tcp_sk(sk);
1276 
1277 	/* If in slow start, ensure cwnd grows to twice what was ACKed. */
1278 	if (tcp_in_slow_start(tp))
1279 		return tp->snd_cwnd < 2 * tp->max_packets_out;
1280 
1281 	return tp->is_cwnd_limited;
1282 }
1283 
1284 /* BBR congestion control needs pacing.
1285  * Same remark for SO_MAX_PACING_RATE.
1286  * sch_fq packet scheduler is efficiently handling pacing,
1287  * but is not always installed/used.
1288  * Return true if TCP stack should pace packets itself.
1289  */
1290 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1291 {
1292 	return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1293 }
1294 
1295 /* Estimates in how many jiffies next packet for this flow can be sent.
1296  * Scheduling a retransmit timer too early would be silly.
1297  */
1298 static inline unsigned long tcp_pacing_delay(const struct sock *sk)
1299 {
1300 	s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
1301 
1302 	return delay > 0 ? nsecs_to_jiffies(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 {
1310 	inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk),
1311 				  max_when);
1312 }
1313 
1314 /* Something is really bad, we could not queue an additional packet,
1315  * because qdisc is full or receiver sent a 0 window, or we are paced.
1316  * We do not want to add fuel to the fire, or abort too early,
1317  * so make sure the timer we arm now is at least 200ms in the future,
1318  * regardless of current icsk_rto value (as it could be ~2ms)
1319  */
1320 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1321 {
1322 	return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1323 }
1324 
1325 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1326 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1327 					    unsigned long max_when)
1328 {
1329 	u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1,
1330 			   inet_csk(sk)->icsk_backoff);
1331 	u64 when = (u64)tcp_probe0_base(sk) << backoff;
1332 
1333 	return (unsigned long)min_t(u64, when, max_when);
1334 }
1335 
1336 static inline void tcp_check_probe_timer(struct sock *sk)
1337 {
1338 	if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1339 		tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1340 				     tcp_probe0_base(sk), TCP_RTO_MAX);
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 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1381 
1382 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1383 {
1384 	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1385 	struct tcp_sock *tp = tcp_sk(sk);
1386 	s32 delta;
1387 
1388 	if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1389 	    ca_ops->cong_control)
1390 		return;
1391 	delta = tcp_jiffies32 - tp->lsndtime;
1392 	if (delta > inet_csk(sk)->icsk_rto)
1393 		tcp_cwnd_restart(sk, delta);
1394 }
1395 
1396 /* Determine a window scaling and initial window to offer. */
1397 void tcp_select_initial_window(const struct sock *sk, int __space,
1398 			       __u32 mss, __u32 *rcv_wnd,
1399 			       __u32 *window_clamp, int wscale_ok,
1400 			       __u8 *rcv_wscale, __u32 init_rcv_wnd);
1401 
1402 static inline int tcp_win_from_space(const struct sock *sk, int space)
1403 {
1404 	int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1405 
1406 	return tcp_adv_win_scale <= 0 ?
1407 		(space>>(-tcp_adv_win_scale)) :
1408 		space - (space>>tcp_adv_win_scale);
1409 }
1410 
1411 /* Note: caller must be prepared to deal with negative returns */
1412 static inline int tcp_space(const struct sock *sk)
1413 {
1414 	return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1415 				  READ_ONCE(sk->sk_backlog.len) -
1416 				  atomic_read(&sk->sk_rmem_alloc));
1417 }
1418 
1419 static inline int tcp_full_space(const struct sock *sk)
1420 {
1421 	return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1422 }
1423 
1424 void tcp_cleanup_rbuf(struct sock *sk, int copied);
1425 
1426 /* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1427  * If 87.5 % (7/8) of the space has been consumed, we want to override
1428  * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1429  * len/truesize ratio.
1430  */
1431 static inline bool tcp_rmem_pressure(const struct sock *sk)
1432 {
1433 	int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1434 	int threshold = rcvbuf - (rcvbuf >> 3);
1435 
1436 	return atomic_read(&sk->sk_rmem_alloc) > threshold;
1437 }
1438 
1439 extern void tcp_openreq_init_rwin(struct request_sock *req,
1440 				  const struct sock *sk_listener,
1441 				  const struct dst_entry *dst);
1442 
1443 void tcp_enter_memory_pressure(struct sock *sk);
1444 void tcp_leave_memory_pressure(struct sock *sk);
1445 
1446 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1447 {
1448 	struct net *net = sock_net((struct sock *)tp);
1449 
1450 	return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1451 }
1452 
1453 static inline int keepalive_time_when(const struct tcp_sock *tp)
1454 {
1455 	struct net *net = sock_net((struct sock *)tp);
1456 
1457 	return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1458 }
1459 
1460 static inline int keepalive_probes(const struct tcp_sock *tp)
1461 {
1462 	struct net *net = sock_net((struct sock *)tp);
1463 
1464 	return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1465 }
1466 
1467 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1468 {
1469 	const struct inet_connection_sock *icsk = &tp->inet_conn;
1470 
1471 	return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1472 			  tcp_jiffies32 - tp->rcv_tstamp);
1473 }
1474 
1475 static inline int tcp_fin_time(const struct sock *sk)
1476 {
1477 	int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1478 	const int rto = inet_csk(sk)->icsk_rto;
1479 
1480 	if (fin_timeout < (rto << 2) - (rto >> 1))
1481 		fin_timeout = (rto << 2) - (rto >> 1);
1482 
1483 	return fin_timeout;
1484 }
1485 
1486 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1487 				  int paws_win)
1488 {
1489 	if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1490 		return true;
1491 	if (unlikely(!time_before32(ktime_get_seconds(),
1492 				    rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1493 		return true;
1494 	/*
1495 	 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1496 	 * then following tcp messages have valid values. Ignore 0 value,
1497 	 * or else 'negative' tsval might forbid us to accept their packets.
1498 	 */
1499 	if (!rx_opt->ts_recent)
1500 		return true;
1501 	return false;
1502 }
1503 
1504 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1505 				   int rst)
1506 {
1507 	if (tcp_paws_check(rx_opt, 0))
1508 		return false;
1509 
1510 	/* RST segments are not recommended to carry timestamp,
1511 	   and, if they do, it is recommended to ignore PAWS because
1512 	   "their cleanup function should take precedence over timestamps."
1513 	   Certainly, it is mistake. It is necessary to understand the reasons
1514 	   of this constraint to relax it: if peer reboots, clock may go
1515 	   out-of-sync and half-open connections will not be reset.
1516 	   Actually, the problem would be not existing if all
1517 	   the implementations followed draft about maintaining clock
1518 	   via reboots. Linux-2.2 DOES NOT!
1519 
1520 	   However, we can relax time bounds for RST segments to MSL.
1521 	 */
1522 	if (rst && !time_before32(ktime_get_seconds(),
1523 				  rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1524 		return false;
1525 	return true;
1526 }
1527 
1528 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1529 			  int mib_idx, u32 *last_oow_ack_time);
1530 
1531 static inline void tcp_mib_init(struct net *net)
1532 {
1533 	/* See RFC 2012 */
1534 	TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1535 	TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1536 	TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1537 	TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1538 }
1539 
1540 /* from STCP */
1541 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1542 {
1543 	tp->lost_skb_hint = NULL;
1544 }
1545 
1546 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1547 {
1548 	tcp_clear_retrans_hints_partial(tp);
1549 	tp->retransmit_skb_hint = NULL;
1550 }
1551 
1552 union tcp_md5_addr {
1553 	struct in_addr  a4;
1554 #if IS_ENABLED(CONFIG_IPV6)
1555 	struct in6_addr	a6;
1556 #endif
1557 };
1558 
1559 /* - key database */
1560 struct tcp_md5sig_key {
1561 	struct hlist_node	node;
1562 	u8			keylen;
1563 	u8			family; /* AF_INET or AF_INET6 */
1564 	u8			prefixlen;
1565 	union tcp_md5_addr	addr;
1566 	int			l3index; /* set if key added with L3 scope */
1567 	u8			key[TCP_MD5SIG_MAXKEYLEN];
1568 	struct rcu_head		rcu;
1569 };
1570 
1571 /* - sock block */
1572 struct tcp_md5sig_info {
1573 	struct hlist_head	head;
1574 	struct rcu_head		rcu;
1575 };
1576 
1577 /* - pseudo header */
1578 struct tcp4_pseudohdr {
1579 	__be32		saddr;
1580 	__be32		daddr;
1581 	__u8		pad;
1582 	__u8		protocol;
1583 	__be16		len;
1584 };
1585 
1586 struct tcp6_pseudohdr {
1587 	struct in6_addr	saddr;
1588 	struct in6_addr daddr;
1589 	__be32		len;
1590 	__be32		protocol;	/* including padding */
1591 };
1592 
1593 union tcp_md5sum_block {
1594 	struct tcp4_pseudohdr ip4;
1595 #if IS_ENABLED(CONFIG_IPV6)
1596 	struct tcp6_pseudohdr ip6;
1597 #endif
1598 };
1599 
1600 /* - pool: digest algorithm, hash description and scratch buffer */
1601 struct tcp_md5sig_pool {
1602 	struct ahash_request	*md5_req;
1603 	void			*scratch;
1604 };
1605 
1606 /* - functions */
1607 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1608 			const struct sock *sk, const struct sk_buff *skb);
1609 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1610 		   int family, u8 prefixlen, int l3index,
1611 		   const u8 *newkey, u8 newkeylen, gfp_t gfp);
1612 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1613 		   int family, u8 prefixlen, int l3index);
1614 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1615 					 const struct sock *addr_sk);
1616 
1617 #ifdef CONFIG_TCP_MD5SIG
1618 #include <linux/jump_label.h>
1619 extern struct static_key_false tcp_md5_needed;
1620 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1621 					   const union tcp_md5_addr *addr,
1622 					   int family);
1623 static inline struct tcp_md5sig_key *
1624 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1625 		  const union tcp_md5_addr *addr, int family)
1626 {
1627 	if (!static_branch_unlikely(&tcp_md5_needed))
1628 		return NULL;
1629 	return __tcp_md5_do_lookup(sk, l3index, addr, family);
1630 }
1631 
1632 #define tcp_twsk_md5_key(twsk)	((twsk)->tw_md5_key)
1633 #else
1634 static inline struct tcp_md5sig_key *
1635 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1636 		  const union tcp_md5_addr *addr, int family)
1637 {
1638 	return NULL;
1639 }
1640 #define tcp_twsk_md5_key(twsk)	NULL
1641 #endif
1642 
1643 bool tcp_alloc_md5sig_pool(void);
1644 
1645 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1646 static inline void tcp_put_md5sig_pool(void)
1647 {
1648 	local_bh_enable();
1649 }
1650 
1651 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1652 			  unsigned int header_len);
1653 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1654 		     const struct tcp_md5sig_key *key);
1655 
1656 /* From tcp_fastopen.c */
1657 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1658 			    struct tcp_fastopen_cookie *cookie);
1659 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1660 			    struct tcp_fastopen_cookie *cookie, bool syn_lost,
1661 			    u16 try_exp);
1662 struct tcp_fastopen_request {
1663 	/* Fast Open cookie. Size 0 means a cookie request */
1664 	struct tcp_fastopen_cookie	cookie;
1665 	struct msghdr			*data;  /* data in MSG_FASTOPEN */
1666 	size_t				size;
1667 	int				copied;	/* queued in tcp_connect() */
1668 	struct ubuf_info		*uarg;
1669 };
1670 void tcp_free_fastopen_req(struct tcp_sock *tp);
1671 void tcp_fastopen_destroy_cipher(struct sock *sk);
1672 void tcp_fastopen_ctx_destroy(struct net *net);
1673 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1674 			      void *primary_key, void *backup_key);
1675 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
1676 			    u64 *key);
1677 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1678 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1679 			      struct request_sock *req,
1680 			      struct tcp_fastopen_cookie *foc,
1681 			      const struct dst_entry *dst);
1682 void tcp_fastopen_init_key_once(struct net *net);
1683 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1684 			     struct tcp_fastopen_cookie *cookie);
1685 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1686 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1687 #define TCP_FASTOPEN_KEY_MAX 2
1688 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1689 	(TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1690 
1691 /* Fastopen key context */
1692 struct tcp_fastopen_context {
1693 	siphash_key_t	key[TCP_FASTOPEN_KEY_MAX];
1694 	int		num;
1695 	struct rcu_head	rcu;
1696 };
1697 
1698 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1699 void tcp_fastopen_active_disable(struct sock *sk);
1700 bool tcp_fastopen_active_should_disable(struct sock *sk);
1701 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1702 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1703 
1704 /* Caller needs to wrap with rcu_read_(un)lock() */
1705 static inline
1706 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1707 {
1708 	struct tcp_fastopen_context *ctx;
1709 
1710 	ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1711 	if (!ctx)
1712 		ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1713 	return ctx;
1714 }
1715 
1716 static inline
1717 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1718 			       const struct tcp_fastopen_cookie *orig)
1719 {
1720 	if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1721 	    orig->len == foc->len &&
1722 	    !memcmp(orig->val, foc->val, foc->len))
1723 		return true;
1724 	return false;
1725 }
1726 
1727 static inline
1728 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1729 {
1730 	return ctx->num;
1731 }
1732 
1733 /* Latencies incurred by various limits for a sender. They are
1734  * chronograph-like stats that are mutually exclusive.
1735  */
1736 enum tcp_chrono {
1737 	TCP_CHRONO_UNSPEC,
1738 	TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1739 	TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1740 	TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1741 	__TCP_CHRONO_MAX,
1742 };
1743 
1744 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1745 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1746 
1747 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1748  * the same memory storage than skb->destructor/_skb_refdst
1749  */
1750 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1751 {
1752 	skb->destructor = NULL;
1753 	skb->_skb_refdst = 0UL;
1754 }
1755 
1756 #define tcp_skb_tsorted_save(skb) {		\
1757 	unsigned long _save = skb->_skb_refdst;	\
1758 	skb->_skb_refdst = 0UL;
1759 
1760 #define tcp_skb_tsorted_restore(skb)		\
1761 	skb->_skb_refdst = _save;		\
1762 }
1763 
1764 void tcp_write_queue_purge(struct sock *sk);
1765 
1766 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1767 {
1768 	return skb_rb_first(&sk->tcp_rtx_queue);
1769 }
1770 
1771 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1772 {
1773 	return skb_rb_last(&sk->tcp_rtx_queue);
1774 }
1775 
1776 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1777 {
1778 	return skb_peek(&sk->sk_write_queue);
1779 }
1780 
1781 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1782 {
1783 	return skb_peek_tail(&sk->sk_write_queue);
1784 }
1785 
1786 #define tcp_for_write_queue_from_safe(skb, tmp, sk)			\
1787 	skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1788 
1789 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1790 {
1791 	return skb_peek(&sk->sk_write_queue);
1792 }
1793 
1794 static inline bool tcp_skb_is_last(const struct sock *sk,
1795 				   const struct sk_buff *skb)
1796 {
1797 	return skb_queue_is_last(&sk->sk_write_queue, skb);
1798 }
1799 
1800 /**
1801  * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
1802  * @sk: socket
1803  *
1804  * Since the write queue can have a temporary empty skb in it,
1805  * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
1806  */
1807 static inline bool tcp_write_queue_empty(const struct sock *sk)
1808 {
1809 	const struct tcp_sock *tp = tcp_sk(sk);
1810 
1811 	return tp->write_seq == tp->snd_nxt;
1812 }
1813 
1814 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1815 {
1816 	return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1817 }
1818 
1819 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1820 {
1821 	return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1822 }
1823 
1824 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1825 {
1826 	__skb_queue_tail(&sk->sk_write_queue, skb);
1827 
1828 	/* Queue it, remembering where we must start sending. */
1829 	if (sk->sk_write_queue.next == skb)
1830 		tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1831 }
1832 
1833 /* Insert new before skb on the write queue of sk.  */
1834 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1835 						  struct sk_buff *skb,
1836 						  struct sock *sk)
1837 {
1838 	__skb_queue_before(&sk->sk_write_queue, skb, new);
1839 }
1840 
1841 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1842 {
1843 	tcp_skb_tsorted_anchor_cleanup(skb);
1844 	__skb_unlink(skb, &sk->sk_write_queue);
1845 }
1846 
1847 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1848 
1849 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1850 {
1851 	tcp_skb_tsorted_anchor_cleanup(skb);
1852 	rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1853 }
1854 
1855 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1856 {
1857 	list_del(&skb->tcp_tsorted_anchor);
1858 	tcp_rtx_queue_unlink(skb, sk);
1859 	sk_wmem_free_skb(sk, skb);
1860 }
1861 
1862 static inline void tcp_push_pending_frames(struct sock *sk)
1863 {
1864 	if (tcp_send_head(sk)) {
1865 		struct tcp_sock *tp = tcp_sk(sk);
1866 
1867 		__tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1868 	}
1869 }
1870 
1871 /* Start sequence of the skb just after the highest skb with SACKed
1872  * bit, valid only if sacked_out > 0 or when the caller has ensured
1873  * validity by itself.
1874  */
1875 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1876 {
1877 	if (!tp->sacked_out)
1878 		return tp->snd_una;
1879 
1880 	if (tp->highest_sack == NULL)
1881 		return tp->snd_nxt;
1882 
1883 	return TCP_SKB_CB(tp->highest_sack)->seq;
1884 }
1885 
1886 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1887 {
1888 	tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1889 }
1890 
1891 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1892 {
1893 	return tcp_sk(sk)->highest_sack;
1894 }
1895 
1896 static inline void tcp_highest_sack_reset(struct sock *sk)
1897 {
1898 	tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1899 }
1900 
1901 /* Called when old skb is about to be deleted and replaced by new skb */
1902 static inline void tcp_highest_sack_replace(struct sock *sk,
1903 					    struct sk_buff *old,
1904 					    struct sk_buff *new)
1905 {
1906 	if (old == tcp_highest_sack(sk))
1907 		tcp_sk(sk)->highest_sack = new;
1908 }
1909 
1910 /* This helper checks if socket has IP_TRANSPARENT set */
1911 static inline bool inet_sk_transparent(const struct sock *sk)
1912 {
1913 	switch (sk->sk_state) {
1914 	case TCP_TIME_WAIT:
1915 		return inet_twsk(sk)->tw_transparent;
1916 	case TCP_NEW_SYN_RECV:
1917 		return inet_rsk(inet_reqsk(sk))->no_srccheck;
1918 	}
1919 	return inet_sk(sk)->transparent;
1920 }
1921 
1922 /* Determines whether this is a thin stream (which may suffer from
1923  * increased latency). Used to trigger latency-reducing mechanisms.
1924  */
1925 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1926 {
1927 	return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1928 }
1929 
1930 /* /proc */
1931 enum tcp_seq_states {
1932 	TCP_SEQ_STATE_LISTENING,
1933 	TCP_SEQ_STATE_ESTABLISHED,
1934 };
1935 
1936 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1937 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1938 void tcp_seq_stop(struct seq_file *seq, void *v);
1939 
1940 struct tcp_seq_afinfo {
1941 	sa_family_t			family;
1942 };
1943 
1944 struct tcp_iter_state {
1945 	struct seq_net_private	p;
1946 	enum tcp_seq_states	state;
1947 	struct sock		*syn_wait_sk;
1948 	struct tcp_seq_afinfo	*bpf_seq_afinfo;
1949 	int			bucket, offset, sbucket, num;
1950 	loff_t			last_pos;
1951 };
1952 
1953 extern struct request_sock_ops tcp_request_sock_ops;
1954 extern struct request_sock_ops tcp6_request_sock_ops;
1955 
1956 void tcp_v4_destroy_sock(struct sock *sk);
1957 
1958 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1959 				netdev_features_t features);
1960 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
1961 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
1962 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
1963 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff));
1964 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb));
1965 int tcp_gro_complete(struct sk_buff *skb);
1966 
1967 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1968 
1969 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1970 {
1971 	struct net *net = sock_net((struct sock *)tp);
1972 	return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1973 }
1974 
1975 bool tcp_stream_memory_free(const struct sock *sk, int wake);
1976 
1977 #ifdef CONFIG_PROC_FS
1978 int tcp4_proc_init(void);
1979 void tcp4_proc_exit(void);
1980 #endif
1981 
1982 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1983 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1984 		     const struct tcp_request_sock_ops *af_ops,
1985 		     struct sock *sk, struct sk_buff *skb);
1986 
1987 /* TCP af-specific functions */
1988 struct tcp_sock_af_ops {
1989 #ifdef CONFIG_TCP_MD5SIG
1990 	struct tcp_md5sig_key	*(*md5_lookup) (const struct sock *sk,
1991 						const struct sock *addr_sk);
1992 	int		(*calc_md5_hash)(char *location,
1993 					 const struct tcp_md5sig_key *md5,
1994 					 const struct sock *sk,
1995 					 const struct sk_buff *skb);
1996 	int		(*md5_parse)(struct sock *sk,
1997 				     int optname,
1998 				     sockptr_t optval,
1999 				     int optlen);
2000 #endif
2001 };
2002 
2003 struct tcp_request_sock_ops {
2004 	u16 mss_clamp;
2005 #ifdef CONFIG_TCP_MD5SIG
2006 	struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
2007 						 const struct sock *addr_sk);
2008 	int		(*calc_md5_hash) (char *location,
2009 					  const struct tcp_md5sig_key *md5,
2010 					  const struct sock *sk,
2011 					  const struct sk_buff *skb);
2012 #endif
2013 #ifdef CONFIG_SYN_COOKIES
2014 	__u32 (*cookie_init_seq)(const struct sk_buff *skb,
2015 				 __u16 *mss);
2016 #endif
2017 	struct dst_entry *(*route_req)(const struct sock *sk,
2018 				       struct sk_buff *skb,
2019 				       struct flowi *fl,
2020 				       struct request_sock *req);
2021 	u32 (*init_seq)(const struct sk_buff *skb);
2022 	u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
2023 	int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2024 			   struct flowi *fl, struct request_sock *req,
2025 			   struct tcp_fastopen_cookie *foc,
2026 			   enum tcp_synack_type synack_type,
2027 			   struct sk_buff *syn_skb);
2028 };
2029 
2030 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2031 #if IS_ENABLED(CONFIG_IPV6)
2032 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2033 #endif
2034 
2035 #ifdef CONFIG_SYN_COOKIES
2036 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2037 					 const struct sock *sk, struct sk_buff *skb,
2038 					 __u16 *mss)
2039 {
2040 	tcp_synq_overflow(sk);
2041 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2042 	return ops->cookie_init_seq(skb, mss);
2043 }
2044 #else
2045 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2046 					 const struct sock *sk, struct sk_buff *skb,
2047 					 __u16 *mss)
2048 {
2049 	return 0;
2050 }
2051 #endif
2052 
2053 int tcpv4_offload_init(void);
2054 
2055 void tcp_v4_init(void);
2056 void tcp_init(void);
2057 
2058 /* tcp_recovery.c */
2059 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2060 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2061 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2062 				u32 reo_wnd);
2063 extern void tcp_rack_mark_lost(struct sock *sk);
2064 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2065 			     u64 xmit_time);
2066 extern void tcp_rack_reo_timeout(struct sock *sk);
2067 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2068 
2069 /* At how many usecs into the future should the RTO fire? */
2070 static inline s64 tcp_rto_delta_us(const struct sock *sk)
2071 {
2072 	const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2073 	u32 rto = inet_csk(sk)->icsk_rto;
2074 	u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2075 
2076 	return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2077 }
2078 
2079 /*
2080  * Save and compile IPv4 options, return a pointer to it
2081  */
2082 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2083 							 struct sk_buff *skb)
2084 {
2085 	const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2086 	struct ip_options_rcu *dopt = NULL;
2087 
2088 	if (opt->optlen) {
2089 		int opt_size = sizeof(*dopt) + opt->optlen;
2090 
2091 		dopt = kmalloc(opt_size, GFP_ATOMIC);
2092 		if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2093 			kfree(dopt);
2094 			dopt = NULL;
2095 		}
2096 	}
2097 	return dopt;
2098 }
2099 
2100 /* locally generated TCP pure ACKs have skb->truesize == 2
2101  * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2102  * This is much faster than dissecting the packet to find out.
2103  * (Think of GRE encapsulations, IPv4, IPv6, ...)
2104  */
2105 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2106 {
2107 	return skb->truesize == 2;
2108 }
2109 
2110 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2111 {
2112 	skb->truesize = 2;
2113 }
2114 
2115 static inline int tcp_inq(struct sock *sk)
2116 {
2117 	struct tcp_sock *tp = tcp_sk(sk);
2118 	int answ;
2119 
2120 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2121 		answ = 0;
2122 	} else if (sock_flag(sk, SOCK_URGINLINE) ||
2123 		   !tp->urg_data ||
2124 		   before(tp->urg_seq, tp->copied_seq) ||
2125 		   !before(tp->urg_seq, tp->rcv_nxt)) {
2126 
2127 		answ = tp->rcv_nxt - tp->copied_seq;
2128 
2129 		/* Subtract 1, if FIN was received */
2130 		if (answ && sock_flag(sk, SOCK_DONE))
2131 			answ--;
2132 	} else {
2133 		answ = tp->urg_seq - tp->copied_seq;
2134 	}
2135 
2136 	return answ;
2137 }
2138 
2139 int tcp_peek_len(struct socket *sock);
2140 
2141 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2142 {
2143 	u16 segs_in;
2144 
2145 	segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2146 	tp->segs_in += segs_in;
2147 	if (skb->len > tcp_hdrlen(skb))
2148 		tp->data_segs_in += segs_in;
2149 }
2150 
2151 /*
2152  * TCP listen path runs lockless.
2153  * We forced "struct sock" to be const qualified to make sure
2154  * we don't modify one of its field by mistake.
2155  * Here, we increment sk_drops which is an atomic_t, so we can safely
2156  * make sock writable again.
2157  */
2158 static inline void tcp_listendrop(const struct sock *sk)
2159 {
2160 	atomic_inc(&((struct sock *)sk)->sk_drops);
2161 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2162 }
2163 
2164 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2165 
2166 /*
2167  * Interface for adding Upper Level Protocols over TCP
2168  */
2169 
2170 #define TCP_ULP_NAME_MAX	16
2171 #define TCP_ULP_MAX		128
2172 #define TCP_ULP_BUF_MAX		(TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2173 
2174 struct tcp_ulp_ops {
2175 	struct list_head	list;
2176 
2177 	/* initialize ulp */
2178 	int (*init)(struct sock *sk);
2179 	/* update ulp */
2180 	void (*update)(struct sock *sk, struct proto *p,
2181 		       void (*write_space)(struct sock *sk));
2182 	/* cleanup ulp */
2183 	void (*release)(struct sock *sk);
2184 	/* diagnostic */
2185 	int (*get_info)(const struct sock *sk, struct sk_buff *skb);
2186 	size_t (*get_info_size)(const struct sock *sk);
2187 	/* clone ulp */
2188 	void (*clone)(const struct request_sock *req, struct sock *newsk,
2189 		      const gfp_t priority);
2190 
2191 	char		name[TCP_ULP_NAME_MAX];
2192 	struct module	*owner;
2193 };
2194 int tcp_register_ulp(struct tcp_ulp_ops *type);
2195 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2196 int tcp_set_ulp(struct sock *sk, const char *name);
2197 void tcp_get_available_ulp(char *buf, size_t len);
2198 void tcp_cleanup_ulp(struct sock *sk);
2199 void tcp_update_ulp(struct sock *sk, struct proto *p,
2200 		    void (*write_space)(struct sock *sk));
2201 
2202 #define MODULE_ALIAS_TCP_ULP(name)				\
2203 	__MODULE_INFO(alias, alias_userspace, name);		\
2204 	__MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2205 
2206 struct sk_msg;
2207 struct sk_psock;
2208 
2209 #ifdef CONFIG_BPF_STREAM_PARSER
2210 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
2211 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2212 #else
2213 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2214 {
2215 }
2216 #endif /* CONFIG_BPF_STREAM_PARSER */
2217 
2218 #ifdef CONFIG_NET_SOCK_MSG
2219 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes,
2220 			  int flags);
2221 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock,
2222 		      struct msghdr *msg, int len, int flags);
2223 #endif /* CONFIG_NET_SOCK_MSG */
2224 
2225 #ifdef CONFIG_CGROUP_BPF
2226 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2227 				      struct sk_buff *skb,
2228 				      unsigned int end_offset)
2229 {
2230 	skops->skb = skb;
2231 	skops->skb_data_end = skb->data + end_offset;
2232 }
2233 #else
2234 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2235 				      struct sk_buff *skb,
2236 				      unsigned int end_offset)
2237 {
2238 }
2239 #endif
2240 
2241 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2242  * is < 0, then the BPF op failed (for example if the loaded BPF
2243  * program does not support the chosen operation or there is no BPF
2244  * program loaded).
2245  */
2246 #ifdef CONFIG_BPF
2247 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2248 {
2249 	struct bpf_sock_ops_kern sock_ops;
2250 	int ret;
2251 
2252 	memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2253 	if (sk_fullsock(sk)) {
2254 		sock_ops.is_fullsock = 1;
2255 		sock_owned_by_me(sk);
2256 	}
2257 
2258 	sock_ops.sk = sk;
2259 	sock_ops.op = op;
2260 	if (nargs > 0)
2261 		memcpy(sock_ops.args, args, nargs * sizeof(*args));
2262 
2263 	ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2264 	if (ret == 0)
2265 		ret = sock_ops.reply;
2266 	else
2267 		ret = -1;
2268 	return ret;
2269 }
2270 
2271 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2272 {
2273 	u32 args[2] = {arg1, arg2};
2274 
2275 	return tcp_call_bpf(sk, op, 2, args);
2276 }
2277 
2278 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2279 				    u32 arg3)
2280 {
2281 	u32 args[3] = {arg1, arg2, arg3};
2282 
2283 	return tcp_call_bpf(sk, op, 3, args);
2284 }
2285 
2286 #else
2287 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2288 {
2289 	return -EPERM;
2290 }
2291 
2292 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2293 {
2294 	return -EPERM;
2295 }
2296 
2297 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2298 				    u32 arg3)
2299 {
2300 	return -EPERM;
2301 }
2302 
2303 #endif
2304 
2305 static inline u32 tcp_timeout_init(struct sock *sk)
2306 {
2307 	int timeout;
2308 
2309 	timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2310 
2311 	if (timeout <= 0)
2312 		timeout = TCP_TIMEOUT_INIT;
2313 	return timeout;
2314 }
2315 
2316 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2317 {
2318 	int rwnd;
2319 
2320 	rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2321 
2322 	if (rwnd < 0)
2323 		rwnd = 0;
2324 	return rwnd;
2325 }
2326 
2327 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2328 {
2329 	return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2330 }
2331 
2332 static inline void tcp_bpf_rtt(struct sock *sk)
2333 {
2334 	if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2335 		tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2336 }
2337 
2338 #if IS_ENABLED(CONFIG_SMC)
2339 extern struct static_key_false tcp_have_smc;
2340 #endif
2341 
2342 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2343 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2344 			     void (*cad)(struct sock *sk, u32 ack_seq));
2345 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2346 void clean_acked_data_flush(void);
2347 #endif
2348 
2349 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2350 static inline void tcp_add_tx_delay(struct sk_buff *skb,
2351 				    const struct tcp_sock *tp)
2352 {
2353 	if (static_branch_unlikely(&tcp_tx_delay_enabled))
2354 		skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2355 }
2356 
2357 /* Compute Earliest Departure Time for some control packets
2358  * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2359  */
2360 static inline u64 tcp_transmit_time(const struct sock *sk)
2361 {
2362 	if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2363 		u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2364 			tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2365 
2366 		return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
2367 	}
2368 	return 0;
2369 }
2370 
2371 #endif	/* _TCP_H */
2372