xref: /linux/include/net/tcp.h (revision 722ecdbce68a87de2d9296f91308f44ea900a039)
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 DECLARE_PER_CPU(unsigned int, tcp_orphan_count);
52 int tcp_orphan_count_sum(void);
53 
54 void tcp_time_wait(struct sock *sk, int state, int timeo);
55 
56 #define MAX_TCP_HEADER	L1_CACHE_ALIGN(128 + MAX_HEADER)
57 #define MAX_TCP_OPTION_SPACE 40
58 #define TCP_MIN_SND_MSS		48
59 #define TCP_MIN_GSO_SIZE	(TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
60 
61 /*
62  * Never offer a window over 32767 without using window scaling. Some
63  * poor stacks do signed 16bit maths!
64  */
65 #define MAX_TCP_WINDOW		32767U
66 
67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68 #define TCP_MIN_MSS		88U
69 
70 /* The initial MTU to use for probing */
71 #define TCP_BASE_MSS		1024
72 
73 /* probing interval, default to 10 minutes as per RFC4821 */
74 #define TCP_PROBE_INTERVAL	600
75 
76 /* Specify interval when tcp mtu probing will stop */
77 #define TCP_PROBE_THRESHOLD	8
78 
79 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
80 #define TCP_FASTRETRANS_THRESH 3
81 
82 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
83 #define TCP_MAX_QUICKACKS	16U
84 
85 /* Maximal number of window scale according to RFC1323 */
86 #define TCP_MAX_WSCALE		14U
87 
88 /* urg_data states */
89 #define TCP_URG_VALID	0x0100
90 #define TCP_URG_NOTYET	0x0200
91 #define TCP_URG_READ	0x0400
92 
93 #define TCP_RETR1	3	/*
94 				 * This is how many retries it does before it
95 				 * tries to figure out if the gateway is
96 				 * down. Minimal RFC value is 3; it corresponds
97 				 * to ~3sec-8min depending on RTO.
98 				 */
99 
100 #define TCP_RETR2	15	/*
101 				 * This should take at least
102 				 * 90 minutes to time out.
103 				 * RFC1122 says that the limit is 100 sec.
104 				 * 15 is ~13-30min depending on RTO.
105 				 */
106 
107 #define TCP_SYN_RETRIES	 6	/* This is how many retries are done
108 				 * when active opening a connection.
109 				 * RFC1122 says the minimum retry MUST
110 				 * be at least 180secs.  Nevertheless
111 				 * this value is corresponding to
112 				 * 63secs of retransmission with the
113 				 * current initial RTO.
114 				 */
115 
116 #define TCP_SYNACK_RETRIES 5	/* This is how may retries are done
117 				 * when passive opening a connection.
118 				 * This is corresponding to 31secs of
119 				 * retransmission with the current
120 				 * initial RTO.
121 				 */
122 
123 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
124 				  * state, about 60 seconds	*/
125 #define TCP_FIN_TIMEOUT	TCP_TIMEWAIT_LEN
126                                  /* BSD style FIN_WAIT2 deadlock breaker.
127 				  * It used to be 3min, new value is 60sec,
128 				  * to combine FIN-WAIT-2 timeout with
129 				  * TIME-WAIT timer.
130 				  */
131 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */
132 
133 #define TCP_DELACK_MAX	((unsigned)(HZ/5))	/* maximal time to delay before sending an ACK */
134 #if HZ >= 100
135 #define TCP_DELACK_MIN	((unsigned)(HZ/25))	/* minimal time to delay before sending an ACK */
136 #define TCP_ATO_MIN	((unsigned)(HZ/25))
137 #else
138 #define TCP_DELACK_MIN	4U
139 #define TCP_ATO_MIN	4U
140 #endif
141 #define TCP_RTO_MAX	((unsigned)(120*HZ))
142 #define TCP_RTO_MIN	((unsigned)(HZ/5))
143 #define TCP_TIMEOUT_MIN	(2U) /* Min timeout for TCP timers in jiffies */
144 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ))	/* RFC6298 2.1 initial RTO value	*/
145 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ))	/* RFC 1122 initial RTO value, now
146 						 * used as a fallback RTO for the
147 						 * initial data transmission if no
148 						 * valid RTT sample has been acquired,
149 						 * most likely due to retrans in 3WHS.
150 						 */
151 
152 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
153 					                 * for local resources.
154 					                 */
155 #define TCP_KEEPALIVE_TIME	(120*60*HZ)	/* two hours */
156 #define TCP_KEEPALIVE_PROBES	9		/* Max of 9 keepalive probes	*/
157 #define TCP_KEEPALIVE_INTVL	(75*HZ)
158 
159 #define MAX_TCP_KEEPIDLE	32767
160 #define MAX_TCP_KEEPINTVL	32767
161 #define MAX_TCP_KEEPCNT		127
162 #define MAX_TCP_SYNCNT		127
163 
164 #define TCP_SYNQ_INTERVAL	(HZ/5)	/* Period of SYNACK timer */
165 
166 #define TCP_PAWS_24DAYS	(60 * 60 * 24 * 24)
167 #define TCP_PAWS_MSL	60		/* Per-host timestamps are invalidated
168 					 * after this time. It should be equal
169 					 * (or greater than) TCP_TIMEWAIT_LEN
170 					 * to provide reliability equal to one
171 					 * provided by timewait state.
172 					 */
173 #define TCP_PAWS_WINDOW	1		/* Replay window for per-host
174 					 * timestamps. It must be less than
175 					 * minimal timewait lifetime.
176 					 */
177 /*
178  *	TCP option
179  */
180 
181 #define TCPOPT_NOP		1	/* Padding */
182 #define TCPOPT_EOL		0	/* End of options */
183 #define TCPOPT_MSS		2	/* Segment size negotiating */
184 #define TCPOPT_WINDOW		3	/* Window scaling */
185 #define TCPOPT_SACK_PERM        4       /* SACK Permitted */
186 #define TCPOPT_SACK             5       /* SACK Block */
187 #define TCPOPT_TIMESTAMP	8	/* Better RTT estimations/PAWS */
188 #define TCPOPT_MD5SIG		19	/* MD5 Signature (RFC2385) */
189 #define TCPOPT_MPTCP		30	/* Multipath TCP (RFC6824) */
190 #define TCPOPT_FASTOPEN		34	/* Fast open (RFC7413) */
191 #define TCPOPT_EXP		254	/* Experimental */
192 /* Magic number to be after the option value for sharing TCP
193  * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
194  */
195 #define TCPOPT_FASTOPEN_MAGIC	0xF989
196 #define TCPOPT_SMC_MAGIC	0xE2D4C3D9
197 
198 /*
199  *     TCP option lengths
200  */
201 
202 #define TCPOLEN_MSS            4
203 #define TCPOLEN_WINDOW         3
204 #define TCPOLEN_SACK_PERM      2
205 #define TCPOLEN_TIMESTAMP      10
206 #define TCPOLEN_MD5SIG         18
207 #define TCPOLEN_FASTOPEN_BASE  2
208 #define TCPOLEN_EXP_FASTOPEN_BASE  4
209 #define TCPOLEN_EXP_SMC_BASE   6
210 
211 /* But this is what stacks really send out. */
212 #define TCPOLEN_TSTAMP_ALIGNED		12
213 #define TCPOLEN_WSCALE_ALIGNED		4
214 #define TCPOLEN_SACKPERM_ALIGNED	4
215 #define TCPOLEN_SACK_BASE		2
216 #define TCPOLEN_SACK_BASE_ALIGNED	4
217 #define TCPOLEN_SACK_PERBLOCK		8
218 #define TCPOLEN_MD5SIG_ALIGNED		20
219 #define TCPOLEN_MSS_ALIGNED		4
220 #define TCPOLEN_EXP_SMC_BASE_ALIGNED	8
221 
222 /* Flags in tp->nonagle */
223 #define TCP_NAGLE_OFF		1	/* Nagle's algo is disabled */
224 #define TCP_NAGLE_CORK		2	/* Socket is corked	    */
225 #define TCP_NAGLE_PUSH		4	/* Cork is overridden for already queued data */
226 
227 /* TCP thin-stream limits */
228 #define TCP_THIN_LINEAR_RETRIES 6       /* After 6 linear retries, do exp. backoff */
229 
230 /* TCP initial congestion window as per rfc6928 */
231 #define TCP_INIT_CWND		10
232 
233 /* Bit Flags for sysctl_tcp_fastopen */
234 #define	TFO_CLIENT_ENABLE	1
235 #define	TFO_SERVER_ENABLE	2
236 #define	TFO_CLIENT_NO_COOKIE	4	/* Data in SYN w/o cookie option */
237 
238 /* Accept SYN data w/o any cookie option */
239 #define	TFO_SERVER_COOKIE_NOT_REQD	0x200
240 
241 /* Force enable TFO on all listeners, i.e., not requiring the
242  * TCP_FASTOPEN socket option.
243  */
244 #define	TFO_SERVER_WO_SOCKOPT1	0x400
245 
246 
247 /* sysctl variables for tcp */
248 extern int sysctl_tcp_max_orphans;
249 extern long sysctl_tcp_mem[3];
250 
251 #define TCP_RACK_LOSS_DETECTION  0x1 /* Use RACK to detect losses */
252 #define TCP_RACK_STATIC_REO_WND  0x2 /* Use static RACK reo wnd */
253 #define TCP_RACK_NO_DUPTHRESH    0x4 /* Do not use DUPACK threshold in RACK */
254 
255 extern atomic_long_t tcp_memory_allocated;
256 extern struct percpu_counter tcp_sockets_allocated;
257 extern unsigned long tcp_memory_pressure;
258 
259 /* optimized version of sk_under_memory_pressure() for TCP sockets */
260 static inline bool tcp_under_memory_pressure(const struct sock *sk)
261 {
262 	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
263 	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
264 		return true;
265 
266 	return READ_ONCE(tcp_memory_pressure);
267 }
268 /*
269  * The next routines deal with comparing 32 bit unsigned ints
270  * and worry about wraparound (automatic with unsigned arithmetic).
271  */
272 
273 static inline bool before(__u32 seq1, __u32 seq2)
274 {
275         return (__s32)(seq1-seq2) < 0;
276 }
277 #define after(seq2, seq1) 	before(seq1, seq2)
278 
279 /* is s2<=s1<=s3 ? */
280 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
281 {
282 	return seq3 - seq2 >= seq1 - seq2;
283 }
284 
285 static inline bool tcp_out_of_memory(struct sock *sk)
286 {
287 	if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
288 	    sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
289 		return true;
290 	return false;
291 }
292 
293 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
294 {
295 	sk_wmem_queued_add(sk, -skb->truesize);
296 	if (!skb_zcopy_pure(skb))
297 		sk_mem_uncharge(sk, skb->truesize);
298 	else
299 		sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb)));
300 	__kfree_skb(skb);
301 }
302 
303 void sk_forced_mem_schedule(struct sock *sk, int size);
304 
305 bool tcp_check_oom(struct sock *sk, int shift);
306 
307 
308 extern struct proto tcp_prot;
309 
310 #define TCP_INC_STATS(net, field)	SNMP_INC_STATS((net)->mib.tcp_statistics, field)
311 #define __TCP_INC_STATS(net, field)	__SNMP_INC_STATS((net)->mib.tcp_statistics, field)
312 #define TCP_DEC_STATS(net, field)	SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
313 #define TCP_ADD_STATS(net, field, val)	SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
314 
315 void tcp_tasklet_init(void);
316 
317 int tcp_v4_err(struct sk_buff *skb, u32);
318 
319 void tcp_shutdown(struct sock *sk, int how);
320 
321 int tcp_v4_early_demux(struct sk_buff *skb);
322 int tcp_v4_rcv(struct sk_buff *skb);
323 
324 void tcp_remove_empty_skb(struct sock *sk);
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 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
351 				     bool force_schedule);
352 
353 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
354 static inline void tcp_dec_quickack_mode(struct sock *sk,
355 					 const unsigned int pkts)
356 {
357 	struct inet_connection_sock *icsk = inet_csk(sk);
358 
359 	if (icsk->icsk_ack.quick) {
360 		if (pkts >= icsk->icsk_ack.quick) {
361 			icsk->icsk_ack.quick = 0;
362 			/* Leaving quickack mode we deflate ATO. */
363 			icsk->icsk_ack.ato   = TCP_ATO_MIN;
364 		} else
365 			icsk->icsk_ack.quick -= pkts;
366 	}
367 }
368 
369 #define	TCP_ECN_OK		1
370 #define	TCP_ECN_QUEUE_CWR	2
371 #define	TCP_ECN_DEMAND_CWR	4
372 #define	TCP_ECN_SEEN		8
373 
374 enum tcp_tw_status {
375 	TCP_TW_SUCCESS = 0,
376 	TCP_TW_RST = 1,
377 	TCP_TW_ACK = 2,
378 	TCP_TW_SYN = 3
379 };
380 
381 
382 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
383 					      struct sk_buff *skb,
384 					      const struct tcphdr *th);
385 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
386 			   struct request_sock *req, bool fastopen,
387 			   bool *lost_race);
388 int tcp_child_process(struct sock *parent, struct sock *child,
389 		      struct sk_buff *skb);
390 void tcp_enter_loss(struct sock *sk);
391 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag);
392 void tcp_clear_retrans(struct tcp_sock *tp);
393 void tcp_update_metrics(struct sock *sk);
394 void tcp_init_metrics(struct sock *sk);
395 void tcp_metrics_init(void);
396 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
397 void __tcp_close(struct sock *sk, long timeout);
398 void tcp_close(struct sock *sk, long timeout);
399 void tcp_init_sock(struct sock *sk);
400 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb);
401 __poll_t tcp_poll(struct file *file, struct socket *sock,
402 		      struct poll_table_struct *wait);
403 int tcp_getsockopt(struct sock *sk, int level, int optname,
404 		   char __user *optval, int __user *optlen);
405 bool tcp_bpf_bypass_getsockopt(int level, int optname);
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,
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_update_recv_tstamps(struct sk_buff *skb,
415 			     struct scm_timestamping_internal *tss);
416 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
417 			struct scm_timestamping_internal *tss);
418 void tcp_data_ready(struct sock *sk);
419 #ifdef CONFIG_MMU
420 int tcp_mmap(struct file *file, struct socket *sock,
421 	     struct vm_area_struct *vma);
422 #endif
423 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
424 		       struct tcp_options_received *opt_rx,
425 		       int estab, struct tcp_fastopen_cookie *foc);
426 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
427 
428 /*
429  *	BPF SKB-less helpers
430  */
431 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
432 			 struct tcphdr *th, u32 *cookie);
433 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
434 			 struct tcphdr *th, u32 *cookie);
435 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
436 			  const struct tcp_request_sock_ops *af_ops,
437 			  struct sock *sk, struct tcphdr *th);
438 /*
439  *	TCP v4 functions exported for the inet6 API
440  */
441 
442 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
443 void tcp_v4_mtu_reduced(struct sock *sk);
444 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
445 void tcp_ld_RTO_revert(struct sock *sk, u32 seq);
446 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
447 struct sock *tcp_create_openreq_child(const struct sock *sk,
448 				      struct request_sock *req,
449 				      struct sk_buff *skb);
450 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
451 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
452 				  struct request_sock *req,
453 				  struct dst_entry *dst,
454 				  struct request_sock *req_unhash,
455 				  bool *own_req);
456 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
457 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
458 int tcp_connect(struct sock *sk);
459 enum tcp_synack_type {
460 	TCP_SYNACK_NORMAL,
461 	TCP_SYNACK_FASTOPEN,
462 	TCP_SYNACK_COOKIE,
463 };
464 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
465 				struct request_sock *req,
466 				struct tcp_fastopen_cookie *foc,
467 				enum tcp_synack_type synack_type,
468 				struct sk_buff *syn_skb);
469 int tcp_disconnect(struct sock *sk, int flags);
470 
471 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
472 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
473 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
474 
475 /* From syncookies.c */
476 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
477 				 struct request_sock *req,
478 				 struct dst_entry *dst, u32 tsoff);
479 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
480 		      u32 cookie);
481 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
482 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
483 					    const struct tcp_request_sock_ops *af_ops,
484 					    struct sock *sk, struct sk_buff *skb);
485 #ifdef CONFIG_SYN_COOKIES
486 
487 /* Syncookies use a monotonic timer which increments every 60 seconds.
488  * This counter is used both as a hash input and partially encoded into
489  * the cookie value.  A cookie is only validated further if the delta
490  * between the current counter value and the encoded one is less than this,
491  * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
492  * the counter advances immediately after a cookie is generated).
493  */
494 #define MAX_SYNCOOKIE_AGE	2
495 #define TCP_SYNCOOKIE_PERIOD	(60 * HZ)
496 #define TCP_SYNCOOKIE_VALID	(MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
497 
498 /* syncookies: remember time of last synqueue overflow
499  * But do not dirty this field too often (once per second is enough)
500  * It is racy as we do not hold a lock, but race is very minor.
501  */
502 static inline void tcp_synq_overflow(const struct sock *sk)
503 {
504 	unsigned int last_overflow;
505 	unsigned int now = jiffies;
506 
507 	if (sk->sk_reuseport) {
508 		struct sock_reuseport *reuse;
509 
510 		reuse = rcu_dereference(sk->sk_reuseport_cb);
511 		if (likely(reuse)) {
512 			last_overflow = READ_ONCE(reuse->synq_overflow_ts);
513 			if (!time_between32(now, last_overflow,
514 					    last_overflow + HZ))
515 				WRITE_ONCE(reuse->synq_overflow_ts, now);
516 			return;
517 		}
518 	}
519 
520 	last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
521 	if (!time_between32(now, last_overflow, last_overflow + HZ))
522 		WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
523 }
524 
525 /* syncookies: no recent synqueue overflow on this listening socket? */
526 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
527 {
528 	unsigned int last_overflow;
529 	unsigned int now = jiffies;
530 
531 	if (sk->sk_reuseport) {
532 		struct sock_reuseport *reuse;
533 
534 		reuse = rcu_dereference(sk->sk_reuseport_cb);
535 		if (likely(reuse)) {
536 			last_overflow = READ_ONCE(reuse->synq_overflow_ts);
537 			return !time_between32(now, last_overflow - HZ,
538 					       last_overflow +
539 					       TCP_SYNCOOKIE_VALID);
540 		}
541 	}
542 
543 	last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
544 
545 	/* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
546 	 * then we're under synflood. However, we have to use
547 	 * 'last_overflow - HZ' as lower bound. That's because a concurrent
548 	 * tcp_synq_overflow() could update .ts_recent_stamp after we read
549 	 * jiffies but before we store .ts_recent_stamp into last_overflow,
550 	 * which could lead to rejecting a valid syncookie.
551 	 */
552 	return !time_between32(now, last_overflow - HZ,
553 			       last_overflow + TCP_SYNCOOKIE_VALID);
554 }
555 
556 static inline u32 tcp_cookie_time(void)
557 {
558 	u64 val = get_jiffies_64();
559 
560 	do_div(val, TCP_SYNCOOKIE_PERIOD);
561 	return val;
562 }
563 
564 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
565 			      u16 *mssp);
566 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
567 u64 cookie_init_timestamp(struct request_sock *req, u64 now);
568 bool cookie_timestamp_decode(const struct net *net,
569 			     struct tcp_options_received *opt);
570 bool cookie_ecn_ok(const struct tcp_options_received *opt,
571 		   const struct net *net, const struct dst_entry *dst);
572 
573 /* From net/ipv6/syncookies.c */
574 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
575 		      u32 cookie);
576 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
577 
578 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
579 			      const struct tcphdr *th, u16 *mssp);
580 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
581 #endif
582 /* tcp_output.c */
583 
584 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb);
585 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb);
586 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
587 			       int nonagle);
588 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
589 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
590 void tcp_retransmit_timer(struct sock *sk);
591 void tcp_xmit_retransmit_queue(struct sock *);
592 void tcp_simple_retransmit(struct sock *);
593 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
594 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
595 enum tcp_queue {
596 	TCP_FRAG_IN_WRITE_QUEUE,
597 	TCP_FRAG_IN_RTX_QUEUE,
598 };
599 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
600 		 struct sk_buff *skb, u32 len,
601 		 unsigned int mss_now, gfp_t gfp);
602 
603 void tcp_send_probe0(struct sock *);
604 void tcp_send_partial(struct sock *);
605 int tcp_write_wakeup(struct sock *, int mib);
606 void tcp_send_fin(struct sock *sk);
607 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
608 int tcp_send_synack(struct sock *);
609 void tcp_push_one(struct sock *, unsigned int mss_now);
610 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
611 void tcp_send_ack(struct sock *sk);
612 void tcp_send_delayed_ack(struct sock *sk);
613 void tcp_send_loss_probe(struct sock *sk);
614 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
615 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
616 			     const struct sk_buff *next_skb);
617 
618 /* tcp_input.c */
619 void tcp_rearm_rto(struct sock *sk);
620 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
621 void tcp_reset(struct sock *sk, struct sk_buff *skb);
622 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
623 void tcp_fin(struct sock *sk);
624 void tcp_check_space(struct sock *sk);
625 
626 /* tcp_timer.c */
627 void tcp_init_xmit_timers(struct sock *);
628 static inline void tcp_clear_xmit_timers(struct sock *sk)
629 {
630 	if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
631 		__sock_put(sk);
632 
633 	if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
634 		__sock_put(sk);
635 
636 	inet_csk_clear_xmit_timers(sk);
637 }
638 
639 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
640 unsigned int tcp_current_mss(struct sock *sk);
641 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when);
642 
643 /* Bound MSS / TSO packet size with the half of the window */
644 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
645 {
646 	int cutoff;
647 
648 	/* When peer uses tiny windows, there is no use in packetizing
649 	 * to sub-MSS pieces for the sake of SWS or making sure there
650 	 * are enough packets in the pipe for fast recovery.
651 	 *
652 	 * On the other hand, for extremely large MSS devices, handling
653 	 * smaller than MSS windows in this way does make sense.
654 	 */
655 	if (tp->max_window > TCP_MSS_DEFAULT)
656 		cutoff = (tp->max_window >> 1);
657 	else
658 		cutoff = tp->max_window;
659 
660 	if (cutoff && pktsize > cutoff)
661 		return max_t(int, cutoff, 68U - tp->tcp_header_len);
662 	else
663 		return pktsize;
664 }
665 
666 /* tcp.c */
667 void tcp_get_info(struct sock *, struct tcp_info *);
668 
669 /* Read 'sendfile()'-style from a TCP socket */
670 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
671 		  sk_read_actor_t recv_actor);
672 
673 void tcp_initialize_rcv_mss(struct sock *sk);
674 
675 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
676 int tcp_mss_to_mtu(struct sock *sk, int mss);
677 void tcp_mtup_init(struct sock *sk);
678 
679 static inline void tcp_bound_rto(const struct sock *sk)
680 {
681 	if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
682 		inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
683 }
684 
685 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
686 {
687 	return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
688 }
689 
690 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
691 {
692 	/* mptcp hooks are only on the slow path */
693 	if (sk_is_mptcp((struct sock *)tp))
694 		return;
695 
696 	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
697 			       ntohl(TCP_FLAG_ACK) |
698 			       snd_wnd);
699 }
700 
701 static inline void tcp_fast_path_on(struct tcp_sock *tp)
702 {
703 	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
704 }
705 
706 static inline void tcp_fast_path_check(struct sock *sk)
707 {
708 	struct tcp_sock *tp = tcp_sk(sk);
709 
710 	if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
711 	    tp->rcv_wnd &&
712 	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
713 	    !tp->urg_data)
714 		tcp_fast_path_on(tp);
715 }
716 
717 /* Compute the actual rto_min value */
718 static inline u32 tcp_rto_min(struct sock *sk)
719 {
720 	const struct dst_entry *dst = __sk_dst_get(sk);
721 	u32 rto_min = inet_csk(sk)->icsk_rto_min;
722 
723 	if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
724 		rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
725 	return rto_min;
726 }
727 
728 static inline u32 tcp_rto_min_us(struct sock *sk)
729 {
730 	return jiffies_to_usecs(tcp_rto_min(sk));
731 }
732 
733 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
734 {
735 	return dst_metric_locked(dst, RTAX_CC_ALGO);
736 }
737 
738 /* Minimum RTT in usec. ~0 means not available. */
739 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
740 {
741 	return minmax_get(&tp->rtt_min);
742 }
743 
744 /* Compute the actual receive window we are currently advertising.
745  * Rcv_nxt can be after the window if our peer push more data
746  * than the offered window.
747  */
748 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
749 {
750 	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
751 
752 	if (win < 0)
753 		win = 0;
754 	return (u32) win;
755 }
756 
757 /* Choose a new window, without checks for shrinking, and without
758  * scaling applied to the result.  The caller does these things
759  * if necessary.  This is a "raw" window selection.
760  */
761 u32 __tcp_select_window(struct sock *sk);
762 
763 void tcp_send_window_probe(struct sock *sk);
764 
765 /* TCP uses 32bit jiffies to save some space.
766  * Note that this is different from tcp_time_stamp, which
767  * historically has been the same until linux-4.13.
768  */
769 #define tcp_jiffies32 ((u32)jiffies)
770 
771 /*
772  * Deliver a 32bit value for TCP timestamp option (RFC 7323)
773  * It is no longer tied to jiffies, but to 1 ms clock.
774  * Note: double check if you want to use tcp_jiffies32 instead of this.
775  */
776 #define TCP_TS_HZ	1000
777 
778 static inline u64 tcp_clock_ns(void)
779 {
780 	return ktime_get_ns();
781 }
782 
783 static inline u64 tcp_clock_us(void)
784 {
785 	return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
786 }
787 
788 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
789 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
790 {
791 	return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
792 }
793 
794 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
795 static inline u32 tcp_ns_to_ts(u64 ns)
796 {
797 	return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
798 }
799 
800 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
801 static inline u32 tcp_time_stamp_raw(void)
802 {
803 	return tcp_ns_to_ts(tcp_clock_ns());
804 }
805 
806 void tcp_mstamp_refresh(struct tcp_sock *tp);
807 
808 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
809 {
810 	return max_t(s64, t1 - t0, 0);
811 }
812 
813 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
814 {
815 	return tcp_ns_to_ts(skb->skb_mstamp_ns);
816 }
817 
818 /* provide the departure time in us unit */
819 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
820 {
821 	return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
822 }
823 
824 
825 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
826 
827 #define TCPHDR_FIN 0x01
828 #define TCPHDR_SYN 0x02
829 #define TCPHDR_RST 0x04
830 #define TCPHDR_PSH 0x08
831 #define TCPHDR_ACK 0x10
832 #define TCPHDR_URG 0x20
833 #define TCPHDR_ECE 0x40
834 #define TCPHDR_CWR 0x80
835 
836 #define TCPHDR_SYN_ECN	(TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
837 
838 /* This is what the send packet queuing engine uses to pass
839  * TCP per-packet control information to the transmission code.
840  * We also store the host-order sequence numbers in here too.
841  * This is 44 bytes if IPV6 is enabled.
842  * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
843  */
844 struct tcp_skb_cb {
845 	__u32		seq;		/* Starting sequence number	*/
846 	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
847 	union {
848 		/* Note : tcp_tw_isn is used in input path only
849 		 *	  (isn chosen by tcp_timewait_state_process())
850 		 *
851 		 * 	  tcp_gso_segs/size are used in write queue only,
852 		 *	  cf tcp_skb_pcount()/tcp_skb_mss()
853 		 */
854 		__u32		tcp_tw_isn;
855 		struct {
856 			u16	tcp_gso_segs;
857 			u16	tcp_gso_size;
858 		};
859 	};
860 	__u8		tcp_flags;	/* TCP header flags. (tcp[13])	*/
861 
862 	__u8		sacked;		/* State flags for SACK.	*/
863 #define TCPCB_SACKED_ACKED	0x01	/* SKB ACK'd by a SACK block	*/
864 #define TCPCB_SACKED_RETRANS	0x02	/* SKB retransmitted		*/
865 #define TCPCB_LOST		0x04	/* SKB is lost			*/
866 #define TCPCB_TAGBITS		0x07	/* All tag bits			*/
867 #define TCPCB_REPAIRED		0x10	/* SKB repaired (no skb_mstamp_ns)	*/
868 #define TCPCB_EVER_RETRANS	0x80	/* Ever retransmitted frame	*/
869 #define TCPCB_RETRANS		(TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
870 				TCPCB_REPAIRED)
871 
872 	__u8		ip_dsfield;	/* IPv4 tos or IPv6 dsfield	*/
873 	__u8		txstamp_ack:1,	/* Record TX timestamp for ack? */
874 			eor:1,		/* Is skb MSG_EOR marked? */
875 			has_rxtstamp:1,	/* SKB has a RX timestamp	*/
876 			unused:5;
877 	__u32		ack_seq;	/* Sequence number ACK'd	*/
878 	union {
879 		struct {
880 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1)
881 			/* There is space for up to 24 bytes */
882 			__u32 is_app_limited:1, /* cwnd not fully used? */
883 			      delivered_ce:20,
884 			      unused:11;
885 			/* pkts S/ACKed so far upon tx of skb, incl retrans: */
886 			__u32 delivered;
887 			/* start of send pipeline phase */
888 			u64 first_tx_mstamp;
889 			/* when we reached the "delivered" count */
890 			u64 delivered_mstamp;
891 		} tx;   /* only used for outgoing skbs */
892 		union {
893 			struct inet_skb_parm	h4;
894 #if IS_ENABLED(CONFIG_IPV6)
895 			struct inet6_skb_parm	h6;
896 #endif
897 		} header;	/* For incoming skbs */
898 	};
899 };
900 
901 #define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
902 
903 extern const struct inet_connection_sock_af_ops ipv4_specific;
904 
905 #if IS_ENABLED(CONFIG_IPV6)
906 /* This is the variant of inet6_iif() that must be used by TCP,
907  * as TCP moves IP6CB into a different location in skb->cb[]
908  */
909 static inline int tcp_v6_iif(const struct sk_buff *skb)
910 {
911 	return TCP_SKB_CB(skb)->header.h6.iif;
912 }
913 
914 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
915 {
916 	bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
917 
918 	return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
919 }
920 
921 /* TCP_SKB_CB reference means this can not be used from early demux */
922 static inline int tcp_v6_sdif(const struct sk_buff *skb)
923 {
924 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
925 	if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
926 		return TCP_SKB_CB(skb)->header.h6.iif;
927 #endif
928 	return 0;
929 }
930 
931 extern const struct inet_connection_sock_af_ops ipv6_specific;
932 
933 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
934 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
935 INDIRECT_CALLABLE_DECLARE(void tcp_v6_early_demux(struct sk_buff *skb));
936 
937 #endif
938 
939 /* TCP_SKB_CB reference means this can not be used from early demux */
940 static inline int tcp_v4_sdif(struct sk_buff *skb)
941 {
942 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
943 	if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
944 		return TCP_SKB_CB(skb)->header.h4.iif;
945 #endif
946 	return 0;
947 }
948 
949 /* Due to TSO, an SKB can be composed of multiple actual
950  * packets.  To keep these tracked properly, we use this.
951  */
952 static inline int tcp_skb_pcount(const struct sk_buff *skb)
953 {
954 	return TCP_SKB_CB(skb)->tcp_gso_segs;
955 }
956 
957 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
958 {
959 	TCP_SKB_CB(skb)->tcp_gso_segs = segs;
960 }
961 
962 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
963 {
964 	TCP_SKB_CB(skb)->tcp_gso_segs += segs;
965 }
966 
967 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
968 static inline int tcp_skb_mss(const struct sk_buff *skb)
969 {
970 	return TCP_SKB_CB(skb)->tcp_gso_size;
971 }
972 
973 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
974 {
975 	return likely(!TCP_SKB_CB(skb)->eor);
976 }
977 
978 static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
979 					const struct sk_buff *from)
980 {
981 	return likely(tcp_skb_can_collapse_to(to) &&
982 		      mptcp_skb_can_collapse(to, from) &&
983 		      skb_pure_zcopy_same(to, from));
984 }
985 
986 /* Events passed to congestion control interface */
987 enum tcp_ca_event {
988 	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
989 	CA_EVENT_CWND_RESTART,	/* congestion window restart */
990 	CA_EVENT_COMPLETE_CWR,	/* end of congestion recovery */
991 	CA_EVENT_LOSS,		/* loss timeout */
992 	CA_EVENT_ECN_NO_CE,	/* ECT set, but not CE marked */
993 	CA_EVENT_ECN_IS_CE,	/* received CE marked IP packet */
994 };
995 
996 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
997 enum tcp_ca_ack_event_flags {
998 	CA_ACK_SLOWPATH		= (1 << 0),	/* In slow path processing */
999 	CA_ACK_WIN_UPDATE	= (1 << 1),	/* ACK updated window */
1000 	CA_ACK_ECE		= (1 << 2),	/* ECE bit is set on ack */
1001 };
1002 
1003 /*
1004  * Interface for adding new TCP congestion control handlers
1005  */
1006 #define TCP_CA_NAME_MAX	16
1007 #define TCP_CA_MAX	128
1008 #define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
1009 
1010 #define TCP_CA_UNSPEC	0
1011 
1012 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1013 #define TCP_CONG_NON_RESTRICTED 0x1
1014 /* Requires ECN/ECT set on all packets */
1015 #define TCP_CONG_NEEDS_ECN	0x2
1016 #define TCP_CONG_MASK	(TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1017 
1018 union tcp_cc_info;
1019 
1020 struct ack_sample {
1021 	u32 pkts_acked;
1022 	s32 rtt_us;
1023 	u32 in_flight;
1024 };
1025 
1026 /* A rate sample measures the number of (original/retransmitted) data
1027  * packets delivered "delivered" over an interval of time "interval_us".
1028  * The tcp_rate.c code fills in the rate sample, and congestion
1029  * control modules that define a cong_control function to run at the end
1030  * of ACK processing can optionally chose to consult this sample when
1031  * setting cwnd and pacing rate.
1032  * A sample is invalid if "delivered" or "interval_us" is negative.
1033  */
1034 struct rate_sample {
1035 	u64  prior_mstamp; /* starting timestamp for interval */
1036 	u32  prior_delivered;	/* tp->delivered at "prior_mstamp" */
1037 	u32  prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */
1038 	s32  delivered;		/* number of packets delivered over interval */
1039 	s32  delivered_ce;	/* number of packets delivered w/ CE marks*/
1040 	long interval_us;	/* time for tp->delivered to incr "delivered" */
1041 	u32 snd_interval_us;	/* snd interval for delivered packets */
1042 	u32 rcv_interval_us;	/* rcv interval for delivered packets */
1043 	long rtt_us;		/* RTT of last (S)ACKed packet (or -1) */
1044 	int  losses;		/* number of packets marked lost upon ACK */
1045 	u32  acked_sacked;	/* number of packets newly (S)ACKed upon ACK */
1046 	u32  prior_in_flight;	/* in flight before this ACK */
1047 	u32  last_end_seq;	/* end_seq of most recently ACKed packet */
1048 	bool is_app_limited;	/* is sample from packet with bubble in pipe? */
1049 	bool is_retrans;	/* is sample from retransmission? */
1050 	bool is_ack_delayed;	/* is this (likely) a delayed ACK? */
1051 };
1052 
1053 struct tcp_congestion_ops {
1054 /* fast path fields are put first to fill one cache line */
1055 
1056 	/* return slow start threshold (required) */
1057 	u32 (*ssthresh)(struct sock *sk);
1058 
1059 	/* do new cwnd calculation (required) */
1060 	void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1061 
1062 	/* call before changing ca_state (optional) */
1063 	void (*set_state)(struct sock *sk, u8 new_state);
1064 
1065 	/* call when cwnd event occurs (optional) */
1066 	void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1067 
1068 	/* call when ack arrives (optional) */
1069 	void (*in_ack_event)(struct sock *sk, u32 flags);
1070 
1071 	/* hook for packet ack accounting (optional) */
1072 	void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1073 
1074 	/* override sysctl_tcp_min_tso_segs */
1075 	u32 (*min_tso_segs)(struct sock *sk);
1076 
1077 	/* call when packets are delivered to update cwnd and pacing rate,
1078 	 * after all the ca_state processing. (optional)
1079 	 */
1080 	void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1081 
1082 
1083 	/* new value of cwnd after loss (required) */
1084 	u32  (*undo_cwnd)(struct sock *sk);
1085 	/* returns the multiplier used in tcp_sndbuf_expand (optional) */
1086 	u32 (*sndbuf_expand)(struct sock *sk);
1087 
1088 /* control/slow paths put last */
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 	struct list_head	list;
1096 	u32			key;
1097 	u32			flags;
1098 
1099 	/* initialize private data (optional) */
1100 	void (*init)(struct sock *sk);
1101 	/* cleanup private data  (optional) */
1102 	void (*release)(struct sock *sk);
1103 } ____cacheline_aligned_in_smp;
1104 
1105 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1106 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1107 
1108 void tcp_assign_congestion_control(struct sock *sk);
1109 void tcp_init_congestion_control(struct sock *sk);
1110 void tcp_cleanup_congestion_control(struct sock *sk);
1111 int tcp_set_default_congestion_control(struct net *net, const char *name);
1112 void tcp_get_default_congestion_control(struct net *net, char *name);
1113 void tcp_get_available_congestion_control(char *buf, size_t len);
1114 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1115 int tcp_set_allowed_congestion_control(char *allowed);
1116 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1117 			       bool cap_net_admin);
1118 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1119 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1120 
1121 u32 tcp_reno_ssthresh(struct sock *sk);
1122 u32 tcp_reno_undo_cwnd(struct sock *sk);
1123 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1124 extern struct tcp_congestion_ops tcp_reno;
1125 
1126 struct tcp_congestion_ops *tcp_ca_find(const char *name);
1127 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1128 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1129 #ifdef CONFIG_INET
1130 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1131 #else
1132 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1133 {
1134 	return NULL;
1135 }
1136 #endif
1137 
1138 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1139 {
1140 	const struct inet_connection_sock *icsk = inet_csk(sk);
1141 
1142 	return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1143 }
1144 
1145 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1146 {
1147 	const struct inet_connection_sock *icsk = inet_csk(sk);
1148 
1149 	if (icsk->icsk_ca_ops->cwnd_event)
1150 		icsk->icsk_ca_ops->cwnd_event(sk, event);
1151 }
1152 
1153 /* From tcp_cong.c */
1154 void tcp_set_ca_state(struct sock *sk, const u8 ca_state);
1155 
1156 /* From tcp_rate.c */
1157 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1158 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1159 			    struct rate_sample *rs);
1160 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1161 		  bool is_sack_reneg, struct rate_sample *rs);
1162 void tcp_rate_check_app_limited(struct sock *sk);
1163 
1164 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2)
1165 {
1166 	return t1 > t2 || (t1 == t2 && after(seq1, seq2));
1167 }
1168 
1169 /* These functions determine how the current flow behaves in respect of SACK
1170  * handling. SACK is negotiated with the peer, and therefore it can vary
1171  * between different flows.
1172  *
1173  * tcp_is_sack - SACK enabled
1174  * tcp_is_reno - No SACK
1175  */
1176 static inline int tcp_is_sack(const struct tcp_sock *tp)
1177 {
1178 	return likely(tp->rx_opt.sack_ok);
1179 }
1180 
1181 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1182 {
1183 	return !tcp_is_sack(tp);
1184 }
1185 
1186 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1187 {
1188 	return tp->sacked_out + tp->lost_out;
1189 }
1190 
1191 /* This determines how many packets are "in the network" to the best
1192  * of our knowledge.  In many cases it is conservative, but where
1193  * detailed information is available from the receiver (via SACK
1194  * blocks etc.) we can make more aggressive calculations.
1195  *
1196  * Use this for decisions involving congestion control, use just
1197  * tp->packets_out to determine if the send queue is empty or not.
1198  *
1199  * Read this equation as:
1200  *
1201  *	"Packets sent once on transmission queue" MINUS
1202  *	"Packets left network, but not honestly ACKed yet" PLUS
1203  *	"Packets fast retransmitted"
1204  */
1205 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1206 {
1207 	return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1208 }
1209 
1210 #define TCP_INFINITE_SSTHRESH	0x7fffffff
1211 
1212 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp)
1213 {
1214 	return tp->snd_cwnd;
1215 }
1216 
1217 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val)
1218 {
1219 	WARN_ON_ONCE((int)val <= 0);
1220 	tp->snd_cwnd = val;
1221 }
1222 
1223 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1224 {
1225 	return tcp_snd_cwnd(tp) < tp->snd_ssthresh;
1226 }
1227 
1228 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1229 {
1230 	return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1231 }
1232 
1233 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1234 {
1235 	return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1236 	       (1 << inet_csk(sk)->icsk_ca_state);
1237 }
1238 
1239 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1240  * The exception is cwnd reduction phase, when cwnd is decreasing towards
1241  * ssthresh.
1242  */
1243 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1244 {
1245 	const struct tcp_sock *tp = tcp_sk(sk);
1246 
1247 	if (tcp_in_cwnd_reduction(sk))
1248 		return tp->snd_ssthresh;
1249 	else
1250 		return max(tp->snd_ssthresh,
1251 			   ((tcp_snd_cwnd(tp) >> 1) +
1252 			    (tcp_snd_cwnd(tp) >> 2)));
1253 }
1254 
1255 /* Use define here intentionally to get WARN_ON location shown at the caller */
1256 #define tcp_verify_left_out(tp)	WARN_ON(tcp_left_out(tp) > tp->packets_out)
1257 
1258 void tcp_enter_cwr(struct sock *sk);
1259 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1260 
1261 /* The maximum number of MSS of available cwnd for which TSO defers
1262  * sending if not using sysctl_tcp_tso_win_divisor.
1263  */
1264 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1265 {
1266 	return 3;
1267 }
1268 
1269 /* Returns end sequence number of the receiver's advertised window */
1270 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1271 {
1272 	return tp->snd_una + tp->snd_wnd;
1273 }
1274 
1275 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1276  * flexible approach. The RFC suggests cwnd should not be raised unless
1277  * it was fully used previously. And that's exactly what we do in
1278  * congestion avoidance mode. But in slow start we allow cwnd to grow
1279  * as long as the application has used half the cwnd.
1280  * Example :
1281  *    cwnd is 10 (IW10), but application sends 9 frames.
1282  *    We allow cwnd to reach 18 when all frames are ACKed.
1283  * This check is safe because it's as aggressive as slow start which already
1284  * risks 100% overshoot. The advantage is that we discourage application to
1285  * either send more filler packets or data to artificially blow up the cwnd
1286  * usage, and allow application-limited process to probe bw more aggressively.
1287  */
1288 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1289 {
1290 	const struct tcp_sock *tp = tcp_sk(sk);
1291 
1292 	/* If in slow start, ensure cwnd grows to twice what was ACKed. */
1293 	if (tcp_in_slow_start(tp))
1294 		return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out;
1295 
1296 	return tp->is_cwnd_limited;
1297 }
1298 
1299 /* BBR congestion control needs pacing.
1300  * Same remark for SO_MAX_PACING_RATE.
1301  * sch_fq packet scheduler is efficiently handling pacing,
1302  * but is not always installed/used.
1303  * Return true if TCP stack should pace packets itself.
1304  */
1305 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1306 {
1307 	return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1308 }
1309 
1310 /* Estimates in how many jiffies next packet for this flow can be sent.
1311  * Scheduling a retransmit timer too early would be silly.
1312  */
1313 static inline unsigned long tcp_pacing_delay(const struct sock *sk)
1314 {
1315 	s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
1316 
1317 	return delay > 0 ? nsecs_to_jiffies(delay) : 0;
1318 }
1319 
1320 static inline void tcp_reset_xmit_timer(struct sock *sk,
1321 					const int what,
1322 					unsigned long when,
1323 					const unsigned long max_when)
1324 {
1325 	inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk),
1326 				  max_when);
1327 }
1328 
1329 /* Something is really bad, we could not queue an additional packet,
1330  * because qdisc is full or receiver sent a 0 window, or we are paced.
1331  * We do not want to add fuel to the fire, or abort too early,
1332  * so make sure the timer we arm now is at least 200ms in the future,
1333  * regardless of current icsk_rto value (as it could be ~2ms)
1334  */
1335 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1336 {
1337 	return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1338 }
1339 
1340 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1341 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1342 					    unsigned long max_when)
1343 {
1344 	u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1,
1345 			   inet_csk(sk)->icsk_backoff);
1346 	u64 when = (u64)tcp_probe0_base(sk) << backoff;
1347 
1348 	return (unsigned long)min_t(u64, when, max_when);
1349 }
1350 
1351 static inline void tcp_check_probe_timer(struct sock *sk)
1352 {
1353 	if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1354 		tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1355 				     tcp_probe0_base(sk), TCP_RTO_MAX);
1356 }
1357 
1358 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1359 {
1360 	tp->snd_wl1 = seq;
1361 }
1362 
1363 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1364 {
1365 	tp->snd_wl1 = seq;
1366 }
1367 
1368 /*
1369  * Calculate(/check) TCP checksum
1370  */
1371 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1372 				   __be32 daddr, __wsum base)
1373 {
1374 	return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1375 }
1376 
1377 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1378 {
1379 	return !skb_csum_unnecessary(skb) &&
1380 		__skb_checksum_complete(skb);
1381 }
1382 
1383 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb,
1384 		     enum skb_drop_reason *reason);
1385 
1386 
1387 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1388 void tcp_set_state(struct sock *sk, int state);
1389 void tcp_done(struct sock *sk);
1390 int tcp_abort(struct sock *sk, int err);
1391 
1392 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1393 {
1394 	rx_opt->dsack = 0;
1395 	rx_opt->num_sacks = 0;
1396 }
1397 
1398 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1399 
1400 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1401 {
1402 	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1403 	struct tcp_sock *tp = tcp_sk(sk);
1404 	s32 delta;
1405 
1406 	if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1407 	    ca_ops->cong_control)
1408 		return;
1409 	delta = tcp_jiffies32 - tp->lsndtime;
1410 	if (delta > inet_csk(sk)->icsk_rto)
1411 		tcp_cwnd_restart(sk, delta);
1412 }
1413 
1414 /* Determine a window scaling and initial window to offer. */
1415 void tcp_select_initial_window(const struct sock *sk, int __space,
1416 			       __u32 mss, __u32 *rcv_wnd,
1417 			       __u32 *window_clamp, int wscale_ok,
1418 			       __u8 *rcv_wscale, __u32 init_rcv_wnd);
1419 
1420 static inline int tcp_win_from_space(const struct sock *sk, int space)
1421 {
1422 	int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1423 
1424 	return tcp_adv_win_scale <= 0 ?
1425 		(space>>(-tcp_adv_win_scale)) :
1426 		space - (space>>tcp_adv_win_scale);
1427 }
1428 
1429 /* Note: caller must be prepared to deal with negative returns */
1430 static inline int tcp_space(const struct sock *sk)
1431 {
1432 	return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1433 				  READ_ONCE(sk->sk_backlog.len) -
1434 				  atomic_read(&sk->sk_rmem_alloc));
1435 }
1436 
1437 static inline int tcp_full_space(const struct sock *sk)
1438 {
1439 	return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1440 }
1441 
1442 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk)
1443 {
1444 	int unused_mem = sk_unused_reserved_mem(sk);
1445 	struct tcp_sock *tp = tcp_sk(sk);
1446 
1447 	tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss);
1448 	if (unused_mem)
1449 		tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh,
1450 					 tcp_win_from_space(sk, unused_mem));
1451 }
1452 
1453 void tcp_cleanup_rbuf(struct sock *sk, int copied);
1454 
1455 /* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1456  * If 87.5 % (7/8) of the space has been consumed, we want to override
1457  * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1458  * len/truesize ratio.
1459  */
1460 static inline bool tcp_rmem_pressure(const struct sock *sk)
1461 {
1462 	int rcvbuf, threshold;
1463 
1464 	if (tcp_under_memory_pressure(sk))
1465 		return true;
1466 
1467 	rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1468 	threshold = rcvbuf - (rcvbuf >> 3);
1469 
1470 	return atomic_read(&sk->sk_rmem_alloc) > threshold;
1471 }
1472 
1473 static inline bool tcp_epollin_ready(const struct sock *sk, int target)
1474 {
1475 	const struct tcp_sock *tp = tcp_sk(sk);
1476 	int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq);
1477 
1478 	if (avail <= 0)
1479 		return false;
1480 
1481 	return (avail >= target) || tcp_rmem_pressure(sk) ||
1482 	       (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss);
1483 }
1484 
1485 extern void tcp_openreq_init_rwin(struct request_sock *req,
1486 				  const struct sock *sk_listener,
1487 				  const struct dst_entry *dst);
1488 
1489 void tcp_enter_memory_pressure(struct sock *sk);
1490 void tcp_leave_memory_pressure(struct sock *sk);
1491 
1492 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1493 {
1494 	struct net *net = sock_net((struct sock *)tp);
1495 
1496 	return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1497 }
1498 
1499 static inline int keepalive_time_when(const struct tcp_sock *tp)
1500 {
1501 	struct net *net = sock_net((struct sock *)tp);
1502 
1503 	return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1504 }
1505 
1506 static inline int keepalive_probes(const struct tcp_sock *tp)
1507 {
1508 	struct net *net = sock_net((struct sock *)tp);
1509 
1510 	return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1511 }
1512 
1513 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1514 {
1515 	const struct inet_connection_sock *icsk = &tp->inet_conn;
1516 
1517 	return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1518 			  tcp_jiffies32 - tp->rcv_tstamp);
1519 }
1520 
1521 static inline int tcp_fin_time(const struct sock *sk)
1522 {
1523 	int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1524 	const int rto = inet_csk(sk)->icsk_rto;
1525 
1526 	if (fin_timeout < (rto << 2) - (rto >> 1))
1527 		fin_timeout = (rto << 2) - (rto >> 1);
1528 
1529 	return fin_timeout;
1530 }
1531 
1532 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1533 				  int paws_win)
1534 {
1535 	if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1536 		return true;
1537 	if (unlikely(!time_before32(ktime_get_seconds(),
1538 				    rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1539 		return true;
1540 	/*
1541 	 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1542 	 * then following tcp messages have valid values. Ignore 0 value,
1543 	 * or else 'negative' tsval might forbid us to accept their packets.
1544 	 */
1545 	if (!rx_opt->ts_recent)
1546 		return true;
1547 	return false;
1548 }
1549 
1550 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1551 				   int rst)
1552 {
1553 	if (tcp_paws_check(rx_opt, 0))
1554 		return false;
1555 
1556 	/* RST segments are not recommended to carry timestamp,
1557 	   and, if they do, it is recommended to ignore PAWS because
1558 	   "their cleanup function should take precedence over timestamps."
1559 	   Certainly, it is mistake. It is necessary to understand the reasons
1560 	   of this constraint to relax it: if peer reboots, clock may go
1561 	   out-of-sync and half-open connections will not be reset.
1562 	   Actually, the problem would be not existing if all
1563 	   the implementations followed draft about maintaining clock
1564 	   via reboots. Linux-2.2 DOES NOT!
1565 
1566 	   However, we can relax time bounds for RST segments to MSL.
1567 	 */
1568 	if (rst && !time_before32(ktime_get_seconds(),
1569 				  rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1570 		return false;
1571 	return true;
1572 }
1573 
1574 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1575 			  int mib_idx, u32 *last_oow_ack_time);
1576 
1577 static inline void tcp_mib_init(struct net *net)
1578 {
1579 	/* See RFC 2012 */
1580 	TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1581 	TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1582 	TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1583 	TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1584 }
1585 
1586 /* from STCP */
1587 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1588 {
1589 	tp->lost_skb_hint = NULL;
1590 }
1591 
1592 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1593 {
1594 	tcp_clear_retrans_hints_partial(tp);
1595 	tp->retransmit_skb_hint = NULL;
1596 }
1597 
1598 union tcp_md5_addr {
1599 	struct in_addr  a4;
1600 #if IS_ENABLED(CONFIG_IPV6)
1601 	struct in6_addr	a6;
1602 #endif
1603 };
1604 
1605 /* - key database */
1606 struct tcp_md5sig_key {
1607 	struct hlist_node	node;
1608 	u8			keylen;
1609 	u8			family; /* AF_INET or AF_INET6 */
1610 	u8			prefixlen;
1611 	u8			flags;
1612 	union tcp_md5_addr	addr;
1613 	int			l3index; /* set if key added with L3 scope */
1614 	u8			key[TCP_MD5SIG_MAXKEYLEN];
1615 	struct rcu_head		rcu;
1616 };
1617 
1618 /* - sock block */
1619 struct tcp_md5sig_info {
1620 	struct hlist_head	head;
1621 	struct rcu_head		rcu;
1622 };
1623 
1624 /* - pseudo header */
1625 struct tcp4_pseudohdr {
1626 	__be32		saddr;
1627 	__be32		daddr;
1628 	__u8		pad;
1629 	__u8		protocol;
1630 	__be16		len;
1631 };
1632 
1633 struct tcp6_pseudohdr {
1634 	struct in6_addr	saddr;
1635 	struct in6_addr daddr;
1636 	__be32		len;
1637 	__be32		protocol;	/* including padding */
1638 };
1639 
1640 union tcp_md5sum_block {
1641 	struct tcp4_pseudohdr ip4;
1642 #if IS_ENABLED(CONFIG_IPV6)
1643 	struct tcp6_pseudohdr ip6;
1644 #endif
1645 };
1646 
1647 /* - pool: digest algorithm, hash description and scratch buffer */
1648 struct tcp_md5sig_pool {
1649 	struct ahash_request	*md5_req;
1650 	void			*scratch;
1651 };
1652 
1653 /* - functions */
1654 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1655 			const struct sock *sk, const struct sk_buff *skb);
1656 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1657 		   int family, u8 prefixlen, int l3index, u8 flags,
1658 		   const u8 *newkey, u8 newkeylen, gfp_t gfp);
1659 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1660 		   int family, u8 prefixlen, int l3index, u8 flags);
1661 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1662 					 const struct sock *addr_sk);
1663 
1664 #ifdef CONFIG_TCP_MD5SIG
1665 #include <linux/jump_label.h>
1666 extern struct static_key_false tcp_md5_needed;
1667 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1668 					   const union tcp_md5_addr *addr,
1669 					   int family);
1670 static inline struct tcp_md5sig_key *
1671 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1672 		  const union tcp_md5_addr *addr, int family)
1673 {
1674 	if (!static_branch_unlikely(&tcp_md5_needed))
1675 		return NULL;
1676 	return __tcp_md5_do_lookup(sk, l3index, addr, family);
1677 }
1678 
1679 enum skb_drop_reason
1680 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
1681 		     const void *saddr, const void *daddr,
1682 		     int family, int dif, int sdif);
1683 
1684 
1685 #define tcp_twsk_md5_key(twsk)	((twsk)->tw_md5_key)
1686 #else
1687 static inline struct tcp_md5sig_key *
1688 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1689 		  const union tcp_md5_addr *addr, int family)
1690 {
1691 	return NULL;
1692 }
1693 
1694 static inline enum skb_drop_reason
1695 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
1696 		     const void *saddr, const void *daddr,
1697 		     int family, int dif, int sdif)
1698 {
1699 	return SKB_NOT_DROPPED_YET;
1700 }
1701 #define tcp_twsk_md5_key(twsk)	NULL
1702 #endif
1703 
1704 bool tcp_alloc_md5sig_pool(void);
1705 
1706 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1707 static inline void tcp_put_md5sig_pool(void)
1708 {
1709 	local_bh_enable();
1710 }
1711 
1712 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1713 			  unsigned int header_len);
1714 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1715 		     const struct tcp_md5sig_key *key);
1716 
1717 /* From tcp_fastopen.c */
1718 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1719 			    struct tcp_fastopen_cookie *cookie);
1720 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1721 			    struct tcp_fastopen_cookie *cookie, bool syn_lost,
1722 			    u16 try_exp);
1723 struct tcp_fastopen_request {
1724 	/* Fast Open cookie. Size 0 means a cookie request */
1725 	struct tcp_fastopen_cookie	cookie;
1726 	struct msghdr			*data;  /* data in MSG_FASTOPEN */
1727 	size_t				size;
1728 	int				copied;	/* queued in tcp_connect() */
1729 	struct ubuf_info		*uarg;
1730 };
1731 void tcp_free_fastopen_req(struct tcp_sock *tp);
1732 void tcp_fastopen_destroy_cipher(struct sock *sk);
1733 void tcp_fastopen_ctx_destroy(struct net *net);
1734 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1735 			      void *primary_key, void *backup_key);
1736 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
1737 			    u64 *key);
1738 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1739 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1740 			      struct request_sock *req,
1741 			      struct tcp_fastopen_cookie *foc,
1742 			      const struct dst_entry *dst);
1743 void tcp_fastopen_init_key_once(struct net *net);
1744 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1745 			     struct tcp_fastopen_cookie *cookie);
1746 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1747 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1748 #define TCP_FASTOPEN_KEY_MAX 2
1749 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1750 	(TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1751 
1752 /* Fastopen key context */
1753 struct tcp_fastopen_context {
1754 	siphash_key_t	key[TCP_FASTOPEN_KEY_MAX];
1755 	int		num;
1756 	struct rcu_head	rcu;
1757 };
1758 
1759 void tcp_fastopen_active_disable(struct sock *sk);
1760 bool tcp_fastopen_active_should_disable(struct sock *sk);
1761 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1762 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1763 
1764 /* Caller needs to wrap with rcu_read_(un)lock() */
1765 static inline
1766 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1767 {
1768 	struct tcp_fastopen_context *ctx;
1769 
1770 	ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1771 	if (!ctx)
1772 		ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1773 	return ctx;
1774 }
1775 
1776 static inline
1777 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1778 			       const struct tcp_fastopen_cookie *orig)
1779 {
1780 	if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1781 	    orig->len == foc->len &&
1782 	    !memcmp(orig->val, foc->val, foc->len))
1783 		return true;
1784 	return false;
1785 }
1786 
1787 static inline
1788 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1789 {
1790 	return ctx->num;
1791 }
1792 
1793 /* Latencies incurred by various limits for a sender. They are
1794  * chronograph-like stats that are mutually exclusive.
1795  */
1796 enum tcp_chrono {
1797 	TCP_CHRONO_UNSPEC,
1798 	TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1799 	TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1800 	TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1801 	__TCP_CHRONO_MAX,
1802 };
1803 
1804 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1805 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1806 
1807 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1808  * the same memory storage than skb->destructor/_skb_refdst
1809  */
1810 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1811 {
1812 	skb->destructor = NULL;
1813 	skb->_skb_refdst = 0UL;
1814 }
1815 
1816 #define tcp_skb_tsorted_save(skb) {		\
1817 	unsigned long _save = skb->_skb_refdst;	\
1818 	skb->_skb_refdst = 0UL;
1819 
1820 #define tcp_skb_tsorted_restore(skb)		\
1821 	skb->_skb_refdst = _save;		\
1822 }
1823 
1824 void tcp_write_queue_purge(struct sock *sk);
1825 
1826 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1827 {
1828 	return skb_rb_first(&sk->tcp_rtx_queue);
1829 }
1830 
1831 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1832 {
1833 	return skb_rb_last(&sk->tcp_rtx_queue);
1834 }
1835 
1836 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1837 {
1838 	return skb_peek_tail(&sk->sk_write_queue);
1839 }
1840 
1841 #define tcp_for_write_queue_from_safe(skb, tmp, sk)			\
1842 	skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1843 
1844 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1845 {
1846 	return skb_peek(&sk->sk_write_queue);
1847 }
1848 
1849 static inline bool tcp_skb_is_last(const struct sock *sk,
1850 				   const struct sk_buff *skb)
1851 {
1852 	return skb_queue_is_last(&sk->sk_write_queue, skb);
1853 }
1854 
1855 /**
1856  * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
1857  * @sk: socket
1858  *
1859  * Since the write queue can have a temporary empty skb in it,
1860  * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
1861  */
1862 static inline bool tcp_write_queue_empty(const struct sock *sk)
1863 {
1864 	const struct tcp_sock *tp = tcp_sk(sk);
1865 
1866 	return tp->write_seq == tp->snd_nxt;
1867 }
1868 
1869 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1870 {
1871 	return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1872 }
1873 
1874 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1875 {
1876 	return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1877 }
1878 
1879 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1880 {
1881 	__skb_queue_tail(&sk->sk_write_queue, skb);
1882 
1883 	/* Queue it, remembering where we must start sending. */
1884 	if (sk->sk_write_queue.next == skb)
1885 		tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1886 }
1887 
1888 /* Insert new before skb on the write queue of sk.  */
1889 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1890 						  struct sk_buff *skb,
1891 						  struct sock *sk)
1892 {
1893 	__skb_queue_before(&sk->sk_write_queue, skb, new);
1894 }
1895 
1896 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1897 {
1898 	tcp_skb_tsorted_anchor_cleanup(skb);
1899 	__skb_unlink(skb, &sk->sk_write_queue);
1900 }
1901 
1902 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1903 
1904 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1905 {
1906 	tcp_skb_tsorted_anchor_cleanup(skb);
1907 	rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1908 }
1909 
1910 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1911 {
1912 	list_del(&skb->tcp_tsorted_anchor);
1913 	tcp_rtx_queue_unlink(skb, sk);
1914 	tcp_wmem_free_skb(sk, skb);
1915 }
1916 
1917 static inline void tcp_push_pending_frames(struct sock *sk)
1918 {
1919 	if (tcp_send_head(sk)) {
1920 		struct tcp_sock *tp = tcp_sk(sk);
1921 
1922 		__tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1923 	}
1924 }
1925 
1926 /* Start sequence of the skb just after the highest skb with SACKed
1927  * bit, valid only if sacked_out > 0 or when the caller has ensured
1928  * validity by itself.
1929  */
1930 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1931 {
1932 	if (!tp->sacked_out)
1933 		return tp->snd_una;
1934 
1935 	if (tp->highest_sack == NULL)
1936 		return tp->snd_nxt;
1937 
1938 	return TCP_SKB_CB(tp->highest_sack)->seq;
1939 }
1940 
1941 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1942 {
1943 	tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1944 }
1945 
1946 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1947 {
1948 	return tcp_sk(sk)->highest_sack;
1949 }
1950 
1951 static inline void tcp_highest_sack_reset(struct sock *sk)
1952 {
1953 	tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1954 }
1955 
1956 /* Called when old skb is about to be deleted and replaced by new skb */
1957 static inline void tcp_highest_sack_replace(struct sock *sk,
1958 					    struct sk_buff *old,
1959 					    struct sk_buff *new)
1960 {
1961 	if (old == tcp_highest_sack(sk))
1962 		tcp_sk(sk)->highest_sack = new;
1963 }
1964 
1965 /* This helper checks if socket has IP_TRANSPARENT set */
1966 static inline bool inet_sk_transparent(const struct sock *sk)
1967 {
1968 	switch (sk->sk_state) {
1969 	case TCP_TIME_WAIT:
1970 		return inet_twsk(sk)->tw_transparent;
1971 	case TCP_NEW_SYN_RECV:
1972 		return inet_rsk(inet_reqsk(sk))->no_srccheck;
1973 	}
1974 	return inet_sk(sk)->transparent;
1975 }
1976 
1977 /* Determines whether this is a thin stream (which may suffer from
1978  * increased latency). Used to trigger latency-reducing mechanisms.
1979  */
1980 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1981 {
1982 	return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1983 }
1984 
1985 /* /proc */
1986 enum tcp_seq_states {
1987 	TCP_SEQ_STATE_LISTENING,
1988 	TCP_SEQ_STATE_ESTABLISHED,
1989 };
1990 
1991 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1992 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1993 void tcp_seq_stop(struct seq_file *seq, void *v);
1994 
1995 struct tcp_seq_afinfo {
1996 	sa_family_t			family;
1997 };
1998 
1999 struct tcp_iter_state {
2000 	struct seq_net_private	p;
2001 	enum tcp_seq_states	state;
2002 	struct sock		*syn_wait_sk;
2003 	int			bucket, offset, sbucket, num;
2004 	loff_t			last_pos;
2005 };
2006 
2007 extern struct request_sock_ops tcp_request_sock_ops;
2008 extern struct request_sock_ops tcp6_request_sock_ops;
2009 
2010 void tcp_v4_destroy_sock(struct sock *sk);
2011 
2012 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
2013 				netdev_features_t features);
2014 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
2015 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
2016 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
2017 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff));
2018 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb));
2019 int tcp_gro_complete(struct sk_buff *skb);
2020 
2021 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
2022 
2023 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
2024 {
2025 	struct net *net = sock_net((struct sock *)tp);
2026 	return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
2027 }
2028 
2029 bool tcp_stream_memory_free(const struct sock *sk, int wake);
2030 
2031 #ifdef CONFIG_PROC_FS
2032 int tcp4_proc_init(void);
2033 void tcp4_proc_exit(void);
2034 #endif
2035 
2036 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
2037 int tcp_conn_request(struct request_sock_ops *rsk_ops,
2038 		     const struct tcp_request_sock_ops *af_ops,
2039 		     struct sock *sk, struct sk_buff *skb);
2040 
2041 /* TCP af-specific functions */
2042 struct tcp_sock_af_ops {
2043 #ifdef CONFIG_TCP_MD5SIG
2044 	struct tcp_md5sig_key	*(*md5_lookup) (const struct sock *sk,
2045 						const struct sock *addr_sk);
2046 	int		(*calc_md5_hash)(char *location,
2047 					 const struct tcp_md5sig_key *md5,
2048 					 const struct sock *sk,
2049 					 const struct sk_buff *skb);
2050 	int		(*md5_parse)(struct sock *sk,
2051 				     int optname,
2052 				     sockptr_t optval,
2053 				     int optlen);
2054 #endif
2055 };
2056 
2057 struct tcp_request_sock_ops {
2058 	u16 mss_clamp;
2059 #ifdef CONFIG_TCP_MD5SIG
2060 	struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
2061 						 const struct sock *addr_sk);
2062 	int		(*calc_md5_hash) (char *location,
2063 					  const struct tcp_md5sig_key *md5,
2064 					  const struct sock *sk,
2065 					  const struct sk_buff *skb);
2066 #endif
2067 #ifdef CONFIG_SYN_COOKIES
2068 	__u32 (*cookie_init_seq)(const struct sk_buff *skb,
2069 				 __u16 *mss);
2070 #endif
2071 	struct dst_entry *(*route_req)(const struct sock *sk,
2072 				       struct sk_buff *skb,
2073 				       struct flowi *fl,
2074 				       struct request_sock *req);
2075 	u32 (*init_seq)(const struct sk_buff *skb);
2076 	u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
2077 	int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2078 			   struct flowi *fl, struct request_sock *req,
2079 			   struct tcp_fastopen_cookie *foc,
2080 			   enum tcp_synack_type synack_type,
2081 			   struct sk_buff *syn_skb);
2082 };
2083 
2084 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2085 #if IS_ENABLED(CONFIG_IPV6)
2086 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2087 #endif
2088 
2089 #ifdef CONFIG_SYN_COOKIES
2090 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2091 					 const struct sock *sk, struct sk_buff *skb,
2092 					 __u16 *mss)
2093 {
2094 	tcp_synq_overflow(sk);
2095 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2096 	return ops->cookie_init_seq(skb, mss);
2097 }
2098 #else
2099 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2100 					 const struct sock *sk, struct sk_buff *skb,
2101 					 __u16 *mss)
2102 {
2103 	return 0;
2104 }
2105 #endif
2106 
2107 int tcpv4_offload_init(void);
2108 
2109 void tcp_v4_init(void);
2110 void tcp_init(void);
2111 
2112 /* tcp_recovery.c */
2113 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2114 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2115 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2116 				u32 reo_wnd);
2117 extern bool tcp_rack_mark_lost(struct sock *sk);
2118 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2119 			     u64 xmit_time);
2120 extern void tcp_rack_reo_timeout(struct sock *sk);
2121 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2122 
2123 /* At how many usecs into the future should the RTO fire? */
2124 static inline s64 tcp_rto_delta_us(const struct sock *sk)
2125 {
2126 	const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2127 	u32 rto = inet_csk(sk)->icsk_rto;
2128 	u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2129 
2130 	return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2131 }
2132 
2133 /*
2134  * Save and compile IPv4 options, return a pointer to it
2135  */
2136 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2137 							 struct sk_buff *skb)
2138 {
2139 	const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2140 	struct ip_options_rcu *dopt = NULL;
2141 
2142 	if (opt->optlen) {
2143 		int opt_size = sizeof(*dopt) + opt->optlen;
2144 
2145 		dopt = kmalloc(opt_size, GFP_ATOMIC);
2146 		if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2147 			kfree(dopt);
2148 			dopt = NULL;
2149 		}
2150 	}
2151 	return dopt;
2152 }
2153 
2154 /* locally generated TCP pure ACKs have skb->truesize == 2
2155  * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2156  * This is much faster than dissecting the packet to find out.
2157  * (Think of GRE encapsulations, IPv4, IPv6, ...)
2158  */
2159 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2160 {
2161 	return skb->truesize == 2;
2162 }
2163 
2164 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2165 {
2166 	skb->truesize = 2;
2167 }
2168 
2169 static inline int tcp_inq(struct sock *sk)
2170 {
2171 	struct tcp_sock *tp = tcp_sk(sk);
2172 	int answ;
2173 
2174 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2175 		answ = 0;
2176 	} else if (sock_flag(sk, SOCK_URGINLINE) ||
2177 		   !tp->urg_data ||
2178 		   before(tp->urg_seq, tp->copied_seq) ||
2179 		   !before(tp->urg_seq, tp->rcv_nxt)) {
2180 
2181 		answ = tp->rcv_nxt - tp->copied_seq;
2182 
2183 		/* Subtract 1, if FIN was received */
2184 		if (answ && sock_flag(sk, SOCK_DONE))
2185 			answ--;
2186 	} else {
2187 		answ = tp->urg_seq - tp->copied_seq;
2188 	}
2189 
2190 	return answ;
2191 }
2192 
2193 int tcp_peek_len(struct socket *sock);
2194 
2195 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2196 {
2197 	u16 segs_in;
2198 
2199 	segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2200 
2201 	/* We update these fields while other threads might
2202 	 * read them from tcp_get_info()
2203 	 */
2204 	WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in);
2205 	if (skb->len > tcp_hdrlen(skb))
2206 		WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in);
2207 }
2208 
2209 /*
2210  * TCP listen path runs lockless.
2211  * We forced "struct sock" to be const qualified to make sure
2212  * we don't modify one of its field by mistake.
2213  * Here, we increment sk_drops which is an atomic_t, so we can safely
2214  * make sock writable again.
2215  */
2216 static inline void tcp_listendrop(const struct sock *sk)
2217 {
2218 	atomic_inc(&((struct sock *)sk)->sk_drops);
2219 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2220 }
2221 
2222 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2223 
2224 /*
2225  * Interface for adding Upper Level Protocols over TCP
2226  */
2227 
2228 #define TCP_ULP_NAME_MAX	16
2229 #define TCP_ULP_MAX		128
2230 #define TCP_ULP_BUF_MAX		(TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2231 
2232 struct tcp_ulp_ops {
2233 	struct list_head	list;
2234 
2235 	/* initialize ulp */
2236 	int (*init)(struct sock *sk);
2237 	/* update ulp */
2238 	void (*update)(struct sock *sk, struct proto *p,
2239 		       void (*write_space)(struct sock *sk));
2240 	/* cleanup ulp */
2241 	void (*release)(struct sock *sk);
2242 	/* diagnostic */
2243 	int (*get_info)(const struct sock *sk, struct sk_buff *skb);
2244 	size_t (*get_info_size)(const struct sock *sk);
2245 	/* clone ulp */
2246 	void (*clone)(const struct request_sock *req, struct sock *newsk,
2247 		      const gfp_t priority);
2248 
2249 	char		name[TCP_ULP_NAME_MAX];
2250 	struct module	*owner;
2251 };
2252 int tcp_register_ulp(struct tcp_ulp_ops *type);
2253 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2254 int tcp_set_ulp(struct sock *sk, const char *name);
2255 void tcp_get_available_ulp(char *buf, size_t len);
2256 void tcp_cleanup_ulp(struct sock *sk);
2257 void tcp_update_ulp(struct sock *sk, struct proto *p,
2258 		    void (*write_space)(struct sock *sk));
2259 
2260 #define MODULE_ALIAS_TCP_ULP(name)				\
2261 	__MODULE_INFO(alias, alias_userspace, name);		\
2262 	__MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2263 
2264 #ifdef CONFIG_NET_SOCK_MSG
2265 struct sk_msg;
2266 struct sk_psock;
2267 
2268 #ifdef CONFIG_BPF_SYSCALL
2269 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
2270 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
2271 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2272 #endif /* CONFIG_BPF_SYSCALL */
2273 
2274 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes,
2275 			  int flags);
2276 #endif /* CONFIG_NET_SOCK_MSG */
2277 
2278 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG)
2279 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2280 {
2281 }
2282 #endif
2283 
2284 #ifdef CONFIG_CGROUP_BPF
2285 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2286 				      struct sk_buff *skb,
2287 				      unsigned int end_offset)
2288 {
2289 	skops->skb = skb;
2290 	skops->skb_data_end = skb->data + end_offset;
2291 }
2292 #else
2293 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2294 				      struct sk_buff *skb,
2295 				      unsigned int end_offset)
2296 {
2297 }
2298 #endif
2299 
2300 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2301  * is < 0, then the BPF op failed (for example if the loaded BPF
2302  * program does not support the chosen operation or there is no BPF
2303  * program loaded).
2304  */
2305 #ifdef CONFIG_BPF
2306 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2307 {
2308 	struct bpf_sock_ops_kern sock_ops;
2309 	int ret;
2310 
2311 	memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2312 	if (sk_fullsock(sk)) {
2313 		sock_ops.is_fullsock = 1;
2314 		sock_owned_by_me(sk);
2315 	}
2316 
2317 	sock_ops.sk = sk;
2318 	sock_ops.op = op;
2319 	if (nargs > 0)
2320 		memcpy(sock_ops.args, args, nargs * sizeof(*args));
2321 
2322 	ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2323 	if (ret == 0)
2324 		ret = sock_ops.reply;
2325 	else
2326 		ret = -1;
2327 	return ret;
2328 }
2329 
2330 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2331 {
2332 	u32 args[2] = {arg1, arg2};
2333 
2334 	return tcp_call_bpf(sk, op, 2, args);
2335 }
2336 
2337 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2338 				    u32 arg3)
2339 {
2340 	u32 args[3] = {arg1, arg2, arg3};
2341 
2342 	return tcp_call_bpf(sk, op, 3, args);
2343 }
2344 
2345 #else
2346 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2347 {
2348 	return -EPERM;
2349 }
2350 
2351 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2352 {
2353 	return -EPERM;
2354 }
2355 
2356 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2357 				    u32 arg3)
2358 {
2359 	return -EPERM;
2360 }
2361 
2362 #endif
2363 
2364 static inline u32 tcp_timeout_init(struct sock *sk)
2365 {
2366 	int timeout;
2367 
2368 	timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2369 
2370 	if (timeout <= 0)
2371 		timeout = TCP_TIMEOUT_INIT;
2372 	return min_t(int, timeout, TCP_RTO_MAX);
2373 }
2374 
2375 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2376 {
2377 	int rwnd;
2378 
2379 	rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2380 
2381 	if (rwnd < 0)
2382 		rwnd = 0;
2383 	return rwnd;
2384 }
2385 
2386 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2387 {
2388 	return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2389 }
2390 
2391 static inline void tcp_bpf_rtt(struct sock *sk)
2392 {
2393 	if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2394 		tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2395 }
2396 
2397 #if IS_ENABLED(CONFIG_SMC)
2398 extern struct static_key_false tcp_have_smc;
2399 #endif
2400 
2401 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2402 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2403 			     void (*cad)(struct sock *sk, u32 ack_seq));
2404 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2405 void clean_acked_data_flush(void);
2406 #endif
2407 
2408 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2409 static inline void tcp_add_tx_delay(struct sk_buff *skb,
2410 				    const struct tcp_sock *tp)
2411 {
2412 	if (static_branch_unlikely(&tcp_tx_delay_enabled))
2413 		skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2414 }
2415 
2416 /* Compute Earliest Departure Time for some control packets
2417  * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2418  */
2419 static inline u64 tcp_transmit_time(const struct sock *sk)
2420 {
2421 	if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2422 		u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2423 			tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2424 
2425 		return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
2426 	}
2427 	return 0;
2428 }
2429 
2430 #endif	/* _TCP_H */
2431