xref: /linux-6.15/include/net/tcp.h (revision 0791c032)
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, <[email protected]>
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/tcp_ao.h>
41 #include <net/inet_ecn.h>
42 #include <net/dst.h>
43 #include <net/mptcp.h>
44 #include <net/xfrm.h>
45 
46 #include <linux/seq_file.h>
47 #include <linux/memcontrol.h>
48 #include <linux/bpf-cgroup.h>
49 #include <linux/siphash.h>
50 
51 extern struct inet_hashinfo tcp_hashinfo;
52 
53 DECLARE_PER_CPU(unsigned int, tcp_orphan_count);
54 int tcp_orphan_count_sum(void);
55 
56 DECLARE_PER_CPU(u32, tcp_tw_isn);
57 
58 void tcp_time_wait(struct sock *sk, int state, int timeo);
59 
60 #define MAX_TCP_HEADER	L1_CACHE_ALIGN(128 + MAX_HEADER)
61 #define MAX_TCP_OPTION_SPACE 40
62 #define TCP_MIN_SND_MSS		48
63 #define TCP_MIN_GSO_SIZE	(TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
64 
65 /*
66  * Never offer a window over 32767 without using window scaling. Some
67  * poor stacks do signed 16bit maths!
68  */
69 #define MAX_TCP_WINDOW		32767U
70 
71 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
72 #define TCP_MIN_MSS		88U
73 
74 /* The initial MTU to use for probing */
75 #define TCP_BASE_MSS		1024
76 
77 /* probing interval, default to 10 minutes as per RFC4821 */
78 #define TCP_PROBE_INTERVAL	600
79 
80 /* Specify interval when tcp mtu probing will stop */
81 #define TCP_PROBE_THRESHOLD	8
82 
83 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
84 #define TCP_FASTRETRANS_THRESH 3
85 
86 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
87 #define TCP_MAX_QUICKACKS	16U
88 
89 /* Maximal number of window scale according to RFC1323 */
90 #define TCP_MAX_WSCALE		14U
91 
92 /* urg_data states */
93 #define TCP_URG_VALID	0x0100
94 #define TCP_URG_NOTYET	0x0200
95 #define TCP_URG_READ	0x0400
96 
97 #define TCP_RETR1	3	/*
98 				 * This is how many retries it does before it
99 				 * tries to figure out if the gateway is
100 				 * down. Minimal RFC value is 3; it corresponds
101 				 * to ~3sec-8min depending on RTO.
102 				 */
103 
104 #define TCP_RETR2	15	/*
105 				 * This should take at least
106 				 * 90 minutes to time out.
107 				 * RFC1122 says that the limit is 100 sec.
108 				 * 15 is ~13-30min depending on RTO.
109 				 */
110 
111 #define TCP_SYN_RETRIES	 6	/* This is how many retries are done
112 				 * when active opening a connection.
113 				 * RFC1122 says the minimum retry MUST
114 				 * be at least 180secs.  Nevertheless
115 				 * this value is corresponding to
116 				 * 63secs of retransmission with the
117 				 * current initial RTO.
118 				 */
119 
120 #define TCP_SYNACK_RETRIES 5	/* This is how may retries are done
121 				 * when passive opening a connection.
122 				 * This is corresponding to 31secs of
123 				 * retransmission with the current
124 				 * initial RTO.
125 				 */
126 
127 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
128 				  * state, about 60 seconds	*/
129 #define TCP_FIN_TIMEOUT	TCP_TIMEWAIT_LEN
130                                  /* BSD style FIN_WAIT2 deadlock breaker.
131 				  * It used to be 3min, new value is 60sec,
132 				  * to combine FIN-WAIT-2 timeout with
133 				  * TIME-WAIT timer.
134 				  */
135 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */
136 
137 #define TCP_DELACK_MAX	((unsigned)(HZ/5))	/* maximal time to delay before sending an ACK */
138 static_assert((1 << ATO_BITS) > TCP_DELACK_MAX);
139 
140 #if HZ >= 100
141 #define TCP_DELACK_MIN	((unsigned)(HZ/25))	/* minimal time to delay before sending an ACK */
142 #define TCP_ATO_MIN	((unsigned)(HZ/25))
143 #else
144 #define TCP_DELACK_MIN	4U
145 #define TCP_ATO_MIN	4U
146 #endif
147 #define TCP_RTO_MAX_SEC 120
148 #define TCP_RTO_MAX	((unsigned)(TCP_RTO_MAX_SEC * HZ))
149 #define TCP_RTO_MIN	((unsigned)(HZ / 5))
150 #define TCP_TIMEOUT_MIN	(2U) /* Min timeout for TCP timers in jiffies */
151 
152 #define TCP_TIMEOUT_MIN_US (2*USEC_PER_MSEC) /* Min TCP timeout in microsecs */
153 
154 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ))	/* RFC6298 2.1 initial RTO value	*/
155 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ))	/* RFC 1122 initial RTO value, now
156 						 * used as a fallback RTO for the
157 						 * initial data transmission if no
158 						 * valid RTT sample has been acquired,
159 						 * most likely due to retrans in 3WHS.
160 						 */
161 
162 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
163 					                 * for local resources.
164 					                 */
165 #define TCP_KEEPALIVE_TIME	(120*60*HZ)	/* two hours */
166 #define TCP_KEEPALIVE_PROBES	9		/* Max of 9 keepalive probes	*/
167 #define TCP_KEEPALIVE_INTVL	(75*HZ)
168 
169 #define MAX_TCP_KEEPIDLE	32767
170 #define MAX_TCP_KEEPINTVL	32767
171 #define MAX_TCP_KEEPCNT		127
172 #define MAX_TCP_SYNCNT		127
173 
174 /* Ensure that TCP PAWS checks are relaxed after ~2147 seconds
175  * to avoid overflows. This assumes a clock smaller than 1 Mhz.
176  * Default clock is 1 Khz, tcp_usec_ts uses 1 Mhz.
177  */
178 #define TCP_PAWS_WRAP (INT_MAX / USEC_PER_SEC)
179 
180 #define TCP_PAWS_MSL	60		/* Per-host timestamps are invalidated
181 					 * after this time. It should be equal
182 					 * (or greater than) TCP_TIMEWAIT_LEN
183 					 * to provide reliability equal to one
184 					 * provided by timewait state.
185 					 */
186 #define TCP_PAWS_WINDOW	1		/* Replay window for per-host
187 					 * timestamps. It must be less than
188 					 * minimal timewait lifetime.
189 					 */
190 /*
191  *	TCP option
192  */
193 
194 #define TCPOPT_NOP		1	/* Padding */
195 #define TCPOPT_EOL		0	/* End of options */
196 #define TCPOPT_MSS		2	/* Segment size negotiating */
197 #define TCPOPT_WINDOW		3	/* Window scaling */
198 #define TCPOPT_SACK_PERM        4       /* SACK Permitted */
199 #define TCPOPT_SACK             5       /* SACK Block */
200 #define TCPOPT_TIMESTAMP	8	/* Better RTT estimations/PAWS */
201 #define TCPOPT_MD5SIG		19	/* MD5 Signature (RFC2385) */
202 #define TCPOPT_AO		29	/* Authentication Option (RFC5925) */
203 #define TCPOPT_MPTCP		30	/* Multipath TCP (RFC6824) */
204 #define TCPOPT_FASTOPEN		34	/* Fast open (RFC7413) */
205 #define TCPOPT_EXP		254	/* Experimental */
206 /* Magic number to be after the option value for sharing TCP
207  * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
208  */
209 #define TCPOPT_FASTOPEN_MAGIC	0xF989
210 #define TCPOPT_SMC_MAGIC	0xE2D4C3D9
211 
212 /*
213  *     TCP option lengths
214  */
215 
216 #define TCPOLEN_MSS            4
217 #define TCPOLEN_WINDOW         3
218 #define TCPOLEN_SACK_PERM      2
219 #define TCPOLEN_TIMESTAMP      10
220 #define TCPOLEN_MD5SIG         18
221 #define TCPOLEN_FASTOPEN_BASE  2
222 #define TCPOLEN_EXP_FASTOPEN_BASE  4
223 #define TCPOLEN_EXP_SMC_BASE   6
224 
225 /* But this is what stacks really send out. */
226 #define TCPOLEN_TSTAMP_ALIGNED		12
227 #define TCPOLEN_WSCALE_ALIGNED		4
228 #define TCPOLEN_SACKPERM_ALIGNED	4
229 #define TCPOLEN_SACK_BASE		2
230 #define TCPOLEN_SACK_BASE_ALIGNED	4
231 #define TCPOLEN_SACK_PERBLOCK		8
232 #define TCPOLEN_MD5SIG_ALIGNED		20
233 #define TCPOLEN_MSS_ALIGNED		4
234 #define TCPOLEN_EXP_SMC_BASE_ALIGNED	8
235 
236 /* Flags in tp->nonagle */
237 #define TCP_NAGLE_OFF		1	/* Nagle's algo is disabled */
238 #define TCP_NAGLE_CORK		2	/* Socket is corked	    */
239 #define TCP_NAGLE_PUSH		4	/* Cork is overridden for already queued data */
240 
241 /* TCP thin-stream limits */
242 #define TCP_THIN_LINEAR_RETRIES 6       /* After 6 linear retries, do exp. backoff */
243 
244 /* TCP initial congestion window as per rfc6928 */
245 #define TCP_INIT_CWND		10
246 
247 /* Bit Flags for sysctl_tcp_fastopen */
248 #define	TFO_CLIENT_ENABLE	1
249 #define	TFO_SERVER_ENABLE	2
250 #define	TFO_CLIENT_NO_COOKIE	4	/* Data in SYN w/o cookie option */
251 
252 /* Accept SYN data w/o any cookie option */
253 #define	TFO_SERVER_COOKIE_NOT_REQD	0x200
254 
255 /* Force enable TFO on all listeners, i.e., not requiring the
256  * TCP_FASTOPEN socket option.
257  */
258 #define	TFO_SERVER_WO_SOCKOPT1	0x400
259 
260 
261 /* sysctl variables for tcp */
262 extern int sysctl_tcp_max_orphans;
263 extern long sysctl_tcp_mem[3];
264 
265 #define TCP_RACK_LOSS_DETECTION  0x1 /* Use RACK to detect losses */
266 #define TCP_RACK_STATIC_REO_WND  0x2 /* Use static RACK reo wnd */
267 #define TCP_RACK_NO_DUPTHRESH    0x4 /* Do not use DUPACK threshold in RACK */
268 
269 extern atomic_long_t tcp_memory_allocated;
270 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc);
271 
272 extern struct percpu_counter tcp_sockets_allocated;
273 extern unsigned long tcp_memory_pressure;
274 
275 /* optimized version of sk_under_memory_pressure() for TCP sockets */
276 static inline bool tcp_under_memory_pressure(const struct sock *sk)
277 {
278 	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
279 	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
280 		return true;
281 
282 	return READ_ONCE(tcp_memory_pressure);
283 }
284 /*
285  * The next routines deal with comparing 32 bit unsigned ints
286  * and worry about wraparound (automatic with unsigned arithmetic).
287  */
288 
289 static inline bool before(__u32 seq1, __u32 seq2)
290 {
291         return (__s32)(seq1-seq2) < 0;
292 }
293 #define after(seq2, seq1) 	before(seq1, seq2)
294 
295 /* is s2<=s1<=s3 ? */
296 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
297 {
298 	return seq3 - seq2 >= seq1 - seq2;
299 }
300 
301 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
302 {
303 	sk_wmem_queued_add(sk, -skb->truesize);
304 	if (!skb_zcopy_pure(skb))
305 		sk_mem_uncharge(sk, skb->truesize);
306 	else
307 		sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb)));
308 	__kfree_skb(skb);
309 }
310 
311 void sk_forced_mem_schedule(struct sock *sk, int size);
312 
313 bool tcp_check_oom(const struct sock *sk, int shift);
314 
315 
316 extern struct proto tcp_prot;
317 
318 #define TCP_INC_STATS(net, field)	SNMP_INC_STATS((net)->mib.tcp_statistics, field)
319 #define __TCP_INC_STATS(net, field)	__SNMP_INC_STATS((net)->mib.tcp_statistics, field)
320 #define TCP_DEC_STATS(net, field)	SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
321 #define TCP_ADD_STATS(net, field, val)	SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
322 
323 void tcp_tasklet_init(void);
324 
325 int tcp_v4_err(struct sk_buff *skb, u32);
326 
327 void tcp_shutdown(struct sock *sk, int how);
328 
329 int tcp_v4_early_demux(struct sk_buff *skb);
330 int tcp_v4_rcv(struct sk_buff *skb);
331 
332 void tcp_remove_empty_skb(struct sock *sk);
333 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
334 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
335 int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied,
336 			 size_t size, struct ubuf_info *uarg);
337 void tcp_splice_eof(struct socket *sock);
338 int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
339 int tcp_wmem_schedule(struct sock *sk, int copy);
340 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
341 	      int size_goal);
342 void tcp_release_cb(struct sock *sk);
343 void tcp_wfree(struct sk_buff *skb);
344 void tcp_write_timer_handler(struct sock *sk);
345 void tcp_delack_timer_handler(struct sock *sk);
346 int tcp_ioctl(struct sock *sk, int cmd, int *karg);
347 enum skb_drop_reason tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
348 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
349 void tcp_rcv_space_adjust(struct sock *sk);
350 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
351 void tcp_twsk_destructor(struct sock *sk);
352 void tcp_twsk_purge(struct list_head *net_exit_list);
353 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
354 			struct pipe_inode_info *pipe, size_t len,
355 			unsigned int flags);
356 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp,
357 				     bool force_schedule);
358 
359 static inline void tcp_dec_quickack_mode(struct sock *sk)
360 {
361 	struct inet_connection_sock *icsk = inet_csk(sk);
362 
363 	if (icsk->icsk_ack.quick) {
364 		/* How many ACKs S/ACKing new data have we sent? */
365 		const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0;
366 
367 		if (pkts >= icsk->icsk_ack.quick) {
368 			icsk->icsk_ack.quick = 0;
369 			/* Leaving quickack mode we deflate ATO. */
370 			icsk->icsk_ack.ato   = TCP_ATO_MIN;
371 		} else
372 			icsk->icsk_ack.quick -= pkts;
373 	}
374 }
375 
376 #define	TCP_ECN_OK		1
377 #define	TCP_ECN_QUEUE_CWR	2
378 #define	TCP_ECN_DEMAND_CWR	4
379 #define	TCP_ECN_SEEN		8
380 
381 enum tcp_tw_status {
382 	TCP_TW_SUCCESS = 0,
383 	TCP_TW_RST = 1,
384 	TCP_TW_ACK = 2,
385 	TCP_TW_SYN = 3
386 };
387 
388 
389 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
390 					      struct sk_buff *skb,
391 					      const struct tcphdr *th,
392 					      u32 *tw_isn);
393 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
394 			   struct request_sock *req, bool fastopen,
395 			   bool *lost_race);
396 enum skb_drop_reason tcp_child_process(struct sock *parent, struct sock *child,
397 				       struct sk_buff *skb);
398 void tcp_enter_loss(struct sock *sk);
399 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag);
400 void tcp_clear_retrans(struct tcp_sock *tp);
401 void tcp_update_metrics(struct sock *sk);
402 void tcp_init_metrics(struct sock *sk);
403 void tcp_metrics_init(void);
404 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
405 void __tcp_close(struct sock *sk, long timeout);
406 void tcp_close(struct sock *sk, long timeout);
407 void tcp_init_sock(struct sock *sk);
408 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb);
409 __poll_t tcp_poll(struct file *file, struct socket *sock,
410 		      struct poll_table_struct *wait);
411 int do_tcp_getsockopt(struct sock *sk, int level,
412 		      int optname, sockptr_t optval, sockptr_t optlen);
413 int tcp_getsockopt(struct sock *sk, int level, int optname,
414 		   char __user *optval, int __user *optlen);
415 bool tcp_bpf_bypass_getsockopt(int level, int optname);
416 int do_tcp_setsockopt(struct sock *sk, int level, int optname,
417 		      sockptr_t optval, unsigned int optlen);
418 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
419 		   unsigned int optlen);
420 void tcp_reset_keepalive_timer(struct sock *sk, unsigned long timeout);
421 void tcp_set_keepalive(struct sock *sk, int val);
422 void tcp_syn_ack_timeout(const struct request_sock *req);
423 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
424 		int flags, int *addr_len);
425 int tcp_set_rcvlowat(struct sock *sk, int val);
426 int tcp_set_window_clamp(struct sock *sk, int val);
427 void tcp_update_recv_tstamps(struct sk_buff *skb,
428 			     struct scm_timestamping_internal *tss);
429 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
430 			struct scm_timestamping_internal *tss);
431 void tcp_data_ready(struct sock *sk);
432 #ifdef CONFIG_MMU
433 int tcp_mmap(struct file *file, struct socket *sock,
434 	     struct vm_area_struct *vma);
435 #endif
436 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
437 		       struct tcp_options_received *opt_rx,
438 		       int estab, struct tcp_fastopen_cookie *foc);
439 
440 /*
441  *	BPF SKB-less helpers
442  */
443 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
444 			 struct tcphdr *th, u32 *cookie);
445 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
446 			 struct tcphdr *th, u32 *cookie);
447 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss);
448 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
449 			  const struct tcp_request_sock_ops *af_ops,
450 			  struct sock *sk, struct tcphdr *th);
451 /*
452  *	TCP v4 functions exported for the inet6 API
453  */
454 
455 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
456 void tcp_v4_mtu_reduced(struct sock *sk);
457 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
458 void tcp_ld_RTO_revert(struct sock *sk, u32 seq);
459 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
460 struct sock *tcp_create_openreq_child(const struct sock *sk,
461 				      struct request_sock *req,
462 				      struct sk_buff *skb);
463 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
464 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
465 				  struct request_sock *req,
466 				  struct dst_entry *dst,
467 				  struct request_sock *req_unhash,
468 				  bool *own_req);
469 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
470 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
471 int tcp_connect(struct sock *sk);
472 enum tcp_synack_type {
473 	TCP_SYNACK_NORMAL,
474 	TCP_SYNACK_FASTOPEN,
475 	TCP_SYNACK_COOKIE,
476 };
477 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
478 				struct request_sock *req,
479 				struct tcp_fastopen_cookie *foc,
480 				enum tcp_synack_type synack_type,
481 				struct sk_buff *syn_skb);
482 int tcp_disconnect(struct sock *sk, int flags);
483 
484 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
485 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
486 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
487 
488 /* From syncookies.c */
489 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
490 				 struct request_sock *req,
491 				 struct dst_entry *dst);
492 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th);
493 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
494 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
495 					    struct sock *sk, struct sk_buff *skb,
496 					    struct tcp_options_received *tcp_opt,
497 					    int mss, u32 tsoff);
498 
499 #if IS_ENABLED(CONFIG_BPF)
500 struct bpf_tcp_req_attrs {
501 	u32 rcv_tsval;
502 	u32 rcv_tsecr;
503 	u16 mss;
504 	u8 rcv_wscale;
505 	u8 snd_wscale;
506 	u8 ecn_ok;
507 	u8 wscale_ok;
508 	u8 sack_ok;
509 	u8 tstamp_ok;
510 	u8 usec_ts_ok;
511 	u8 reserved[3];
512 };
513 #endif
514 
515 #ifdef CONFIG_SYN_COOKIES
516 
517 /* Syncookies use a monotonic timer which increments every 60 seconds.
518  * This counter is used both as a hash input and partially encoded into
519  * the cookie value.  A cookie is only validated further if the delta
520  * between the current counter value and the encoded one is less than this,
521  * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
522  * the counter advances immediately after a cookie is generated).
523  */
524 #define MAX_SYNCOOKIE_AGE	2
525 #define TCP_SYNCOOKIE_PERIOD	(60 * HZ)
526 #define TCP_SYNCOOKIE_VALID	(MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
527 
528 /* syncookies: remember time of last synqueue overflow
529  * But do not dirty this field too often (once per second is enough)
530  * It is racy as we do not hold a lock, but race is very minor.
531  */
532 static inline void tcp_synq_overflow(const struct sock *sk)
533 {
534 	unsigned int last_overflow;
535 	unsigned int now = jiffies;
536 
537 	if (sk->sk_reuseport) {
538 		struct sock_reuseport *reuse;
539 
540 		reuse = rcu_dereference(sk->sk_reuseport_cb);
541 		if (likely(reuse)) {
542 			last_overflow = READ_ONCE(reuse->synq_overflow_ts);
543 			if (!time_between32(now, last_overflow,
544 					    last_overflow + HZ))
545 				WRITE_ONCE(reuse->synq_overflow_ts, now);
546 			return;
547 		}
548 	}
549 
550 	last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
551 	if (!time_between32(now, last_overflow, last_overflow + HZ))
552 		WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now);
553 }
554 
555 /* syncookies: no recent synqueue overflow on this listening socket? */
556 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
557 {
558 	unsigned int last_overflow;
559 	unsigned int now = jiffies;
560 
561 	if (sk->sk_reuseport) {
562 		struct sock_reuseport *reuse;
563 
564 		reuse = rcu_dereference(sk->sk_reuseport_cb);
565 		if (likely(reuse)) {
566 			last_overflow = READ_ONCE(reuse->synq_overflow_ts);
567 			return !time_between32(now, last_overflow - HZ,
568 					       last_overflow +
569 					       TCP_SYNCOOKIE_VALID);
570 		}
571 	}
572 
573 	last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
574 
575 	/* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
576 	 * then we're under synflood. However, we have to use
577 	 * 'last_overflow - HZ' as lower bound. That's because a concurrent
578 	 * tcp_synq_overflow() could update .ts_recent_stamp after we read
579 	 * jiffies but before we store .ts_recent_stamp into last_overflow,
580 	 * which could lead to rejecting a valid syncookie.
581 	 */
582 	return !time_between32(now, last_overflow - HZ,
583 			       last_overflow + TCP_SYNCOOKIE_VALID);
584 }
585 
586 static inline u32 tcp_cookie_time(void)
587 {
588 	u64 val = get_jiffies_64();
589 
590 	do_div(val, TCP_SYNCOOKIE_PERIOD);
591 	return val;
592 }
593 
594 /* Convert one nsec 64bit timestamp to ts (ms or usec resolution) */
595 static inline u64 tcp_ns_to_ts(bool usec_ts, u64 val)
596 {
597 	if (usec_ts)
598 		return div_u64(val, NSEC_PER_USEC);
599 
600 	return div_u64(val, NSEC_PER_MSEC);
601 }
602 
603 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
604 			      u16 *mssp);
605 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
606 u64 cookie_init_timestamp(struct request_sock *req, u64 now);
607 bool cookie_timestamp_decode(const struct net *net,
608 			     struct tcp_options_received *opt);
609 
610 static inline bool cookie_ecn_ok(const struct net *net, const struct dst_entry *dst)
611 {
612 	return READ_ONCE(net->ipv4.sysctl_tcp_ecn) ||
613 		dst_feature(dst, RTAX_FEATURE_ECN);
614 }
615 
616 #if IS_ENABLED(CONFIG_BPF)
617 static inline bool cookie_bpf_ok(struct sk_buff *skb)
618 {
619 	return skb->sk;
620 }
621 
622 struct request_sock *cookie_bpf_check(struct sock *sk, struct sk_buff *skb);
623 #else
624 static inline bool cookie_bpf_ok(struct sk_buff *skb)
625 {
626 	return false;
627 }
628 
629 static inline struct request_sock *cookie_bpf_check(struct net *net, struct sock *sk,
630 						    struct sk_buff *skb)
631 {
632 	return NULL;
633 }
634 #endif
635 
636 /* From net/ipv6/syncookies.c */
637 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th);
638 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
639 
640 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
641 			      const struct tcphdr *th, u16 *mssp);
642 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
643 #endif
644 /* tcp_output.c */
645 
646 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb);
647 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb);
648 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
649 			       int nonagle);
650 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
651 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
652 void tcp_retransmit_timer(struct sock *sk);
653 void tcp_xmit_retransmit_queue(struct sock *);
654 void tcp_simple_retransmit(struct sock *);
655 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
656 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
657 enum tcp_queue {
658 	TCP_FRAG_IN_WRITE_QUEUE,
659 	TCP_FRAG_IN_RTX_QUEUE,
660 };
661 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
662 		 struct sk_buff *skb, u32 len,
663 		 unsigned int mss_now, gfp_t gfp);
664 
665 void tcp_send_probe0(struct sock *);
666 int tcp_write_wakeup(struct sock *, int mib);
667 void tcp_send_fin(struct sock *sk);
668 void tcp_send_active_reset(struct sock *sk, gfp_t priority,
669 			   enum sk_rst_reason reason);
670 int tcp_send_synack(struct sock *);
671 void tcp_push_one(struct sock *, unsigned int mss_now);
672 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
673 void tcp_send_ack(struct sock *sk);
674 void tcp_send_delayed_ack(struct sock *sk);
675 void tcp_send_loss_probe(struct sock *sk);
676 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
677 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
678 			     const struct sk_buff *next_skb);
679 
680 /* tcp_input.c */
681 void tcp_rearm_rto(struct sock *sk);
682 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
683 void tcp_done_with_error(struct sock *sk, int err);
684 void tcp_reset(struct sock *sk, struct sk_buff *skb);
685 void tcp_fin(struct sock *sk);
686 void tcp_check_space(struct sock *sk);
687 void tcp_sack_compress_send_ack(struct sock *sk);
688 
689 static inline void tcp_cleanup_skb(struct sk_buff *skb)
690 {
691 	skb_dst_drop(skb);
692 	secpath_reset(skb);
693 }
694 
695 static inline void tcp_add_receive_queue(struct sock *sk, struct sk_buff *skb)
696 {
697 	DEBUG_NET_WARN_ON_ONCE(skb_dst(skb));
698 	DEBUG_NET_WARN_ON_ONCE(secpath_exists(skb));
699 	__skb_queue_tail(&sk->sk_receive_queue, skb);
700 }
701 
702 /* tcp_timer.c */
703 void tcp_init_xmit_timers(struct sock *);
704 static inline void tcp_clear_xmit_timers(struct sock *sk)
705 {
706 	if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
707 		__sock_put(sk);
708 
709 	if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
710 		__sock_put(sk);
711 
712 	inet_csk_clear_xmit_timers(sk);
713 }
714 
715 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
716 unsigned int tcp_current_mss(struct sock *sk);
717 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when);
718 
719 /* Bound MSS / TSO packet size with the half of the window */
720 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
721 {
722 	int cutoff;
723 
724 	/* When peer uses tiny windows, there is no use in packetizing
725 	 * to sub-MSS pieces for the sake of SWS or making sure there
726 	 * are enough packets in the pipe for fast recovery.
727 	 *
728 	 * On the other hand, for extremely large MSS devices, handling
729 	 * smaller than MSS windows in this way does make sense.
730 	 */
731 	if (tp->max_window > TCP_MSS_DEFAULT)
732 		cutoff = (tp->max_window >> 1);
733 	else
734 		cutoff = tp->max_window;
735 
736 	if (cutoff && pktsize > cutoff)
737 		return max_t(int, cutoff, 68U - tp->tcp_header_len);
738 	else
739 		return pktsize;
740 }
741 
742 /* tcp.c */
743 void tcp_get_info(struct sock *, struct tcp_info *);
744 
745 /* Read 'sendfile()'-style from a TCP socket */
746 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
747 		  sk_read_actor_t recv_actor);
748 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
749 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off);
750 void tcp_read_done(struct sock *sk, size_t len);
751 
752 void tcp_initialize_rcv_mss(struct sock *sk);
753 
754 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
755 int tcp_mss_to_mtu(struct sock *sk, int mss);
756 void tcp_mtup_init(struct sock *sk);
757 
758 static inline unsigned int tcp_rto_max(const struct sock *sk)
759 {
760 	return READ_ONCE(inet_csk(sk)->icsk_rto_max);
761 }
762 
763 static inline void tcp_bound_rto(struct sock *sk)
764 {
765 	inet_csk(sk)->icsk_rto = min(inet_csk(sk)->icsk_rto, tcp_rto_max(sk));
766 }
767 
768 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
769 {
770 	return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
771 }
772 
773 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
774 {
775 	/* mptcp hooks are only on the slow path */
776 	if (sk_is_mptcp((struct sock *)tp))
777 		return;
778 
779 	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
780 			       ntohl(TCP_FLAG_ACK) |
781 			       snd_wnd);
782 }
783 
784 static inline void tcp_fast_path_on(struct tcp_sock *tp)
785 {
786 	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
787 }
788 
789 static inline void tcp_fast_path_check(struct sock *sk)
790 {
791 	struct tcp_sock *tp = tcp_sk(sk);
792 
793 	if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
794 	    tp->rcv_wnd &&
795 	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
796 	    !tp->urg_data)
797 		tcp_fast_path_on(tp);
798 }
799 
800 u32 tcp_delack_max(const struct sock *sk);
801 
802 /* Compute the actual rto_min value */
803 static inline u32 tcp_rto_min(const struct sock *sk)
804 {
805 	const struct dst_entry *dst = __sk_dst_get(sk);
806 	u32 rto_min = inet_csk(sk)->icsk_rto_min;
807 
808 	if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
809 		rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
810 	return rto_min;
811 }
812 
813 static inline u32 tcp_rto_min_us(const struct sock *sk)
814 {
815 	return jiffies_to_usecs(tcp_rto_min(sk));
816 }
817 
818 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
819 {
820 	return dst_metric_locked(dst, RTAX_CC_ALGO);
821 }
822 
823 /* Minimum RTT in usec. ~0 means not available. */
824 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
825 {
826 	return minmax_get(&tp->rtt_min);
827 }
828 
829 /* Compute the actual receive window we are currently advertising.
830  * Rcv_nxt can be after the window if our peer push more data
831  * than the offered window.
832  */
833 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
834 {
835 	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
836 
837 	if (win < 0)
838 		win = 0;
839 	return (u32) win;
840 }
841 
842 /* Choose a new window, without checks for shrinking, and without
843  * scaling applied to the result.  The caller does these things
844  * if necessary.  This is a "raw" window selection.
845  */
846 u32 __tcp_select_window(struct sock *sk);
847 
848 void tcp_send_window_probe(struct sock *sk);
849 
850 /* TCP uses 32bit jiffies to save some space.
851  * Note that this is different from tcp_time_stamp, which
852  * historically has been the same until linux-4.13.
853  */
854 #define tcp_jiffies32 ((u32)jiffies)
855 
856 /*
857  * Deliver a 32bit value for TCP timestamp option (RFC 7323)
858  * It is no longer tied to jiffies, but to 1 ms clock.
859  * Note: double check if you want to use tcp_jiffies32 instead of this.
860  */
861 #define TCP_TS_HZ	1000
862 
863 static inline u64 tcp_clock_ns(void)
864 {
865 	return ktime_get_ns();
866 }
867 
868 static inline u64 tcp_clock_us(void)
869 {
870 	return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
871 }
872 
873 static inline u64 tcp_clock_ms(void)
874 {
875 	return div_u64(tcp_clock_ns(), NSEC_PER_MSEC);
876 }
877 
878 /* TCP Timestamp included in TS option (RFC 1323) can either use ms
879  * or usec resolution. Each socket carries a flag to select one or other
880  * resolution, as the route attribute could change anytime.
881  * Each flow must stick to initial resolution.
882  */
883 static inline u32 tcp_clock_ts(bool usec_ts)
884 {
885 	return usec_ts ? tcp_clock_us() : tcp_clock_ms();
886 }
887 
888 static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp)
889 {
890 	return div_u64(tp->tcp_mstamp, USEC_PER_MSEC);
891 }
892 
893 static inline u32 tcp_time_stamp_ts(const struct tcp_sock *tp)
894 {
895 	if (tp->tcp_usec_ts)
896 		return tp->tcp_mstamp;
897 	return tcp_time_stamp_ms(tp);
898 }
899 
900 void tcp_mstamp_refresh(struct tcp_sock *tp);
901 
902 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
903 {
904 	return max_t(s64, t1 - t0, 0);
905 }
906 
907 /* provide the departure time in us unit */
908 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
909 {
910 	return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
911 }
912 
913 /* Provide skb TSval in usec or ms unit */
914 static inline u32 tcp_skb_timestamp_ts(bool usec_ts, const struct sk_buff *skb)
915 {
916 	if (usec_ts)
917 		return tcp_skb_timestamp_us(skb);
918 
919 	return div_u64(skb->skb_mstamp_ns, NSEC_PER_MSEC);
920 }
921 
922 static inline u32 tcp_tw_tsval(const struct tcp_timewait_sock *tcptw)
923 {
924 	return tcp_clock_ts(tcptw->tw_sk.tw_usec_ts) + tcptw->tw_ts_offset;
925 }
926 
927 static inline u32 tcp_rsk_tsval(const struct tcp_request_sock *treq)
928 {
929 	return tcp_clock_ts(treq->req_usec_ts) + treq->ts_off;
930 }
931 
932 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
933 
934 #define TCPHDR_FIN 0x01
935 #define TCPHDR_SYN 0x02
936 #define TCPHDR_RST 0x04
937 #define TCPHDR_PSH 0x08
938 #define TCPHDR_ACK 0x10
939 #define TCPHDR_URG 0x20
940 #define TCPHDR_ECE 0x40
941 #define TCPHDR_CWR 0x80
942 
943 #define TCPHDR_SYN_ECN	(TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
944 
945 /* State flags for sacked in struct tcp_skb_cb */
946 enum tcp_skb_cb_sacked_flags {
947 	TCPCB_SACKED_ACKED	= (1 << 0),	/* SKB ACK'd by a SACK block	*/
948 	TCPCB_SACKED_RETRANS	= (1 << 1),	/* SKB retransmitted		*/
949 	TCPCB_LOST		= (1 << 2),	/* SKB is lost			*/
950 	TCPCB_TAGBITS		= (TCPCB_SACKED_ACKED | TCPCB_SACKED_RETRANS |
951 				   TCPCB_LOST),	/* All tag bits			*/
952 	TCPCB_REPAIRED		= (1 << 4),	/* SKB repaired (no skb_mstamp_ns)	*/
953 	TCPCB_EVER_RETRANS	= (1 << 7),	/* Ever retransmitted frame	*/
954 	TCPCB_RETRANS		= (TCPCB_SACKED_RETRANS | TCPCB_EVER_RETRANS |
955 				   TCPCB_REPAIRED),
956 };
957 
958 /* This is what the send packet queuing engine uses to pass
959  * TCP per-packet control information to the transmission code.
960  * We also store the host-order sequence numbers in here too.
961  * This is 44 bytes if IPV6 is enabled.
962  * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
963  */
964 struct tcp_skb_cb {
965 	__u32		seq;		/* Starting sequence number	*/
966 	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
967 	union {
968 		/* Note :
969 		 * 	  tcp_gso_segs/size are used in write queue only,
970 		 *	  cf tcp_skb_pcount()/tcp_skb_mss()
971 		 */
972 		struct {
973 			u16	tcp_gso_segs;
974 			u16	tcp_gso_size;
975 		};
976 	};
977 	__u8		tcp_flags;	/* TCP header flags. (tcp[13])	*/
978 
979 	__u8		sacked;		/* State flags for SACK.	*/
980 	__u8		ip_dsfield;	/* IPv4 tos or IPv6 dsfield	*/
981 #define TSTAMP_ACK_SK	0x1
982 #define TSTAMP_ACK_BPF	0x2
983 	__u8		txstamp_ack:2,	/* Record TX timestamp for ack? */
984 			eor:1,		/* Is skb MSG_EOR marked? */
985 			has_rxtstamp:1,	/* SKB has a RX timestamp	*/
986 			unused:4;
987 	__u32		ack_seq;	/* Sequence number ACK'd	*/
988 	union {
989 		struct {
990 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1)
991 			/* There is space for up to 24 bytes */
992 			__u32 is_app_limited:1, /* cwnd not fully used? */
993 			      delivered_ce:20,
994 			      unused:11;
995 			/* pkts S/ACKed so far upon tx of skb, incl retrans: */
996 			__u32 delivered;
997 			/* start of send pipeline phase */
998 			u64 first_tx_mstamp;
999 			/* when we reached the "delivered" count */
1000 			u64 delivered_mstamp;
1001 		} tx;   /* only used for outgoing skbs */
1002 		union {
1003 			struct inet_skb_parm	h4;
1004 #if IS_ENABLED(CONFIG_IPV6)
1005 			struct inet6_skb_parm	h6;
1006 #endif
1007 		} header;	/* For incoming skbs */
1008 	};
1009 };
1010 
1011 #define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
1012 
1013 extern const struct inet_connection_sock_af_ops ipv4_specific;
1014 
1015 #if IS_ENABLED(CONFIG_IPV6)
1016 /* This is the variant of inet6_iif() that must be used by TCP,
1017  * as TCP moves IP6CB into a different location in skb->cb[]
1018  */
1019 static inline int tcp_v6_iif(const struct sk_buff *skb)
1020 {
1021 	return TCP_SKB_CB(skb)->header.h6.iif;
1022 }
1023 
1024 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
1025 {
1026 	bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
1027 
1028 	return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
1029 }
1030 
1031 /* TCP_SKB_CB reference means this can not be used from early demux */
1032 static inline int tcp_v6_sdif(const struct sk_buff *skb)
1033 {
1034 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
1035 	if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
1036 		return TCP_SKB_CB(skb)->header.h6.iif;
1037 #endif
1038 	return 0;
1039 }
1040 
1041 extern const struct inet_connection_sock_af_ops ipv6_specific;
1042 
1043 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
1044 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
1045 void tcp_v6_early_demux(struct sk_buff *skb);
1046 
1047 #endif
1048 
1049 /* TCP_SKB_CB reference means this can not be used from early demux */
1050 static inline int tcp_v4_sdif(struct sk_buff *skb)
1051 {
1052 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
1053 	if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
1054 		return TCP_SKB_CB(skb)->header.h4.iif;
1055 #endif
1056 	return 0;
1057 }
1058 
1059 /* Due to TSO, an SKB can be composed of multiple actual
1060  * packets.  To keep these tracked properly, we use this.
1061  */
1062 static inline int tcp_skb_pcount(const struct sk_buff *skb)
1063 {
1064 	return TCP_SKB_CB(skb)->tcp_gso_segs;
1065 }
1066 
1067 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
1068 {
1069 	TCP_SKB_CB(skb)->tcp_gso_segs = segs;
1070 }
1071 
1072 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
1073 {
1074 	TCP_SKB_CB(skb)->tcp_gso_segs += segs;
1075 }
1076 
1077 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
1078 static inline int tcp_skb_mss(const struct sk_buff *skb)
1079 {
1080 	return TCP_SKB_CB(skb)->tcp_gso_size;
1081 }
1082 
1083 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
1084 {
1085 	return likely(!TCP_SKB_CB(skb)->eor);
1086 }
1087 
1088 static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
1089 					const struct sk_buff *from)
1090 {
1091 	/* skb_cmp_decrypted() not needed, use tcp_write_collapse_fence() */
1092 	return likely(tcp_skb_can_collapse_to(to) &&
1093 		      mptcp_skb_can_collapse(to, from) &&
1094 		      skb_pure_zcopy_same(to, from) &&
1095 		      skb_frags_readable(to) == skb_frags_readable(from));
1096 }
1097 
1098 static inline bool tcp_skb_can_collapse_rx(const struct sk_buff *to,
1099 					   const struct sk_buff *from)
1100 {
1101 	return likely(mptcp_skb_can_collapse(to, from) &&
1102 		      !skb_cmp_decrypted(to, from));
1103 }
1104 
1105 /* Events passed to congestion control interface */
1106 enum tcp_ca_event {
1107 	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
1108 	CA_EVENT_CWND_RESTART,	/* congestion window restart */
1109 	CA_EVENT_COMPLETE_CWR,	/* end of congestion recovery */
1110 	CA_EVENT_LOSS,		/* loss timeout */
1111 	CA_EVENT_ECN_NO_CE,	/* ECT set, but not CE marked */
1112 	CA_EVENT_ECN_IS_CE,	/* received CE marked IP packet */
1113 };
1114 
1115 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
1116 enum tcp_ca_ack_event_flags {
1117 	CA_ACK_SLOWPATH		= (1 << 0),	/* In slow path processing */
1118 	CA_ACK_WIN_UPDATE	= (1 << 1),	/* ACK updated window */
1119 	CA_ACK_ECE		= (1 << 2),	/* ECE bit is set on ack */
1120 };
1121 
1122 /*
1123  * Interface for adding new TCP congestion control handlers
1124  */
1125 #define TCP_CA_NAME_MAX	16
1126 #define TCP_CA_MAX	128
1127 #define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
1128 
1129 #define TCP_CA_UNSPEC	0
1130 
1131 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1132 #define TCP_CONG_NON_RESTRICTED 0x1
1133 /* Requires ECN/ECT set on all packets */
1134 #define TCP_CONG_NEEDS_ECN	0x2
1135 #define TCP_CONG_MASK	(TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1136 
1137 union tcp_cc_info;
1138 
1139 struct ack_sample {
1140 	u32 pkts_acked;
1141 	s32 rtt_us;
1142 	u32 in_flight;
1143 };
1144 
1145 /* A rate sample measures the number of (original/retransmitted) data
1146  * packets delivered "delivered" over an interval of time "interval_us".
1147  * The tcp_rate.c code fills in the rate sample, and congestion
1148  * control modules that define a cong_control function to run at the end
1149  * of ACK processing can optionally chose to consult this sample when
1150  * setting cwnd and pacing rate.
1151  * A sample is invalid if "delivered" or "interval_us" is negative.
1152  */
1153 struct rate_sample {
1154 	u64  prior_mstamp; /* starting timestamp for interval */
1155 	u32  prior_delivered;	/* tp->delivered at "prior_mstamp" */
1156 	u32  prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */
1157 	s32  delivered;		/* number of packets delivered over interval */
1158 	s32  delivered_ce;	/* number of packets delivered w/ CE marks*/
1159 	long interval_us;	/* time for tp->delivered to incr "delivered" */
1160 	u32 snd_interval_us;	/* snd interval for delivered packets */
1161 	u32 rcv_interval_us;	/* rcv interval for delivered packets */
1162 	long rtt_us;		/* RTT of last (S)ACKed packet (or -1) */
1163 	int  losses;		/* number of packets marked lost upon ACK */
1164 	u32  acked_sacked;	/* number of packets newly (S)ACKed upon ACK */
1165 	u32  prior_in_flight;	/* in flight before this ACK */
1166 	u32  last_end_seq;	/* end_seq of most recently ACKed packet */
1167 	bool is_app_limited;	/* is sample from packet with bubble in pipe? */
1168 	bool is_retrans;	/* is sample from retransmission? */
1169 	bool is_ack_delayed;	/* is this (likely) a delayed ACK? */
1170 };
1171 
1172 struct tcp_congestion_ops {
1173 /* fast path fields are put first to fill one cache line */
1174 
1175 	/* return slow start threshold (required) */
1176 	u32 (*ssthresh)(struct sock *sk);
1177 
1178 	/* do new cwnd calculation (required) */
1179 	void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1180 
1181 	/* call before changing ca_state (optional) */
1182 	void (*set_state)(struct sock *sk, u8 new_state);
1183 
1184 	/* call when cwnd event occurs (optional) */
1185 	void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1186 
1187 	/* call when ack arrives (optional) */
1188 	void (*in_ack_event)(struct sock *sk, u32 flags);
1189 
1190 	/* hook for packet ack accounting (optional) */
1191 	void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1192 
1193 	/* override sysctl_tcp_min_tso_segs */
1194 	u32 (*min_tso_segs)(struct sock *sk);
1195 
1196 	/* call when packets are delivered to update cwnd and pacing rate,
1197 	 * after all the ca_state processing. (optional)
1198 	 */
1199 	void (*cong_control)(struct sock *sk, u32 ack, int flag, const struct rate_sample *rs);
1200 
1201 
1202 	/* new value of cwnd after loss (required) */
1203 	u32  (*undo_cwnd)(struct sock *sk);
1204 	/* returns the multiplier used in tcp_sndbuf_expand (optional) */
1205 	u32 (*sndbuf_expand)(struct sock *sk);
1206 
1207 /* control/slow paths put last */
1208 	/* get info for inet_diag (optional) */
1209 	size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1210 			   union tcp_cc_info *info);
1211 
1212 	char 			name[TCP_CA_NAME_MAX];
1213 	struct module		*owner;
1214 	struct list_head	list;
1215 	u32			key;
1216 	u32			flags;
1217 
1218 	/* initialize private data (optional) */
1219 	void (*init)(struct sock *sk);
1220 	/* cleanup private data  (optional) */
1221 	void (*release)(struct sock *sk);
1222 } ____cacheline_aligned_in_smp;
1223 
1224 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1225 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1226 int tcp_update_congestion_control(struct tcp_congestion_ops *type,
1227 				  struct tcp_congestion_ops *old_type);
1228 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca);
1229 
1230 void tcp_assign_congestion_control(struct sock *sk);
1231 void tcp_init_congestion_control(struct sock *sk);
1232 void tcp_cleanup_congestion_control(struct sock *sk);
1233 int tcp_set_default_congestion_control(struct net *net, const char *name);
1234 void tcp_get_default_congestion_control(struct net *net, char *name);
1235 void tcp_get_available_congestion_control(char *buf, size_t len);
1236 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1237 int tcp_set_allowed_congestion_control(char *allowed);
1238 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1239 			       bool cap_net_admin);
1240 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1241 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1242 
1243 u32 tcp_reno_ssthresh(struct sock *sk);
1244 u32 tcp_reno_undo_cwnd(struct sock *sk);
1245 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1246 extern struct tcp_congestion_ops tcp_reno;
1247 
1248 struct tcp_congestion_ops *tcp_ca_find(const char *name);
1249 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1250 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
1251 #ifdef CONFIG_INET
1252 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1253 #else
1254 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1255 {
1256 	return NULL;
1257 }
1258 #endif
1259 
1260 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1261 {
1262 	const struct inet_connection_sock *icsk = inet_csk(sk);
1263 
1264 	return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1265 }
1266 
1267 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1268 {
1269 	const struct inet_connection_sock *icsk = inet_csk(sk);
1270 
1271 	if (icsk->icsk_ca_ops->cwnd_event)
1272 		icsk->icsk_ca_ops->cwnd_event(sk, event);
1273 }
1274 
1275 /* From tcp_cong.c */
1276 void tcp_set_ca_state(struct sock *sk, const u8 ca_state);
1277 
1278 /* From tcp_rate.c */
1279 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1280 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1281 			    struct rate_sample *rs);
1282 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1283 		  bool is_sack_reneg, struct rate_sample *rs);
1284 void tcp_rate_check_app_limited(struct sock *sk);
1285 
1286 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2)
1287 {
1288 	return t1 > t2 || (t1 == t2 && after(seq1, seq2));
1289 }
1290 
1291 /* These functions determine how the current flow behaves in respect of SACK
1292  * handling. SACK is negotiated with the peer, and therefore it can vary
1293  * between different flows.
1294  *
1295  * tcp_is_sack - SACK enabled
1296  * tcp_is_reno - No SACK
1297  */
1298 static inline int tcp_is_sack(const struct tcp_sock *tp)
1299 {
1300 	return likely(tp->rx_opt.sack_ok);
1301 }
1302 
1303 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1304 {
1305 	return !tcp_is_sack(tp);
1306 }
1307 
1308 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1309 {
1310 	return tp->sacked_out + tp->lost_out;
1311 }
1312 
1313 /* This determines how many packets are "in the network" to the best
1314  * of our knowledge.  In many cases it is conservative, but where
1315  * detailed information is available from the receiver (via SACK
1316  * blocks etc.) we can make more aggressive calculations.
1317  *
1318  * Use this for decisions involving congestion control, use just
1319  * tp->packets_out to determine if the send queue is empty or not.
1320  *
1321  * Read this equation as:
1322  *
1323  *	"Packets sent once on transmission queue" MINUS
1324  *	"Packets left network, but not honestly ACKed yet" PLUS
1325  *	"Packets fast retransmitted"
1326  */
1327 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1328 {
1329 	return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1330 }
1331 
1332 #define TCP_INFINITE_SSTHRESH	0x7fffffff
1333 
1334 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp)
1335 {
1336 	return tp->snd_cwnd;
1337 }
1338 
1339 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val)
1340 {
1341 	WARN_ON_ONCE((int)val <= 0);
1342 	tp->snd_cwnd = val;
1343 }
1344 
1345 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1346 {
1347 	return tcp_snd_cwnd(tp) < tp->snd_ssthresh;
1348 }
1349 
1350 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1351 {
1352 	return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1353 }
1354 
1355 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1356 {
1357 	return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1358 	       (1 << inet_csk(sk)->icsk_ca_state);
1359 }
1360 
1361 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1362  * The exception is cwnd reduction phase, when cwnd is decreasing towards
1363  * ssthresh.
1364  */
1365 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1366 {
1367 	const struct tcp_sock *tp = tcp_sk(sk);
1368 
1369 	if (tcp_in_cwnd_reduction(sk))
1370 		return tp->snd_ssthresh;
1371 	else
1372 		return max(tp->snd_ssthresh,
1373 			   ((tcp_snd_cwnd(tp) >> 1) +
1374 			    (tcp_snd_cwnd(tp) >> 2)));
1375 }
1376 
1377 /* Use define here intentionally to get WARN_ON location shown at the caller */
1378 #define tcp_verify_left_out(tp)	WARN_ON(tcp_left_out(tp) > tp->packets_out)
1379 
1380 void tcp_enter_cwr(struct sock *sk);
1381 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1382 
1383 /* The maximum number of MSS of available cwnd for which TSO defers
1384  * sending if not using sysctl_tcp_tso_win_divisor.
1385  */
1386 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1387 {
1388 	return 3;
1389 }
1390 
1391 /* Returns end sequence number of the receiver's advertised window */
1392 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1393 {
1394 	return tp->snd_una + tp->snd_wnd;
1395 }
1396 
1397 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1398  * flexible approach. The RFC suggests cwnd should not be raised unless
1399  * it was fully used previously. And that's exactly what we do in
1400  * congestion avoidance mode. But in slow start we allow cwnd to grow
1401  * as long as the application has used half the cwnd.
1402  * Example :
1403  *    cwnd is 10 (IW10), but application sends 9 frames.
1404  *    We allow cwnd to reach 18 when all frames are ACKed.
1405  * This check is safe because it's as aggressive as slow start which already
1406  * risks 100% overshoot. The advantage is that we discourage application to
1407  * either send more filler packets or data to artificially blow up the cwnd
1408  * usage, and allow application-limited process to probe bw more aggressively.
1409  */
1410 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1411 {
1412 	const struct tcp_sock *tp = tcp_sk(sk);
1413 
1414 	if (tp->is_cwnd_limited)
1415 		return true;
1416 
1417 	/* If in slow start, ensure cwnd grows to twice what was ACKed. */
1418 	if (tcp_in_slow_start(tp))
1419 		return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out;
1420 
1421 	return false;
1422 }
1423 
1424 /* BBR congestion control needs pacing.
1425  * Same remark for SO_MAX_PACING_RATE.
1426  * sch_fq packet scheduler is efficiently handling pacing,
1427  * but is not always installed/used.
1428  * Return true if TCP stack should pace packets itself.
1429  */
1430 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1431 {
1432 	return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1433 }
1434 
1435 /* Estimates in how many jiffies next packet for this flow can be sent.
1436  * Scheduling a retransmit timer too early would be silly.
1437  */
1438 static inline unsigned long tcp_pacing_delay(const struct sock *sk)
1439 {
1440 	s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
1441 
1442 	return delay > 0 ? nsecs_to_jiffies(delay) : 0;
1443 }
1444 
1445 static inline void tcp_reset_xmit_timer(struct sock *sk,
1446 					const int what,
1447 					unsigned long when,
1448 					bool pace_delay)
1449 {
1450 	if (pace_delay)
1451 		when += tcp_pacing_delay(sk);
1452 	inet_csk_reset_xmit_timer(sk, what, when,
1453 				  tcp_rto_max(sk));
1454 }
1455 
1456 /* Something is really bad, we could not queue an additional packet,
1457  * because qdisc is full or receiver sent a 0 window, or we are paced.
1458  * We do not want to add fuel to the fire, or abort too early,
1459  * so make sure the timer we arm now is at least 200ms in the future,
1460  * regardless of current icsk_rto value (as it could be ~2ms)
1461  */
1462 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1463 {
1464 	return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1465 }
1466 
1467 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1468 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1469 					    unsigned long max_when)
1470 {
1471 	u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1,
1472 			   inet_csk(sk)->icsk_backoff);
1473 	u64 when = (u64)tcp_probe0_base(sk) << backoff;
1474 
1475 	return (unsigned long)min_t(u64, when, max_when);
1476 }
1477 
1478 static inline void tcp_check_probe_timer(struct sock *sk)
1479 {
1480 	if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1481 		tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1482 				     tcp_probe0_base(sk), true);
1483 }
1484 
1485 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1486 {
1487 	tp->snd_wl1 = seq;
1488 }
1489 
1490 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1491 {
1492 	tp->snd_wl1 = seq;
1493 }
1494 
1495 /*
1496  * Calculate(/check) TCP checksum
1497  */
1498 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1499 				   __be32 daddr, __wsum base)
1500 {
1501 	return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1502 }
1503 
1504 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1505 {
1506 	return !skb_csum_unnecessary(skb) &&
1507 		__skb_checksum_complete(skb);
1508 }
1509 
1510 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb,
1511 		     enum skb_drop_reason *reason);
1512 
1513 
1514 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1515 void tcp_set_state(struct sock *sk, int state);
1516 void tcp_done(struct sock *sk);
1517 int tcp_abort(struct sock *sk, int err);
1518 
1519 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1520 {
1521 	rx_opt->dsack = 0;
1522 	rx_opt->num_sacks = 0;
1523 }
1524 
1525 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1526 
1527 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1528 {
1529 	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1530 	struct tcp_sock *tp = tcp_sk(sk);
1531 	s32 delta;
1532 
1533 	if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) ||
1534 	    tp->packets_out || ca_ops->cong_control)
1535 		return;
1536 	delta = tcp_jiffies32 - tp->lsndtime;
1537 	if (delta > inet_csk(sk)->icsk_rto)
1538 		tcp_cwnd_restart(sk, delta);
1539 }
1540 
1541 /* Determine a window scaling and initial window to offer. */
1542 void tcp_select_initial_window(const struct sock *sk, int __space,
1543 			       __u32 mss, __u32 *rcv_wnd,
1544 			       __u32 *window_clamp, int wscale_ok,
1545 			       __u8 *rcv_wscale, __u32 init_rcv_wnd);
1546 
1547 static inline int __tcp_win_from_space(u8 scaling_ratio, int space)
1548 {
1549 	s64 scaled_space = (s64)space * scaling_ratio;
1550 
1551 	return scaled_space >> TCP_RMEM_TO_WIN_SCALE;
1552 }
1553 
1554 static inline int tcp_win_from_space(const struct sock *sk, int space)
1555 {
1556 	return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space);
1557 }
1558 
1559 /* inverse of __tcp_win_from_space() */
1560 static inline int __tcp_space_from_win(u8 scaling_ratio, int win)
1561 {
1562 	u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE;
1563 
1564 	do_div(val, scaling_ratio);
1565 	return val;
1566 }
1567 
1568 static inline int tcp_space_from_win(const struct sock *sk, int win)
1569 {
1570 	return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win);
1571 }
1572 
1573 /* Assume a 50% default for skb->len/skb->truesize ratio.
1574  * This may be adjusted later in tcp_measure_rcv_mss().
1575  */
1576 #define TCP_DEFAULT_SCALING_RATIO (1 << (TCP_RMEM_TO_WIN_SCALE - 1))
1577 
1578 static inline void tcp_scaling_ratio_init(struct sock *sk)
1579 {
1580 	tcp_sk(sk)->scaling_ratio = TCP_DEFAULT_SCALING_RATIO;
1581 }
1582 
1583 /* Note: caller must be prepared to deal with negative returns */
1584 static inline int tcp_space(const struct sock *sk)
1585 {
1586 	return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1587 				  READ_ONCE(sk->sk_backlog.len) -
1588 				  atomic_read(&sk->sk_rmem_alloc));
1589 }
1590 
1591 static inline int tcp_full_space(const struct sock *sk)
1592 {
1593 	return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1594 }
1595 
1596 static inline void __tcp_adjust_rcv_ssthresh(struct sock *sk, u32 new_ssthresh)
1597 {
1598 	int unused_mem = sk_unused_reserved_mem(sk);
1599 	struct tcp_sock *tp = tcp_sk(sk);
1600 
1601 	tp->rcv_ssthresh = min(tp->rcv_ssthresh, new_ssthresh);
1602 	if (unused_mem)
1603 		tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh,
1604 					 tcp_win_from_space(sk, unused_mem));
1605 }
1606 
1607 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk)
1608 {
1609 	__tcp_adjust_rcv_ssthresh(sk, 4U * tcp_sk(sk)->advmss);
1610 }
1611 
1612 void tcp_cleanup_rbuf(struct sock *sk, int copied);
1613 void __tcp_cleanup_rbuf(struct sock *sk, int copied);
1614 
1615 
1616 /* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1617  * If 87.5 % (7/8) of the space has been consumed, we want to override
1618  * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1619  * len/truesize ratio.
1620  */
1621 static inline bool tcp_rmem_pressure(const struct sock *sk)
1622 {
1623 	int rcvbuf, threshold;
1624 
1625 	if (tcp_under_memory_pressure(sk))
1626 		return true;
1627 
1628 	rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1629 	threshold = rcvbuf - (rcvbuf >> 3);
1630 
1631 	return atomic_read(&sk->sk_rmem_alloc) > threshold;
1632 }
1633 
1634 static inline bool tcp_epollin_ready(const struct sock *sk, int target)
1635 {
1636 	const struct tcp_sock *tp = tcp_sk(sk);
1637 	int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq);
1638 
1639 	if (avail <= 0)
1640 		return false;
1641 
1642 	return (avail >= target) || tcp_rmem_pressure(sk) ||
1643 	       (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss);
1644 }
1645 
1646 extern void tcp_openreq_init_rwin(struct request_sock *req,
1647 				  const struct sock *sk_listener,
1648 				  const struct dst_entry *dst);
1649 
1650 void tcp_enter_memory_pressure(struct sock *sk);
1651 void tcp_leave_memory_pressure(struct sock *sk);
1652 
1653 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1654 {
1655 	struct net *net = sock_net((struct sock *)tp);
1656 	int val;
1657 
1658 	/* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl()
1659 	 * and do_tcp_setsockopt().
1660 	 */
1661 	val = READ_ONCE(tp->keepalive_intvl);
1662 
1663 	return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl);
1664 }
1665 
1666 static inline int keepalive_time_when(const struct tcp_sock *tp)
1667 {
1668 	struct net *net = sock_net((struct sock *)tp);
1669 	int val;
1670 
1671 	/* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */
1672 	val = READ_ONCE(tp->keepalive_time);
1673 
1674 	return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time);
1675 }
1676 
1677 static inline int keepalive_probes(const struct tcp_sock *tp)
1678 {
1679 	struct net *net = sock_net((struct sock *)tp);
1680 	int val;
1681 
1682 	/* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt()
1683 	 * and do_tcp_setsockopt().
1684 	 */
1685 	val = READ_ONCE(tp->keepalive_probes);
1686 
1687 	return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes);
1688 }
1689 
1690 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1691 {
1692 	const struct inet_connection_sock *icsk = &tp->inet_conn;
1693 
1694 	return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1695 			  tcp_jiffies32 - tp->rcv_tstamp);
1696 }
1697 
1698 static inline int tcp_fin_time(const struct sock *sk)
1699 {
1700 	int fin_timeout = tcp_sk(sk)->linger2 ? :
1701 		READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout);
1702 	const int rto = inet_csk(sk)->icsk_rto;
1703 
1704 	if (fin_timeout < (rto << 2) - (rto >> 1))
1705 		fin_timeout = (rto << 2) - (rto >> 1);
1706 
1707 	return fin_timeout;
1708 }
1709 
1710 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1711 				  int paws_win)
1712 {
1713 	if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1714 		return true;
1715 	if (unlikely(!time_before32(ktime_get_seconds(),
1716 				    rx_opt->ts_recent_stamp + TCP_PAWS_WRAP)))
1717 		return true;
1718 	/*
1719 	 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1720 	 * then following tcp messages have valid values. Ignore 0 value,
1721 	 * or else 'negative' tsval might forbid us to accept their packets.
1722 	 */
1723 	if (!rx_opt->ts_recent)
1724 		return true;
1725 	return false;
1726 }
1727 
1728 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1729 				   int rst)
1730 {
1731 	if (tcp_paws_check(rx_opt, 0))
1732 		return false;
1733 
1734 	/* RST segments are not recommended to carry timestamp,
1735 	   and, if they do, it is recommended to ignore PAWS because
1736 	   "their cleanup function should take precedence over timestamps."
1737 	   Certainly, it is mistake. It is necessary to understand the reasons
1738 	   of this constraint to relax it: if peer reboots, clock may go
1739 	   out-of-sync and half-open connections will not be reset.
1740 	   Actually, the problem would be not existing if all
1741 	   the implementations followed draft about maintaining clock
1742 	   via reboots. Linux-2.2 DOES NOT!
1743 
1744 	   However, we can relax time bounds for RST segments to MSL.
1745 	 */
1746 	if (rst && !time_before32(ktime_get_seconds(),
1747 				  rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1748 		return false;
1749 	return true;
1750 }
1751 
1752 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1753 			  int mib_idx, u32 *last_oow_ack_time);
1754 
1755 static inline void tcp_mib_init(struct net *net)
1756 {
1757 	/* See RFC 2012 */
1758 	TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1759 	TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1760 	TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1761 	TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1762 }
1763 
1764 /* from STCP */
1765 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1766 {
1767 	tp->lost_skb_hint = NULL;
1768 }
1769 
1770 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1771 {
1772 	tcp_clear_retrans_hints_partial(tp);
1773 	tp->retransmit_skb_hint = NULL;
1774 }
1775 
1776 #define tcp_md5_addr tcp_ao_addr
1777 
1778 /* - key database */
1779 struct tcp_md5sig_key {
1780 	struct hlist_node	node;
1781 	u8			keylen;
1782 	u8			family; /* AF_INET or AF_INET6 */
1783 	u8			prefixlen;
1784 	u8			flags;
1785 	union tcp_md5_addr	addr;
1786 	int			l3index; /* set if key added with L3 scope */
1787 	u8			key[TCP_MD5SIG_MAXKEYLEN];
1788 	struct rcu_head		rcu;
1789 };
1790 
1791 /* - sock block */
1792 struct tcp_md5sig_info {
1793 	struct hlist_head	head;
1794 	struct rcu_head		rcu;
1795 };
1796 
1797 /* - pseudo header */
1798 struct tcp4_pseudohdr {
1799 	__be32		saddr;
1800 	__be32		daddr;
1801 	__u8		pad;
1802 	__u8		protocol;
1803 	__be16		len;
1804 };
1805 
1806 struct tcp6_pseudohdr {
1807 	struct in6_addr	saddr;
1808 	struct in6_addr daddr;
1809 	__be32		len;
1810 	__be32		protocol;	/* including padding */
1811 };
1812 
1813 union tcp_md5sum_block {
1814 	struct tcp4_pseudohdr ip4;
1815 #if IS_ENABLED(CONFIG_IPV6)
1816 	struct tcp6_pseudohdr ip6;
1817 #endif
1818 };
1819 
1820 /*
1821  * struct tcp_sigpool - per-CPU pool of ahash_requests
1822  * @scratch: per-CPU temporary area, that can be used between
1823  *	     tcp_sigpool_start() and tcp_sigpool_end() to perform
1824  *	     crypto request
1825  * @req: pre-allocated ahash request
1826  */
1827 struct tcp_sigpool {
1828 	void *scratch;
1829 	struct ahash_request *req;
1830 };
1831 
1832 int tcp_sigpool_alloc_ahash(const char *alg, size_t scratch_size);
1833 void tcp_sigpool_get(unsigned int id);
1834 void tcp_sigpool_release(unsigned int id);
1835 int tcp_sigpool_hash_skb_data(struct tcp_sigpool *hp,
1836 			      const struct sk_buff *skb,
1837 			      unsigned int header_len);
1838 
1839 /**
1840  * tcp_sigpool_start - disable bh and start using tcp_sigpool_ahash
1841  * @id: tcp_sigpool that was previously allocated by tcp_sigpool_alloc_ahash()
1842  * @c: returned tcp_sigpool for usage (uninitialized on failure)
1843  *
1844  * Returns: 0 on success, error otherwise.
1845  */
1846 int tcp_sigpool_start(unsigned int id, struct tcp_sigpool *c);
1847 /**
1848  * tcp_sigpool_end - enable bh and stop using tcp_sigpool
1849  * @c: tcp_sigpool context that was returned by tcp_sigpool_start()
1850  */
1851 void tcp_sigpool_end(struct tcp_sigpool *c);
1852 size_t tcp_sigpool_algo(unsigned int id, char *buf, size_t buf_len);
1853 /* - functions */
1854 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1855 			const struct sock *sk, const struct sk_buff *skb);
1856 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1857 		   int family, u8 prefixlen, int l3index, u8 flags,
1858 		   const u8 *newkey, u8 newkeylen);
1859 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr,
1860 		     int family, u8 prefixlen, int l3index,
1861 		     struct tcp_md5sig_key *key);
1862 
1863 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1864 		   int family, u8 prefixlen, int l3index, u8 flags);
1865 void tcp_clear_md5_list(struct sock *sk);
1866 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1867 					 const struct sock *addr_sk);
1868 
1869 #ifdef CONFIG_TCP_MD5SIG
1870 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1871 					   const union tcp_md5_addr *addr,
1872 					   int family, bool any_l3index);
1873 static inline struct tcp_md5sig_key *
1874 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1875 		  const union tcp_md5_addr *addr, int family)
1876 {
1877 	if (!static_branch_unlikely(&tcp_md5_needed.key))
1878 		return NULL;
1879 	return __tcp_md5_do_lookup(sk, l3index, addr, family, false);
1880 }
1881 
1882 static inline struct tcp_md5sig_key *
1883 tcp_md5_do_lookup_any_l3index(const struct sock *sk,
1884 			      const union tcp_md5_addr *addr, int family)
1885 {
1886 	if (!static_branch_unlikely(&tcp_md5_needed.key))
1887 		return NULL;
1888 	return __tcp_md5_do_lookup(sk, 0, addr, family, true);
1889 }
1890 
1891 #define tcp_twsk_md5_key(twsk)	((twsk)->tw_md5_key)
1892 #else
1893 static inline struct tcp_md5sig_key *
1894 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1895 		  const union tcp_md5_addr *addr, int family)
1896 {
1897 	return NULL;
1898 }
1899 
1900 static inline struct tcp_md5sig_key *
1901 tcp_md5_do_lookup_any_l3index(const struct sock *sk,
1902 			      const union tcp_md5_addr *addr, int family)
1903 {
1904 	return NULL;
1905 }
1906 
1907 #define tcp_twsk_md5_key(twsk)	NULL
1908 #endif
1909 
1910 int tcp_md5_alloc_sigpool(void);
1911 void tcp_md5_release_sigpool(void);
1912 void tcp_md5_add_sigpool(void);
1913 extern int tcp_md5_sigpool_id;
1914 
1915 int tcp_md5_hash_key(struct tcp_sigpool *hp,
1916 		     const struct tcp_md5sig_key *key);
1917 
1918 /* From tcp_fastopen.c */
1919 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1920 			    struct tcp_fastopen_cookie *cookie);
1921 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1922 			    struct tcp_fastopen_cookie *cookie, bool syn_lost,
1923 			    u16 try_exp);
1924 struct tcp_fastopen_request {
1925 	/* Fast Open cookie. Size 0 means a cookie request */
1926 	struct tcp_fastopen_cookie	cookie;
1927 	struct msghdr			*data;  /* data in MSG_FASTOPEN */
1928 	size_t				size;
1929 	int				copied;	/* queued in tcp_connect() */
1930 	struct ubuf_info		*uarg;
1931 };
1932 void tcp_free_fastopen_req(struct tcp_sock *tp);
1933 void tcp_fastopen_destroy_cipher(struct sock *sk);
1934 void tcp_fastopen_ctx_destroy(struct net *net);
1935 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1936 			      void *primary_key, void *backup_key);
1937 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
1938 			    u64 *key);
1939 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1940 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1941 			      struct request_sock *req,
1942 			      struct tcp_fastopen_cookie *foc,
1943 			      const struct dst_entry *dst);
1944 void tcp_fastopen_init_key_once(struct net *net);
1945 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1946 			     struct tcp_fastopen_cookie *cookie);
1947 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1948 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1949 #define TCP_FASTOPEN_KEY_MAX 2
1950 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1951 	(TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1952 
1953 /* Fastopen key context */
1954 struct tcp_fastopen_context {
1955 	siphash_key_t	key[TCP_FASTOPEN_KEY_MAX];
1956 	int		num;
1957 	struct rcu_head	rcu;
1958 };
1959 
1960 void tcp_fastopen_active_disable(struct sock *sk);
1961 bool tcp_fastopen_active_should_disable(struct sock *sk);
1962 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1963 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1964 
1965 /* Caller needs to wrap with rcu_read_(un)lock() */
1966 static inline
1967 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1968 {
1969 	struct tcp_fastopen_context *ctx;
1970 
1971 	ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1972 	if (!ctx)
1973 		ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1974 	return ctx;
1975 }
1976 
1977 static inline
1978 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1979 			       const struct tcp_fastopen_cookie *orig)
1980 {
1981 	if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1982 	    orig->len == foc->len &&
1983 	    !memcmp(orig->val, foc->val, foc->len))
1984 		return true;
1985 	return false;
1986 }
1987 
1988 static inline
1989 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1990 {
1991 	return ctx->num;
1992 }
1993 
1994 /* Latencies incurred by various limits for a sender. They are
1995  * chronograph-like stats that are mutually exclusive.
1996  */
1997 enum tcp_chrono {
1998 	TCP_CHRONO_UNSPEC,
1999 	TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
2000 	TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
2001 	TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
2002 	__TCP_CHRONO_MAX,
2003 };
2004 
2005 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
2006 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
2007 
2008 /* This helper is needed, because skb->tcp_tsorted_anchor uses
2009  * the same memory storage than skb->destructor/_skb_refdst
2010  */
2011 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
2012 {
2013 	skb->destructor = NULL;
2014 	skb->_skb_refdst = 0UL;
2015 }
2016 
2017 #define tcp_skb_tsorted_save(skb) {		\
2018 	unsigned long _save = skb->_skb_refdst;	\
2019 	skb->_skb_refdst = 0UL;
2020 
2021 #define tcp_skb_tsorted_restore(skb)		\
2022 	skb->_skb_refdst = _save;		\
2023 }
2024 
2025 void tcp_write_queue_purge(struct sock *sk);
2026 
2027 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
2028 {
2029 	return skb_rb_first(&sk->tcp_rtx_queue);
2030 }
2031 
2032 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
2033 {
2034 	return skb_rb_last(&sk->tcp_rtx_queue);
2035 }
2036 
2037 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
2038 {
2039 	return skb_peek_tail(&sk->sk_write_queue);
2040 }
2041 
2042 #define tcp_for_write_queue_from_safe(skb, tmp, sk)			\
2043 	skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
2044 
2045 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
2046 {
2047 	return skb_peek(&sk->sk_write_queue);
2048 }
2049 
2050 static inline bool tcp_skb_is_last(const struct sock *sk,
2051 				   const struct sk_buff *skb)
2052 {
2053 	return skb_queue_is_last(&sk->sk_write_queue, skb);
2054 }
2055 
2056 /**
2057  * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
2058  * @sk: socket
2059  *
2060  * Since the write queue can have a temporary empty skb in it,
2061  * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
2062  */
2063 static inline bool tcp_write_queue_empty(const struct sock *sk)
2064 {
2065 	const struct tcp_sock *tp = tcp_sk(sk);
2066 
2067 	return tp->write_seq == tp->snd_nxt;
2068 }
2069 
2070 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
2071 {
2072 	return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
2073 }
2074 
2075 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
2076 {
2077 	return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
2078 }
2079 
2080 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
2081 {
2082 	__skb_queue_tail(&sk->sk_write_queue, skb);
2083 
2084 	/* Queue it, remembering where we must start sending. */
2085 	if (sk->sk_write_queue.next == skb)
2086 		tcp_chrono_start(sk, TCP_CHRONO_BUSY);
2087 }
2088 
2089 /* Insert new before skb on the write queue of sk.  */
2090 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
2091 						  struct sk_buff *skb,
2092 						  struct sock *sk)
2093 {
2094 	__skb_queue_before(&sk->sk_write_queue, skb, new);
2095 }
2096 
2097 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
2098 {
2099 	tcp_skb_tsorted_anchor_cleanup(skb);
2100 	__skb_unlink(skb, &sk->sk_write_queue);
2101 }
2102 
2103 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
2104 
2105 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
2106 {
2107 	tcp_skb_tsorted_anchor_cleanup(skb);
2108 	rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
2109 }
2110 
2111 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
2112 {
2113 	list_del(&skb->tcp_tsorted_anchor);
2114 	tcp_rtx_queue_unlink(skb, sk);
2115 	tcp_wmem_free_skb(sk, skb);
2116 }
2117 
2118 static inline void tcp_write_collapse_fence(struct sock *sk)
2119 {
2120 	struct sk_buff *skb = tcp_write_queue_tail(sk);
2121 
2122 	if (skb)
2123 		TCP_SKB_CB(skb)->eor = 1;
2124 }
2125 
2126 static inline void tcp_push_pending_frames(struct sock *sk)
2127 {
2128 	if (tcp_send_head(sk)) {
2129 		struct tcp_sock *tp = tcp_sk(sk);
2130 
2131 		__tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
2132 	}
2133 }
2134 
2135 /* Start sequence of the skb just after the highest skb with SACKed
2136  * bit, valid only if sacked_out > 0 or when the caller has ensured
2137  * validity by itself.
2138  */
2139 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
2140 {
2141 	if (!tp->sacked_out)
2142 		return tp->snd_una;
2143 
2144 	if (tp->highest_sack == NULL)
2145 		return tp->snd_nxt;
2146 
2147 	return TCP_SKB_CB(tp->highest_sack)->seq;
2148 }
2149 
2150 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
2151 {
2152 	tcp_sk(sk)->highest_sack = skb_rb_next(skb);
2153 }
2154 
2155 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
2156 {
2157 	return tcp_sk(sk)->highest_sack;
2158 }
2159 
2160 static inline void tcp_highest_sack_reset(struct sock *sk)
2161 {
2162 	tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
2163 }
2164 
2165 /* Called when old skb is about to be deleted and replaced by new skb */
2166 static inline void tcp_highest_sack_replace(struct sock *sk,
2167 					    struct sk_buff *old,
2168 					    struct sk_buff *new)
2169 {
2170 	if (old == tcp_highest_sack(sk))
2171 		tcp_sk(sk)->highest_sack = new;
2172 }
2173 
2174 /* This helper checks if socket has IP_TRANSPARENT set */
2175 static inline bool inet_sk_transparent(const struct sock *sk)
2176 {
2177 	switch (sk->sk_state) {
2178 	case TCP_TIME_WAIT:
2179 		return inet_twsk(sk)->tw_transparent;
2180 	case TCP_NEW_SYN_RECV:
2181 		return inet_rsk(inet_reqsk(sk))->no_srccheck;
2182 	}
2183 	return inet_test_bit(TRANSPARENT, sk);
2184 }
2185 
2186 /* Determines whether this is a thin stream (which may suffer from
2187  * increased latency). Used to trigger latency-reducing mechanisms.
2188  */
2189 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
2190 {
2191 	return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
2192 }
2193 
2194 /* /proc */
2195 enum tcp_seq_states {
2196 	TCP_SEQ_STATE_LISTENING,
2197 	TCP_SEQ_STATE_ESTABLISHED,
2198 };
2199 
2200 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
2201 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
2202 void tcp_seq_stop(struct seq_file *seq, void *v);
2203 
2204 struct tcp_seq_afinfo {
2205 	sa_family_t			family;
2206 };
2207 
2208 struct tcp_iter_state {
2209 	struct seq_net_private	p;
2210 	enum tcp_seq_states	state;
2211 	struct sock		*syn_wait_sk;
2212 	int			bucket, offset, sbucket, num;
2213 	loff_t			last_pos;
2214 };
2215 
2216 extern struct request_sock_ops tcp_request_sock_ops;
2217 extern struct request_sock_ops tcp6_request_sock_ops;
2218 
2219 void tcp_v4_destroy_sock(struct sock *sk);
2220 
2221 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
2222 				netdev_features_t features);
2223 struct tcphdr *tcp_gro_pull_header(struct sk_buff *skb);
2224 struct sk_buff *tcp_gro_lookup(struct list_head *head, struct tcphdr *th);
2225 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb,
2226 				struct tcphdr *th);
2227 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
2228 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
2229 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff));
2230 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb));
2231 #ifdef CONFIG_INET
2232 void tcp_gro_complete(struct sk_buff *skb);
2233 #else
2234 static inline void tcp_gro_complete(struct sk_buff *skb) { }
2235 #endif
2236 
2237 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
2238 
2239 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
2240 {
2241 	struct net *net = sock_net((struct sock *)tp);
2242 	u32 val;
2243 
2244 	val = READ_ONCE(tp->notsent_lowat);
2245 
2246 	return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
2247 }
2248 
2249 bool tcp_stream_memory_free(const struct sock *sk, int wake);
2250 
2251 #ifdef CONFIG_PROC_FS
2252 int tcp4_proc_init(void);
2253 void tcp4_proc_exit(void);
2254 #endif
2255 
2256 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
2257 int tcp_conn_request(struct request_sock_ops *rsk_ops,
2258 		     const struct tcp_request_sock_ops *af_ops,
2259 		     struct sock *sk, struct sk_buff *skb);
2260 
2261 /* TCP af-specific functions */
2262 struct tcp_sock_af_ops {
2263 #ifdef CONFIG_TCP_MD5SIG
2264 	struct tcp_md5sig_key	*(*md5_lookup) (const struct sock *sk,
2265 						const struct sock *addr_sk);
2266 	int		(*calc_md5_hash)(char *location,
2267 					 const struct tcp_md5sig_key *md5,
2268 					 const struct sock *sk,
2269 					 const struct sk_buff *skb);
2270 	int		(*md5_parse)(struct sock *sk,
2271 				     int optname,
2272 				     sockptr_t optval,
2273 				     int optlen);
2274 #endif
2275 #ifdef CONFIG_TCP_AO
2276 	int (*ao_parse)(struct sock *sk, int optname, sockptr_t optval, int optlen);
2277 	struct tcp_ao_key *(*ao_lookup)(const struct sock *sk,
2278 					struct sock *addr_sk,
2279 					int sndid, int rcvid);
2280 	int (*ao_calc_key_sk)(struct tcp_ao_key *mkt, u8 *key,
2281 			      const struct sock *sk,
2282 			      __be32 sisn, __be32 disn, bool send);
2283 	int (*calc_ao_hash)(char *location, struct tcp_ao_key *ao,
2284 			    const struct sock *sk, const struct sk_buff *skb,
2285 			    const u8 *tkey, int hash_offset, u32 sne);
2286 #endif
2287 };
2288 
2289 struct tcp_request_sock_ops {
2290 	u16 mss_clamp;
2291 #ifdef CONFIG_TCP_MD5SIG
2292 	struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
2293 						 const struct sock *addr_sk);
2294 	int		(*calc_md5_hash) (char *location,
2295 					  const struct tcp_md5sig_key *md5,
2296 					  const struct sock *sk,
2297 					  const struct sk_buff *skb);
2298 #endif
2299 #ifdef CONFIG_TCP_AO
2300 	struct tcp_ao_key *(*ao_lookup)(const struct sock *sk,
2301 					struct request_sock *req,
2302 					int sndid, int rcvid);
2303 	int (*ao_calc_key)(struct tcp_ao_key *mkt, u8 *key, struct request_sock *sk);
2304 	int (*ao_synack_hash)(char *ao_hash, struct tcp_ao_key *mkt,
2305 			      struct request_sock *req, const struct sk_buff *skb,
2306 			      int hash_offset, u32 sne);
2307 #endif
2308 #ifdef CONFIG_SYN_COOKIES
2309 	__u32 (*cookie_init_seq)(const struct sk_buff *skb,
2310 				 __u16 *mss);
2311 #endif
2312 	struct dst_entry *(*route_req)(const struct sock *sk,
2313 				       struct sk_buff *skb,
2314 				       struct flowi *fl,
2315 				       struct request_sock *req,
2316 				       u32 tw_isn);
2317 	u32 (*init_seq)(const struct sk_buff *skb);
2318 	u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
2319 	int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2320 			   struct flowi *fl, struct request_sock *req,
2321 			   struct tcp_fastopen_cookie *foc,
2322 			   enum tcp_synack_type synack_type,
2323 			   struct sk_buff *syn_skb);
2324 };
2325 
2326 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2327 #if IS_ENABLED(CONFIG_IPV6)
2328 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2329 #endif
2330 
2331 #ifdef CONFIG_SYN_COOKIES
2332 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2333 					 const struct sock *sk, struct sk_buff *skb,
2334 					 __u16 *mss)
2335 {
2336 	tcp_synq_overflow(sk);
2337 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2338 	return ops->cookie_init_seq(skb, mss);
2339 }
2340 #else
2341 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2342 					 const struct sock *sk, struct sk_buff *skb,
2343 					 __u16 *mss)
2344 {
2345 	return 0;
2346 }
2347 #endif
2348 
2349 struct tcp_key {
2350 	union {
2351 		struct {
2352 			struct tcp_ao_key *ao_key;
2353 			char *traffic_key;
2354 			u32 sne;
2355 			u8 rcv_next;
2356 		};
2357 		struct tcp_md5sig_key *md5_key;
2358 	};
2359 	enum {
2360 		TCP_KEY_NONE = 0,
2361 		TCP_KEY_MD5,
2362 		TCP_KEY_AO,
2363 	} type;
2364 };
2365 
2366 static inline void tcp_get_current_key(const struct sock *sk,
2367 				       struct tcp_key *out)
2368 {
2369 #if defined(CONFIG_TCP_AO) || defined(CONFIG_TCP_MD5SIG)
2370 	const struct tcp_sock *tp = tcp_sk(sk);
2371 #endif
2372 
2373 #ifdef CONFIG_TCP_AO
2374 	if (static_branch_unlikely(&tcp_ao_needed.key)) {
2375 		struct tcp_ao_info *ao;
2376 
2377 		ao = rcu_dereference_protected(tp->ao_info,
2378 					       lockdep_sock_is_held(sk));
2379 		if (ao) {
2380 			out->ao_key = READ_ONCE(ao->current_key);
2381 			out->type = TCP_KEY_AO;
2382 			return;
2383 		}
2384 	}
2385 #endif
2386 #ifdef CONFIG_TCP_MD5SIG
2387 	if (static_branch_unlikely(&tcp_md5_needed.key) &&
2388 	    rcu_access_pointer(tp->md5sig_info)) {
2389 		out->md5_key = tp->af_specific->md5_lookup(sk, sk);
2390 		if (out->md5_key) {
2391 			out->type = TCP_KEY_MD5;
2392 			return;
2393 		}
2394 	}
2395 #endif
2396 	out->type = TCP_KEY_NONE;
2397 }
2398 
2399 static inline bool tcp_key_is_md5(const struct tcp_key *key)
2400 {
2401 	if (static_branch_tcp_md5())
2402 		return key->type == TCP_KEY_MD5;
2403 	return false;
2404 }
2405 
2406 static inline bool tcp_key_is_ao(const struct tcp_key *key)
2407 {
2408 	if (static_branch_tcp_ao())
2409 		return key->type == TCP_KEY_AO;
2410 	return false;
2411 }
2412 
2413 int tcpv4_offload_init(void);
2414 
2415 void tcp_v4_init(void);
2416 void tcp_init(void);
2417 
2418 /* tcp_recovery.c */
2419 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2420 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2421 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2422 				u32 reo_wnd);
2423 extern bool tcp_rack_mark_lost(struct sock *sk);
2424 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2425 			     u64 xmit_time);
2426 extern void tcp_rack_reo_timeout(struct sock *sk);
2427 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2428 
2429 /* tcp_plb.c */
2430 
2431 /*
2432  * Scaling factor for fractions in PLB. For example, tcp_plb_update_state
2433  * expects cong_ratio which represents fraction of traffic that experienced
2434  * congestion over a single RTT. In order to avoid floating point operations,
2435  * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in.
2436  */
2437 #define TCP_PLB_SCALE 8
2438 
2439 /* State for PLB (Protective Load Balancing) for a single TCP connection. */
2440 struct tcp_plb_state {
2441 	u8	consec_cong_rounds:5, /* consecutive congested rounds */
2442 		unused:3;
2443 	u32	pause_until; /* jiffies32 when PLB can resume rerouting */
2444 };
2445 
2446 static inline void tcp_plb_init(const struct sock *sk,
2447 				struct tcp_plb_state *plb)
2448 {
2449 	plb->consec_cong_rounds = 0;
2450 	plb->pause_until = 0;
2451 }
2452 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb,
2453 			  const int cong_ratio);
2454 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb);
2455 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb);
2456 
2457 static inline void tcp_warn_once(const struct sock *sk, bool cond, const char *str)
2458 {
2459 	WARN_ONCE(cond,
2460 		  "%scwn:%u out:%u sacked:%u lost:%u retrans:%u tlp_high_seq:%u sk_state:%u ca_state:%u advmss:%u mss_cache:%u pmtu:%u\n",
2461 		  str,
2462 		  tcp_snd_cwnd(tcp_sk(sk)),
2463 		  tcp_sk(sk)->packets_out, tcp_sk(sk)->sacked_out,
2464 		  tcp_sk(sk)->lost_out, tcp_sk(sk)->retrans_out,
2465 		  tcp_sk(sk)->tlp_high_seq, sk->sk_state,
2466 		  inet_csk(sk)->icsk_ca_state,
2467 		  tcp_sk(sk)->advmss, tcp_sk(sk)->mss_cache,
2468 		  inet_csk(sk)->icsk_pmtu_cookie);
2469 }
2470 
2471 /* At how many usecs into the future should the RTO fire? */
2472 static inline s64 tcp_rto_delta_us(const struct sock *sk)
2473 {
2474 	const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2475 	u32 rto = inet_csk(sk)->icsk_rto;
2476 
2477 	if (likely(skb)) {
2478 		u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2479 
2480 		return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2481 	} else {
2482 		tcp_warn_once(sk, 1, "rtx queue empty: ");
2483 		return jiffies_to_usecs(rto);
2484 	}
2485 
2486 }
2487 
2488 /*
2489  * Save and compile IPv4 options, return a pointer to it
2490  */
2491 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2492 							 struct sk_buff *skb)
2493 {
2494 	const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2495 	struct ip_options_rcu *dopt = NULL;
2496 
2497 	if (opt->optlen) {
2498 		int opt_size = sizeof(*dopt) + opt->optlen;
2499 
2500 		dopt = kmalloc(opt_size, GFP_ATOMIC);
2501 		if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2502 			kfree(dopt);
2503 			dopt = NULL;
2504 		}
2505 	}
2506 	return dopt;
2507 }
2508 
2509 /* locally generated TCP pure ACKs have skb->truesize == 2
2510  * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2511  * This is much faster than dissecting the packet to find out.
2512  * (Think of GRE encapsulations, IPv4, IPv6, ...)
2513  */
2514 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2515 {
2516 	return skb->truesize == 2;
2517 }
2518 
2519 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2520 {
2521 	skb->truesize = 2;
2522 }
2523 
2524 static inline int tcp_inq(struct sock *sk)
2525 {
2526 	struct tcp_sock *tp = tcp_sk(sk);
2527 	int answ;
2528 
2529 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2530 		answ = 0;
2531 	} else if (sock_flag(sk, SOCK_URGINLINE) ||
2532 		   !tp->urg_data ||
2533 		   before(tp->urg_seq, tp->copied_seq) ||
2534 		   !before(tp->urg_seq, tp->rcv_nxt)) {
2535 
2536 		answ = tp->rcv_nxt - tp->copied_seq;
2537 
2538 		/* Subtract 1, if FIN was received */
2539 		if (answ && sock_flag(sk, SOCK_DONE))
2540 			answ--;
2541 	} else {
2542 		answ = tp->urg_seq - tp->copied_seq;
2543 	}
2544 
2545 	return answ;
2546 }
2547 
2548 int tcp_peek_len(struct socket *sock);
2549 
2550 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2551 {
2552 	u16 segs_in;
2553 
2554 	segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2555 
2556 	/* We update these fields while other threads might
2557 	 * read them from tcp_get_info()
2558 	 */
2559 	WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in);
2560 	if (skb->len > tcp_hdrlen(skb))
2561 		WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in);
2562 }
2563 
2564 /*
2565  * TCP listen path runs lockless.
2566  * We forced "struct sock" to be const qualified to make sure
2567  * we don't modify one of its field by mistake.
2568  * Here, we increment sk_drops which is an atomic_t, so we can safely
2569  * make sock writable again.
2570  */
2571 static inline void tcp_listendrop(const struct sock *sk)
2572 {
2573 	atomic_inc(&((struct sock *)sk)->sk_drops);
2574 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2575 }
2576 
2577 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2578 
2579 /*
2580  * Interface for adding Upper Level Protocols over TCP
2581  */
2582 
2583 #define TCP_ULP_NAME_MAX	16
2584 #define TCP_ULP_MAX		128
2585 #define TCP_ULP_BUF_MAX		(TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2586 
2587 struct tcp_ulp_ops {
2588 	struct list_head	list;
2589 
2590 	/* initialize ulp */
2591 	int (*init)(struct sock *sk);
2592 	/* update ulp */
2593 	void (*update)(struct sock *sk, struct proto *p,
2594 		       void (*write_space)(struct sock *sk));
2595 	/* cleanup ulp */
2596 	void (*release)(struct sock *sk);
2597 	/* diagnostic */
2598 	int (*get_info)(struct sock *sk, struct sk_buff *skb);
2599 	size_t (*get_info_size)(const struct sock *sk);
2600 	/* clone ulp */
2601 	void (*clone)(const struct request_sock *req, struct sock *newsk,
2602 		      const gfp_t priority);
2603 
2604 	char		name[TCP_ULP_NAME_MAX];
2605 	struct module	*owner;
2606 };
2607 int tcp_register_ulp(struct tcp_ulp_ops *type);
2608 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2609 int tcp_set_ulp(struct sock *sk, const char *name);
2610 void tcp_get_available_ulp(char *buf, size_t len);
2611 void tcp_cleanup_ulp(struct sock *sk);
2612 void tcp_update_ulp(struct sock *sk, struct proto *p,
2613 		    void (*write_space)(struct sock *sk));
2614 
2615 #define MODULE_ALIAS_TCP_ULP(name)				\
2616 	__MODULE_INFO(alias, alias_userspace, name);		\
2617 	__MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2618 
2619 #ifdef CONFIG_NET_SOCK_MSG
2620 struct sk_msg;
2621 struct sk_psock;
2622 
2623 #ifdef CONFIG_BPF_SYSCALL
2624 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
2625 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2626 #endif /* CONFIG_BPF_SYSCALL */
2627 
2628 #ifdef CONFIG_INET
2629 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb);
2630 #else
2631 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb)
2632 {
2633 }
2634 #endif
2635 
2636 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress,
2637 			  struct sk_msg *msg, u32 bytes, int flags);
2638 #endif /* CONFIG_NET_SOCK_MSG */
2639 
2640 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG)
2641 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2642 {
2643 }
2644 #endif
2645 
2646 #ifdef CONFIG_CGROUP_BPF
2647 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2648 				      struct sk_buff *skb,
2649 				      unsigned int end_offset)
2650 {
2651 	skops->skb = skb;
2652 	skops->skb_data_end = skb->data + end_offset;
2653 }
2654 #else
2655 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2656 				      struct sk_buff *skb,
2657 				      unsigned int end_offset)
2658 {
2659 }
2660 #endif
2661 
2662 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2663  * is < 0, then the BPF op failed (for example if the loaded BPF
2664  * program does not support the chosen operation or there is no BPF
2665  * program loaded).
2666  */
2667 #ifdef CONFIG_BPF
2668 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2669 {
2670 	struct bpf_sock_ops_kern sock_ops;
2671 	int ret;
2672 
2673 	memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2674 	if (sk_fullsock(sk)) {
2675 		sock_ops.is_fullsock = 1;
2676 		sock_ops.is_locked_tcp_sock = 1;
2677 		sock_owned_by_me(sk);
2678 	}
2679 
2680 	sock_ops.sk = sk;
2681 	sock_ops.op = op;
2682 	if (nargs > 0)
2683 		memcpy(sock_ops.args, args, nargs * sizeof(*args));
2684 
2685 	ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2686 	if (ret == 0)
2687 		ret = sock_ops.reply;
2688 	else
2689 		ret = -1;
2690 	return ret;
2691 }
2692 
2693 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2694 {
2695 	u32 args[2] = {arg1, arg2};
2696 
2697 	return tcp_call_bpf(sk, op, 2, args);
2698 }
2699 
2700 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2701 				    u32 arg3)
2702 {
2703 	u32 args[3] = {arg1, arg2, arg3};
2704 
2705 	return tcp_call_bpf(sk, op, 3, args);
2706 }
2707 
2708 #else
2709 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2710 {
2711 	return -EPERM;
2712 }
2713 
2714 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2715 {
2716 	return -EPERM;
2717 }
2718 
2719 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2720 				    u32 arg3)
2721 {
2722 	return -EPERM;
2723 }
2724 
2725 #endif
2726 
2727 static inline u32 tcp_timeout_init(struct sock *sk)
2728 {
2729 	int timeout;
2730 
2731 	timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2732 
2733 	if (timeout <= 0)
2734 		timeout = TCP_TIMEOUT_INIT;
2735 	return min_t(int, timeout, TCP_RTO_MAX);
2736 }
2737 
2738 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2739 {
2740 	int rwnd;
2741 
2742 	rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2743 
2744 	if (rwnd < 0)
2745 		rwnd = 0;
2746 	return rwnd;
2747 }
2748 
2749 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2750 {
2751 	return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2752 }
2753 
2754 static inline void tcp_bpf_rtt(struct sock *sk, long mrtt, u32 srtt)
2755 {
2756 	if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2757 		tcp_call_bpf_2arg(sk, BPF_SOCK_OPS_RTT_CB, mrtt, srtt);
2758 }
2759 
2760 #if IS_ENABLED(CONFIG_SMC)
2761 extern struct static_key_false tcp_have_smc;
2762 #endif
2763 
2764 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2765 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2766 			     void (*cad)(struct sock *sk, u32 ack_seq));
2767 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2768 void clean_acked_data_flush(void);
2769 #endif
2770 
2771 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2772 static inline void tcp_add_tx_delay(struct sk_buff *skb,
2773 				    const struct tcp_sock *tp)
2774 {
2775 	if (static_branch_unlikely(&tcp_tx_delay_enabled))
2776 		skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2777 }
2778 
2779 /* Compute Earliest Departure Time for some control packets
2780  * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2781  */
2782 static inline u64 tcp_transmit_time(const struct sock *sk)
2783 {
2784 	if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2785 		u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2786 			tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2787 
2788 		return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
2789 	}
2790 	return 0;
2791 }
2792 
2793 static inline int tcp_parse_auth_options(const struct tcphdr *th,
2794 		const u8 **md5_hash, const struct tcp_ao_hdr **aoh)
2795 {
2796 	const u8 *md5_tmp, *ao_tmp;
2797 	int ret;
2798 
2799 	ret = tcp_do_parse_auth_options(th, &md5_tmp, &ao_tmp);
2800 	if (ret)
2801 		return ret;
2802 
2803 	if (md5_hash)
2804 		*md5_hash = md5_tmp;
2805 
2806 	if (aoh) {
2807 		if (!ao_tmp)
2808 			*aoh = NULL;
2809 		else
2810 			*aoh = (struct tcp_ao_hdr *)(ao_tmp - 2);
2811 	}
2812 
2813 	return 0;
2814 }
2815 
2816 static inline bool tcp_ao_required(struct sock *sk, const void *saddr,
2817 				   int family, int l3index, bool stat_inc)
2818 {
2819 #ifdef CONFIG_TCP_AO
2820 	struct tcp_ao_info *ao_info;
2821 	struct tcp_ao_key *ao_key;
2822 
2823 	if (!static_branch_unlikely(&tcp_ao_needed.key))
2824 		return false;
2825 
2826 	ao_info = rcu_dereference_check(tcp_sk(sk)->ao_info,
2827 					lockdep_sock_is_held(sk));
2828 	if (!ao_info)
2829 		return false;
2830 
2831 	ao_key = tcp_ao_do_lookup(sk, l3index, saddr, family, -1, -1);
2832 	if (ao_info->ao_required || ao_key) {
2833 		if (stat_inc) {
2834 			NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOREQUIRED);
2835 			atomic64_inc(&ao_info->counters.ao_required);
2836 		}
2837 		return true;
2838 	}
2839 #endif
2840 	return false;
2841 }
2842 
2843 enum skb_drop_reason tcp_inbound_hash(struct sock *sk,
2844 		const struct request_sock *req, const struct sk_buff *skb,
2845 		const void *saddr, const void *daddr,
2846 		int family, int dif, int sdif);
2847 
2848 #endif	/* _TCP_H */
2849