xref: /linux/include/net/tcp.h (revision 2151003e773c7e7dba4d64bed4bfc483681b5f6a)
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		Definitions for the TCP module.
8  *
9  * Version:	@(#)tcp.h	1.0.5	05/23/93
10  *
11  * Authors:	Ross Biro
12  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13  */
14 #ifndef _TCP_H
15 #define _TCP_H
16 
17 #define FASTRETRANS_DEBUG 1
18 
19 #include <linux/list.h>
20 #include <linux/tcp.h>
21 #include <linux/bug.h>
22 #include <linux/slab.h>
23 #include <linux/cache.h>
24 #include <linux/percpu.h>
25 #include <linux/skbuff.h>
26 #include <linux/kref.h>
27 #include <linux/ktime.h>
28 #include <linux/indirect_call_wrapper.h>
29 
30 #include <net/inet_connection_sock.h>
31 #include <net/inet_timewait_sock.h>
32 #include <net/inet_hashtables.h>
33 #include <net/checksum.h>
34 #include <net/request_sock.h>
35 #include <net/sock_reuseport.h>
36 #include <net/sock.h>
37 #include <net/snmp.h>
38 #include <net/ip.h>
39 #include <net/tcp_states.h>
40 #include <net/tcp_ao.h>
41 #include <net/inet_ecn.h>
42 #include <net/dst.h>
43 #include <net/mptcp.h>
44 
45 #include <linux/seq_file.h>
46 #include <linux/memcontrol.h>
47 #include <linux/bpf-cgroup.h>
48 #include <linux/siphash.h>
49 
50 extern struct inet_hashinfo tcp_hashinfo;
51 
52 DECLARE_PER_CPU(unsigned int, tcp_orphan_count);
53 int tcp_orphan_count_sum(void);
54 
55 DECLARE_PER_CPU(u32, tcp_tw_isn);
56 
57 void tcp_time_wait(struct sock *sk, int state, int timeo);
58 
59 #define MAX_TCP_HEADER	L1_CACHE_ALIGN(128 + MAX_HEADER)
60 #define MAX_TCP_OPTION_SPACE 40
61 #define TCP_MIN_SND_MSS		48
62 #define TCP_MIN_GSO_SIZE	(TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
63 
64 /*
65  * Never offer a window over 32767 without using window scaling. Some
66  * poor stacks do signed 16bit maths!
67  */
68 #define MAX_TCP_WINDOW		32767U
69 
70 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
71 #define TCP_MIN_MSS		88U
72 
73 /* The initial MTU to use for probing */
74 #define TCP_BASE_MSS		1024
75 
76 /* probing interval, default to 10 minutes as per RFC4821 */
77 #define TCP_PROBE_INTERVAL	600
78 
79 /* Specify interval when tcp mtu probing will stop */
80 #define TCP_PROBE_THRESHOLD	8
81 
82 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
83 #define TCP_FASTRETRANS_THRESH 3
84 
85 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
86 #define TCP_MAX_QUICKACKS	16U
87 
88 /* Maximal number of window scale according to RFC1323 */
89 #define TCP_MAX_WSCALE		14U
90 
91 /* urg_data states */
92 #define TCP_URG_VALID	0x0100
93 #define TCP_URG_NOTYET	0x0200
94 #define TCP_URG_READ	0x0400
95 
96 #define TCP_RETR1	3	/*
97 				 * This is how many retries it does before it
98 				 * tries to figure out if the gateway is
99 				 * down. Minimal RFC value is 3; it corresponds
100 				 * to ~3sec-8min depending on RTO.
101 				 */
102 
103 #define TCP_RETR2	15	/*
104 				 * This should take at least
105 				 * 90 minutes to time out.
106 				 * RFC1122 says that the limit is 100 sec.
107 				 * 15 is ~13-30min depending on RTO.
108 				 */
109 
110 #define TCP_SYN_RETRIES	 6	/* This is how many retries are done
111 				 * when active opening a connection.
112 				 * RFC1122 says the minimum retry MUST
113 				 * be at least 180secs.  Nevertheless
114 				 * this value is corresponding to
115 				 * 63secs of retransmission with the
116 				 * current initial RTO.
117 				 */
118 
119 #define TCP_SYNACK_RETRIES 5	/* This is how may retries are done
120 				 * when passive opening a connection.
121 				 * This is corresponding to 31secs of
122 				 * retransmission with the current
123 				 * initial RTO.
124 				 */
125 
126 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
127 				  * state, about 60 seconds	*/
128 #define TCP_FIN_TIMEOUT	TCP_TIMEWAIT_LEN
129                                  /* BSD style FIN_WAIT2 deadlock breaker.
130 				  * It used to be 3min, new value is 60sec,
131 				  * to combine FIN-WAIT-2 timeout with
132 				  * TIME-WAIT timer.
133 				  */
134 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */
135 
136 #define TCP_DELACK_MAX	((unsigned)(HZ/5))	/* maximal time to delay before sending an ACK */
137 static_assert((1 << ATO_BITS) > TCP_DELACK_MAX);
138 
139 #if HZ >= 100
140 #define TCP_DELACK_MIN	((unsigned)(HZ/25))	/* minimal time to delay before sending an ACK */
141 #define TCP_ATO_MIN	((unsigned)(HZ/25))
142 #else
143 #define TCP_DELACK_MIN	4U
144 #define TCP_ATO_MIN	4U
145 #endif
146 #define TCP_RTO_MAX_SEC 120
147 #define TCP_RTO_MAX	((unsigned)(TCP_RTO_MAX_SEC * HZ))
148 #define TCP_RTO_MIN	((unsigned)(HZ / 5))
149 #define TCP_TIMEOUT_MIN	(2U) /* Min timeout for TCP timers in jiffies */
150 
151 #define TCP_TIMEOUT_MIN_US (2*USEC_PER_MSEC) /* Min TCP timeout in microsecs */
152 
153 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ))	/* RFC6298 2.1 initial RTO value	*/
154 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ))	/* RFC 1122 initial RTO value, now
155 						 * used as a fallback RTO for the
156 						 * initial data transmission if no
157 						 * valid RTT sample has been acquired,
158 						 * most likely due to retrans in 3WHS.
159 						 */
160 
161 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
162 					                 * for local resources.
163 					                 */
164 #define TCP_KEEPALIVE_TIME	(120*60*HZ)	/* two hours */
165 #define TCP_KEEPALIVE_PROBES	9		/* Max of 9 keepalive probes	*/
166 #define TCP_KEEPALIVE_INTVL	(75*HZ)
167 
168 #define MAX_TCP_KEEPIDLE	32767
169 #define MAX_TCP_KEEPINTVL	32767
170 #define MAX_TCP_KEEPCNT		127
171 #define MAX_TCP_SYNCNT		127
172 
173 /* Ensure that TCP PAWS checks are relaxed after ~2147 seconds
174  * to avoid overflows. This assumes a clock smaller than 1 Mhz.
175  * Default clock is 1 Khz, tcp_usec_ts uses 1 Mhz.
176  */
177 #define TCP_PAWS_WRAP (INT_MAX / USEC_PER_SEC)
178 
179 #define TCP_PAWS_MSL	60		/* Per-host timestamps are invalidated
180 					 * after this time. It should be equal
181 					 * (or greater than) TCP_TIMEWAIT_LEN
182 					 * to provide reliability equal to one
183 					 * provided by timewait state.
184 					 */
185 #define TCP_PAWS_WINDOW	1		/* Replay window for per-host
186 					 * timestamps. It must be less than
187 					 * minimal timewait lifetime.
188 					 */
189 /*
190  *	TCP option
191  */
192 
193 #define TCPOPT_NOP		1	/* Padding */
194 #define TCPOPT_EOL		0	/* End of options */
195 #define TCPOPT_MSS		2	/* Segment size negotiating */
196 #define TCPOPT_WINDOW		3	/* Window scaling */
197 #define TCPOPT_SACK_PERM        4       /* SACK Permitted */
198 #define TCPOPT_SACK             5       /* SACK Block */
199 #define TCPOPT_TIMESTAMP	8	/* Better RTT estimations/PAWS */
200 #define TCPOPT_MD5SIG		19	/* MD5 Signature (RFC2385) */
201 #define TCPOPT_AO		29	/* Authentication Option (RFC5925) */
202 #define TCPOPT_MPTCP		30	/* Multipath TCP (RFC6824) */
203 #define TCPOPT_FASTOPEN		34	/* Fast open (RFC7413) */
204 #define TCPOPT_EXP		254	/* Experimental */
205 /* Magic number to be after the option value for sharing TCP
206  * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
207  */
208 #define TCPOPT_FASTOPEN_MAGIC	0xF989
209 #define TCPOPT_SMC_MAGIC	0xE2D4C3D9
210 
211 /*
212  *     TCP option lengths
213  */
214 
215 #define TCPOLEN_MSS            4
216 #define TCPOLEN_WINDOW         3
217 #define TCPOLEN_SACK_PERM      2
218 #define TCPOLEN_TIMESTAMP      10
219 #define TCPOLEN_MD5SIG         18
220 #define TCPOLEN_FASTOPEN_BASE  2
221 #define TCPOLEN_EXP_FASTOPEN_BASE  4
222 #define TCPOLEN_EXP_SMC_BASE   6
223 
224 /* But this is what stacks really send out. */
225 #define TCPOLEN_TSTAMP_ALIGNED		12
226 #define TCPOLEN_WSCALE_ALIGNED		4
227 #define TCPOLEN_SACKPERM_ALIGNED	4
228 #define TCPOLEN_SACK_BASE		2
229 #define TCPOLEN_SACK_BASE_ALIGNED	4
230 #define TCPOLEN_SACK_PERBLOCK		8
231 #define TCPOLEN_MD5SIG_ALIGNED		20
232 #define TCPOLEN_MSS_ALIGNED		4
233 #define TCPOLEN_EXP_SMC_BASE_ALIGNED	8
234 
235 /* Flags in tp->nonagle */
236 #define TCP_NAGLE_OFF		1	/* Nagle's algo is disabled */
237 #define TCP_NAGLE_CORK		2	/* Socket is corked	    */
238 #define TCP_NAGLE_PUSH		4	/* Cork is overridden for already queued data */
239 
240 /* TCP thin-stream limits */
241 #define TCP_THIN_LINEAR_RETRIES 6       /* After 6 linear retries, do exp. backoff */
242 
243 /* TCP initial congestion window as per rfc6928 */
244 #define TCP_INIT_CWND		10
245 
246 /* Bit Flags for sysctl_tcp_fastopen */
247 #define	TFO_CLIENT_ENABLE	1
248 #define	TFO_SERVER_ENABLE	2
249 #define	TFO_CLIENT_NO_COOKIE	4	/* Data in SYN w/o cookie option */
250 
251 /* Accept SYN data w/o any cookie option */
252 #define	TFO_SERVER_COOKIE_NOT_REQD	0x200
253 
254 /* Force enable TFO on all listeners, i.e., not requiring the
255  * TCP_FASTOPEN socket option.
256  */
257 #define	TFO_SERVER_WO_SOCKOPT1	0x400
258 
259 
260 /* sysctl variables for tcp */
261 extern int sysctl_tcp_max_orphans;
262 extern long sysctl_tcp_mem[3];
263 
264 #define TCP_RACK_LOSS_DETECTION  0x1 /* Use RACK to detect losses */
265 #define TCP_RACK_STATIC_REO_WND  0x2 /* Use static RACK reo wnd */
266 #define TCP_RACK_NO_DUPTHRESH    0x4 /* Do not use DUPACK threshold in RACK */
267 
268 extern atomic_long_t tcp_memory_allocated;
269 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc);
270 
271 extern struct percpu_counter tcp_sockets_allocated;
272 extern unsigned long tcp_memory_pressure;
273 
274 /* optimized version of sk_under_memory_pressure() for TCP sockets */
275 static inline bool tcp_under_memory_pressure(const struct sock *sk)
276 {
277 	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
278 	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
279 		return true;
280 
281 	return READ_ONCE(tcp_memory_pressure);
282 }
283 /*
284  * The next routines deal with comparing 32 bit unsigned ints
285  * and worry about wraparound (automatic with unsigned arithmetic).
286  */
287 
288 static inline bool before(__u32 seq1, __u32 seq2)
289 {
290         return (__s32)(seq1-seq2) < 0;
291 }
292 #define after(seq2, seq1) 	before(seq1, seq2)
293 
294 /* is s2<=s1<=s3 ? */
295 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
296 {
297 	return seq3 - seq2 >= seq1 - seq2;
298 }
299 
300 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
301 {
302 	sk_wmem_queued_add(sk, -skb->truesize);
303 	if (!skb_zcopy_pure(skb))
304 		sk_mem_uncharge(sk, skb->truesize);
305 	else
306 		sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb)));
307 	__kfree_skb(skb);
308 }
309 
310 void sk_forced_mem_schedule(struct sock *sk, int size);
311 
312 bool tcp_check_oom(const struct sock *sk, int shift);
313 
314 
315 extern struct proto tcp_prot;
316 
317 #define TCP_INC_STATS(net, field)	SNMP_INC_STATS((net)->mib.tcp_statistics, field)
318 #define __TCP_INC_STATS(net, field)	__SNMP_INC_STATS((net)->mib.tcp_statistics, field)
319 #define TCP_DEC_STATS(net, field)	SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
320 #define TCP_ADD_STATS(net, field, val)	SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
321 
322 void tcp_tasklet_init(void);
323 
324 int tcp_v4_err(struct sk_buff *skb, u32);
325 
326 void tcp_shutdown(struct sock *sk, int how);
327 
328 int tcp_v4_early_demux(struct sk_buff *skb);
329 int tcp_v4_rcv(struct sk_buff *skb);
330 
331 void tcp_remove_empty_skb(struct sock *sk);
332 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
333 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
334 int tcp_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied,
335 			 size_t size, struct ubuf_info *uarg);
336 void tcp_splice_eof(struct socket *sock);
337 int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
338 int tcp_wmem_schedule(struct sock *sk, int copy);
339 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
340 	      int size_goal);
341 void tcp_release_cb(struct sock *sk);
342 void tcp_wfree(struct sk_buff *skb);
343 void tcp_write_timer_handler(struct sock *sk);
344 void tcp_delack_timer_handler(struct sock *sk);
345 int tcp_ioctl(struct sock *sk, int cmd, int *karg);
346 enum skb_drop_reason tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
347 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
348 void tcp_rcv_space_adjust(struct sock *sk);
349 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
350 void tcp_twsk_destructor(struct sock *sk);
351 void tcp_twsk_purge(struct list_head *net_exit_list);
352 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
353 			struct pipe_inode_info *pipe, size_t len,
354 			unsigned int flags);
355 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, gfp_t gfp,
356 				     bool force_schedule);
357 
358 static inline void tcp_dec_quickack_mode(struct sock *sk)
359 {
360 	struct inet_connection_sock *icsk = inet_csk(sk);
361 
362 	if (icsk->icsk_ack.quick) {
363 		/* How many ACKs S/ACKing new data have we sent? */
364 		const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0;
365 
366 		if (pkts >= icsk->icsk_ack.quick) {
367 			icsk->icsk_ack.quick = 0;
368 			/* Leaving quickack mode we deflate ATO. */
369 			icsk->icsk_ack.ato   = TCP_ATO_MIN;
370 		} else
371 			icsk->icsk_ack.quick -= pkts;
372 	}
373 }
374 
375 #define	TCP_ECN_OK		1
376 #define	TCP_ECN_QUEUE_CWR	2
377 #define	TCP_ECN_DEMAND_CWR	4
378 #define	TCP_ECN_SEEN		8
379 
380 enum tcp_tw_status {
381 	TCP_TW_SUCCESS = 0,
382 	TCP_TW_RST = 1,
383 	TCP_TW_ACK = 2,
384 	TCP_TW_SYN = 3
385 };
386 
387 
388 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
389 					      struct sk_buff *skb,
390 					      const struct tcphdr *th,
391 					      u32 *tw_isn);
392 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
393 			   struct request_sock *req, bool fastopen,
394 			   bool *lost_race);
395 enum skb_drop_reason tcp_child_process(struct sock *parent, struct sock *child,
396 				       struct sk_buff *skb);
397 void tcp_enter_loss(struct sock *sk);
398 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag);
399 void tcp_clear_retrans(struct tcp_sock *tp);
400 void tcp_update_metrics(struct sock *sk);
401 void tcp_init_metrics(struct sock *sk);
402 void tcp_metrics_init(void);
403 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
404 void __tcp_close(struct sock *sk, long timeout);
405 void tcp_close(struct sock *sk, long timeout);
406 void tcp_init_sock(struct sock *sk);
407 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb);
408 __poll_t tcp_poll(struct file *file, struct socket *sock,
409 		      struct poll_table_struct *wait);
410 int do_tcp_getsockopt(struct sock *sk, int level,
411 		      int optname, sockptr_t optval, sockptr_t optlen);
412 int tcp_getsockopt(struct sock *sk, int level, int optname,
413 		   char __user *optval, int __user *optlen);
414 bool tcp_bpf_bypass_getsockopt(int level, int optname);
415 int do_tcp_setsockopt(struct sock *sk, int level, int optname,
416 		      sockptr_t optval, unsigned int optlen);
417 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
418 		   unsigned int optlen);
419 void tcp_reset_keepalive_timer(struct sock *sk, unsigned long timeout);
420 void tcp_set_keepalive(struct sock *sk, int val);
421 void tcp_syn_ack_timeout(const struct request_sock *req);
422 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
423 		int flags, int *addr_len);
424 int tcp_set_rcvlowat(struct sock *sk, int val);
425 int tcp_set_window_clamp(struct sock *sk, int val);
426 void tcp_update_recv_tstamps(struct sk_buff *skb,
427 			     struct scm_timestamping_internal *tss);
428 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
429 			struct scm_timestamping_internal *tss);
430 void tcp_data_ready(struct sock *sk);
431 #ifdef CONFIG_MMU
432 int tcp_mmap(struct file *file, struct socket *sock,
433 	     struct vm_area_struct *vma);
434 #endif
435 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
436 		       struct tcp_options_received *opt_rx,
437 		       int estab, struct tcp_fastopen_cookie *foc);
438 
439 /*
440  *	BPF SKB-less helpers
441  */
442 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
443 			 struct tcphdr *th, u32 *cookie);
444 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
445 			 struct tcphdr *th, u32 *cookie);
446 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss);
447 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
448 			  const struct tcp_request_sock_ops *af_ops,
449 			  struct sock *sk, struct tcphdr *th);
450 /*
451  *	TCP v4 functions exported for the inet6 API
452  */
453 
454 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
455 void tcp_v4_mtu_reduced(struct sock *sk);
456 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
457 void tcp_ld_RTO_revert(struct sock *sk, u32 seq);
458 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
459 struct sock *tcp_create_openreq_child(const struct sock *sk,
460 				      struct request_sock *req,
461 				      struct sk_buff *skb);
462 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
463 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
464 				  struct request_sock *req,
465 				  struct dst_entry *dst,
466 				  struct request_sock *req_unhash,
467 				  bool *own_req);
468 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
469 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
470 int tcp_connect(struct sock *sk);
471 enum tcp_synack_type {
472 	TCP_SYNACK_NORMAL,
473 	TCP_SYNACK_FASTOPEN,
474 	TCP_SYNACK_COOKIE,
475 };
476 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
477 				struct request_sock *req,
478 				struct tcp_fastopen_cookie *foc,
479 				enum tcp_synack_type synack_type,
480 				struct sk_buff *syn_skb);
481 int tcp_disconnect(struct sock *sk, int flags);
482 
483 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
484 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
485 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
486 
487 /* From syncookies.c */
488 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
489 				 struct request_sock *req,
490 				 struct dst_entry *dst);
491 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th);
492 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
493 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
494 					    struct sock *sk, struct sk_buff *skb,
495 					    struct tcp_options_received *tcp_opt,
496 					    int mss, u32 tsoff);
497 
498 #if IS_ENABLED(CONFIG_BPF)
499 struct bpf_tcp_req_attrs {
500 	u32 rcv_tsval;
501 	u32 rcv_tsecr;
502 	u16 mss;
503 	u8 rcv_wscale;
504 	u8 snd_wscale;
505 	u8 ecn_ok;
506 	u8 wscale_ok;
507 	u8 sack_ok;
508 	u8 tstamp_ok;
509 	u8 usec_ts_ok;
510 	u8 reserved[3];
511 };
512 #endif
513 
514 #ifdef CONFIG_SYN_COOKIES
515 
516 /* Syncookies use a monotonic timer which increments every 60 seconds.
517  * This counter is used both as a hash input and partially encoded into
518  * the cookie value.  A cookie is only validated further if the delta
519  * between the current counter value and the encoded one is less than this,
520  * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
521  * the counter advances immediately after a cookie is generated).
522  */
523 #define MAX_SYNCOOKIE_AGE	2
524 #define TCP_SYNCOOKIE_PERIOD	(60 * HZ)
525 #define TCP_SYNCOOKIE_VALID	(MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
526 
527 /* syncookies: remember time of last synqueue overflow
528  * But do not dirty this field too often (once per second is enough)
529  * It is racy as we do not hold a lock, but race is very minor.
530  */
531 static inline void tcp_synq_overflow(const struct sock *sk)
532 {
533 	unsigned int last_overflow;
534 	unsigned int now = jiffies;
535 
536 	if (sk->sk_reuseport) {
537 		struct sock_reuseport *reuse;
538 
539 		reuse = rcu_dereference(sk->sk_reuseport_cb);
540 		if (likely(reuse)) {
541 			last_overflow = READ_ONCE(reuse->synq_overflow_ts);
542 			if (!time_between32(now, last_overflow,
543 					    last_overflow + HZ))
544 				WRITE_ONCE(reuse->synq_overflow_ts, now);
545 			return;
546 		}
547 	}
548 
549 	last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
550 	if (!time_between32(now, last_overflow, last_overflow + HZ))
551 		WRITE_ONCE(tcp_sk_rw(sk)->rx_opt.ts_recent_stamp, now);
552 }
553 
554 /* syncookies: no recent synqueue overflow on this listening socket? */
555 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
556 {
557 	unsigned int last_overflow;
558 	unsigned int now = jiffies;
559 
560 	if (sk->sk_reuseport) {
561 		struct sock_reuseport *reuse;
562 
563 		reuse = rcu_dereference(sk->sk_reuseport_cb);
564 		if (likely(reuse)) {
565 			last_overflow = READ_ONCE(reuse->synq_overflow_ts);
566 			return !time_between32(now, last_overflow - HZ,
567 					       last_overflow +
568 					       TCP_SYNCOOKIE_VALID);
569 		}
570 	}
571 
572 	last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
573 
574 	/* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
575 	 * then we're under synflood. However, we have to use
576 	 * 'last_overflow - HZ' as lower bound. That's because a concurrent
577 	 * tcp_synq_overflow() could update .ts_recent_stamp after we read
578 	 * jiffies but before we store .ts_recent_stamp into last_overflow,
579 	 * which could lead to rejecting a valid syncookie.
580 	 */
581 	return !time_between32(now, last_overflow - HZ,
582 			       last_overflow + TCP_SYNCOOKIE_VALID);
583 }
584 
585 static inline u32 tcp_cookie_time(void)
586 {
587 	u64 val = get_jiffies_64();
588 
589 	do_div(val, TCP_SYNCOOKIE_PERIOD);
590 	return val;
591 }
592 
593 /* Convert one nsec 64bit timestamp to ts (ms or usec resolution) */
594 static inline u64 tcp_ns_to_ts(bool usec_ts, u64 val)
595 {
596 	if (usec_ts)
597 		return div_u64(val, NSEC_PER_USEC);
598 
599 	return div_u64(val, NSEC_PER_MSEC);
600 }
601 
602 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
603 			      u16 *mssp);
604 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
605 u64 cookie_init_timestamp(struct request_sock *req, u64 now);
606 bool cookie_timestamp_decode(const struct net *net,
607 			     struct tcp_options_received *opt);
608 
609 static inline bool cookie_ecn_ok(const struct net *net, const struct dst_entry *dst)
610 {
611 	return READ_ONCE(net->ipv4.sysctl_tcp_ecn) ||
612 		dst_feature(dst, RTAX_FEATURE_ECN);
613 }
614 
615 #if IS_ENABLED(CONFIG_BPF)
616 static inline bool cookie_bpf_ok(struct sk_buff *skb)
617 {
618 	return skb->sk;
619 }
620 
621 struct request_sock *cookie_bpf_check(struct sock *sk, struct sk_buff *skb);
622 #else
623 static inline bool cookie_bpf_ok(struct sk_buff *skb)
624 {
625 	return false;
626 }
627 
628 static inline struct request_sock *cookie_bpf_check(struct net *net, struct sock *sk,
629 						    struct sk_buff *skb)
630 {
631 	return NULL;
632 }
633 #endif
634 
635 /* From net/ipv6/syncookies.c */
636 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th);
637 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
638 
639 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
640 			      const struct tcphdr *th, u16 *mssp);
641 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
642 #endif
643 /* tcp_output.c */
644 
645 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb);
646 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb);
647 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
648 			       int nonagle);
649 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
650 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
651 void tcp_retransmit_timer(struct sock *sk);
652 void tcp_xmit_retransmit_queue(struct sock *);
653 void tcp_simple_retransmit(struct sock *);
654 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
655 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
656 enum tcp_queue {
657 	TCP_FRAG_IN_WRITE_QUEUE,
658 	TCP_FRAG_IN_RTX_QUEUE,
659 };
660 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
661 		 struct sk_buff *skb, u32 len,
662 		 unsigned int mss_now, gfp_t gfp);
663 
664 void tcp_send_probe0(struct sock *);
665 int tcp_write_wakeup(struct sock *, int mib);
666 void tcp_send_fin(struct sock *sk);
667 void tcp_send_active_reset(struct sock *sk, gfp_t priority,
668 			   enum sk_rst_reason reason);
669 int tcp_send_synack(struct sock *);
670 void tcp_push_one(struct sock *, unsigned int mss_now);
671 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
672 void tcp_send_ack(struct sock *sk);
673 void tcp_send_delayed_ack(struct sock *sk);
674 void tcp_send_loss_probe(struct sock *sk);
675 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
676 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
677 			     const struct sk_buff *next_skb);
678 
679 /* tcp_input.c */
680 void tcp_rearm_rto(struct sock *sk);
681 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
682 void tcp_done_with_error(struct sock *sk, int err);
683 void tcp_reset(struct sock *sk, struct sk_buff *skb);
684 void tcp_fin(struct sock *sk);
685 void tcp_check_space(struct sock *sk);
686 void tcp_sack_compress_send_ack(struct sock *sk);
687 
688 /* tcp_timer.c */
689 void tcp_init_xmit_timers(struct sock *);
690 static inline void tcp_clear_xmit_timers(struct sock *sk)
691 {
692 	if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
693 		__sock_put(sk);
694 
695 	if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
696 		__sock_put(sk);
697 
698 	inet_csk_clear_xmit_timers(sk);
699 }
700 
701 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
702 unsigned int tcp_current_mss(struct sock *sk);
703 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when);
704 
705 /* Bound MSS / TSO packet size with the half of the window */
706 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
707 {
708 	int cutoff;
709 
710 	/* When peer uses tiny windows, there is no use in packetizing
711 	 * to sub-MSS pieces for the sake of SWS or making sure there
712 	 * are enough packets in the pipe for fast recovery.
713 	 *
714 	 * On the other hand, for extremely large MSS devices, handling
715 	 * smaller than MSS windows in this way does make sense.
716 	 */
717 	if (tp->max_window > TCP_MSS_DEFAULT)
718 		cutoff = (tp->max_window >> 1);
719 	else
720 		cutoff = tp->max_window;
721 
722 	if (cutoff && pktsize > cutoff)
723 		return max_t(int, cutoff, 68U - tp->tcp_header_len);
724 	else
725 		return pktsize;
726 }
727 
728 /* tcp.c */
729 void tcp_get_info(struct sock *, struct tcp_info *);
730 
731 /* Read 'sendfile()'-style from a TCP socket */
732 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
733 		  sk_read_actor_t recv_actor);
734 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
735 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off);
736 void tcp_read_done(struct sock *sk, size_t len);
737 
738 void tcp_initialize_rcv_mss(struct sock *sk);
739 
740 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
741 int tcp_mss_to_mtu(struct sock *sk, int mss);
742 void tcp_mtup_init(struct sock *sk);
743 
744 static inline unsigned int tcp_rto_max(const struct sock *sk)
745 {
746 	return READ_ONCE(inet_csk(sk)->icsk_rto_max);
747 }
748 
749 static inline void tcp_bound_rto(struct sock *sk)
750 {
751 	inet_csk(sk)->icsk_rto = min(inet_csk(sk)->icsk_rto, tcp_rto_max(sk));
752 }
753 
754 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
755 {
756 	return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
757 }
758 
759 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
760 {
761 	/* mptcp hooks are only on the slow path */
762 	if (sk_is_mptcp((struct sock *)tp))
763 		return;
764 
765 	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
766 			       ntohl(TCP_FLAG_ACK) |
767 			       snd_wnd);
768 }
769 
770 static inline void tcp_fast_path_on(struct tcp_sock *tp)
771 {
772 	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
773 }
774 
775 static inline void tcp_fast_path_check(struct sock *sk)
776 {
777 	struct tcp_sock *tp = tcp_sk(sk);
778 
779 	if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
780 	    tp->rcv_wnd &&
781 	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
782 	    !tp->urg_data)
783 		tcp_fast_path_on(tp);
784 }
785 
786 u32 tcp_delack_max(const struct sock *sk);
787 
788 /* Compute the actual rto_min value */
789 static inline u32 tcp_rto_min(const struct sock *sk)
790 {
791 	const struct dst_entry *dst = __sk_dst_get(sk);
792 	u32 rto_min = inet_csk(sk)->icsk_rto_min;
793 
794 	if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
795 		rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
796 	return rto_min;
797 }
798 
799 static inline u32 tcp_rto_min_us(const struct sock *sk)
800 {
801 	return jiffies_to_usecs(tcp_rto_min(sk));
802 }
803 
804 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
805 {
806 	return dst_metric_locked(dst, RTAX_CC_ALGO);
807 }
808 
809 /* Minimum RTT in usec. ~0 means not available. */
810 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
811 {
812 	return minmax_get(&tp->rtt_min);
813 }
814 
815 /* Compute the actual receive window we are currently advertising.
816  * Rcv_nxt can be after the window if our peer push more data
817  * than the offered window.
818  */
819 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
820 {
821 	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
822 
823 	if (win < 0)
824 		win = 0;
825 	return (u32) win;
826 }
827 
828 /* Choose a new window, without checks for shrinking, and without
829  * scaling applied to the result.  The caller does these things
830  * if necessary.  This is a "raw" window selection.
831  */
832 u32 __tcp_select_window(struct sock *sk);
833 
834 void tcp_send_window_probe(struct sock *sk);
835 
836 /* TCP uses 32bit jiffies to save some space.
837  * Note that this is different from tcp_time_stamp, which
838  * historically has been the same until linux-4.13.
839  */
840 #define tcp_jiffies32 ((u32)jiffies)
841 
842 /*
843  * Deliver a 32bit value for TCP timestamp option (RFC 7323)
844  * It is no longer tied to jiffies, but to 1 ms clock.
845  * Note: double check if you want to use tcp_jiffies32 instead of this.
846  */
847 #define TCP_TS_HZ	1000
848 
849 static inline u64 tcp_clock_ns(void)
850 {
851 	return ktime_get_ns();
852 }
853 
854 static inline u64 tcp_clock_us(void)
855 {
856 	return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
857 }
858 
859 static inline u64 tcp_clock_ms(void)
860 {
861 	return div_u64(tcp_clock_ns(), NSEC_PER_MSEC);
862 }
863 
864 /* TCP Timestamp included in TS option (RFC 1323) can either use ms
865  * or usec resolution. Each socket carries a flag to select one or other
866  * resolution, as the route attribute could change anytime.
867  * Each flow must stick to initial resolution.
868  */
869 static inline u32 tcp_clock_ts(bool usec_ts)
870 {
871 	return usec_ts ? tcp_clock_us() : tcp_clock_ms();
872 }
873 
874 static inline u32 tcp_time_stamp_ms(const struct tcp_sock *tp)
875 {
876 	return div_u64(tp->tcp_mstamp, USEC_PER_MSEC);
877 }
878 
879 static inline u32 tcp_time_stamp_ts(const struct tcp_sock *tp)
880 {
881 	if (tp->tcp_usec_ts)
882 		return tp->tcp_mstamp;
883 	return tcp_time_stamp_ms(tp);
884 }
885 
886 void tcp_mstamp_refresh(struct tcp_sock *tp);
887 
888 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
889 {
890 	return max_t(s64, t1 - t0, 0);
891 }
892 
893 /* provide the departure time in us unit */
894 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
895 {
896 	return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
897 }
898 
899 /* Provide skb TSval in usec or ms unit */
900 static inline u32 tcp_skb_timestamp_ts(bool usec_ts, const struct sk_buff *skb)
901 {
902 	if (usec_ts)
903 		return tcp_skb_timestamp_us(skb);
904 
905 	return div_u64(skb->skb_mstamp_ns, NSEC_PER_MSEC);
906 }
907 
908 static inline u32 tcp_tw_tsval(const struct tcp_timewait_sock *tcptw)
909 {
910 	return tcp_clock_ts(tcptw->tw_sk.tw_usec_ts) + tcptw->tw_ts_offset;
911 }
912 
913 static inline u32 tcp_rsk_tsval(const struct tcp_request_sock *treq)
914 {
915 	return tcp_clock_ts(treq->req_usec_ts) + treq->ts_off;
916 }
917 
918 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
919 
920 #define TCPHDR_FIN 0x01
921 #define TCPHDR_SYN 0x02
922 #define TCPHDR_RST 0x04
923 #define TCPHDR_PSH 0x08
924 #define TCPHDR_ACK 0x10
925 #define TCPHDR_URG 0x20
926 #define TCPHDR_ECE 0x40
927 #define TCPHDR_CWR 0x80
928 
929 #define TCPHDR_SYN_ECN	(TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
930 
931 /* State flags for sacked in struct tcp_skb_cb */
932 enum tcp_skb_cb_sacked_flags {
933 	TCPCB_SACKED_ACKED	= (1 << 0),	/* SKB ACK'd by a SACK block	*/
934 	TCPCB_SACKED_RETRANS	= (1 << 1),	/* SKB retransmitted		*/
935 	TCPCB_LOST		= (1 << 2),	/* SKB is lost			*/
936 	TCPCB_TAGBITS		= (TCPCB_SACKED_ACKED | TCPCB_SACKED_RETRANS |
937 				   TCPCB_LOST),	/* All tag bits			*/
938 	TCPCB_REPAIRED		= (1 << 4),	/* SKB repaired (no skb_mstamp_ns)	*/
939 	TCPCB_EVER_RETRANS	= (1 << 7),	/* Ever retransmitted frame	*/
940 	TCPCB_RETRANS		= (TCPCB_SACKED_RETRANS | TCPCB_EVER_RETRANS |
941 				   TCPCB_REPAIRED),
942 };
943 
944 /* This is what the send packet queuing engine uses to pass
945  * TCP per-packet control information to the transmission code.
946  * We also store the host-order sequence numbers in here too.
947  * This is 44 bytes if IPV6 is enabled.
948  * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
949  */
950 struct tcp_skb_cb {
951 	__u32		seq;		/* Starting sequence number	*/
952 	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
953 	union {
954 		/* Note :
955 		 * 	  tcp_gso_segs/size are used in write queue only,
956 		 *	  cf tcp_skb_pcount()/tcp_skb_mss()
957 		 */
958 		struct {
959 			u16	tcp_gso_segs;
960 			u16	tcp_gso_size;
961 		};
962 	};
963 	__u8		tcp_flags;	/* TCP header flags. (tcp[13])	*/
964 
965 	__u8		sacked;		/* State flags for SACK.	*/
966 	__u8		ip_dsfield;	/* IPv4 tos or IPv6 dsfield	*/
967 	__u8		txstamp_ack:1,	/* Record TX timestamp for ack? */
968 			eor:1,		/* Is skb MSG_EOR marked? */
969 			has_rxtstamp:1,	/* SKB has a RX timestamp	*/
970 			unused:5;
971 	__u32		ack_seq;	/* Sequence number ACK'd	*/
972 	union {
973 		struct {
974 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1)
975 			/* There is space for up to 24 bytes */
976 			__u32 is_app_limited:1, /* cwnd not fully used? */
977 			      delivered_ce:20,
978 			      unused:11;
979 			/* pkts S/ACKed so far upon tx of skb, incl retrans: */
980 			__u32 delivered;
981 			/* start of send pipeline phase */
982 			u64 first_tx_mstamp;
983 			/* when we reached the "delivered" count */
984 			u64 delivered_mstamp;
985 		} tx;   /* only used for outgoing skbs */
986 		union {
987 			struct inet_skb_parm	h4;
988 #if IS_ENABLED(CONFIG_IPV6)
989 			struct inet6_skb_parm	h6;
990 #endif
991 		} header;	/* For incoming skbs */
992 	};
993 };
994 
995 #define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
996 
997 extern const struct inet_connection_sock_af_ops ipv4_specific;
998 
999 #if IS_ENABLED(CONFIG_IPV6)
1000 /* This is the variant of inet6_iif() that must be used by TCP,
1001  * as TCP moves IP6CB into a different location in skb->cb[]
1002  */
1003 static inline int tcp_v6_iif(const struct sk_buff *skb)
1004 {
1005 	return TCP_SKB_CB(skb)->header.h6.iif;
1006 }
1007 
1008 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
1009 {
1010 	bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
1011 
1012 	return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
1013 }
1014 
1015 /* TCP_SKB_CB reference means this can not be used from early demux */
1016 static inline int tcp_v6_sdif(const struct sk_buff *skb)
1017 {
1018 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
1019 	if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
1020 		return TCP_SKB_CB(skb)->header.h6.iif;
1021 #endif
1022 	return 0;
1023 }
1024 
1025 extern const struct inet_connection_sock_af_ops ipv6_specific;
1026 
1027 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
1028 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
1029 void tcp_v6_early_demux(struct sk_buff *skb);
1030 
1031 #endif
1032 
1033 /* TCP_SKB_CB reference means this can not be used from early demux */
1034 static inline int tcp_v4_sdif(struct sk_buff *skb)
1035 {
1036 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
1037 	if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
1038 		return TCP_SKB_CB(skb)->header.h4.iif;
1039 #endif
1040 	return 0;
1041 }
1042 
1043 /* Due to TSO, an SKB can be composed of multiple actual
1044  * packets.  To keep these tracked properly, we use this.
1045  */
1046 static inline int tcp_skb_pcount(const struct sk_buff *skb)
1047 {
1048 	return TCP_SKB_CB(skb)->tcp_gso_segs;
1049 }
1050 
1051 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
1052 {
1053 	TCP_SKB_CB(skb)->tcp_gso_segs = segs;
1054 }
1055 
1056 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
1057 {
1058 	TCP_SKB_CB(skb)->tcp_gso_segs += segs;
1059 }
1060 
1061 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
1062 static inline int tcp_skb_mss(const struct sk_buff *skb)
1063 {
1064 	return TCP_SKB_CB(skb)->tcp_gso_size;
1065 }
1066 
1067 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
1068 {
1069 	return likely(!TCP_SKB_CB(skb)->eor);
1070 }
1071 
1072 static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
1073 					const struct sk_buff *from)
1074 {
1075 	/* skb_cmp_decrypted() not needed, use tcp_write_collapse_fence() */
1076 	return likely(tcp_skb_can_collapse_to(to) &&
1077 		      mptcp_skb_can_collapse(to, from) &&
1078 		      skb_pure_zcopy_same(to, from) &&
1079 		      skb_frags_readable(to) == skb_frags_readable(from));
1080 }
1081 
1082 static inline bool tcp_skb_can_collapse_rx(const struct sk_buff *to,
1083 					   const struct sk_buff *from)
1084 {
1085 	return likely(mptcp_skb_can_collapse(to, from) &&
1086 		      !skb_cmp_decrypted(to, from));
1087 }
1088 
1089 /* Events passed to congestion control interface */
1090 enum tcp_ca_event {
1091 	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
1092 	CA_EVENT_CWND_RESTART,	/* congestion window restart */
1093 	CA_EVENT_COMPLETE_CWR,	/* end of congestion recovery */
1094 	CA_EVENT_LOSS,		/* loss timeout */
1095 	CA_EVENT_ECN_NO_CE,	/* ECT set, but not CE marked */
1096 	CA_EVENT_ECN_IS_CE,	/* received CE marked IP packet */
1097 };
1098 
1099 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
1100 enum tcp_ca_ack_event_flags {
1101 	CA_ACK_SLOWPATH		= (1 << 0),	/* In slow path processing */
1102 	CA_ACK_WIN_UPDATE	= (1 << 1),	/* ACK updated window */
1103 	CA_ACK_ECE		= (1 << 2),	/* ECE bit is set on ack */
1104 };
1105 
1106 /*
1107  * Interface for adding new TCP congestion control handlers
1108  */
1109 #define TCP_CA_NAME_MAX	16
1110 #define TCP_CA_MAX	128
1111 #define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
1112 
1113 #define TCP_CA_UNSPEC	0
1114 
1115 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1116 #define TCP_CONG_NON_RESTRICTED 0x1
1117 /* Requires ECN/ECT set on all packets */
1118 #define TCP_CONG_NEEDS_ECN	0x2
1119 #define TCP_CONG_MASK	(TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1120 
1121 union tcp_cc_info;
1122 
1123 struct ack_sample {
1124 	u32 pkts_acked;
1125 	s32 rtt_us;
1126 	u32 in_flight;
1127 };
1128 
1129 /* A rate sample measures the number of (original/retransmitted) data
1130  * packets delivered "delivered" over an interval of time "interval_us".
1131  * The tcp_rate.c code fills in the rate sample, and congestion
1132  * control modules that define a cong_control function to run at the end
1133  * of ACK processing can optionally chose to consult this sample when
1134  * setting cwnd and pacing rate.
1135  * A sample is invalid if "delivered" or "interval_us" is negative.
1136  */
1137 struct rate_sample {
1138 	u64  prior_mstamp; /* starting timestamp for interval */
1139 	u32  prior_delivered;	/* tp->delivered at "prior_mstamp" */
1140 	u32  prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */
1141 	s32  delivered;		/* number of packets delivered over interval */
1142 	s32  delivered_ce;	/* number of packets delivered w/ CE marks*/
1143 	long interval_us;	/* time for tp->delivered to incr "delivered" */
1144 	u32 snd_interval_us;	/* snd interval for delivered packets */
1145 	u32 rcv_interval_us;	/* rcv interval for delivered packets */
1146 	long rtt_us;		/* RTT of last (S)ACKed packet (or -1) */
1147 	int  losses;		/* number of packets marked lost upon ACK */
1148 	u32  acked_sacked;	/* number of packets newly (S)ACKed upon ACK */
1149 	u32  prior_in_flight;	/* in flight before this ACK */
1150 	u32  last_end_seq;	/* end_seq of most recently ACKed packet */
1151 	bool is_app_limited;	/* is sample from packet with bubble in pipe? */
1152 	bool is_retrans;	/* is sample from retransmission? */
1153 	bool is_ack_delayed;	/* is this (likely) a delayed ACK? */
1154 };
1155 
1156 struct tcp_congestion_ops {
1157 /* fast path fields are put first to fill one cache line */
1158 
1159 	/* return slow start threshold (required) */
1160 	u32 (*ssthresh)(struct sock *sk);
1161 
1162 	/* do new cwnd calculation (required) */
1163 	void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1164 
1165 	/* call before changing ca_state (optional) */
1166 	void (*set_state)(struct sock *sk, u8 new_state);
1167 
1168 	/* call when cwnd event occurs (optional) */
1169 	void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1170 
1171 	/* call when ack arrives (optional) */
1172 	void (*in_ack_event)(struct sock *sk, u32 flags);
1173 
1174 	/* hook for packet ack accounting (optional) */
1175 	void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1176 
1177 	/* override sysctl_tcp_min_tso_segs */
1178 	u32 (*min_tso_segs)(struct sock *sk);
1179 
1180 	/* call when packets are delivered to update cwnd and pacing rate,
1181 	 * after all the ca_state processing. (optional)
1182 	 */
1183 	void (*cong_control)(struct sock *sk, u32 ack, int flag, const struct rate_sample *rs);
1184 
1185 
1186 	/* new value of cwnd after loss (required) */
1187 	u32  (*undo_cwnd)(struct sock *sk);
1188 	/* returns the multiplier used in tcp_sndbuf_expand (optional) */
1189 	u32 (*sndbuf_expand)(struct sock *sk);
1190 
1191 /* control/slow paths put last */
1192 	/* get info for inet_diag (optional) */
1193 	size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1194 			   union tcp_cc_info *info);
1195 
1196 	char 			name[TCP_CA_NAME_MAX];
1197 	struct module		*owner;
1198 	struct list_head	list;
1199 	u32			key;
1200 	u32			flags;
1201 
1202 	/* initialize private data (optional) */
1203 	void (*init)(struct sock *sk);
1204 	/* cleanup private data  (optional) */
1205 	void (*release)(struct sock *sk);
1206 } ____cacheline_aligned_in_smp;
1207 
1208 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1209 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1210 int tcp_update_congestion_control(struct tcp_congestion_ops *type,
1211 				  struct tcp_congestion_ops *old_type);
1212 int tcp_validate_congestion_control(struct tcp_congestion_ops *ca);
1213 
1214 void tcp_assign_congestion_control(struct sock *sk);
1215 void tcp_init_congestion_control(struct sock *sk);
1216 void tcp_cleanup_congestion_control(struct sock *sk);
1217 int tcp_set_default_congestion_control(struct net *net, const char *name);
1218 void tcp_get_default_congestion_control(struct net *net, char *name);
1219 void tcp_get_available_congestion_control(char *buf, size_t len);
1220 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1221 int tcp_set_allowed_congestion_control(char *allowed);
1222 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1223 			       bool cap_net_admin);
1224 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1225 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1226 
1227 u32 tcp_reno_ssthresh(struct sock *sk);
1228 u32 tcp_reno_undo_cwnd(struct sock *sk);
1229 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1230 extern struct tcp_congestion_ops tcp_reno;
1231 
1232 struct tcp_congestion_ops *tcp_ca_find(const char *name);
1233 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1234 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
1235 #ifdef CONFIG_INET
1236 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1237 #else
1238 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1239 {
1240 	return NULL;
1241 }
1242 #endif
1243 
1244 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1245 {
1246 	const struct inet_connection_sock *icsk = inet_csk(sk);
1247 
1248 	return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1249 }
1250 
1251 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1252 {
1253 	const struct inet_connection_sock *icsk = inet_csk(sk);
1254 
1255 	if (icsk->icsk_ca_ops->cwnd_event)
1256 		icsk->icsk_ca_ops->cwnd_event(sk, event);
1257 }
1258 
1259 /* From tcp_cong.c */
1260 void tcp_set_ca_state(struct sock *sk, const u8 ca_state);
1261 
1262 /* From tcp_rate.c */
1263 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1264 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1265 			    struct rate_sample *rs);
1266 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1267 		  bool is_sack_reneg, struct rate_sample *rs);
1268 void tcp_rate_check_app_limited(struct sock *sk);
1269 
1270 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2)
1271 {
1272 	return t1 > t2 || (t1 == t2 && after(seq1, seq2));
1273 }
1274 
1275 /* These functions determine how the current flow behaves in respect of SACK
1276  * handling. SACK is negotiated with the peer, and therefore it can vary
1277  * between different flows.
1278  *
1279  * tcp_is_sack - SACK enabled
1280  * tcp_is_reno - No SACK
1281  */
1282 static inline int tcp_is_sack(const struct tcp_sock *tp)
1283 {
1284 	return likely(tp->rx_opt.sack_ok);
1285 }
1286 
1287 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1288 {
1289 	return !tcp_is_sack(tp);
1290 }
1291 
1292 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1293 {
1294 	return tp->sacked_out + tp->lost_out;
1295 }
1296 
1297 /* This determines how many packets are "in the network" to the best
1298  * of our knowledge.  In many cases it is conservative, but where
1299  * detailed information is available from the receiver (via SACK
1300  * blocks etc.) we can make more aggressive calculations.
1301  *
1302  * Use this for decisions involving congestion control, use just
1303  * tp->packets_out to determine if the send queue is empty or not.
1304  *
1305  * Read this equation as:
1306  *
1307  *	"Packets sent once on transmission queue" MINUS
1308  *	"Packets left network, but not honestly ACKed yet" PLUS
1309  *	"Packets fast retransmitted"
1310  */
1311 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1312 {
1313 	return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1314 }
1315 
1316 #define TCP_INFINITE_SSTHRESH	0x7fffffff
1317 
1318 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp)
1319 {
1320 	return tp->snd_cwnd;
1321 }
1322 
1323 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val)
1324 {
1325 	WARN_ON_ONCE((int)val <= 0);
1326 	tp->snd_cwnd = val;
1327 }
1328 
1329 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1330 {
1331 	return tcp_snd_cwnd(tp) < tp->snd_ssthresh;
1332 }
1333 
1334 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1335 {
1336 	return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1337 }
1338 
1339 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1340 {
1341 	return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1342 	       (1 << inet_csk(sk)->icsk_ca_state);
1343 }
1344 
1345 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1346  * The exception is cwnd reduction phase, when cwnd is decreasing towards
1347  * ssthresh.
1348  */
1349 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1350 {
1351 	const struct tcp_sock *tp = tcp_sk(sk);
1352 
1353 	if (tcp_in_cwnd_reduction(sk))
1354 		return tp->snd_ssthresh;
1355 	else
1356 		return max(tp->snd_ssthresh,
1357 			   ((tcp_snd_cwnd(tp) >> 1) +
1358 			    (tcp_snd_cwnd(tp) >> 2)));
1359 }
1360 
1361 /* Use define here intentionally to get WARN_ON location shown at the caller */
1362 #define tcp_verify_left_out(tp)	WARN_ON(tcp_left_out(tp) > tp->packets_out)
1363 
1364 void tcp_enter_cwr(struct sock *sk);
1365 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1366 
1367 /* The maximum number of MSS of available cwnd for which TSO defers
1368  * sending if not using sysctl_tcp_tso_win_divisor.
1369  */
1370 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1371 {
1372 	return 3;
1373 }
1374 
1375 /* Returns end sequence number of the receiver's advertised window */
1376 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1377 {
1378 	return tp->snd_una + tp->snd_wnd;
1379 }
1380 
1381 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1382  * flexible approach. The RFC suggests cwnd should not be raised unless
1383  * it was fully used previously. And that's exactly what we do in
1384  * congestion avoidance mode. But in slow start we allow cwnd to grow
1385  * as long as the application has used half the cwnd.
1386  * Example :
1387  *    cwnd is 10 (IW10), but application sends 9 frames.
1388  *    We allow cwnd to reach 18 when all frames are ACKed.
1389  * This check is safe because it's as aggressive as slow start which already
1390  * risks 100% overshoot. The advantage is that we discourage application to
1391  * either send more filler packets or data to artificially blow up the cwnd
1392  * usage, and allow application-limited process to probe bw more aggressively.
1393  */
1394 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1395 {
1396 	const struct tcp_sock *tp = tcp_sk(sk);
1397 
1398 	if (tp->is_cwnd_limited)
1399 		return true;
1400 
1401 	/* If in slow start, ensure cwnd grows to twice what was ACKed. */
1402 	if (tcp_in_slow_start(tp))
1403 		return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out;
1404 
1405 	return false;
1406 }
1407 
1408 /* BBR congestion control needs pacing.
1409  * Same remark for SO_MAX_PACING_RATE.
1410  * sch_fq packet scheduler is efficiently handling pacing,
1411  * but is not always installed/used.
1412  * Return true if TCP stack should pace packets itself.
1413  */
1414 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1415 {
1416 	return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1417 }
1418 
1419 /* Estimates in how many jiffies next packet for this flow can be sent.
1420  * Scheduling a retransmit timer too early would be silly.
1421  */
1422 static inline unsigned long tcp_pacing_delay(const struct sock *sk)
1423 {
1424 	s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
1425 
1426 	return delay > 0 ? nsecs_to_jiffies(delay) : 0;
1427 }
1428 
1429 static inline void tcp_reset_xmit_timer(struct sock *sk,
1430 					const int what,
1431 					unsigned long when,
1432 					bool pace_delay)
1433 {
1434 	if (pace_delay)
1435 		when += tcp_pacing_delay(sk);
1436 	inet_csk_reset_xmit_timer(sk, what, when,
1437 				  tcp_rto_max(sk));
1438 }
1439 
1440 /* Something is really bad, we could not queue an additional packet,
1441  * because qdisc is full or receiver sent a 0 window, or we are paced.
1442  * We do not want to add fuel to the fire, or abort too early,
1443  * so make sure the timer we arm now is at least 200ms in the future,
1444  * regardless of current icsk_rto value (as it could be ~2ms)
1445  */
1446 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1447 {
1448 	return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1449 }
1450 
1451 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1452 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1453 					    unsigned long max_when)
1454 {
1455 	u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1,
1456 			   inet_csk(sk)->icsk_backoff);
1457 	u64 when = (u64)tcp_probe0_base(sk) << backoff;
1458 
1459 	return (unsigned long)min_t(u64, when, max_when);
1460 }
1461 
1462 static inline void tcp_check_probe_timer(struct sock *sk)
1463 {
1464 	if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1465 		tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1466 				     tcp_probe0_base(sk), true);
1467 }
1468 
1469 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1470 {
1471 	tp->snd_wl1 = seq;
1472 }
1473 
1474 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1475 {
1476 	tp->snd_wl1 = seq;
1477 }
1478 
1479 /*
1480  * Calculate(/check) TCP checksum
1481  */
1482 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1483 				   __be32 daddr, __wsum base)
1484 {
1485 	return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1486 }
1487 
1488 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1489 {
1490 	return !skb_csum_unnecessary(skb) &&
1491 		__skb_checksum_complete(skb);
1492 }
1493 
1494 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb,
1495 		     enum skb_drop_reason *reason);
1496 
1497 
1498 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1499 void tcp_set_state(struct sock *sk, int state);
1500 void tcp_done(struct sock *sk);
1501 int tcp_abort(struct sock *sk, int err);
1502 
1503 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1504 {
1505 	rx_opt->dsack = 0;
1506 	rx_opt->num_sacks = 0;
1507 }
1508 
1509 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1510 
1511 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1512 {
1513 	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1514 	struct tcp_sock *tp = tcp_sk(sk);
1515 	s32 delta;
1516 
1517 	if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) ||
1518 	    tp->packets_out || ca_ops->cong_control)
1519 		return;
1520 	delta = tcp_jiffies32 - tp->lsndtime;
1521 	if (delta > inet_csk(sk)->icsk_rto)
1522 		tcp_cwnd_restart(sk, delta);
1523 }
1524 
1525 /* Determine a window scaling and initial window to offer. */
1526 void tcp_select_initial_window(const struct sock *sk, int __space,
1527 			       __u32 mss, __u32 *rcv_wnd,
1528 			       __u32 *window_clamp, int wscale_ok,
1529 			       __u8 *rcv_wscale, __u32 init_rcv_wnd);
1530 
1531 static inline int __tcp_win_from_space(u8 scaling_ratio, int space)
1532 {
1533 	s64 scaled_space = (s64)space * scaling_ratio;
1534 
1535 	return scaled_space >> TCP_RMEM_TO_WIN_SCALE;
1536 }
1537 
1538 static inline int tcp_win_from_space(const struct sock *sk, int space)
1539 {
1540 	return __tcp_win_from_space(tcp_sk(sk)->scaling_ratio, space);
1541 }
1542 
1543 /* inverse of __tcp_win_from_space() */
1544 static inline int __tcp_space_from_win(u8 scaling_ratio, int win)
1545 {
1546 	u64 val = (u64)win << TCP_RMEM_TO_WIN_SCALE;
1547 
1548 	do_div(val, scaling_ratio);
1549 	return val;
1550 }
1551 
1552 static inline int tcp_space_from_win(const struct sock *sk, int win)
1553 {
1554 	return __tcp_space_from_win(tcp_sk(sk)->scaling_ratio, win);
1555 }
1556 
1557 /* Assume a 50% default for skb->len/skb->truesize ratio.
1558  * This may be adjusted later in tcp_measure_rcv_mss().
1559  */
1560 #define TCP_DEFAULT_SCALING_RATIO (1 << (TCP_RMEM_TO_WIN_SCALE - 1))
1561 
1562 static inline void tcp_scaling_ratio_init(struct sock *sk)
1563 {
1564 	tcp_sk(sk)->scaling_ratio = TCP_DEFAULT_SCALING_RATIO;
1565 }
1566 
1567 /* Note: caller must be prepared to deal with negative returns */
1568 static inline int tcp_space(const struct sock *sk)
1569 {
1570 	return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1571 				  READ_ONCE(sk->sk_backlog.len) -
1572 				  atomic_read(&sk->sk_rmem_alloc));
1573 }
1574 
1575 static inline int tcp_full_space(const struct sock *sk)
1576 {
1577 	return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1578 }
1579 
1580 static inline void __tcp_adjust_rcv_ssthresh(struct sock *sk, u32 new_ssthresh)
1581 {
1582 	int unused_mem = sk_unused_reserved_mem(sk);
1583 	struct tcp_sock *tp = tcp_sk(sk);
1584 
1585 	tp->rcv_ssthresh = min(tp->rcv_ssthresh, new_ssthresh);
1586 	if (unused_mem)
1587 		tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh,
1588 					 tcp_win_from_space(sk, unused_mem));
1589 }
1590 
1591 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk)
1592 {
1593 	__tcp_adjust_rcv_ssthresh(sk, 4U * tcp_sk(sk)->advmss);
1594 }
1595 
1596 void tcp_cleanup_rbuf(struct sock *sk, int copied);
1597 void __tcp_cleanup_rbuf(struct sock *sk, int copied);
1598 
1599 
1600 /* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1601  * If 87.5 % (7/8) of the space has been consumed, we want to override
1602  * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1603  * len/truesize ratio.
1604  */
1605 static inline bool tcp_rmem_pressure(const struct sock *sk)
1606 {
1607 	int rcvbuf, threshold;
1608 
1609 	if (tcp_under_memory_pressure(sk))
1610 		return true;
1611 
1612 	rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1613 	threshold = rcvbuf - (rcvbuf >> 3);
1614 
1615 	return atomic_read(&sk->sk_rmem_alloc) > threshold;
1616 }
1617 
1618 static inline bool tcp_epollin_ready(const struct sock *sk, int target)
1619 {
1620 	const struct tcp_sock *tp = tcp_sk(sk);
1621 	int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq);
1622 
1623 	if (avail <= 0)
1624 		return false;
1625 
1626 	return (avail >= target) || tcp_rmem_pressure(sk) ||
1627 	       (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss);
1628 }
1629 
1630 extern void tcp_openreq_init_rwin(struct request_sock *req,
1631 				  const struct sock *sk_listener,
1632 				  const struct dst_entry *dst);
1633 
1634 void tcp_enter_memory_pressure(struct sock *sk);
1635 void tcp_leave_memory_pressure(struct sock *sk);
1636 
1637 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1638 {
1639 	struct net *net = sock_net((struct sock *)tp);
1640 	int val;
1641 
1642 	/* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl()
1643 	 * and do_tcp_setsockopt().
1644 	 */
1645 	val = READ_ONCE(tp->keepalive_intvl);
1646 
1647 	return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl);
1648 }
1649 
1650 static inline int keepalive_time_when(const struct tcp_sock *tp)
1651 {
1652 	struct net *net = sock_net((struct sock *)tp);
1653 	int val;
1654 
1655 	/* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */
1656 	val = READ_ONCE(tp->keepalive_time);
1657 
1658 	return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time);
1659 }
1660 
1661 static inline int keepalive_probes(const struct tcp_sock *tp)
1662 {
1663 	struct net *net = sock_net((struct sock *)tp);
1664 	int val;
1665 
1666 	/* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt()
1667 	 * and do_tcp_setsockopt().
1668 	 */
1669 	val = READ_ONCE(tp->keepalive_probes);
1670 
1671 	return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes);
1672 }
1673 
1674 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1675 {
1676 	const struct inet_connection_sock *icsk = &tp->inet_conn;
1677 
1678 	return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1679 			  tcp_jiffies32 - tp->rcv_tstamp);
1680 }
1681 
1682 static inline int tcp_fin_time(const struct sock *sk)
1683 {
1684 	int fin_timeout = tcp_sk(sk)->linger2 ? :
1685 		READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout);
1686 	const int rto = inet_csk(sk)->icsk_rto;
1687 
1688 	if (fin_timeout < (rto << 2) - (rto >> 1))
1689 		fin_timeout = (rto << 2) - (rto >> 1);
1690 
1691 	return fin_timeout;
1692 }
1693 
1694 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1695 				  int paws_win)
1696 {
1697 	if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1698 		return true;
1699 	if (unlikely(!time_before32(ktime_get_seconds(),
1700 				    rx_opt->ts_recent_stamp + TCP_PAWS_WRAP)))
1701 		return true;
1702 	/*
1703 	 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1704 	 * then following tcp messages have valid values. Ignore 0 value,
1705 	 * or else 'negative' tsval might forbid us to accept their packets.
1706 	 */
1707 	if (!rx_opt->ts_recent)
1708 		return true;
1709 	return false;
1710 }
1711 
1712 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1713 				   int rst)
1714 {
1715 	if (tcp_paws_check(rx_opt, 0))
1716 		return false;
1717 
1718 	/* RST segments are not recommended to carry timestamp,
1719 	   and, if they do, it is recommended to ignore PAWS because
1720 	   "their cleanup function should take precedence over timestamps."
1721 	   Certainly, it is mistake. It is necessary to understand the reasons
1722 	   of this constraint to relax it: if peer reboots, clock may go
1723 	   out-of-sync and half-open connections will not be reset.
1724 	   Actually, the problem would be not existing if all
1725 	   the implementations followed draft about maintaining clock
1726 	   via reboots. Linux-2.2 DOES NOT!
1727 
1728 	   However, we can relax time bounds for RST segments to MSL.
1729 	 */
1730 	if (rst && !time_before32(ktime_get_seconds(),
1731 				  rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1732 		return false;
1733 	return true;
1734 }
1735 
1736 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1737 			  int mib_idx, u32 *last_oow_ack_time);
1738 
1739 static inline void tcp_mib_init(struct net *net)
1740 {
1741 	/* See RFC 2012 */
1742 	TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1743 	TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1744 	TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1745 	TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1746 }
1747 
1748 /* from STCP */
1749 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1750 {
1751 	tp->lost_skb_hint = NULL;
1752 }
1753 
1754 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1755 {
1756 	tcp_clear_retrans_hints_partial(tp);
1757 	tp->retransmit_skb_hint = NULL;
1758 }
1759 
1760 #define tcp_md5_addr tcp_ao_addr
1761 
1762 /* - key database */
1763 struct tcp_md5sig_key {
1764 	struct hlist_node	node;
1765 	u8			keylen;
1766 	u8			family; /* AF_INET or AF_INET6 */
1767 	u8			prefixlen;
1768 	u8			flags;
1769 	union tcp_md5_addr	addr;
1770 	int			l3index; /* set if key added with L3 scope */
1771 	u8			key[TCP_MD5SIG_MAXKEYLEN];
1772 	struct rcu_head		rcu;
1773 };
1774 
1775 /* - sock block */
1776 struct tcp_md5sig_info {
1777 	struct hlist_head	head;
1778 	struct rcu_head		rcu;
1779 };
1780 
1781 /* - pseudo header */
1782 struct tcp4_pseudohdr {
1783 	__be32		saddr;
1784 	__be32		daddr;
1785 	__u8		pad;
1786 	__u8		protocol;
1787 	__be16		len;
1788 };
1789 
1790 struct tcp6_pseudohdr {
1791 	struct in6_addr	saddr;
1792 	struct in6_addr daddr;
1793 	__be32		len;
1794 	__be32		protocol;	/* including padding */
1795 };
1796 
1797 union tcp_md5sum_block {
1798 	struct tcp4_pseudohdr ip4;
1799 #if IS_ENABLED(CONFIG_IPV6)
1800 	struct tcp6_pseudohdr ip6;
1801 #endif
1802 };
1803 
1804 /*
1805  * struct tcp_sigpool - per-CPU pool of ahash_requests
1806  * @scratch: per-CPU temporary area, that can be used between
1807  *	     tcp_sigpool_start() and tcp_sigpool_end() to perform
1808  *	     crypto request
1809  * @req: pre-allocated ahash request
1810  */
1811 struct tcp_sigpool {
1812 	void *scratch;
1813 	struct ahash_request *req;
1814 };
1815 
1816 int tcp_sigpool_alloc_ahash(const char *alg, size_t scratch_size);
1817 void tcp_sigpool_get(unsigned int id);
1818 void tcp_sigpool_release(unsigned int id);
1819 int tcp_sigpool_hash_skb_data(struct tcp_sigpool *hp,
1820 			      const struct sk_buff *skb,
1821 			      unsigned int header_len);
1822 
1823 /**
1824  * tcp_sigpool_start - disable bh and start using tcp_sigpool_ahash
1825  * @id: tcp_sigpool that was previously allocated by tcp_sigpool_alloc_ahash()
1826  * @c: returned tcp_sigpool for usage (uninitialized on failure)
1827  *
1828  * Returns: 0 on success, error otherwise.
1829  */
1830 int tcp_sigpool_start(unsigned int id, struct tcp_sigpool *c);
1831 /**
1832  * tcp_sigpool_end - enable bh and stop using tcp_sigpool
1833  * @c: tcp_sigpool context that was returned by tcp_sigpool_start()
1834  */
1835 void tcp_sigpool_end(struct tcp_sigpool *c);
1836 size_t tcp_sigpool_algo(unsigned int id, char *buf, size_t buf_len);
1837 /* - functions */
1838 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1839 			const struct sock *sk, const struct sk_buff *skb);
1840 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1841 		   int family, u8 prefixlen, int l3index, u8 flags,
1842 		   const u8 *newkey, u8 newkeylen);
1843 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr,
1844 		     int family, u8 prefixlen, int l3index,
1845 		     struct tcp_md5sig_key *key);
1846 
1847 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1848 		   int family, u8 prefixlen, int l3index, u8 flags);
1849 void tcp_clear_md5_list(struct sock *sk);
1850 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1851 					 const struct sock *addr_sk);
1852 
1853 #ifdef CONFIG_TCP_MD5SIG
1854 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1855 					   const union tcp_md5_addr *addr,
1856 					   int family, bool any_l3index);
1857 static inline struct tcp_md5sig_key *
1858 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1859 		  const union tcp_md5_addr *addr, int family)
1860 {
1861 	if (!static_branch_unlikely(&tcp_md5_needed.key))
1862 		return NULL;
1863 	return __tcp_md5_do_lookup(sk, l3index, addr, family, false);
1864 }
1865 
1866 static inline struct tcp_md5sig_key *
1867 tcp_md5_do_lookup_any_l3index(const struct sock *sk,
1868 			      const union tcp_md5_addr *addr, int family)
1869 {
1870 	if (!static_branch_unlikely(&tcp_md5_needed.key))
1871 		return NULL;
1872 	return __tcp_md5_do_lookup(sk, 0, addr, family, true);
1873 }
1874 
1875 #define tcp_twsk_md5_key(twsk)	((twsk)->tw_md5_key)
1876 #else
1877 static inline struct tcp_md5sig_key *
1878 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1879 		  const union tcp_md5_addr *addr, int family)
1880 {
1881 	return NULL;
1882 }
1883 
1884 static inline struct tcp_md5sig_key *
1885 tcp_md5_do_lookup_any_l3index(const struct sock *sk,
1886 			      const union tcp_md5_addr *addr, int family)
1887 {
1888 	return NULL;
1889 }
1890 
1891 #define tcp_twsk_md5_key(twsk)	NULL
1892 #endif
1893 
1894 int tcp_md5_alloc_sigpool(void);
1895 void tcp_md5_release_sigpool(void);
1896 void tcp_md5_add_sigpool(void);
1897 extern int tcp_md5_sigpool_id;
1898 
1899 int tcp_md5_hash_key(struct tcp_sigpool *hp,
1900 		     const struct tcp_md5sig_key *key);
1901 
1902 /* From tcp_fastopen.c */
1903 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1904 			    struct tcp_fastopen_cookie *cookie);
1905 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1906 			    struct tcp_fastopen_cookie *cookie, bool syn_lost,
1907 			    u16 try_exp);
1908 struct tcp_fastopen_request {
1909 	/* Fast Open cookie. Size 0 means a cookie request */
1910 	struct tcp_fastopen_cookie	cookie;
1911 	struct msghdr			*data;  /* data in MSG_FASTOPEN */
1912 	size_t				size;
1913 	int				copied;	/* queued in tcp_connect() */
1914 	struct ubuf_info		*uarg;
1915 };
1916 void tcp_free_fastopen_req(struct tcp_sock *tp);
1917 void tcp_fastopen_destroy_cipher(struct sock *sk);
1918 void tcp_fastopen_ctx_destroy(struct net *net);
1919 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1920 			      void *primary_key, void *backup_key);
1921 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
1922 			    u64 *key);
1923 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1924 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1925 			      struct request_sock *req,
1926 			      struct tcp_fastopen_cookie *foc,
1927 			      const struct dst_entry *dst);
1928 void tcp_fastopen_init_key_once(struct net *net);
1929 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1930 			     struct tcp_fastopen_cookie *cookie);
1931 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1932 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1933 #define TCP_FASTOPEN_KEY_MAX 2
1934 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1935 	(TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1936 
1937 /* Fastopen key context */
1938 struct tcp_fastopen_context {
1939 	siphash_key_t	key[TCP_FASTOPEN_KEY_MAX];
1940 	int		num;
1941 	struct rcu_head	rcu;
1942 };
1943 
1944 void tcp_fastopen_active_disable(struct sock *sk);
1945 bool tcp_fastopen_active_should_disable(struct sock *sk);
1946 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1947 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1948 
1949 /* Caller needs to wrap with rcu_read_(un)lock() */
1950 static inline
1951 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1952 {
1953 	struct tcp_fastopen_context *ctx;
1954 
1955 	ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1956 	if (!ctx)
1957 		ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1958 	return ctx;
1959 }
1960 
1961 static inline
1962 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1963 			       const struct tcp_fastopen_cookie *orig)
1964 {
1965 	if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1966 	    orig->len == foc->len &&
1967 	    !memcmp(orig->val, foc->val, foc->len))
1968 		return true;
1969 	return false;
1970 }
1971 
1972 static inline
1973 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1974 {
1975 	return ctx->num;
1976 }
1977 
1978 /* Latencies incurred by various limits for a sender. They are
1979  * chronograph-like stats that are mutually exclusive.
1980  */
1981 enum tcp_chrono {
1982 	TCP_CHRONO_UNSPEC,
1983 	TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1984 	TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1985 	TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1986 	__TCP_CHRONO_MAX,
1987 };
1988 
1989 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1990 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1991 
1992 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1993  * the same memory storage than skb->destructor/_skb_refdst
1994  */
1995 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1996 {
1997 	skb->destructor = NULL;
1998 	skb->_skb_refdst = 0UL;
1999 }
2000 
2001 #define tcp_skb_tsorted_save(skb) {		\
2002 	unsigned long _save = skb->_skb_refdst;	\
2003 	skb->_skb_refdst = 0UL;
2004 
2005 #define tcp_skb_tsorted_restore(skb)		\
2006 	skb->_skb_refdst = _save;		\
2007 }
2008 
2009 void tcp_write_queue_purge(struct sock *sk);
2010 
2011 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
2012 {
2013 	return skb_rb_first(&sk->tcp_rtx_queue);
2014 }
2015 
2016 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
2017 {
2018 	return skb_rb_last(&sk->tcp_rtx_queue);
2019 }
2020 
2021 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
2022 {
2023 	return skb_peek_tail(&sk->sk_write_queue);
2024 }
2025 
2026 #define tcp_for_write_queue_from_safe(skb, tmp, sk)			\
2027 	skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
2028 
2029 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
2030 {
2031 	return skb_peek(&sk->sk_write_queue);
2032 }
2033 
2034 static inline bool tcp_skb_is_last(const struct sock *sk,
2035 				   const struct sk_buff *skb)
2036 {
2037 	return skb_queue_is_last(&sk->sk_write_queue, skb);
2038 }
2039 
2040 /**
2041  * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
2042  * @sk: socket
2043  *
2044  * Since the write queue can have a temporary empty skb in it,
2045  * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
2046  */
2047 static inline bool tcp_write_queue_empty(const struct sock *sk)
2048 {
2049 	const struct tcp_sock *tp = tcp_sk(sk);
2050 
2051 	return tp->write_seq == tp->snd_nxt;
2052 }
2053 
2054 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
2055 {
2056 	return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
2057 }
2058 
2059 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
2060 {
2061 	return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
2062 }
2063 
2064 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
2065 {
2066 	__skb_queue_tail(&sk->sk_write_queue, skb);
2067 
2068 	/* Queue it, remembering where we must start sending. */
2069 	if (sk->sk_write_queue.next == skb)
2070 		tcp_chrono_start(sk, TCP_CHRONO_BUSY);
2071 }
2072 
2073 /* Insert new before skb on the write queue of sk.  */
2074 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
2075 						  struct sk_buff *skb,
2076 						  struct sock *sk)
2077 {
2078 	__skb_queue_before(&sk->sk_write_queue, skb, new);
2079 }
2080 
2081 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
2082 {
2083 	tcp_skb_tsorted_anchor_cleanup(skb);
2084 	__skb_unlink(skb, &sk->sk_write_queue);
2085 }
2086 
2087 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
2088 
2089 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
2090 {
2091 	tcp_skb_tsorted_anchor_cleanup(skb);
2092 	rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
2093 }
2094 
2095 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
2096 {
2097 	list_del(&skb->tcp_tsorted_anchor);
2098 	tcp_rtx_queue_unlink(skb, sk);
2099 	tcp_wmem_free_skb(sk, skb);
2100 }
2101 
2102 static inline void tcp_write_collapse_fence(struct sock *sk)
2103 {
2104 	struct sk_buff *skb = tcp_write_queue_tail(sk);
2105 
2106 	if (skb)
2107 		TCP_SKB_CB(skb)->eor = 1;
2108 }
2109 
2110 static inline void tcp_push_pending_frames(struct sock *sk)
2111 {
2112 	if (tcp_send_head(sk)) {
2113 		struct tcp_sock *tp = tcp_sk(sk);
2114 
2115 		__tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
2116 	}
2117 }
2118 
2119 /* Start sequence of the skb just after the highest skb with SACKed
2120  * bit, valid only if sacked_out > 0 or when the caller has ensured
2121  * validity by itself.
2122  */
2123 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
2124 {
2125 	if (!tp->sacked_out)
2126 		return tp->snd_una;
2127 
2128 	if (tp->highest_sack == NULL)
2129 		return tp->snd_nxt;
2130 
2131 	return TCP_SKB_CB(tp->highest_sack)->seq;
2132 }
2133 
2134 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
2135 {
2136 	tcp_sk(sk)->highest_sack = skb_rb_next(skb);
2137 }
2138 
2139 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
2140 {
2141 	return tcp_sk(sk)->highest_sack;
2142 }
2143 
2144 static inline void tcp_highest_sack_reset(struct sock *sk)
2145 {
2146 	tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
2147 }
2148 
2149 /* Called when old skb is about to be deleted and replaced by new skb */
2150 static inline void tcp_highest_sack_replace(struct sock *sk,
2151 					    struct sk_buff *old,
2152 					    struct sk_buff *new)
2153 {
2154 	if (old == tcp_highest_sack(sk))
2155 		tcp_sk(sk)->highest_sack = new;
2156 }
2157 
2158 /* This helper checks if socket has IP_TRANSPARENT set */
2159 static inline bool inet_sk_transparent(const struct sock *sk)
2160 {
2161 	switch (sk->sk_state) {
2162 	case TCP_TIME_WAIT:
2163 		return inet_twsk(sk)->tw_transparent;
2164 	case TCP_NEW_SYN_RECV:
2165 		return inet_rsk(inet_reqsk(sk))->no_srccheck;
2166 	}
2167 	return inet_test_bit(TRANSPARENT, sk);
2168 }
2169 
2170 /* Determines whether this is a thin stream (which may suffer from
2171  * increased latency). Used to trigger latency-reducing mechanisms.
2172  */
2173 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
2174 {
2175 	return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
2176 }
2177 
2178 /* /proc */
2179 enum tcp_seq_states {
2180 	TCP_SEQ_STATE_LISTENING,
2181 	TCP_SEQ_STATE_ESTABLISHED,
2182 };
2183 
2184 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
2185 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
2186 void tcp_seq_stop(struct seq_file *seq, void *v);
2187 
2188 struct tcp_seq_afinfo {
2189 	sa_family_t			family;
2190 };
2191 
2192 struct tcp_iter_state {
2193 	struct seq_net_private	p;
2194 	enum tcp_seq_states	state;
2195 	struct sock		*syn_wait_sk;
2196 	int			bucket, offset, sbucket, num;
2197 	loff_t			last_pos;
2198 };
2199 
2200 extern struct request_sock_ops tcp_request_sock_ops;
2201 extern struct request_sock_ops tcp6_request_sock_ops;
2202 
2203 void tcp_v4_destroy_sock(struct sock *sk);
2204 
2205 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
2206 				netdev_features_t features);
2207 struct tcphdr *tcp_gro_pull_header(struct sk_buff *skb);
2208 struct sk_buff *tcp_gro_lookup(struct list_head *head, struct tcphdr *th);
2209 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb,
2210 				struct tcphdr *th);
2211 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
2212 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
2213 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff));
2214 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb));
2215 #ifdef CONFIG_INET
2216 void tcp_gro_complete(struct sk_buff *skb);
2217 #else
2218 static inline void tcp_gro_complete(struct sk_buff *skb) { }
2219 #endif
2220 
2221 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
2222 
2223 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
2224 {
2225 	struct net *net = sock_net((struct sock *)tp);
2226 	u32 val;
2227 
2228 	val = READ_ONCE(tp->notsent_lowat);
2229 
2230 	return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
2231 }
2232 
2233 bool tcp_stream_memory_free(const struct sock *sk, int wake);
2234 
2235 #ifdef CONFIG_PROC_FS
2236 int tcp4_proc_init(void);
2237 void tcp4_proc_exit(void);
2238 #endif
2239 
2240 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
2241 int tcp_conn_request(struct request_sock_ops *rsk_ops,
2242 		     const struct tcp_request_sock_ops *af_ops,
2243 		     struct sock *sk, struct sk_buff *skb);
2244 
2245 /* TCP af-specific functions */
2246 struct tcp_sock_af_ops {
2247 #ifdef CONFIG_TCP_MD5SIG
2248 	struct tcp_md5sig_key	*(*md5_lookup) (const struct sock *sk,
2249 						const struct sock *addr_sk);
2250 	int		(*calc_md5_hash)(char *location,
2251 					 const struct tcp_md5sig_key *md5,
2252 					 const struct sock *sk,
2253 					 const struct sk_buff *skb);
2254 	int		(*md5_parse)(struct sock *sk,
2255 				     int optname,
2256 				     sockptr_t optval,
2257 				     int optlen);
2258 #endif
2259 #ifdef CONFIG_TCP_AO
2260 	int (*ao_parse)(struct sock *sk, int optname, sockptr_t optval, int optlen);
2261 	struct tcp_ao_key *(*ao_lookup)(const struct sock *sk,
2262 					struct sock *addr_sk,
2263 					int sndid, int rcvid);
2264 	int (*ao_calc_key_sk)(struct tcp_ao_key *mkt, u8 *key,
2265 			      const struct sock *sk,
2266 			      __be32 sisn, __be32 disn, bool send);
2267 	int (*calc_ao_hash)(char *location, struct tcp_ao_key *ao,
2268 			    const struct sock *sk, const struct sk_buff *skb,
2269 			    const u8 *tkey, int hash_offset, u32 sne);
2270 #endif
2271 };
2272 
2273 struct tcp_request_sock_ops {
2274 	u16 mss_clamp;
2275 #ifdef CONFIG_TCP_MD5SIG
2276 	struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
2277 						 const struct sock *addr_sk);
2278 	int		(*calc_md5_hash) (char *location,
2279 					  const struct tcp_md5sig_key *md5,
2280 					  const struct sock *sk,
2281 					  const struct sk_buff *skb);
2282 #endif
2283 #ifdef CONFIG_TCP_AO
2284 	struct tcp_ao_key *(*ao_lookup)(const struct sock *sk,
2285 					struct request_sock *req,
2286 					int sndid, int rcvid);
2287 	int (*ao_calc_key)(struct tcp_ao_key *mkt, u8 *key, struct request_sock *sk);
2288 	int (*ao_synack_hash)(char *ao_hash, struct tcp_ao_key *mkt,
2289 			      struct request_sock *req, const struct sk_buff *skb,
2290 			      int hash_offset, u32 sne);
2291 #endif
2292 #ifdef CONFIG_SYN_COOKIES
2293 	__u32 (*cookie_init_seq)(const struct sk_buff *skb,
2294 				 __u16 *mss);
2295 #endif
2296 	struct dst_entry *(*route_req)(const struct sock *sk,
2297 				       struct sk_buff *skb,
2298 				       struct flowi *fl,
2299 				       struct request_sock *req,
2300 				       u32 tw_isn);
2301 	u32 (*init_seq)(const struct sk_buff *skb);
2302 	u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
2303 	int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2304 			   struct flowi *fl, struct request_sock *req,
2305 			   struct tcp_fastopen_cookie *foc,
2306 			   enum tcp_synack_type synack_type,
2307 			   struct sk_buff *syn_skb);
2308 };
2309 
2310 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2311 #if IS_ENABLED(CONFIG_IPV6)
2312 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2313 #endif
2314 
2315 #ifdef CONFIG_SYN_COOKIES
2316 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2317 					 const struct sock *sk, struct sk_buff *skb,
2318 					 __u16 *mss)
2319 {
2320 	tcp_synq_overflow(sk);
2321 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2322 	return ops->cookie_init_seq(skb, mss);
2323 }
2324 #else
2325 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2326 					 const struct sock *sk, struct sk_buff *skb,
2327 					 __u16 *mss)
2328 {
2329 	return 0;
2330 }
2331 #endif
2332 
2333 struct tcp_key {
2334 	union {
2335 		struct {
2336 			struct tcp_ao_key *ao_key;
2337 			char *traffic_key;
2338 			u32 sne;
2339 			u8 rcv_next;
2340 		};
2341 		struct tcp_md5sig_key *md5_key;
2342 	};
2343 	enum {
2344 		TCP_KEY_NONE = 0,
2345 		TCP_KEY_MD5,
2346 		TCP_KEY_AO,
2347 	} type;
2348 };
2349 
2350 static inline void tcp_get_current_key(const struct sock *sk,
2351 				       struct tcp_key *out)
2352 {
2353 #if defined(CONFIG_TCP_AO) || defined(CONFIG_TCP_MD5SIG)
2354 	const struct tcp_sock *tp = tcp_sk(sk);
2355 #endif
2356 
2357 #ifdef CONFIG_TCP_AO
2358 	if (static_branch_unlikely(&tcp_ao_needed.key)) {
2359 		struct tcp_ao_info *ao;
2360 
2361 		ao = rcu_dereference_protected(tp->ao_info,
2362 					       lockdep_sock_is_held(sk));
2363 		if (ao) {
2364 			out->ao_key = READ_ONCE(ao->current_key);
2365 			out->type = TCP_KEY_AO;
2366 			return;
2367 		}
2368 	}
2369 #endif
2370 #ifdef CONFIG_TCP_MD5SIG
2371 	if (static_branch_unlikely(&tcp_md5_needed.key) &&
2372 	    rcu_access_pointer(tp->md5sig_info)) {
2373 		out->md5_key = tp->af_specific->md5_lookup(sk, sk);
2374 		if (out->md5_key) {
2375 			out->type = TCP_KEY_MD5;
2376 			return;
2377 		}
2378 	}
2379 #endif
2380 	out->type = TCP_KEY_NONE;
2381 }
2382 
2383 static inline bool tcp_key_is_md5(const struct tcp_key *key)
2384 {
2385 	if (static_branch_tcp_md5())
2386 		return key->type == TCP_KEY_MD5;
2387 	return false;
2388 }
2389 
2390 static inline bool tcp_key_is_ao(const struct tcp_key *key)
2391 {
2392 	if (static_branch_tcp_ao())
2393 		return key->type == TCP_KEY_AO;
2394 	return false;
2395 }
2396 
2397 int tcpv4_offload_init(void);
2398 
2399 void tcp_v4_init(void);
2400 void tcp_init(void);
2401 
2402 /* tcp_recovery.c */
2403 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2404 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2405 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2406 				u32 reo_wnd);
2407 extern bool tcp_rack_mark_lost(struct sock *sk);
2408 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2409 			     u64 xmit_time);
2410 extern void tcp_rack_reo_timeout(struct sock *sk);
2411 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2412 
2413 /* tcp_plb.c */
2414 
2415 /*
2416  * Scaling factor for fractions in PLB. For example, tcp_plb_update_state
2417  * expects cong_ratio which represents fraction of traffic that experienced
2418  * congestion over a single RTT. In order to avoid floating point operations,
2419  * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in.
2420  */
2421 #define TCP_PLB_SCALE 8
2422 
2423 /* State for PLB (Protective Load Balancing) for a single TCP connection. */
2424 struct tcp_plb_state {
2425 	u8	consec_cong_rounds:5, /* consecutive congested rounds */
2426 		unused:3;
2427 	u32	pause_until; /* jiffies32 when PLB can resume rerouting */
2428 };
2429 
2430 static inline void tcp_plb_init(const struct sock *sk,
2431 				struct tcp_plb_state *plb)
2432 {
2433 	plb->consec_cong_rounds = 0;
2434 	plb->pause_until = 0;
2435 }
2436 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb,
2437 			  const int cong_ratio);
2438 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb);
2439 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb);
2440 
2441 static inline void tcp_warn_once(const struct sock *sk, bool cond, const char *str)
2442 {
2443 	WARN_ONCE(cond,
2444 		  "%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",
2445 		  str,
2446 		  tcp_snd_cwnd(tcp_sk(sk)),
2447 		  tcp_sk(sk)->packets_out, tcp_sk(sk)->sacked_out,
2448 		  tcp_sk(sk)->lost_out, tcp_sk(sk)->retrans_out,
2449 		  tcp_sk(sk)->tlp_high_seq, sk->sk_state,
2450 		  inet_csk(sk)->icsk_ca_state,
2451 		  tcp_sk(sk)->advmss, tcp_sk(sk)->mss_cache,
2452 		  inet_csk(sk)->icsk_pmtu_cookie);
2453 }
2454 
2455 /* At how many usecs into the future should the RTO fire? */
2456 static inline s64 tcp_rto_delta_us(const struct sock *sk)
2457 {
2458 	const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2459 	u32 rto = inet_csk(sk)->icsk_rto;
2460 
2461 	if (likely(skb)) {
2462 		u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2463 
2464 		return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2465 	} else {
2466 		tcp_warn_once(sk, 1, "rtx queue empty: ");
2467 		return jiffies_to_usecs(rto);
2468 	}
2469 
2470 }
2471 
2472 /*
2473  * Save and compile IPv4 options, return a pointer to it
2474  */
2475 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2476 							 struct sk_buff *skb)
2477 {
2478 	const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2479 	struct ip_options_rcu *dopt = NULL;
2480 
2481 	if (opt->optlen) {
2482 		int opt_size = sizeof(*dopt) + opt->optlen;
2483 
2484 		dopt = kmalloc(opt_size, GFP_ATOMIC);
2485 		if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2486 			kfree(dopt);
2487 			dopt = NULL;
2488 		}
2489 	}
2490 	return dopt;
2491 }
2492 
2493 /* locally generated TCP pure ACKs have skb->truesize == 2
2494  * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2495  * This is much faster than dissecting the packet to find out.
2496  * (Think of GRE encapsulations, IPv4, IPv6, ...)
2497  */
2498 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2499 {
2500 	return skb->truesize == 2;
2501 }
2502 
2503 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2504 {
2505 	skb->truesize = 2;
2506 }
2507 
2508 static inline int tcp_inq(struct sock *sk)
2509 {
2510 	struct tcp_sock *tp = tcp_sk(sk);
2511 	int answ;
2512 
2513 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2514 		answ = 0;
2515 	} else if (sock_flag(sk, SOCK_URGINLINE) ||
2516 		   !tp->urg_data ||
2517 		   before(tp->urg_seq, tp->copied_seq) ||
2518 		   !before(tp->urg_seq, tp->rcv_nxt)) {
2519 
2520 		answ = tp->rcv_nxt - tp->copied_seq;
2521 
2522 		/* Subtract 1, if FIN was received */
2523 		if (answ && sock_flag(sk, SOCK_DONE))
2524 			answ--;
2525 	} else {
2526 		answ = tp->urg_seq - tp->copied_seq;
2527 	}
2528 
2529 	return answ;
2530 }
2531 
2532 int tcp_peek_len(struct socket *sock);
2533 
2534 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2535 {
2536 	u16 segs_in;
2537 
2538 	segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2539 
2540 	/* We update these fields while other threads might
2541 	 * read them from tcp_get_info()
2542 	 */
2543 	WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in);
2544 	if (skb->len > tcp_hdrlen(skb))
2545 		WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in);
2546 }
2547 
2548 /*
2549  * TCP listen path runs lockless.
2550  * We forced "struct sock" to be const qualified to make sure
2551  * we don't modify one of its field by mistake.
2552  * Here, we increment sk_drops which is an atomic_t, so we can safely
2553  * make sock writable again.
2554  */
2555 static inline void tcp_listendrop(const struct sock *sk)
2556 {
2557 	atomic_inc(&((struct sock *)sk)->sk_drops);
2558 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2559 }
2560 
2561 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2562 
2563 /*
2564  * Interface for adding Upper Level Protocols over TCP
2565  */
2566 
2567 #define TCP_ULP_NAME_MAX	16
2568 #define TCP_ULP_MAX		128
2569 #define TCP_ULP_BUF_MAX		(TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2570 
2571 struct tcp_ulp_ops {
2572 	struct list_head	list;
2573 
2574 	/* initialize ulp */
2575 	int (*init)(struct sock *sk);
2576 	/* update ulp */
2577 	void (*update)(struct sock *sk, struct proto *p,
2578 		       void (*write_space)(struct sock *sk));
2579 	/* cleanup ulp */
2580 	void (*release)(struct sock *sk);
2581 	/* diagnostic */
2582 	int (*get_info)(struct sock *sk, struct sk_buff *skb);
2583 	size_t (*get_info_size)(const struct sock *sk);
2584 	/* clone ulp */
2585 	void (*clone)(const struct request_sock *req, struct sock *newsk,
2586 		      const gfp_t priority);
2587 
2588 	char		name[TCP_ULP_NAME_MAX];
2589 	struct module	*owner;
2590 };
2591 int tcp_register_ulp(struct tcp_ulp_ops *type);
2592 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2593 int tcp_set_ulp(struct sock *sk, const char *name);
2594 void tcp_get_available_ulp(char *buf, size_t len);
2595 void tcp_cleanup_ulp(struct sock *sk);
2596 void tcp_update_ulp(struct sock *sk, struct proto *p,
2597 		    void (*write_space)(struct sock *sk));
2598 
2599 #define MODULE_ALIAS_TCP_ULP(name)				\
2600 	__MODULE_INFO(alias, alias_userspace, name);		\
2601 	__MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2602 
2603 #ifdef CONFIG_NET_SOCK_MSG
2604 struct sk_msg;
2605 struct sk_psock;
2606 
2607 #ifdef CONFIG_BPF_SYSCALL
2608 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
2609 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2610 #endif /* CONFIG_BPF_SYSCALL */
2611 
2612 #ifdef CONFIG_INET
2613 void tcp_eat_skb(struct sock *sk, struct sk_buff *skb);
2614 #else
2615 static inline void tcp_eat_skb(struct sock *sk, struct sk_buff *skb)
2616 {
2617 }
2618 #endif
2619 
2620 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress,
2621 			  struct sk_msg *msg, u32 bytes, int flags);
2622 #endif /* CONFIG_NET_SOCK_MSG */
2623 
2624 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG)
2625 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2626 {
2627 }
2628 #endif
2629 
2630 #ifdef CONFIG_CGROUP_BPF
2631 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2632 				      struct sk_buff *skb,
2633 				      unsigned int end_offset)
2634 {
2635 	skops->skb = skb;
2636 	skops->skb_data_end = skb->data + end_offset;
2637 }
2638 #else
2639 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2640 				      struct sk_buff *skb,
2641 				      unsigned int end_offset)
2642 {
2643 }
2644 #endif
2645 
2646 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2647  * is < 0, then the BPF op failed (for example if the loaded BPF
2648  * program does not support the chosen operation or there is no BPF
2649  * program loaded).
2650  */
2651 #ifdef CONFIG_BPF
2652 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2653 {
2654 	struct bpf_sock_ops_kern sock_ops;
2655 	int ret;
2656 
2657 	memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2658 	if (sk_fullsock(sk)) {
2659 		sock_ops.is_fullsock = 1;
2660 		sock_owned_by_me(sk);
2661 	}
2662 
2663 	sock_ops.sk = sk;
2664 	sock_ops.op = op;
2665 	if (nargs > 0)
2666 		memcpy(sock_ops.args, args, nargs * sizeof(*args));
2667 
2668 	ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2669 	if (ret == 0)
2670 		ret = sock_ops.reply;
2671 	else
2672 		ret = -1;
2673 	return ret;
2674 }
2675 
2676 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2677 {
2678 	u32 args[2] = {arg1, arg2};
2679 
2680 	return tcp_call_bpf(sk, op, 2, args);
2681 }
2682 
2683 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2684 				    u32 arg3)
2685 {
2686 	u32 args[3] = {arg1, arg2, arg3};
2687 
2688 	return tcp_call_bpf(sk, op, 3, args);
2689 }
2690 
2691 #else
2692 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2693 {
2694 	return -EPERM;
2695 }
2696 
2697 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2698 {
2699 	return -EPERM;
2700 }
2701 
2702 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2703 				    u32 arg3)
2704 {
2705 	return -EPERM;
2706 }
2707 
2708 #endif
2709 
2710 static inline u32 tcp_timeout_init(struct sock *sk)
2711 {
2712 	int timeout;
2713 
2714 	timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2715 
2716 	if (timeout <= 0)
2717 		timeout = TCP_TIMEOUT_INIT;
2718 	return min_t(int, timeout, TCP_RTO_MAX);
2719 }
2720 
2721 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2722 {
2723 	int rwnd;
2724 
2725 	rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2726 
2727 	if (rwnd < 0)
2728 		rwnd = 0;
2729 	return rwnd;
2730 }
2731 
2732 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2733 {
2734 	return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2735 }
2736 
2737 static inline void tcp_bpf_rtt(struct sock *sk, long mrtt, u32 srtt)
2738 {
2739 	if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2740 		tcp_call_bpf_2arg(sk, BPF_SOCK_OPS_RTT_CB, mrtt, srtt);
2741 }
2742 
2743 #if IS_ENABLED(CONFIG_SMC)
2744 extern struct static_key_false tcp_have_smc;
2745 #endif
2746 
2747 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2748 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2749 			     void (*cad)(struct sock *sk, u32 ack_seq));
2750 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2751 void clean_acked_data_flush(void);
2752 #endif
2753 
2754 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2755 static inline void tcp_add_tx_delay(struct sk_buff *skb,
2756 				    const struct tcp_sock *tp)
2757 {
2758 	if (static_branch_unlikely(&tcp_tx_delay_enabled))
2759 		skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2760 }
2761 
2762 /* Compute Earliest Departure Time for some control packets
2763  * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2764  */
2765 static inline u64 tcp_transmit_time(const struct sock *sk)
2766 {
2767 	if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2768 		u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2769 			tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2770 
2771 		return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
2772 	}
2773 	return 0;
2774 }
2775 
2776 static inline int tcp_parse_auth_options(const struct tcphdr *th,
2777 		const u8 **md5_hash, const struct tcp_ao_hdr **aoh)
2778 {
2779 	const u8 *md5_tmp, *ao_tmp;
2780 	int ret;
2781 
2782 	ret = tcp_do_parse_auth_options(th, &md5_tmp, &ao_tmp);
2783 	if (ret)
2784 		return ret;
2785 
2786 	if (md5_hash)
2787 		*md5_hash = md5_tmp;
2788 
2789 	if (aoh) {
2790 		if (!ao_tmp)
2791 			*aoh = NULL;
2792 		else
2793 			*aoh = (struct tcp_ao_hdr *)(ao_tmp - 2);
2794 	}
2795 
2796 	return 0;
2797 }
2798 
2799 static inline bool tcp_ao_required(struct sock *sk, const void *saddr,
2800 				   int family, int l3index, bool stat_inc)
2801 {
2802 #ifdef CONFIG_TCP_AO
2803 	struct tcp_ao_info *ao_info;
2804 	struct tcp_ao_key *ao_key;
2805 
2806 	if (!static_branch_unlikely(&tcp_ao_needed.key))
2807 		return false;
2808 
2809 	ao_info = rcu_dereference_check(tcp_sk(sk)->ao_info,
2810 					lockdep_sock_is_held(sk));
2811 	if (!ao_info)
2812 		return false;
2813 
2814 	ao_key = tcp_ao_do_lookup(sk, l3index, saddr, family, -1, -1);
2815 	if (ao_info->ao_required || ao_key) {
2816 		if (stat_inc) {
2817 			NET_INC_STATS(sock_net(sk), LINUX_MIB_TCPAOREQUIRED);
2818 			atomic64_inc(&ao_info->counters.ao_required);
2819 		}
2820 		return true;
2821 	}
2822 #endif
2823 	return false;
2824 }
2825 
2826 enum skb_drop_reason tcp_inbound_hash(struct sock *sk,
2827 		const struct request_sock *req, const struct sk_buff *skb,
2828 		const void *saddr, const void *daddr,
2829 		int family, int dif, int sdif);
2830 
2831 #endif	/* _TCP_H */
2832