1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3  * INET		An implementation of the TCP/IP protocol suite for the LINUX
4  *		operating system.  INET is implemented using the  BSD Socket
5  *		interface as the means of communication with the user level.
6  *
7  *		Definitions for the TCP module.
8  *
9  * Version:	@(#)tcp.h	1.0.5	05/23/93
10  *
11  * Authors:	Ross Biro
12  *		Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
13  */
14 #ifndef _TCP_H
15 #define _TCP_H
16 
17 #define FASTRETRANS_DEBUG 1
18 
19 #include <linux/list.h>
20 #include <linux/tcp.h>
21 #include <linux/bug.h>
22 #include <linux/slab.h>
23 #include <linux/cache.h>
24 #include <linux/percpu.h>
25 #include <linux/skbuff.h>
26 #include <linux/kref.h>
27 #include <linux/ktime.h>
28 #include <linux/indirect_call_wrapper.h>
29 
30 #include <net/inet_connection_sock.h>
31 #include <net/inet_timewait_sock.h>
32 #include <net/inet_hashtables.h>
33 #include <net/checksum.h>
34 #include <net/request_sock.h>
35 #include <net/sock_reuseport.h>
36 #include <net/sock.h>
37 #include <net/snmp.h>
38 #include <net/ip.h>
39 #include <net/tcp_states.h>
40 #include <net/inet_ecn.h>
41 #include <net/dst.h>
42 #include <net/mptcp.h>
43 #ifdef CONFIG_NEWIP
44 #include <linux/nip.h> /* NIP */
45 #endif
46 #include <linux/seq_file.h>
47 #include <linux/memcontrol.h>
48 #include <linux/bpf-cgroup.h>
49 #include <linux/siphash.h>
50 
51 extern struct inet_hashinfo tcp_hashinfo;
52 
53 DECLARE_PER_CPU(unsigned int, tcp_orphan_count);
54 int tcp_orphan_count_sum(void);
55 
56 void tcp_time_wait(struct sock *sk, int state, int timeo);
57 
58 #define MAX_TCP_HEADER	L1_CACHE_ALIGN(128 + MAX_HEADER)
59 #define MAX_TCP_OPTION_SPACE 40
60 #define TCP_MIN_SND_MSS		48
61 #define TCP_MIN_GSO_SIZE	(TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
62 
63 /*
64  * Never offer a window over 32767 without using window scaling. Some
65  * poor stacks do signed 16bit maths!
66  */
67 #define MAX_TCP_WINDOW		32767U
68 
69 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
70 #define TCP_MIN_MSS		88U
71 
72 /* The initial MTU to use for probing */
73 #define TCP_BASE_MSS		1024
74 
75 /* probing interval, default to 10 minutes as per RFC4821 */
76 #define TCP_PROBE_INTERVAL	600
77 
78 /* Specify interval when tcp mtu probing will stop */
79 #define TCP_PROBE_THRESHOLD	8
80 
81 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
82 #define TCP_FASTRETRANS_THRESH 3
83 
84 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
85 #define TCP_MAX_QUICKACKS	16U
86 
87 /* Maximal number of window scale according to RFC1323 */
88 #define TCP_MAX_WSCALE		14U
89 
90 /* urg_data states */
91 #define TCP_URG_VALID	0x0100
92 #define TCP_URG_NOTYET	0x0200
93 #define TCP_URG_READ	0x0400
94 
95 #define TCP_RETR1	3	/*
96 				 * This is how many retries it does before it
97 				 * tries to figure out if the gateway is
98 				 * down. Minimal RFC value is 3; it corresponds
99 				 * to ~3sec-8min depending on RTO.
100 				 */
101 
102 #define TCP_RETR2	15	/*
103 				 * This should take at least
104 				 * 90 minutes to time out.
105 				 * RFC1122 says that the limit is 100 sec.
106 				 * 15 is ~13-30min depending on RTO.
107 				 */
108 
109 #define TCP_SYN_RETRIES	 6	/* This is how many retries are done
110 				 * when active opening a connection.
111 				 * RFC1122 says the minimum retry MUST
112 				 * be at least 180secs.  Nevertheless
113 				 * this value is corresponding to
114 				 * 63secs of retransmission with the
115 				 * current initial RTO.
116 				 */
117 
118 #define TCP_SYNACK_RETRIES 5	/* This is how may retries are done
119 				 * when passive opening a connection.
120 				 * This is corresponding to 31secs of
121 				 * retransmission with the current
122 				 * initial RTO.
123 				 */
124 
125 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
126 				  * state, about 60 seconds	*/
127 #define TCP_FIN_TIMEOUT	TCP_TIMEWAIT_LEN
128                                  /* BSD style FIN_WAIT2 deadlock breaker.
129 				  * It used to be 3min, new value is 60sec,
130 				  * to combine FIN-WAIT-2 timeout with
131 				  * TIME-WAIT timer.
132 				  */
133 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */
134 
135 #define TCP_DELACK_MAX	((unsigned)(HZ/5))	/* maximal time to delay before sending an ACK */
136 #if HZ >= 100
137 #define TCP_DELACK_MIN	((unsigned)(HZ/25))	/* minimal time to delay before sending an ACK */
138 #define TCP_ATO_MIN	((unsigned)(HZ/25))
139 #else
140 #define TCP_DELACK_MIN	4U
141 #define TCP_ATO_MIN	4U
142 #endif
143 #define TCP_RTO_MAX	((unsigned)(120*HZ))
144 #define TCP_RTO_MIN	((unsigned)(HZ/5))
145 #define TCP_TIMEOUT_MIN	(2U) /* Min timeout for TCP timers in jiffies */
146 
147 #define TCP_TIMEOUT_MIN_US (2*USEC_PER_MSEC) /* Min TCP timeout in microsecs */
148 
149 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ))	/* RFC6298 2.1 initial RTO value	*/
150 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ))	/* RFC 1122 initial RTO value, now
151 						 * used as a fallback RTO for the
152 						 * initial data transmission if no
153 						 * valid RTT sample has been acquired,
154 						 * most likely due to retrans in 3WHS.
155 						 */
156 
157 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
158 					                 * for local resources.
159 					                 */
160 #define TCP_KEEPALIVE_TIME	(120*60*HZ)	/* two hours */
161 #define TCP_KEEPALIVE_PROBES	9		/* Max of 9 keepalive probes	*/
162 #define TCP_KEEPALIVE_INTVL	(75*HZ)
163 
164 #define MAX_TCP_KEEPIDLE	32767
165 #define MAX_TCP_KEEPINTVL	32767
166 #define MAX_TCP_KEEPCNT		127
167 #define MAX_TCP_SYNCNT		127
168 
169 #define TCP_SYNQ_INTERVAL	(HZ/5)	/* Period of SYNACK timer */
170 
171 #define TCP_PAWS_24DAYS	(60 * 60 * 24 * 24)
172 #define TCP_PAWS_MSL	60		/* Per-host timestamps are invalidated
173 					 * after this time. It should be equal
174 					 * (or greater than) TCP_TIMEWAIT_LEN
175 					 * to provide reliability equal to one
176 					 * provided by timewait state.
177 					 */
178 #define TCP_PAWS_WINDOW	1		/* Replay window for per-host
179 					 * timestamps. It must be less than
180 					 * minimal timewait lifetime.
181 					 */
182 /*
183  *	TCP option
184  */
185 
186 #define TCPOPT_NOP		1	/* Padding */
187 #define TCPOPT_EOL		0	/* End of options */
188 #define TCPOPT_MSS		2	/* Segment size negotiating */
189 #define TCPOPT_WINDOW		3	/* Window scaling */
190 #define TCPOPT_SACK_PERM        4       /* SACK Permitted */
191 #define TCPOPT_SACK             5       /* SACK Block */
192 #define TCPOPT_TIMESTAMP	8	/* Better RTT estimations/PAWS */
193 #define TCPOPT_MD5SIG		19	/* MD5 Signature (RFC2385) */
194 #define TCPOPT_MPTCP		30	/* Multipath TCP (RFC6824) */
195 #define TCPOPT_FASTOPEN		34	/* Fast open (RFC7413) */
196 #define TCPOPT_EXP		254	/* Experimental */
197 /* Magic number to be after the option value for sharing TCP
198  * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
199  */
200 #define TCPOPT_FASTOPEN_MAGIC	0xF989
201 #define TCPOPT_SMC_MAGIC	0xE2D4C3D9
202 
203 /*
204  *     TCP option lengths
205  */
206 
207 #define TCPOLEN_MSS            4
208 #define TCPOLEN_WINDOW         3
209 #define TCPOLEN_SACK_PERM      2
210 #define TCPOLEN_TIMESTAMP      10
211 #define TCPOLEN_MD5SIG         18
212 #define TCPOLEN_FASTOPEN_BASE  2
213 #define TCPOLEN_EXP_FASTOPEN_BASE  4
214 #define TCPOLEN_EXP_SMC_BASE   6
215 
216 /* But this is what stacks really send out. */
217 #define TCPOLEN_TSTAMP_ALIGNED		12
218 #define TCPOLEN_WSCALE_ALIGNED		4
219 #define TCPOLEN_SACKPERM_ALIGNED	4
220 #define TCPOLEN_SACK_BASE		2
221 #define TCPOLEN_SACK_BASE_ALIGNED	4
222 #define TCPOLEN_SACK_PERBLOCK		8
223 #define TCPOLEN_MD5SIG_ALIGNED		20
224 #define TCPOLEN_MSS_ALIGNED		4
225 #define TCPOLEN_EXP_SMC_BASE_ALIGNED	8
226 
227 /* Flags in tp->nonagle */
228 #define TCP_NAGLE_OFF		1	/* Nagle's algo is disabled */
229 #define TCP_NAGLE_CORK		2	/* Socket is corked	    */
230 #define TCP_NAGLE_PUSH		4	/* Cork is overridden for already queued data */
231 
232 /* TCP thin-stream limits */
233 #define TCP_THIN_LINEAR_RETRIES 6       /* After 6 linear retries, do exp. backoff */
234 
235 /* TCP initial congestion window as per rfc6928 */
236 #define TCP_INIT_CWND		10
237 
238 /* Bit Flags for sysctl_tcp_fastopen */
239 #define	TFO_CLIENT_ENABLE	1
240 #define	TFO_SERVER_ENABLE	2
241 #define	TFO_CLIENT_NO_COOKIE	4	/* Data in SYN w/o cookie option */
242 
243 /* Accept SYN data w/o any cookie option */
244 #define	TFO_SERVER_COOKIE_NOT_REQD	0x200
245 
246 /* Force enable TFO on all listeners, i.e., not requiring the
247  * TCP_FASTOPEN socket option.
248  */
249 #define	TFO_SERVER_WO_SOCKOPT1	0x400
250 
251 
252 /* sysctl variables for tcp */
253 extern int sysctl_tcp_max_orphans;
254 extern long sysctl_tcp_mem[3];
255 
256 #define TCP_RACK_LOSS_DETECTION  0x1 /* Use RACK to detect losses */
257 #define TCP_RACK_STATIC_REO_WND  0x2 /* Use static RACK reo wnd */
258 #define TCP_RACK_NO_DUPTHRESH    0x4 /* Do not use DUPACK threshold in RACK */
259 
260 extern atomic_long_t tcp_memory_allocated;
261 extern struct percpu_counter tcp_sockets_allocated;
262 extern unsigned long tcp_memory_pressure;
263 
264 /* optimized version of sk_under_memory_pressure() for TCP sockets */
tcp_under_memory_pressure(const struct sock *sk)265 static inline bool tcp_under_memory_pressure(const struct sock *sk)
266 {
267 	if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
268 	    mem_cgroup_under_socket_pressure(sk->sk_memcg))
269 		return true;
270 
271 	return READ_ONCE(tcp_memory_pressure);
272 }
273 /*
274  * The next routines deal with comparing 32 bit unsigned ints
275  * and worry about wraparound (automatic with unsigned arithmetic).
276  */
277 
before(__u32 seq1, __u32 seq2)278 static inline bool before(__u32 seq1, __u32 seq2)
279 {
280         return (__s32)(seq1-seq2) < 0;
281 }
282 #define after(seq2, seq1) 	before(seq1, seq2)
283 
284 /* is s2<=s1<=s3 ? */
between(__u32 seq1, __u32 seq2, __u32 seq3)285 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
286 {
287 	return seq3 - seq2 >= seq1 - seq2;
288 }
289 
tcp_out_of_memory(struct sock *sk)290 static inline bool tcp_out_of_memory(struct sock *sk)
291 {
292 	if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
293 	    sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
294 		return true;
295 	return false;
296 }
297 
298 void sk_forced_mem_schedule(struct sock *sk, int size);
299 
300 bool tcp_check_oom(struct sock *sk, int shift);
301 
302 
303 extern struct proto tcp_prot;
304 
305 #define TCP_INC_STATS(net, field)	SNMP_INC_STATS((net)->mib.tcp_statistics, field)
306 #define __TCP_INC_STATS(net, field)	__SNMP_INC_STATS((net)->mib.tcp_statistics, field)
307 #define TCP_DEC_STATS(net, field)	SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
308 #define TCP_ADD_STATS(net, field, val)	SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
309 
310 void tcp_tasklet_init(void);
311 
312 int tcp_v4_err(struct sk_buff *skb, u32);
313 
314 void tcp_shutdown(struct sock *sk, int how);
315 
316 int tcp_v4_early_demux(struct sk_buff *skb);
317 int tcp_v4_rcv(struct sk_buff *skb);
318 
319 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
320 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
321 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
322 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
323 		 int flags);
324 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
325 			size_t size, int flags);
326 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
327 		 size_t size, int flags);
328 int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
329 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
330 	      int size_goal);
331 void tcp_release_cb(struct sock *sk);
332 void tcp_wfree(struct sk_buff *skb);
333 void tcp_write_timer_handler(struct sock *sk);
334 void tcp_delack_timer_handler(struct sock *sk);
335 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
336 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
337 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
338 void tcp_rcv_space_adjust(struct sock *sk);
339 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
340 void tcp_twsk_destructor(struct sock *sk);
341 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
342 			struct pipe_inode_info *pipe, size_t len,
343 			unsigned int flags);
344 
tcp_dec_quickack_mode(struct sock *sk)345 static inline void tcp_dec_quickack_mode(struct sock *sk)
346 {
347 	struct inet_connection_sock *icsk = inet_csk(sk);
348 
349 	if (icsk->icsk_ack.quick) {
350 		/* How many ACKs S/ACKing new data have we sent? */
351 		const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0;
352 
353 		if (pkts >= icsk->icsk_ack.quick) {
354 			icsk->icsk_ack.quick = 0;
355 			/* Leaving quickack mode we deflate ATO. */
356 			icsk->icsk_ack.ato   = TCP_ATO_MIN;
357 		} else
358 			icsk->icsk_ack.quick -= pkts;
359 	}
360 }
361 
362 #define	TCP_ECN_OK		1
363 #define	TCP_ECN_QUEUE_CWR	2
364 #define	TCP_ECN_DEMAND_CWR	4
365 #define	TCP_ECN_SEEN		8
366 
367 enum tcp_tw_status {
368 	TCP_TW_SUCCESS = 0,
369 	TCP_TW_RST = 1,
370 	TCP_TW_ACK = 2,
371 	TCP_TW_SYN = 3
372 };
373 
374 
375 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
376 					      struct sk_buff *skb,
377 					      const struct tcphdr *th);
378 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
379 			   struct request_sock *req, bool fastopen,
380 			   bool *lost_race);
381 int tcp_child_process(struct sock *parent, struct sock *child,
382 		      struct sk_buff *skb);
383 void tcp_enter_loss(struct sock *sk);
384 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
385 void tcp_clear_retrans(struct tcp_sock *tp);
386 void tcp_update_metrics(struct sock *sk);
387 void tcp_init_metrics(struct sock *sk);
388 void tcp_metrics_init(void);
389 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
390 void __tcp_close(struct sock *sk, long timeout);
391 void tcp_close(struct sock *sk, long timeout);
392 void tcp_init_sock(struct sock *sk);
393 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb);
394 __poll_t tcp_poll(struct file *file, struct socket *sock,
395 		      struct poll_table_struct *wait);
396 int tcp_getsockopt(struct sock *sk, int level, int optname,
397 		   char __user *optval, int __user *optlen);
398 bool tcp_bpf_bypass_getsockopt(int level, int optname);
399 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
400 		   unsigned int optlen);
401 void tcp_set_keepalive(struct sock *sk, int val);
402 void tcp_syn_ack_timeout(const struct request_sock *req);
403 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
404 		int flags, int *addr_len);
405 int tcp_set_rcvlowat(struct sock *sk, int val);
406 void tcp_data_ready(struct sock *sk);
407 #ifdef CONFIG_MMU
408 int tcp_mmap(struct file *file, struct socket *sock,
409 	     struct vm_area_struct *vma);
410 #endif
411 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
412 		       struct tcp_options_received *opt_rx,
413 		       int estab, struct tcp_fastopen_cookie *foc);
414 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
415 
416 /*
417  *	BPF SKB-less helpers
418  */
419 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
420 			 struct tcphdr *th, u32 *cookie);
421 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
422 			 struct tcphdr *th, u32 *cookie);
423 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
424 			  const struct tcp_request_sock_ops *af_ops,
425 			  struct sock *sk, struct tcphdr *th);
426 /*
427  *	TCP v4 functions exported for the inet6 API
428  */
429 
430 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
431 void tcp_v4_mtu_reduced(struct sock *sk);
432 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
433 void tcp_ld_RTO_revert(struct sock *sk, u32 seq);
434 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
435 struct sock *tcp_create_openreq_child(const struct sock *sk,
436 				      struct request_sock *req,
437 				      struct sk_buff *skb);
438 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
439 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
440 				  struct request_sock *req,
441 				  struct dst_entry *dst,
442 				  struct request_sock *req_unhash,
443 				  bool *own_req);
444 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
445 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
446 int tcp_connect(struct sock *sk);
447 enum tcp_synack_type {
448 	TCP_SYNACK_NORMAL,
449 	TCP_SYNACK_FASTOPEN,
450 	TCP_SYNACK_COOKIE,
451 };
452 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
453 				struct request_sock *req,
454 				struct tcp_fastopen_cookie *foc,
455 				enum tcp_synack_type synack_type,
456 				struct sk_buff *syn_skb);
457 int tcp_disconnect(struct sock *sk, int flags);
458 
459 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
460 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
461 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
462 
463 /* From syncookies.c */
464 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
465 				 struct request_sock *req,
466 				 struct dst_entry *dst, u32 tsoff);
467 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
468 		      u32 cookie);
469 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
470 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
471 					    const struct tcp_request_sock_ops *af_ops,
472 					    struct sock *sk, struct sk_buff *skb);
473 #ifdef CONFIG_SYN_COOKIES
474 
475 /* Syncookies use a monotonic timer which increments every 60 seconds.
476  * This counter is used both as a hash input and partially encoded into
477  * the cookie value.  A cookie is only validated further if the delta
478  * between the current counter value and the encoded one is less than this,
479  * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
480  * the counter advances immediately after a cookie is generated).
481  */
482 #define MAX_SYNCOOKIE_AGE	2
483 #define TCP_SYNCOOKIE_PERIOD	(60 * HZ)
484 #define TCP_SYNCOOKIE_VALID	(MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
485 
486 /* syncookies: remember time of last synqueue overflow
487  * But do not dirty this field too often (once per second is enough)
488  * It is racy as we do not hold a lock, but race is very minor.
489  */
tcp_synq_overflow(const struct sock *sk)490 static inline void tcp_synq_overflow(const struct sock *sk)
491 {
492 	unsigned int last_overflow;
493 	unsigned int now = jiffies;
494 
495 	if (sk->sk_reuseport) {
496 		struct sock_reuseport *reuse;
497 
498 		reuse = rcu_dereference(sk->sk_reuseport_cb);
499 		if (likely(reuse)) {
500 			last_overflow = READ_ONCE(reuse->synq_overflow_ts);
501 			if (!time_between32(now, last_overflow,
502 					    last_overflow + HZ))
503 				WRITE_ONCE(reuse->synq_overflow_ts, now);
504 			return;
505 		}
506 	}
507 
508 	last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
509 	if (!time_between32(now, last_overflow, last_overflow + HZ))
510 		WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
511 }
512 
513 /* syncookies: no recent synqueue overflow on this listening socket? */
tcp_synq_no_recent_overflow(const struct sock *sk)514 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
515 {
516 	unsigned int last_overflow;
517 	unsigned int now = jiffies;
518 
519 	if (sk->sk_reuseport) {
520 		struct sock_reuseport *reuse;
521 
522 		reuse = rcu_dereference(sk->sk_reuseport_cb);
523 		if (likely(reuse)) {
524 			last_overflow = READ_ONCE(reuse->synq_overflow_ts);
525 			return !time_between32(now, last_overflow - HZ,
526 					       last_overflow +
527 					       TCP_SYNCOOKIE_VALID);
528 		}
529 	}
530 
531 	last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
532 
533 	/* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
534 	 * then we're under synflood. However, we have to use
535 	 * 'last_overflow - HZ' as lower bound. That's because a concurrent
536 	 * tcp_synq_overflow() could update .ts_recent_stamp after we read
537 	 * jiffies but before we store .ts_recent_stamp into last_overflow,
538 	 * which could lead to rejecting a valid syncookie.
539 	 */
540 	return !time_between32(now, last_overflow - HZ,
541 			       last_overflow + TCP_SYNCOOKIE_VALID);
542 }
543 
tcp_cookie_time(void)544 static inline u32 tcp_cookie_time(void)
545 {
546 	u64 val = get_jiffies_64();
547 
548 	do_div(val, TCP_SYNCOOKIE_PERIOD);
549 	return val;
550 }
551 
552 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
553 			      u16 *mssp);
554 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
555 u64 cookie_init_timestamp(struct request_sock *req, u64 now);
556 bool cookie_timestamp_decode(const struct net *net,
557 			     struct tcp_options_received *opt);
558 bool cookie_ecn_ok(const struct tcp_options_received *opt,
559 		   const struct net *net, const struct dst_entry *dst);
560 
561 /* From net/ipv6/syncookies.c */
562 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
563 		      u32 cookie);
564 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
565 
566 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
567 			      const struct tcphdr *th, u16 *mssp);
568 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
569 #endif
570 /* tcp_output.c */
571 
572 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
573 			       int nonagle);
574 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
575 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
576 void tcp_retransmit_timer(struct sock *sk);
577 void tcp_xmit_retransmit_queue(struct sock *);
578 void tcp_simple_retransmit(struct sock *);
579 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
580 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
581 enum tcp_queue {
582 	TCP_FRAG_IN_WRITE_QUEUE,
583 	TCP_FRAG_IN_RTX_QUEUE,
584 };
585 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
586 		 struct sk_buff *skb, u32 len,
587 		 unsigned int mss_now, gfp_t gfp);
588 
589 void tcp_send_probe0(struct sock *);
590 void tcp_send_partial(struct sock *);
591 int tcp_write_wakeup(struct sock *, int mib);
592 void tcp_send_fin(struct sock *sk);
593 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
594 int tcp_send_synack(struct sock *);
595 void tcp_push_one(struct sock *, unsigned int mss_now);
596 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
597 void tcp_send_ack(struct sock *sk);
598 void tcp_send_delayed_ack(struct sock *sk);
599 void tcp_send_loss_probe(struct sock *sk);
600 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
601 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
602 			     const struct sk_buff *next_skb);
603 
604 /* tcp_input.c */
605 void tcp_rearm_rto(struct sock *sk);
606 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
607 void tcp_reset(struct sock *sk);
608 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
609 void tcp_fin(struct sock *sk);
610 void tcp_check_space(struct sock *sk);
611 
612 /* tcp_timer.c */
613 void tcp_init_xmit_timers(struct sock *);
tcp_clear_xmit_timers(struct sock *sk)614 static inline void tcp_clear_xmit_timers(struct sock *sk)
615 {
616 	if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
617 		__sock_put(sk);
618 
619 	if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
620 		__sock_put(sk);
621 
622 	inet_csk_clear_xmit_timers(sk);
623 }
624 
625 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
626 unsigned int tcp_current_mss(struct sock *sk);
627 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when);
628 
629 /* Bound MSS / TSO packet size with the half of the window */
tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)630 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
631 {
632 	int cutoff;
633 
634 	/* When peer uses tiny windows, there is no use in packetizing
635 	 * to sub-MSS pieces for the sake of SWS or making sure there
636 	 * are enough packets in the pipe for fast recovery.
637 	 *
638 	 * On the other hand, for extremely large MSS devices, handling
639 	 * smaller than MSS windows in this way does make sense.
640 	 */
641 	if (tp->max_window > TCP_MSS_DEFAULT)
642 		cutoff = (tp->max_window >> 1);
643 	else
644 		cutoff = tp->max_window;
645 
646 	if (cutoff && pktsize > cutoff)
647 		return max_t(int, cutoff, 68U - tp->tcp_header_len);
648 	else
649 		return pktsize;
650 }
651 
652 /* tcp.c */
653 void tcp_get_info(struct sock *, struct tcp_info *);
654 
655 /* Read 'sendfile()'-style from a TCP socket */
656 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
657 		  sk_read_actor_t recv_actor);
658 
659 void tcp_initialize_rcv_mss(struct sock *sk);
660 
661 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
662 int tcp_mss_to_mtu(struct sock *sk, int mss);
663 void tcp_mtup_init(struct sock *sk);
664 
tcp_bound_rto(const struct sock *sk)665 static inline void tcp_bound_rto(const struct sock *sk)
666 {
667 	if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
668 		inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
669 }
670 
__tcp_set_rto(const struct tcp_sock *tp)671 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
672 {
673 	return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
674 }
675 
__tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)676 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
677 {
678 	/* mptcp hooks are only on the slow path */
679 	if (sk_is_mptcp((struct sock *)tp))
680 		return;
681 
682 	tp->pred_flags = htonl((tp->tcp_header_len << 26) |
683 			       ntohl(TCP_FLAG_ACK) |
684 			       snd_wnd);
685 }
686 
tcp_fast_path_on(struct tcp_sock *tp)687 static inline void tcp_fast_path_on(struct tcp_sock *tp)
688 {
689 	__tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
690 }
691 
tcp_fast_path_check(struct sock *sk)692 static inline void tcp_fast_path_check(struct sock *sk)
693 {
694 	struct tcp_sock *tp = tcp_sk(sk);
695 
696 	if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
697 	    tp->rcv_wnd &&
698 	    atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
699 	    !tp->urg_data)
700 		tcp_fast_path_on(tp);
701 }
702 
703 /* Compute the actual rto_min value */
tcp_rto_min(struct sock *sk)704 static inline u32 tcp_rto_min(struct sock *sk)
705 {
706 	const struct dst_entry *dst = __sk_dst_get(sk);
707 	u32 rto_min = inet_csk(sk)->icsk_rto_min;
708 
709 	if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
710 		rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
711 	return rto_min;
712 }
713 
tcp_rto_min_us(struct sock *sk)714 static inline u32 tcp_rto_min_us(struct sock *sk)
715 {
716 	return jiffies_to_usecs(tcp_rto_min(sk));
717 }
718 
tcp_ca_dst_locked(const struct dst_entry *dst)719 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
720 {
721 	return dst_metric_locked(dst, RTAX_CC_ALGO);
722 }
723 
724 /* Minimum RTT in usec. ~0 means not available. */
tcp_min_rtt(const struct tcp_sock *tp)725 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
726 {
727 	return minmax_get(&tp->rtt_min);
728 }
729 
730 /* Compute the actual receive window we are currently advertising.
731  * Rcv_nxt can be after the window if our peer push more data
732  * than the offered window.
733  */
tcp_receive_window(const struct tcp_sock *tp)734 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
735 {
736 	s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
737 
738 	if (win < 0)
739 		win = 0;
740 	return (u32) win;
741 }
742 
743 /* Choose a new window, without checks for shrinking, and without
744  * scaling applied to the result.  The caller does these things
745  * if necessary.  This is a "raw" window selection.
746  */
747 u32 __tcp_select_window(struct sock *sk);
748 
749 void tcp_send_window_probe(struct sock *sk);
750 
751 /* TCP uses 32bit jiffies to save some space.
752  * Note that this is different from tcp_time_stamp, which
753  * historically has been the same until linux-4.13.
754  */
755 #define tcp_jiffies32 ((u32)jiffies)
756 
757 /*
758  * Deliver a 32bit value for TCP timestamp option (RFC 7323)
759  * It is no longer tied to jiffies, but to 1 ms clock.
760  * Note: double check if you want to use tcp_jiffies32 instead of this.
761  */
762 #define TCP_TS_HZ	1000
763 
tcp_clock_ns(void)764 static inline u64 tcp_clock_ns(void)
765 {
766 	return ktime_get_ns();
767 }
768 
tcp_clock_us(void)769 static inline u64 tcp_clock_us(void)
770 {
771 	return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
772 }
773 
774 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
tcp_time_stamp(const struct tcp_sock *tp)775 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
776 {
777 	return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
778 }
779 
780 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
tcp_ns_to_ts(u64 ns)781 static inline u64 tcp_ns_to_ts(u64 ns)
782 {
783 	return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
784 }
785 
786 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
tcp_time_stamp_raw(void)787 static inline u32 tcp_time_stamp_raw(void)
788 {
789 	return tcp_ns_to_ts(tcp_clock_ns());
790 }
791 
792 void tcp_mstamp_refresh(struct tcp_sock *tp);
793 
tcp_stamp_us_delta(u64 t1, u64 t0)794 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
795 {
796 	return max_t(s64, t1 - t0, 0);
797 }
798 
tcp_skb_timestamp(const struct sk_buff *skb)799 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
800 {
801 	return tcp_ns_to_ts(skb->skb_mstamp_ns);
802 }
803 
804 /* provide the departure time in us unit */
tcp_skb_timestamp_us(const struct sk_buff *skb)805 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
806 {
807 	return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
808 }
809 
810 
811 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
812 
813 #define TCPHDR_FIN 0x01
814 #define TCPHDR_SYN 0x02
815 #define TCPHDR_RST 0x04
816 #define TCPHDR_PSH 0x08
817 #define TCPHDR_ACK 0x10
818 #define TCPHDR_URG 0x20
819 #define TCPHDR_ECE 0x40
820 #define TCPHDR_CWR 0x80
821 
822 #define TCPHDR_SYN_ECN	(TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
823 
824 /* This is what the send packet queuing engine uses to pass
825  * TCP per-packet control information to the transmission code.
826  * We also store the host-order sequence numbers in here too.
827  * This is 44 bytes if IPV6 is enabled.
828  * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
829  */
830 struct tcp_skb_cb {
831 	__u32		seq;		/* Starting sequence number	*/
832 	__u32		end_seq;	/* SEQ + FIN + SYN + datalen	*/
833 	union {
834 		/* Note : tcp_tw_isn is used in input path only
835 		 *	  (isn chosen by tcp_timewait_state_process())
836 		 *
837 		 * 	  tcp_gso_segs/size are used in write queue only,
838 		 *	  cf tcp_skb_pcount()/tcp_skb_mss()
839 		 */
840 		__u32		tcp_tw_isn;
841 		struct {
842 			u16	tcp_gso_segs;
843 			u16	tcp_gso_size;
844 		};
845 	};
846 	__u8		tcp_flags;	/* TCP header flags. (tcp[13])	*/
847 
848 	__u8		sacked;		/* State flags for SACK.	*/
849 #define TCPCB_SACKED_ACKED	0x01	/* SKB ACK'd by a SACK block	*/
850 #define TCPCB_SACKED_RETRANS	0x02	/* SKB retransmitted		*/
851 #define TCPCB_LOST		0x04	/* SKB is lost			*/
852 #define TCPCB_TAGBITS		0x07	/* All tag bits			*/
853 #define TCPCB_REPAIRED		0x10	/* SKB repaired (no skb_mstamp_ns)	*/
854 #define TCPCB_EVER_RETRANS	0x80	/* Ever retransmitted frame	*/
855 #define TCPCB_RETRANS		(TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
856 				TCPCB_REPAIRED)
857 
858 	__u8		ip_dsfield;	/* IPv4 tos or IPv6 dsfield	*/
859 	__u8		txstamp_ack:1,	/* Record TX timestamp for ack? */
860 			eor:1,		/* Is skb MSG_EOR marked? */
861 			has_rxtstamp:1,	/* SKB has a RX timestamp	*/
862 			unused:5;
863 	__u32		ack_seq;	/* Sequence number ACK'd	*/
864 	union {
865 		struct {
866 			/* There is space for up to 24 bytes */
867 			__u32 in_flight:30,/* Bytes in flight at transmit */
868 			      is_app_limited:1, /* cwnd not fully used? */
869 			      unused:1;
870 			/* pkts S/ACKed so far upon tx of skb, incl retrans: */
871 			__u32 delivered;
872 			/* start of send pipeline phase */
873 			u64 first_tx_mstamp;
874 			/* when we reached the "delivered" count */
875 			u64 delivered_mstamp;
876 		} tx;   /* only used for outgoing skbs */
877 		union {
878 			struct inet_skb_parm	h4;
879 #if IS_ENABLED(CONFIG_IPV6)
880 			struct inet6_skb_parm	h6;
881 #endif
882 #if IS_ENABLED(CONFIG_NEWIP)
883 			struct ninet_skb_parm	hnip; /* NIP */
884 #endif
885 		} header;	/* For incoming skbs */
886 		struct {
887 			__u32 flags;
888 			struct sock *sk_redir;
889 			void *data_end;
890 		} bpf;
891 	};
892 };
893 
894 #define TCP_SKB_CB(__skb)	((struct tcp_skb_cb *)&((__skb)->cb[0]))
895 
bpf_compute_data_end_sk_skb(struct sk_buff *skb)896 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb)
897 {
898 	TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb);
899 }
900 
tcp_skb_bpf_ingress(const struct sk_buff *skb)901 static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb)
902 {
903 	return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS;
904 }
905 
tcp_skb_bpf_redirect_fetch(struct sk_buff *skb)906 static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb)
907 {
908 	return TCP_SKB_CB(skb)->bpf.sk_redir;
909 }
910 
tcp_skb_bpf_redirect_clear(struct sk_buff *skb)911 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb)
912 {
913 	TCP_SKB_CB(skb)->bpf.sk_redir = NULL;
914 }
915 
916 extern const struct inet_connection_sock_af_ops ipv4_specific;
917 
918 #if IS_ENABLED(CONFIG_IPV6)
919 /* This is the variant of inet6_iif() that must be used by TCP,
920  * as TCP moves IP6CB into a different location in skb->cb[]
921  */
tcp_v6_iif(const struct sk_buff *skb)922 static inline int tcp_v6_iif(const struct sk_buff *skb)
923 {
924 	return TCP_SKB_CB(skb)->header.h6.iif;
925 }
926 
tcp_v6_iif_l3_slave(const struct sk_buff *skb)927 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
928 {
929 	bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
930 
931 	return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
932 }
933 
934 /* TCP_SKB_CB reference means this can not be used from early demux */
tcp_v6_sdif(const struct sk_buff *skb)935 static inline int tcp_v6_sdif(const struct sk_buff *skb)
936 {
937 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
938 	if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
939 		return TCP_SKB_CB(skb)->header.h6.iif;
940 #endif
941 	return 0;
942 }
943 
944 extern const struct inet_connection_sock_af_ops ipv6_specific;
945 
946 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
947 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
948 void tcp_v6_early_demux(struct sk_buff *skb);
949 
950 #endif
951 
952 /* TCP_SKB_CB reference means this can not be used from early demux */
tcp_v4_sdif(struct sk_buff *skb)953 static inline int tcp_v4_sdif(struct sk_buff *skb)
954 {
955 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
956 	if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
957 		return TCP_SKB_CB(skb)->header.h4.iif;
958 #endif
959 	return 0;
960 }
961 
962 /* Due to TSO, an SKB can be composed of multiple actual
963  * packets.  To keep these tracked properly, we use this.
964  */
tcp_skb_pcount(const struct sk_buff *skb)965 static inline int tcp_skb_pcount(const struct sk_buff *skb)
966 {
967 	return TCP_SKB_CB(skb)->tcp_gso_segs;
968 }
969 
tcp_skb_pcount_set(struct sk_buff *skb, int segs)970 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
971 {
972 	TCP_SKB_CB(skb)->tcp_gso_segs = segs;
973 }
974 
tcp_skb_pcount_add(struct sk_buff *skb, int segs)975 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
976 {
977 	TCP_SKB_CB(skb)->tcp_gso_segs += segs;
978 }
979 
980 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
tcp_skb_mss(const struct sk_buff *skb)981 static inline int tcp_skb_mss(const struct sk_buff *skb)
982 {
983 	return TCP_SKB_CB(skb)->tcp_gso_size;
984 }
985 
tcp_skb_can_collapse_to(const struct sk_buff *skb)986 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
987 {
988 	return likely(!TCP_SKB_CB(skb)->eor);
989 }
990 
tcp_skb_can_collapse(const struct sk_buff *to, const struct sk_buff *from)991 static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
992 					const struct sk_buff *from)
993 {
994 	return likely(tcp_skb_can_collapse_to(to) &&
995 		      mptcp_skb_can_collapse(to, from));
996 }
997 
998 /* Events passed to congestion control interface */
999 enum tcp_ca_event {
1000 	CA_EVENT_TX_START,	/* first transmit when no packets in flight */
1001 	CA_EVENT_CWND_RESTART,	/* congestion window restart */
1002 	CA_EVENT_COMPLETE_CWR,	/* end of congestion recovery */
1003 	CA_EVENT_LOSS,		/* loss timeout */
1004 	CA_EVENT_ECN_NO_CE,	/* ECT set, but not CE marked */
1005 	CA_EVENT_ECN_IS_CE,	/* received CE marked IP packet */
1006 };
1007 
1008 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
1009 enum tcp_ca_ack_event_flags {
1010 	CA_ACK_SLOWPATH		= (1 << 0),	/* In slow path processing */
1011 	CA_ACK_WIN_UPDATE	= (1 << 1),	/* ACK updated window */
1012 	CA_ACK_ECE		= (1 << 2),	/* ECE bit is set on ack */
1013 };
1014 
1015 /*
1016  * Interface for adding new TCP congestion control handlers
1017  */
1018 #define TCP_CA_NAME_MAX	16
1019 #define TCP_CA_MAX	128
1020 #define TCP_CA_BUF_MAX	(TCP_CA_NAME_MAX*TCP_CA_MAX)
1021 
1022 #define TCP_CA_UNSPEC	0
1023 
1024 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1025 #define TCP_CONG_NON_RESTRICTED 0x1
1026 /* Requires ECN/ECT set on all packets */
1027 #define TCP_CONG_NEEDS_ECN	0x2
1028 #define TCP_CONG_MASK	(TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1029 
1030 union tcp_cc_info;
1031 
1032 struct ack_sample {
1033 	u32 pkts_acked;
1034 	s32 rtt_us;
1035 	u32 in_flight;
1036 };
1037 
1038 /* A rate sample measures the number of (original/retransmitted) data
1039  * packets delivered "delivered" over an interval of time "interval_us".
1040  * The tcp_rate.c code fills in the rate sample, and congestion
1041  * control modules that define a cong_control function to run at the end
1042  * of ACK processing can optionally chose to consult this sample when
1043  * setting cwnd and pacing rate.
1044  * A sample is invalid if "delivered" or "interval_us" is negative.
1045  */
1046 struct rate_sample {
1047 	u64  prior_mstamp; /* starting timestamp for interval */
1048 	u32  prior_delivered;	/* tp->delivered at "prior_mstamp" */
1049 	s32  delivered;		/* number of packets delivered over interval */
1050 	long interval_us;	/* time for tp->delivered to incr "delivered" */
1051 	u32 snd_interval_us;	/* snd interval for delivered packets */
1052 	u32 rcv_interval_us;	/* rcv interval for delivered packets */
1053 	long rtt_us;		/* RTT of last (S)ACKed packet (or -1) */
1054 	int  losses;		/* number of packets marked lost upon ACK */
1055 	u32  acked_sacked;	/* number of packets newly (S)ACKed upon ACK */
1056 	u32  prior_in_flight;	/* in flight before this ACK */
1057 	u32  last_end_seq;	/* end_seq of most recently ACKed packet */
1058 	bool is_app_limited;	/* is sample from packet with bubble in pipe? */
1059 	bool is_retrans;	/* is sample from retransmission? */
1060 	bool is_ack_delayed;	/* is this (likely) a delayed ACK? */
1061 };
1062 
1063 struct tcp_congestion_ops {
1064 	struct list_head	list;
1065 	u32 key;
1066 	u32 flags;
1067 
1068 	/* initialize private data (optional) */
1069 	void (*init)(struct sock *sk);
1070 	/* cleanup private data  (optional) */
1071 	void (*release)(struct sock *sk);
1072 
1073 	/* return slow start threshold (required) */
1074 	u32 (*ssthresh)(struct sock *sk);
1075 	/* do new cwnd calculation (required) */
1076 	void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1077 	/* call before changing ca_state (optional) */
1078 	void (*set_state)(struct sock *sk, u8 new_state);
1079 	/* call when cwnd event occurs (optional) */
1080 	void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1081 	/* call when ack arrives (optional) */
1082 	void (*in_ack_event)(struct sock *sk, u32 flags);
1083 	/* new value of cwnd after loss (required) */
1084 	u32  (*undo_cwnd)(struct sock *sk);
1085 	/* hook for packet ack accounting (optional) */
1086 	void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1087 	/* override sysctl_tcp_min_tso_segs */
1088 	u32 (*min_tso_segs)(struct sock *sk);
1089 	/* returns the multiplier used in tcp_sndbuf_expand (optional) */
1090 	u32 (*sndbuf_expand)(struct sock *sk);
1091 	/* call when packets are delivered to update cwnd and pacing rate,
1092 	 * after all the ca_state processing. (optional)
1093 	 */
1094 	void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1095 	/* get info for inet_diag (optional) */
1096 	size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1097 			   union tcp_cc_info *info);
1098 
1099 	char 		name[TCP_CA_NAME_MAX];
1100 	struct module 	*owner;
1101 };
1102 
1103 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1104 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1105 
1106 void tcp_assign_congestion_control(struct sock *sk);
1107 void tcp_init_congestion_control(struct sock *sk);
1108 void tcp_cleanup_congestion_control(struct sock *sk);
1109 int tcp_set_default_congestion_control(struct net *net, const char *name);
1110 void tcp_get_default_congestion_control(struct net *net, char *name);
1111 void tcp_get_available_congestion_control(char *buf, size_t len);
1112 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1113 int tcp_set_allowed_congestion_control(char *allowed);
1114 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1115 			       bool cap_net_admin);
1116 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1117 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1118 
1119 u32 tcp_reno_ssthresh(struct sock *sk);
1120 u32 tcp_reno_undo_cwnd(struct sock *sk);
1121 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1122 extern struct tcp_congestion_ops tcp_reno;
1123 
1124 struct tcp_congestion_ops *tcp_ca_find(const char *name);
1125 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1126 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1127 #ifdef CONFIG_INET
1128 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1129 #else
tcp_ca_get_name_by_key(u32 key, char *buffer)1130 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1131 {
1132 	return NULL;
1133 }
1134 #endif
1135 
tcp_ca_needs_ecn(const struct sock *sk)1136 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1137 {
1138 	const struct inet_connection_sock *icsk = inet_csk(sk);
1139 
1140 	return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1141 }
1142 
tcp_set_ca_state(struct sock *sk, const u8 ca_state)1143 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1144 {
1145 	struct inet_connection_sock *icsk = inet_csk(sk);
1146 
1147 	if (icsk->icsk_ca_ops->set_state)
1148 		icsk->icsk_ca_ops->set_state(sk, ca_state);
1149 	icsk->icsk_ca_state = ca_state;
1150 }
1151 
tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)1152 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1153 {
1154 	const struct inet_connection_sock *icsk = inet_csk(sk);
1155 
1156 	if (icsk->icsk_ca_ops->cwnd_event)
1157 		icsk->icsk_ca_ops->cwnd_event(sk, event);
1158 }
1159 
1160 /* From tcp_rate.c */
1161 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1162 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1163 			    struct rate_sample *rs);
1164 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1165 		  bool is_sack_reneg, struct rate_sample *rs);
1166 void tcp_rate_check_app_limited(struct sock *sk);
1167 
tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2)1168 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2)
1169 {
1170 	return t1 > t2 || (t1 == t2 && after(seq1, seq2));
1171 }
1172 
1173 /* These functions determine how the current flow behaves in respect of SACK
1174  * handling. SACK is negotiated with the peer, and therefore it can vary
1175  * between different flows.
1176  *
1177  * tcp_is_sack - SACK enabled
1178  * tcp_is_reno - No SACK
1179  */
tcp_is_sack(const struct tcp_sock *tp)1180 static inline int tcp_is_sack(const struct tcp_sock *tp)
1181 {
1182 	return likely(tp->rx_opt.sack_ok);
1183 }
1184 
tcp_is_reno(const struct tcp_sock *tp)1185 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1186 {
1187 	return !tcp_is_sack(tp);
1188 }
1189 
tcp_left_out(const struct tcp_sock *tp)1190 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1191 {
1192 	return tp->sacked_out + tp->lost_out;
1193 }
1194 
1195 /* This determines how many packets are "in the network" to the best
1196  * of our knowledge.  In many cases it is conservative, but where
1197  * detailed information is available from the receiver (via SACK
1198  * blocks etc.) we can make more aggressive calculations.
1199  *
1200  * Use this for decisions involving congestion control, use just
1201  * tp->packets_out to determine if the send queue is empty or not.
1202  *
1203  * Read this equation as:
1204  *
1205  *	"Packets sent once on transmission queue" MINUS
1206  *	"Packets left network, but not honestly ACKed yet" PLUS
1207  *	"Packets fast retransmitted"
1208  */
tcp_packets_in_flight(const struct tcp_sock *tp)1209 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1210 {
1211 	return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1212 }
1213 
1214 #define TCP_INFINITE_SSTHRESH	0x7fffffff
1215 
tcp_in_slow_start(const struct tcp_sock *tp)1216 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1217 {
1218 	return tp->snd_cwnd < tp->snd_ssthresh;
1219 }
1220 
tcp_in_initial_slowstart(const struct tcp_sock *tp)1221 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1222 {
1223 	return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1224 }
1225 
tcp_in_cwnd_reduction(const struct sock *sk)1226 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1227 {
1228 	return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1229 	       (1 << inet_csk(sk)->icsk_ca_state);
1230 }
1231 
1232 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1233  * The exception is cwnd reduction phase, when cwnd is decreasing towards
1234  * ssthresh.
1235  */
tcp_current_ssthresh(const struct sock *sk)1236 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1237 {
1238 	const struct tcp_sock *tp = tcp_sk(sk);
1239 
1240 	if (tcp_in_cwnd_reduction(sk))
1241 		return tp->snd_ssthresh;
1242 	else
1243 		return max(tp->snd_ssthresh,
1244 			   ((tp->snd_cwnd >> 1) +
1245 			    (tp->snd_cwnd >> 2)));
1246 }
1247 
1248 /* Use define here intentionally to get WARN_ON location shown at the caller */
1249 #define tcp_verify_left_out(tp)	WARN_ON(tcp_left_out(tp) > tp->packets_out)
1250 
1251 void tcp_enter_cwr(struct sock *sk);
1252 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1253 
1254 /* The maximum number of MSS of available cwnd for which TSO defers
1255  * sending if not using sysctl_tcp_tso_win_divisor.
1256  */
tcp_max_tso_deferred_mss(const struct tcp_sock *tp)1257 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1258 {
1259 	return 3;
1260 }
1261 
1262 /* Returns end sequence number of the receiver's advertised window */
tcp_wnd_end(const struct tcp_sock *tp)1263 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1264 {
1265 	return tp->snd_una + tp->snd_wnd;
1266 }
1267 
1268 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1269  * flexible approach. The RFC suggests cwnd should not be raised unless
1270  * it was fully used previously. And that's exactly what we do in
1271  * congestion avoidance mode. But in slow start we allow cwnd to grow
1272  * as long as the application has used half the cwnd.
1273  * Example :
1274  *    cwnd is 10 (IW10), but application sends 9 frames.
1275  *    We allow cwnd to reach 18 when all frames are ACKed.
1276  * This check is safe because it's as aggressive as slow start which already
1277  * risks 100% overshoot. The advantage is that we discourage application to
1278  * either send more filler packets or data to artificially blow up the cwnd
1279  * usage, and allow application-limited process to probe bw more aggressively.
1280  */
tcp_is_cwnd_limited(const struct sock *sk)1281 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1282 {
1283 	const struct tcp_sock *tp = tcp_sk(sk);
1284 
1285 	if (tp->is_cwnd_limited)
1286 		return true;
1287 
1288 	/* If in slow start, ensure cwnd grows to twice what was ACKed. */
1289 	if (tcp_in_slow_start(tp))
1290 		return tp->snd_cwnd < 2 * tp->max_packets_out;
1291 
1292 	return false;
1293 }
1294 
1295 /* BBR congestion control needs pacing.
1296  * Same remark for SO_MAX_PACING_RATE.
1297  * sch_fq packet scheduler is efficiently handling pacing,
1298  * but is not always installed/used.
1299  * Return true if TCP stack should pace packets itself.
1300  */
tcp_needs_internal_pacing(const struct sock *sk)1301 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1302 {
1303 	return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1304 }
1305 
1306 /* Estimates in how many jiffies next packet for this flow can be sent.
1307  * Scheduling a retransmit timer too early would be silly.
1308  */
tcp_pacing_delay(const struct sock *sk)1309 static inline unsigned long tcp_pacing_delay(const struct sock *sk)
1310 {
1311 	s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
1312 
1313 	return delay > 0 ? nsecs_to_jiffies(delay) : 0;
1314 }
1315 
tcp_reset_xmit_timer(struct sock *sk, const int what, unsigned long when, const unsigned long max_when)1316 static inline void tcp_reset_xmit_timer(struct sock *sk,
1317 					const int what,
1318 					unsigned long when,
1319 					const unsigned long max_when)
1320 {
1321 	inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk),
1322 				  max_when);
1323 }
1324 
1325 /* Something is really bad, we could not queue an additional packet,
1326  * because qdisc is full or receiver sent a 0 window, or we are paced.
1327  * We do not want to add fuel to the fire, or abort too early,
1328  * so make sure the timer we arm now is at least 200ms in the future,
1329  * regardless of current icsk_rto value (as it could be ~2ms)
1330  */
tcp_probe0_base(const struct sock *sk)1331 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1332 {
1333 	return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1334 }
1335 
1336 /* Variant of inet_csk_rto_backoff() used for zero window probes */
tcp_probe0_when(const struct sock *sk, unsigned long max_when)1337 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1338 					    unsigned long max_when)
1339 {
1340 	u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1341 
1342 	return (unsigned long)min_t(u64, when, max_when);
1343 }
1344 
tcp_check_probe_timer(struct sock *sk)1345 static inline void tcp_check_probe_timer(struct sock *sk)
1346 {
1347 	if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1348 		tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1349 				     tcp_probe0_base(sk), TCP_RTO_MAX);
1350 }
1351 
tcp_init_wl(struct tcp_sock *tp, u32 seq)1352 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1353 {
1354 	tp->snd_wl1 = seq;
1355 }
1356 
tcp_update_wl(struct tcp_sock *tp, u32 seq)1357 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1358 {
1359 	tp->snd_wl1 = seq;
1360 }
1361 
1362 /*
1363  * Calculate(/check) TCP checksum
1364  */
tcp_v4_check(int len, __be32 saddr, __be32 daddr, __wsum base)1365 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1366 				   __be32 daddr, __wsum base)
1367 {
1368 	return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1369 }
1370 
tcp_checksum_complete(struct sk_buff *skb)1371 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1372 {
1373 	return !skb_csum_unnecessary(skb) &&
1374 		__skb_checksum_complete(skb);
1375 }
1376 
1377 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1378 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1379 void tcp_set_state(struct sock *sk, int state);
1380 void tcp_done(struct sock *sk);
1381 int tcp_abort(struct sock *sk, int err);
1382 
tcp_sack_reset(struct tcp_options_received *rx_opt)1383 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1384 {
1385 	rx_opt->dsack = 0;
1386 	rx_opt->num_sacks = 0;
1387 }
1388 
1389 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1390 
tcp_slow_start_after_idle_check(struct sock *sk)1391 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1392 {
1393 	const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1394 	struct tcp_sock *tp = tcp_sk(sk);
1395 	s32 delta;
1396 
1397 	if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) ||
1398 	    tp->packets_out || ca_ops->cong_control)
1399 		return;
1400 	delta = tcp_jiffies32 - tp->lsndtime;
1401 	if (delta > inet_csk(sk)->icsk_rto)
1402 		tcp_cwnd_restart(sk, delta);
1403 }
1404 
1405 /* Determine a window scaling and initial window to offer. */
1406 void tcp_select_initial_window(const struct sock *sk, int __space,
1407 			       __u32 mss, __u32 *rcv_wnd,
1408 			       __u32 *window_clamp, int wscale_ok,
1409 			       __u8 *rcv_wscale, __u32 init_rcv_wnd);
1410 
tcp_win_from_space(const struct sock *sk, int space)1411 static inline int tcp_win_from_space(const struct sock *sk, int space)
1412 {
1413 	int tcp_adv_win_scale = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale);
1414 
1415 	return tcp_adv_win_scale <= 0 ?
1416 		(space>>(-tcp_adv_win_scale)) :
1417 		space - (space>>tcp_adv_win_scale);
1418 }
1419 
1420 /* Note: caller must be prepared to deal with negative returns */
tcp_space(const struct sock *sk)1421 static inline int tcp_space(const struct sock *sk)
1422 {
1423 	return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1424 				  READ_ONCE(sk->sk_backlog.len) -
1425 				  atomic_read(&sk->sk_rmem_alloc));
1426 }
1427 
tcp_full_space(const struct sock *sk)1428 static inline int tcp_full_space(const struct sock *sk)
1429 {
1430 	return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1431 }
1432 
1433 void tcp_cleanup_rbuf(struct sock *sk, int copied);
1434 
1435 /* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1436  * If 87.5 % (7/8) of the space has been consumed, we want to override
1437  * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1438  * len/truesize ratio.
1439  */
tcp_rmem_pressure(const struct sock *sk)1440 static inline bool tcp_rmem_pressure(const struct sock *sk)
1441 {
1442 	int rcvbuf, threshold;
1443 
1444 	if (tcp_under_memory_pressure(sk))
1445 		return true;
1446 
1447 	rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1448 	threshold = rcvbuf - (rcvbuf >> 3);
1449 
1450 	return atomic_read(&sk->sk_rmem_alloc) > threshold;
1451 }
1452 
1453 extern void tcp_openreq_init_rwin(struct request_sock *req,
1454 				  const struct sock *sk_listener,
1455 				  const struct dst_entry *dst);
1456 
1457 void tcp_enter_memory_pressure(struct sock *sk);
1458 void tcp_leave_memory_pressure(struct sock *sk);
1459 
keepalive_intvl_when(const struct tcp_sock *tp)1460 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1461 {
1462 	struct net *net = sock_net((struct sock *)tp);
1463 	int val;
1464 
1465 	/* Paired with WRITE_ONCE() in tcp_sock_set_keepintvl()
1466 	 * and do_tcp_setsockopt().
1467 	 */
1468 	val = READ_ONCE(tp->keepalive_intvl);
1469 
1470 	return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl);
1471 }
1472 
keepalive_time_when(const struct tcp_sock *tp)1473 static inline int keepalive_time_when(const struct tcp_sock *tp)
1474 {
1475 	struct net *net = sock_net((struct sock *)tp);
1476 	int val;
1477 
1478 	/* Paired with WRITE_ONCE() in tcp_sock_set_keepidle_locked() */
1479 	val = READ_ONCE(tp->keepalive_time);
1480 
1481 	return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time);
1482 }
1483 
keepalive_probes(const struct tcp_sock *tp)1484 static inline int keepalive_probes(const struct tcp_sock *tp)
1485 {
1486 	struct net *net = sock_net((struct sock *)tp);
1487 	int val;
1488 
1489 	/* Paired with WRITE_ONCE() in tcp_sock_set_keepcnt()
1490 	 * and do_tcp_setsockopt().
1491 	 */
1492 	val = READ_ONCE(tp->keepalive_probes);
1493 
1494 	return val ? : READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes);
1495 }
1496 
keepalive_time_elapsed(const struct tcp_sock *tp)1497 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1498 {
1499 	const struct inet_connection_sock *icsk = &tp->inet_conn;
1500 
1501 	return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1502 			  tcp_jiffies32 - tp->rcv_tstamp);
1503 }
1504 
tcp_fin_time(const struct sock *sk)1505 static inline int tcp_fin_time(const struct sock *sk)
1506 {
1507 	int fin_timeout = tcp_sk(sk)->linger2 ? :
1508 		READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout);
1509 	const int rto = inet_csk(sk)->icsk_rto;
1510 
1511 	if (fin_timeout < (rto << 2) - (rto >> 1))
1512 		fin_timeout = (rto << 2) - (rto >> 1);
1513 
1514 	return fin_timeout;
1515 }
1516 
tcp_paws_check(const struct tcp_options_received *rx_opt, int paws_win)1517 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1518 				  int paws_win)
1519 {
1520 	if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1521 		return true;
1522 	if (unlikely(!time_before32(ktime_get_seconds(),
1523 				    rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1524 		return true;
1525 	/*
1526 	 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1527 	 * then following tcp messages have valid values. Ignore 0 value,
1528 	 * or else 'negative' tsval might forbid us to accept their packets.
1529 	 */
1530 	if (!rx_opt->ts_recent)
1531 		return true;
1532 	return false;
1533 }
1534 
tcp_paws_reject(const struct tcp_options_received *rx_opt, int rst)1535 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1536 				   int rst)
1537 {
1538 	if (tcp_paws_check(rx_opt, 0))
1539 		return false;
1540 
1541 	/* RST segments are not recommended to carry timestamp,
1542 	   and, if they do, it is recommended to ignore PAWS because
1543 	   "their cleanup function should take precedence over timestamps."
1544 	   Certainly, it is mistake. It is necessary to understand the reasons
1545 	   of this constraint to relax it: if peer reboots, clock may go
1546 	   out-of-sync and half-open connections will not be reset.
1547 	   Actually, the problem would be not existing if all
1548 	   the implementations followed draft about maintaining clock
1549 	   via reboots. Linux-2.2 DOES NOT!
1550 
1551 	   However, we can relax time bounds for RST segments to MSL.
1552 	 */
1553 	if (rst && !time_before32(ktime_get_seconds(),
1554 				  rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1555 		return false;
1556 	return true;
1557 }
1558 
1559 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1560 			  int mib_idx, u32 *last_oow_ack_time);
1561 
tcp_mib_init(struct net *net)1562 static inline void tcp_mib_init(struct net *net)
1563 {
1564 	/* See RFC 2012 */
1565 	TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1566 	TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1567 	TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1568 	TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1569 }
1570 
1571 /* from STCP */
tcp_clear_retrans_hints_partial(struct tcp_sock *tp)1572 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1573 {
1574 	tp->lost_skb_hint = NULL;
1575 }
1576 
tcp_clear_all_retrans_hints(struct tcp_sock *tp)1577 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1578 {
1579 	tcp_clear_retrans_hints_partial(tp);
1580 	tp->retransmit_skb_hint = NULL;
1581 }
1582 
1583 union tcp_md5_addr {
1584 	struct in_addr  a4;
1585 #if IS_ENABLED(CONFIG_IPV6)
1586 	struct in6_addr	a6;
1587 #endif
1588 };
1589 
1590 /* - key database */
1591 struct tcp_md5sig_key {
1592 	struct hlist_node	node;
1593 	u8			keylen;
1594 	u8			family; /* AF_INET or AF_INET6 */
1595 	u8			prefixlen;
1596 	union tcp_md5_addr	addr;
1597 	int			l3index; /* set if key added with L3 scope */
1598 	u8			key[TCP_MD5SIG_MAXKEYLEN];
1599 	struct rcu_head		rcu;
1600 };
1601 
1602 /* - sock block */
1603 struct tcp_md5sig_info {
1604 	struct hlist_head	head;
1605 	struct rcu_head		rcu;
1606 };
1607 
1608 /* - pseudo header */
1609 struct tcp4_pseudohdr {
1610 	__be32		saddr;
1611 	__be32		daddr;
1612 	__u8		pad;
1613 	__u8		protocol;
1614 	__be16		len;
1615 };
1616 
1617 struct tcp6_pseudohdr {
1618 	struct in6_addr	saddr;
1619 	struct in6_addr daddr;
1620 	__be32		len;
1621 	__be32		protocol;	/* including padding */
1622 };
1623 
1624 union tcp_md5sum_block {
1625 	struct tcp4_pseudohdr ip4;
1626 #if IS_ENABLED(CONFIG_IPV6)
1627 	struct tcp6_pseudohdr ip6;
1628 #endif
1629 };
1630 
1631 /* - pool: digest algorithm, hash description and scratch buffer */
1632 struct tcp_md5sig_pool {
1633 	struct ahash_request	*md5_req;
1634 	void			*scratch;
1635 };
1636 
1637 /* - functions */
1638 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1639 			const struct sock *sk, const struct sk_buff *skb);
1640 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1641 		   int family, u8 prefixlen, int l3index,
1642 		   const u8 *newkey, u8 newkeylen, gfp_t gfp);
1643 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1644 		   int family, u8 prefixlen, int l3index);
1645 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1646 					 const struct sock *addr_sk);
1647 
1648 #ifdef CONFIG_TCP_MD5SIG
1649 #include <linux/jump_label.h>
1650 extern struct static_key_false tcp_md5_needed;
1651 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1652 					   const union tcp_md5_addr *addr,
1653 					   int family);
1654 static inline struct tcp_md5sig_key *
tcp_md5_do_lookup(const struct sock *sk, int l3index, const union tcp_md5_addr *addr, int family)1655 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1656 		  const union tcp_md5_addr *addr, int family)
1657 {
1658 	if (!static_branch_unlikely(&tcp_md5_needed))
1659 		return NULL;
1660 	return __tcp_md5_do_lookup(sk, l3index, addr, family);
1661 }
1662 
1663 #define tcp_twsk_md5_key(twsk)	((twsk)->tw_md5_key)
1664 #else
1665 static inline struct tcp_md5sig_key *
tcp_md5_do_lookup(const struct sock *sk, int l3index, const union tcp_md5_addr *addr, int family)1666 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1667 		  const union tcp_md5_addr *addr, int family)
1668 {
1669 	return NULL;
1670 }
1671 #define tcp_twsk_md5_key(twsk)	NULL
1672 #endif
1673 
1674 bool tcp_alloc_md5sig_pool(void);
1675 
1676 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
tcp_put_md5sig_pool(void)1677 static inline void tcp_put_md5sig_pool(void)
1678 {
1679 	local_bh_enable();
1680 }
1681 
1682 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1683 			  unsigned int header_len);
1684 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1685 		     const struct tcp_md5sig_key *key);
1686 
1687 /* From tcp_fastopen.c */
1688 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1689 			    struct tcp_fastopen_cookie *cookie);
1690 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1691 			    struct tcp_fastopen_cookie *cookie, bool syn_lost,
1692 			    u16 try_exp);
1693 struct tcp_fastopen_request {
1694 	/* Fast Open cookie. Size 0 means a cookie request */
1695 	struct tcp_fastopen_cookie	cookie;
1696 	struct msghdr			*data;  /* data in MSG_FASTOPEN */
1697 	size_t				size;
1698 	int				copied;	/* queued in tcp_connect() */
1699 	struct ubuf_info		*uarg;
1700 };
1701 void tcp_free_fastopen_req(struct tcp_sock *tp);
1702 void tcp_fastopen_destroy_cipher(struct sock *sk);
1703 void tcp_fastopen_ctx_destroy(struct net *net);
1704 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1705 			      void *primary_key, void *backup_key);
1706 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
1707 			    u64 *key);
1708 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1709 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1710 			      struct request_sock *req,
1711 			      struct tcp_fastopen_cookie *foc,
1712 			      const struct dst_entry *dst);
1713 void tcp_fastopen_init_key_once(struct net *net);
1714 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1715 			     struct tcp_fastopen_cookie *cookie);
1716 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1717 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1718 #define TCP_FASTOPEN_KEY_MAX 2
1719 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1720 	(TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1721 
1722 /* Fastopen key context */
1723 struct tcp_fastopen_context {
1724 	siphash_key_t	key[TCP_FASTOPEN_KEY_MAX];
1725 	int		num;
1726 	struct rcu_head	rcu;
1727 };
1728 
1729 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1730 void tcp_fastopen_active_disable(struct sock *sk);
1731 bool tcp_fastopen_active_should_disable(struct sock *sk);
1732 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1733 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1734 
1735 /* Caller needs to wrap with rcu_read_(un)lock() */
1736 static inline
tcp_fastopen_get_ctx(const struct sock *sk)1737 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1738 {
1739 	struct tcp_fastopen_context *ctx;
1740 
1741 	ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1742 	if (!ctx)
1743 		ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1744 	return ctx;
1745 }
1746 
1747 static inline
tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc, const struct tcp_fastopen_cookie *orig)1748 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1749 			       const struct tcp_fastopen_cookie *orig)
1750 {
1751 	if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1752 	    orig->len == foc->len &&
1753 	    !memcmp(orig->val, foc->val, foc->len))
1754 		return true;
1755 	return false;
1756 }
1757 
1758 static inline
tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)1759 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1760 {
1761 	return ctx->num;
1762 }
1763 
1764 /* Latencies incurred by various limits for a sender. They are
1765  * chronograph-like stats that are mutually exclusive.
1766  */
1767 enum tcp_chrono {
1768 	TCP_CHRONO_UNSPEC,
1769 	TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1770 	TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1771 	TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1772 	__TCP_CHRONO_MAX,
1773 };
1774 
1775 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1776 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1777 
1778 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1779  * the same memory storage than skb->destructor/_skb_refdst
1780  */
tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)1781 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1782 {
1783 	skb->destructor = NULL;
1784 	skb->_skb_refdst = 0UL;
1785 }
1786 
1787 #define tcp_skb_tsorted_save(skb) {		\
1788 	unsigned long _save = skb->_skb_refdst;	\
1789 	skb->_skb_refdst = 0UL;
1790 
1791 #define tcp_skb_tsorted_restore(skb)		\
1792 	skb->_skb_refdst = _save;		\
1793 }
1794 
1795 void tcp_write_queue_purge(struct sock *sk);
1796 
tcp_rtx_queue_head(const struct sock *sk)1797 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1798 {
1799 	return skb_rb_first(&sk->tcp_rtx_queue);
1800 }
1801 
tcp_rtx_queue_tail(const struct sock *sk)1802 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1803 {
1804 	return skb_rb_last(&sk->tcp_rtx_queue);
1805 }
1806 
tcp_write_queue_head(const struct sock *sk)1807 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1808 {
1809 	return skb_peek(&sk->sk_write_queue);
1810 }
1811 
tcp_write_queue_tail(const struct sock *sk)1812 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1813 {
1814 	return skb_peek_tail(&sk->sk_write_queue);
1815 }
1816 
1817 #define tcp_for_write_queue_from_safe(skb, tmp, sk)			\
1818 	skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1819 
tcp_send_head(const struct sock *sk)1820 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1821 {
1822 	return skb_peek(&sk->sk_write_queue);
1823 }
1824 
tcp_skb_is_last(const struct sock *sk, const struct sk_buff *skb)1825 static inline bool tcp_skb_is_last(const struct sock *sk,
1826 				   const struct sk_buff *skb)
1827 {
1828 	return skb_queue_is_last(&sk->sk_write_queue, skb);
1829 }
1830 
1831 /**
1832  * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
1833  * @sk: socket
1834  *
1835  * Since the write queue can have a temporary empty skb in it,
1836  * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
1837  */
tcp_write_queue_empty(const struct sock *sk)1838 static inline bool tcp_write_queue_empty(const struct sock *sk)
1839 {
1840 	const struct tcp_sock *tp = tcp_sk(sk);
1841 
1842 	return tp->write_seq == tp->snd_nxt;
1843 }
1844 
tcp_rtx_queue_empty(const struct sock *sk)1845 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1846 {
1847 	return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1848 }
1849 
tcp_rtx_and_write_queues_empty(const struct sock *sk)1850 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1851 {
1852 	return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1853 }
1854 
tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)1855 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1856 {
1857 	__skb_queue_tail(&sk->sk_write_queue, skb);
1858 
1859 	/* Queue it, remembering where we must start sending. */
1860 	if (sk->sk_write_queue.next == skb)
1861 		tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1862 }
1863 
1864 /* Insert new before skb on the write queue of sk.  */
tcp_insert_write_queue_before(struct sk_buff *new, struct sk_buff *skb, struct sock *sk)1865 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1866 						  struct sk_buff *skb,
1867 						  struct sock *sk)
1868 {
1869 	__skb_queue_before(&sk->sk_write_queue, skb, new);
1870 }
1871 
tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)1872 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1873 {
1874 	tcp_skb_tsorted_anchor_cleanup(skb);
1875 	__skb_unlink(skb, &sk->sk_write_queue);
1876 }
1877 
1878 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1879 
tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)1880 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1881 {
1882 	tcp_skb_tsorted_anchor_cleanup(skb);
1883 	rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1884 }
1885 
tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)1886 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1887 {
1888 	list_del(&skb->tcp_tsorted_anchor);
1889 	tcp_rtx_queue_unlink(skb, sk);
1890 	sk_wmem_free_skb(sk, skb);
1891 }
1892 
tcp_push_pending_frames(struct sock *sk)1893 static inline void tcp_push_pending_frames(struct sock *sk)
1894 {
1895 	if (tcp_send_head(sk)) {
1896 		struct tcp_sock *tp = tcp_sk(sk);
1897 
1898 		__tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1899 	}
1900 }
1901 
1902 /* Start sequence of the skb just after the highest skb with SACKed
1903  * bit, valid only if sacked_out > 0 or when the caller has ensured
1904  * validity by itself.
1905  */
tcp_highest_sack_seq(struct tcp_sock *tp)1906 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1907 {
1908 	if (!tp->sacked_out)
1909 		return tp->snd_una;
1910 
1911 	if (tp->highest_sack == NULL)
1912 		return tp->snd_nxt;
1913 
1914 	return TCP_SKB_CB(tp->highest_sack)->seq;
1915 }
1916 
tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)1917 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1918 {
1919 	tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1920 }
1921 
tcp_highest_sack(struct sock *sk)1922 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1923 {
1924 	return tcp_sk(sk)->highest_sack;
1925 }
1926 
tcp_highest_sack_reset(struct sock *sk)1927 static inline void tcp_highest_sack_reset(struct sock *sk)
1928 {
1929 	tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1930 }
1931 
1932 /* Called when old skb is about to be deleted and replaced by new skb */
tcp_highest_sack_replace(struct sock *sk, struct sk_buff *old, struct sk_buff *new)1933 static inline void tcp_highest_sack_replace(struct sock *sk,
1934 					    struct sk_buff *old,
1935 					    struct sk_buff *new)
1936 {
1937 	if (old == tcp_highest_sack(sk))
1938 		tcp_sk(sk)->highest_sack = new;
1939 }
1940 
1941 /* This helper checks if socket has IP_TRANSPARENT set */
inet_sk_transparent(const struct sock *sk)1942 static inline bool inet_sk_transparent(const struct sock *sk)
1943 {
1944 	switch (sk->sk_state) {
1945 	case TCP_TIME_WAIT:
1946 		return inet_twsk(sk)->tw_transparent;
1947 	case TCP_NEW_SYN_RECV:
1948 		return inet_rsk(inet_reqsk(sk))->no_srccheck;
1949 	}
1950 	return inet_sk(sk)->transparent;
1951 }
1952 
1953 /* Determines whether this is a thin stream (which may suffer from
1954  * increased latency). Used to trigger latency-reducing mechanisms.
1955  */
tcp_stream_is_thin(struct tcp_sock *tp)1956 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1957 {
1958 	return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1959 }
1960 
1961 /* /proc */
1962 enum tcp_seq_states {
1963 	TCP_SEQ_STATE_LISTENING,
1964 	TCP_SEQ_STATE_ESTABLISHED,
1965 };
1966 
1967 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1968 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1969 void tcp_seq_stop(struct seq_file *seq, void *v);
1970 
1971 struct tcp_seq_afinfo {
1972 	sa_family_t			family;
1973 };
1974 
1975 struct tcp_iter_state {
1976 	struct seq_net_private	p;
1977 	enum tcp_seq_states	state;
1978 	struct sock		*syn_wait_sk;
1979 	struct tcp_seq_afinfo	*bpf_seq_afinfo;
1980 	int			bucket, offset, sbucket, num;
1981 	loff_t			last_pos;
1982 };
1983 
1984 extern struct request_sock_ops tcp_request_sock_ops;
1985 extern struct request_sock_ops tcp6_request_sock_ops;
1986 
1987 void tcp_v4_destroy_sock(struct sock *sk);
1988 
1989 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1990 				netdev_features_t features);
1991 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
1992 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
1993 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
1994 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff));
1995 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb));
1996 int tcp_gro_complete(struct sk_buff *skb);
1997 
1998 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1999 
tcp_notsent_lowat(const struct tcp_sock *tp)2000 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
2001 {
2002 	struct net *net = sock_net((struct sock *)tp);
2003 	u32 val;
2004 
2005 	val = READ_ONCE(tp->notsent_lowat);
2006 
2007 	return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
2008 }
2009 
2010 /* @wake is one when sk_stream_write_space() calls us.
2011  * This sends EPOLLOUT only if notsent_bytes is half the limit.
2012  * This mimics the strategy used in sock_def_write_space().
2013  */
tcp_stream_memory_free(const struct sock *sk, int wake)2014 static inline bool tcp_stream_memory_free(const struct sock *sk, int wake)
2015 {
2016 	const struct tcp_sock *tp = tcp_sk(sk);
2017 	u32 notsent_bytes = READ_ONCE(tp->write_seq) -
2018 			    READ_ONCE(tp->snd_nxt);
2019 
2020 	return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
2021 }
2022 
2023 #ifdef CONFIG_PROC_FS
2024 int tcp4_proc_init(void);
2025 void tcp4_proc_exit(void);
2026 #endif
2027 
2028 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
2029 int tcp_conn_request(struct request_sock_ops *rsk_ops,
2030 		     const struct tcp_request_sock_ops *af_ops,
2031 		     struct sock *sk, struct sk_buff *skb);
2032 
2033 /* TCP af-specific functions */
2034 struct tcp_sock_af_ops {
2035 #ifdef CONFIG_TCP_MD5SIG
2036 	struct tcp_md5sig_key	*(*md5_lookup) (const struct sock *sk,
2037 						const struct sock *addr_sk);
2038 	int		(*calc_md5_hash)(char *location,
2039 					 const struct tcp_md5sig_key *md5,
2040 					 const struct sock *sk,
2041 					 const struct sk_buff *skb);
2042 	int		(*md5_parse)(struct sock *sk,
2043 				     int optname,
2044 				     sockptr_t optval,
2045 				     int optlen);
2046 #endif
2047 };
2048 
2049 struct tcp_request_sock_ops {
2050 	u16 mss_clamp;
2051 #ifdef CONFIG_TCP_MD5SIG
2052 	struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
2053 						 const struct sock *addr_sk);
2054 	int		(*calc_md5_hash) (char *location,
2055 					  const struct tcp_md5sig_key *md5,
2056 					  const struct sock *sk,
2057 					  const struct sk_buff *skb);
2058 #endif
2059 	void (*init_req)(struct request_sock *req,
2060 			 const struct sock *sk_listener,
2061 			 struct sk_buff *skb);
2062 #ifdef CONFIG_SYN_COOKIES
2063 	__u32 (*cookie_init_seq)(const struct sk_buff *skb,
2064 				 __u16 *mss);
2065 #endif
2066 	struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
2067 				       const struct request_sock *req);
2068 	u32 (*init_seq)(const struct sk_buff *skb);
2069 	u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
2070 	int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2071 			   struct flowi *fl, struct request_sock *req,
2072 			   struct tcp_fastopen_cookie *foc,
2073 			   enum tcp_synack_type synack_type,
2074 			   struct sk_buff *syn_skb);
2075 };
2076 
2077 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2078 #if IS_ENABLED(CONFIG_IPV6)
2079 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2080 #endif
2081 
2082 #ifdef CONFIG_SYN_COOKIES
cookie_init_sequence(const struct tcp_request_sock_ops *ops, const struct sock *sk, struct sk_buff *skb, __u16 *mss)2083 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2084 					 const struct sock *sk, struct sk_buff *skb,
2085 					 __u16 *mss)
2086 {
2087 	tcp_synq_overflow(sk);
2088 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2089 	return ops->cookie_init_seq(skb, mss);
2090 }
2091 #else
cookie_init_sequence(const struct tcp_request_sock_ops *ops, const struct sock *sk, struct sk_buff *skb, __u16 *mss)2092 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2093 					 const struct sock *sk, struct sk_buff *skb,
2094 					 __u16 *mss)
2095 {
2096 	return 0;
2097 }
2098 #endif
2099 
2100 int tcpv4_offload_init(void);
2101 
2102 void tcp_v4_init(void);
2103 void tcp_init(void);
2104 
2105 /* tcp_recovery.c */
2106 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2107 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2108 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2109 				u32 reo_wnd);
2110 extern bool tcp_rack_mark_lost(struct sock *sk);
2111 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2112 			     u64 xmit_time);
2113 extern void tcp_rack_reo_timeout(struct sock *sk);
2114 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2115 
2116 /* At how many usecs into the future should the RTO fire? */
tcp_rto_delta_us(const struct sock *sk)2117 static inline s64 tcp_rto_delta_us(const struct sock *sk)
2118 {
2119 	const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2120 	u32 rto = inet_csk(sk)->icsk_rto;
2121 	u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2122 
2123 	return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2124 }
2125 
2126 /*
2127  * Save and compile IPv4 options, return a pointer to it
2128  */
tcp_v4_save_options(struct net *net, struct sk_buff *skb)2129 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2130 							 struct sk_buff *skb)
2131 {
2132 	const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2133 	struct ip_options_rcu *dopt = NULL;
2134 
2135 	if (opt->optlen) {
2136 		int opt_size = sizeof(*dopt) + opt->optlen;
2137 
2138 		dopt = kmalloc(opt_size, GFP_ATOMIC);
2139 		if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2140 			kfree(dopt);
2141 			dopt = NULL;
2142 		}
2143 	}
2144 	return dopt;
2145 }
2146 
2147 /* locally generated TCP pure ACKs have skb->truesize == 2
2148  * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2149  * This is much faster than dissecting the packet to find out.
2150  * (Think of GRE encapsulations, IPv4, IPv6, ...)
2151  */
skb_is_tcp_pure_ack(const struct sk_buff *skb)2152 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2153 {
2154 	return skb->truesize == 2;
2155 }
2156 
skb_set_tcp_pure_ack(struct sk_buff *skb)2157 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2158 {
2159 	skb->truesize = 2;
2160 }
2161 
tcp_inq(struct sock *sk)2162 static inline int tcp_inq(struct sock *sk)
2163 {
2164 	struct tcp_sock *tp = tcp_sk(sk);
2165 	int answ;
2166 
2167 	if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2168 		answ = 0;
2169 	} else if (sock_flag(sk, SOCK_URGINLINE) ||
2170 		   !tp->urg_data ||
2171 		   before(tp->urg_seq, tp->copied_seq) ||
2172 		   !before(tp->urg_seq, tp->rcv_nxt)) {
2173 
2174 		answ = tp->rcv_nxt - tp->copied_seq;
2175 
2176 		/* Subtract 1, if FIN was received */
2177 		if (answ && sock_flag(sk, SOCK_DONE))
2178 			answ--;
2179 	} else {
2180 		answ = tp->urg_seq - tp->copied_seq;
2181 	}
2182 
2183 	return answ;
2184 }
2185 
2186 int tcp_peek_len(struct socket *sock);
2187 
tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)2188 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2189 {
2190 	u16 segs_in;
2191 
2192 	segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2193 	tp->segs_in += segs_in;
2194 	if (skb->len > tcp_hdrlen(skb))
2195 		tp->data_segs_in += segs_in;
2196 }
2197 
2198 /*
2199  * TCP listen path runs lockless.
2200  * We forced "struct sock" to be const qualified to make sure
2201  * we don't modify one of its field by mistake.
2202  * Here, we increment sk_drops which is an atomic_t, so we can safely
2203  * make sock writable again.
2204  */
tcp_listendrop(const struct sock *sk)2205 static inline void tcp_listendrop(const struct sock *sk)
2206 {
2207 	atomic_inc(&((struct sock *)sk)->sk_drops);
2208 	__NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2209 }
2210 
2211 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2212 
2213 /*
2214  * Interface for adding Upper Level Protocols over TCP
2215  */
2216 
2217 #define TCP_ULP_NAME_MAX	16
2218 #define TCP_ULP_MAX		128
2219 #define TCP_ULP_BUF_MAX		(TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2220 
2221 struct tcp_ulp_ops {
2222 	struct list_head	list;
2223 
2224 	/* initialize ulp */
2225 	int (*init)(struct sock *sk);
2226 	/* update ulp */
2227 	void (*update)(struct sock *sk, struct proto *p,
2228 		       void (*write_space)(struct sock *sk));
2229 	/* cleanup ulp */
2230 	void (*release)(struct sock *sk);
2231 	/* diagnostic */
2232 	int (*get_info)(const struct sock *sk, struct sk_buff *skb);
2233 	size_t (*get_info_size)(const struct sock *sk);
2234 	/* clone ulp */
2235 	void (*clone)(const struct request_sock *req, struct sock *newsk,
2236 		      const gfp_t priority);
2237 
2238 	char		name[TCP_ULP_NAME_MAX];
2239 	struct module	*owner;
2240 };
2241 int tcp_register_ulp(struct tcp_ulp_ops *type);
2242 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2243 int tcp_set_ulp(struct sock *sk, const char *name);
2244 void tcp_get_available_ulp(char *buf, size_t len);
2245 void tcp_cleanup_ulp(struct sock *sk);
2246 void tcp_update_ulp(struct sock *sk, struct proto *p,
2247 		    void (*write_space)(struct sock *sk));
2248 
2249 #define MODULE_ALIAS_TCP_ULP(name)				\
2250 	__MODULE_INFO(alias, alias_userspace, name);		\
2251 	__MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2252 
2253 struct sk_msg;
2254 struct sk_psock;
2255 
2256 #ifdef CONFIG_BPF_STREAM_PARSER
2257 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
2258 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2259 #else
tcp_bpf_clone(const struct sock *sk, struct sock *newsk)2260 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2261 {
2262 }
2263 #endif /* CONFIG_BPF_STREAM_PARSER */
2264 
2265 #ifdef CONFIG_NET_SOCK_MSG
2266 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes,
2267 			  int flags);
2268 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock,
2269 		      struct msghdr *msg, int len, int flags);
2270 #endif /* CONFIG_NET_SOCK_MSG */
2271 
2272 #ifdef CONFIG_CGROUP_BPF
bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, struct sk_buff *skb, unsigned int end_offset)2273 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2274 				      struct sk_buff *skb,
2275 				      unsigned int end_offset)
2276 {
2277 	skops->skb = skb;
2278 	skops->skb_data_end = skb->data + end_offset;
2279 }
2280 #else
bpf_skops_init_skb(struct bpf_sock_ops_kern *skops, struct sk_buff *skb, unsigned int end_offset)2281 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2282 				      struct sk_buff *skb,
2283 				      unsigned int end_offset)
2284 {
2285 }
2286 #endif
2287 
2288 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2289  * is < 0, then the BPF op failed (for example if the loaded BPF
2290  * program does not support the chosen operation or there is no BPF
2291  * program loaded).
2292  */
2293 #ifdef CONFIG_BPF
tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)2294 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2295 {
2296 	struct bpf_sock_ops_kern sock_ops;
2297 	int ret;
2298 
2299 	memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2300 	if (sk_fullsock(sk)) {
2301 		sock_ops.is_fullsock = 1;
2302 		sock_owned_by_me(sk);
2303 	}
2304 
2305 	sock_ops.sk = sk;
2306 	sock_ops.op = op;
2307 	if (nargs > 0)
2308 		memcpy(sock_ops.args, args, nargs * sizeof(*args));
2309 
2310 	ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2311 	if (ret == 0)
2312 		ret = sock_ops.reply;
2313 	else
2314 		ret = -1;
2315 	return ret;
2316 }
2317 
tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)2318 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2319 {
2320 	u32 args[2] = {arg1, arg2};
2321 
2322 	return tcp_call_bpf(sk, op, 2, args);
2323 }
2324 
tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, u32 arg3)2325 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2326 				    u32 arg3)
2327 {
2328 	u32 args[3] = {arg1, arg2, arg3};
2329 
2330 	return tcp_call_bpf(sk, op, 3, args);
2331 }
2332 
2333 #else
tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)2334 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2335 {
2336 	return -EPERM;
2337 }
2338 
tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)2339 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2340 {
2341 	return -EPERM;
2342 }
2343 
tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2, u32 arg3)2344 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2345 				    u32 arg3)
2346 {
2347 	return -EPERM;
2348 }
2349 
2350 #endif
2351 
tcp_timeout_init(struct sock *sk)2352 static inline u32 tcp_timeout_init(struct sock *sk)
2353 {
2354 	int timeout;
2355 
2356 	timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2357 
2358 	if (timeout <= 0)
2359 		timeout = TCP_TIMEOUT_INIT;
2360 	return timeout;
2361 }
2362 
tcp_rwnd_init_bpf(struct sock *sk)2363 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2364 {
2365 	int rwnd;
2366 
2367 	rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2368 
2369 	if (rwnd < 0)
2370 		rwnd = 0;
2371 	return rwnd;
2372 }
2373 
tcp_bpf_ca_needs_ecn(struct sock *sk)2374 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2375 {
2376 	return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2377 }
2378 
tcp_bpf_rtt(struct sock *sk)2379 static inline void tcp_bpf_rtt(struct sock *sk)
2380 {
2381 	if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2382 		tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2383 }
2384 
2385 #if IS_ENABLED(CONFIG_SMC)
2386 extern struct static_key_false tcp_have_smc;
2387 #endif
2388 
2389 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2390 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2391 			     void (*cad)(struct sock *sk, u32 ack_seq));
2392 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2393 void clean_acked_data_flush(void);
2394 #endif
2395 
2396 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
tcp_add_tx_delay(struct sk_buff *skb, const struct tcp_sock *tp)2397 static inline void tcp_add_tx_delay(struct sk_buff *skb,
2398 				    const struct tcp_sock *tp)
2399 {
2400 	if (static_branch_unlikely(&tcp_tx_delay_enabled))
2401 		skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2402 }
2403 
2404 /* Compute Earliest Departure Time for some control packets
2405  * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2406  */
tcp_transmit_time(const struct sock *sk)2407 static inline u64 tcp_transmit_time(const struct sock *sk)
2408 {
2409 	if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2410 		u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2411 			tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2412 
2413 		return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;
2414 	}
2415 	return 0;
2416 }
2417 
2418 #endif	/* _TCP_H */
2419