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