1 /*
2 * INET An implementation of the TCP/IP protocol suite for the LINUX
3 * operating system. INET is implemented using the BSD Socket
4 * interface as the means of communication with the user level.
5 *
6 * Definitions for the TCP module.
7 *
8 * Version: @(#)tcp.h 1.0.5 05/23/93
9 *
10 * Authors: Ross Biro
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 *
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License
15 * as published by the Free Software Foundation; either version
16 * 2 of the License, or (at your option) any later version.
17 */
18 #ifndef _TCP_H
19 #define _TCP_H
20
21 #define FASTRETRANS_DEBUG 1
22
23 #include <linux/list.h>
24 #include <linux/tcp.h>
25 #include <linux/bug.h>
26 #include <linux/slab.h>
27 #include <linux/cache.h>
28 #include <linux/percpu.h>
29 #include <linux/skbuff.h>
30 #include <linux/cryptohash.h>
31 #include <linux/kref.h>
32 #include <linux/ktime.h>
33
34 #include <net/inet_connection_sock.h>
35 #include <net/inet_timewait_sock.h>
36 #include <net/inet_hashtables.h>
37 #include <net/checksum.h>
38 #include <net/request_sock.h>
39 #include <net/sock_reuseport.h>
40 #include <net/sock.h>
41 #include <net/snmp.h>
42 #include <net/ip.h>
43 #include <net/tcp_states.h>
44 #include <net/inet_ecn.h>
45 #include <net/dst.h>
46
47 #include <linux/seq_file.h>
48 #include <linux/memcontrol.h>
49 #include <linux/bpf-cgroup.h>
50
51 extern struct inet_hashinfo tcp_hashinfo;
52
53 extern struct percpu_counter tcp_orphan_count;
54 void tcp_time_wait(struct sock *sk, int state, int timeo);
55
56 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER)
57 #define MAX_TCP_OPTION_SPACE 40
58 #define TCP_MIN_SND_MSS 48
59 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
60
61 /*
62 * Never offer a window over 32767 without using window scaling. Some
63 * poor stacks do signed 16bit maths!
64 */
65 #define MAX_TCP_WINDOW 32767U
66
67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68 #define TCP_MIN_MSS 88U
69
70 /* The least MTU to use for probing */
71 #define TCP_BASE_MSS 1024
72
73 /* probing interval, default to 10 minutes as per RFC4821 */
74 #define TCP_PROBE_INTERVAL 600
75
76 /* Specify interval when tcp mtu probing will stop */
77 #define TCP_PROBE_THRESHOLD 8
78
79 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
80 #define TCP_FASTRETRANS_THRESH 3
81
82 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
83 #define TCP_MAX_QUICKACKS 16U
84
85 /* Maximal number of window scale according to RFC1323 */
86 #define TCP_MAX_WSCALE 14U
87
88 /* urg_data states */
89 #define TCP_URG_VALID 0x0100
90 #define TCP_URG_NOTYET 0x0200
91 #define TCP_URG_READ 0x0400
92
93 #define TCP_RETR1 3 /*
94 * This is how many retries it does before it
95 * tries to figure out if the gateway is
96 * down. Minimal RFC value is 3; it corresponds
97 * to ~3sec-8min depending on RTO.
98 */
99
100 #define TCP_RETR2 15 /*
101 * This should take at least
102 * 90 minutes to time out.
103 * RFC1122 says that the limit is 100 sec.
104 * 15 is ~13-30min depending on RTO.
105 */
106
107 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
108 * when active opening a connection.
109 * RFC1122 says the minimum retry MUST
110 * be at least 180secs. Nevertheless
111 * this value is corresponding to
112 * 63secs of retransmission with the
113 * current initial RTO.
114 */
115
116 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
117 * when passive opening a connection.
118 * This is corresponding to 31secs of
119 * retransmission with the current
120 * initial RTO.
121 */
122
123 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
124 * state, about 60 seconds */
125 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
126 /* BSD style FIN_WAIT2 deadlock breaker.
127 * It used to be 3min, new value is 60sec,
128 * to combine FIN-WAIT-2 timeout with
129 * TIME-WAIT timer.
130 */
131 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */
132
133 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
134 #if HZ >= 100
135 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
136 #define TCP_ATO_MIN ((unsigned)(HZ/25))
137 #else
138 #define TCP_DELACK_MIN 4U
139 #define TCP_ATO_MIN 4U
140 #endif
141 #define TCP_RTO_MAX ((unsigned)(120*HZ))
142 #define TCP_RTO_MIN ((unsigned)(HZ/5))
143 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
144 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
145 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
146 * used as a fallback RTO for the
147 * initial data transmission if no
148 * valid RTT sample has been acquired,
149 * most likely due to retrans in 3WHS.
150 */
151
152 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
153 * for local resources.
154 */
155 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
156 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
157 #define TCP_KEEPALIVE_INTVL (75*HZ)
158
159 #define MAX_TCP_KEEPIDLE 32767
160 #define MAX_TCP_KEEPINTVL 32767
161 #define MAX_TCP_KEEPCNT 127
162 #define MAX_TCP_SYNCNT 127
163
164 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
165
166 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
167 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
168 * after this time. It should be equal
169 * (or greater than) TCP_TIMEWAIT_LEN
170 * to provide reliability equal to one
171 * provided by timewait state.
172 */
173 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
174 * timestamps. It must be less than
175 * minimal timewait lifetime.
176 */
177 /*
178 * TCP option
179 */
180
181 #define TCPOPT_NOP 1 /* Padding */
182 #define TCPOPT_EOL 0 /* End of options */
183 #define TCPOPT_MSS 2 /* Segment size negotiating */
184 #define TCPOPT_WINDOW 3 /* Window scaling */
185 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
186 #define TCPOPT_SACK 5 /* SACK Block */
187 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
188 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
189 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
190 #define TCPOPT_EXP 254 /* Experimental */
191 /* Magic number to be after the option value for sharing TCP
192 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
193 */
194 #define TCPOPT_FASTOPEN_MAGIC 0xF989
195 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9
196
197 /*
198 * TCP option lengths
199 */
200
201 #define TCPOLEN_MSS 4
202 #define TCPOLEN_WINDOW 3
203 #define TCPOLEN_SACK_PERM 2
204 #define TCPOLEN_TIMESTAMP 10
205 #define TCPOLEN_MD5SIG 18
206 #define TCPOLEN_FASTOPEN_BASE 2
207 #define TCPOLEN_EXP_FASTOPEN_BASE 4
208 #define TCPOLEN_EXP_SMC_BASE 6
209
210 /* But this is what stacks really send out. */
211 #define TCPOLEN_TSTAMP_ALIGNED 12
212 #define TCPOLEN_WSCALE_ALIGNED 4
213 #define TCPOLEN_SACKPERM_ALIGNED 4
214 #define TCPOLEN_SACK_BASE 2
215 #define TCPOLEN_SACK_BASE_ALIGNED 4
216 #define TCPOLEN_SACK_PERBLOCK 8
217 #define TCPOLEN_MD5SIG_ALIGNED 20
218 #define TCPOLEN_MSS_ALIGNED 4
219 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
220
221 /* Flags in tp->nonagle */
222 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
223 #define TCP_NAGLE_CORK 2 /* Socket is corked */
224 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
225
226 /* TCP thin-stream limits */
227 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
228
229 /* TCP initial congestion window as per rfc6928 */
230 #define TCP_INIT_CWND 10
231
232 /* Bit Flags for sysctl_tcp_fastopen */
233 #define TFO_CLIENT_ENABLE 1
234 #define TFO_SERVER_ENABLE 2
235 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
236
237 /* Accept SYN data w/o any cookie option */
238 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
239
240 /* Force enable TFO on all listeners, i.e., not requiring the
241 * TCP_FASTOPEN socket option.
242 */
243 #define TFO_SERVER_WO_SOCKOPT1 0x400
244
245
246 /* sysctl variables for tcp */
247 extern int sysctl_tcp_max_orphans;
248 extern long sysctl_tcp_mem[3];
249
250 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
251 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
252 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */
253
254 extern atomic_long_t tcp_memory_allocated;
255 extern struct percpu_counter tcp_sockets_allocated;
256 extern unsigned long tcp_memory_pressure;
257
258 /* optimized version of sk_under_memory_pressure() for TCP sockets */
tcp_under_memory_pressure(const struct sock * sk)259 static inline bool tcp_under_memory_pressure(const struct sock *sk)
260 {
261 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
262 mem_cgroup_under_socket_pressure(sk->sk_memcg))
263 return true;
264
265 return READ_ONCE(tcp_memory_pressure);
266 }
267 /*
268 * The next routines deal with comparing 32 bit unsigned ints
269 * and worry about wraparound (automatic with unsigned arithmetic).
270 */
271
before(__u32 seq1,__u32 seq2)272 static inline bool before(__u32 seq1, __u32 seq2)
273 {
274 return (__s32)(seq1-seq2) < 0;
275 }
276 #define after(seq2, seq1) before(seq1, seq2)
277
278 /* is s2<=s1<=s3 ? */
between(__u32 seq1,__u32 seq2,__u32 seq3)279 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
280 {
281 return seq3 - seq2 >= seq1 - seq2;
282 }
283
tcp_out_of_memory(struct sock * sk)284 static inline bool tcp_out_of_memory(struct sock *sk)
285 {
286 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
287 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
288 return true;
289 return false;
290 }
291
292 void sk_forced_mem_schedule(struct sock *sk, int size);
293
tcp_too_many_orphans(struct sock * sk,int shift)294 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
295 {
296 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
297 int orphans = percpu_counter_read_positive(ocp);
298
299 if (orphans << shift > sysctl_tcp_max_orphans) {
300 orphans = percpu_counter_sum_positive(ocp);
301 if (orphans << shift > sysctl_tcp_max_orphans)
302 return true;
303 }
304 return false;
305 }
306
307 bool tcp_check_oom(struct sock *sk, int shift);
308
309
310 extern struct proto tcp_prot;
311
312 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
313 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
314 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
315 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
316
317 void tcp_tasklet_init(void);
318
319 void tcp_v4_err(struct sk_buff *skb, u32);
320
321 void tcp_shutdown(struct sock *sk, int how);
322
323 int tcp_v4_early_demux(struct sk_buff *skb);
324 int tcp_v4_rcv(struct sk_buff *skb);
325
326 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
327 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
328 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
329 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
330 int flags);
331 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
332 size_t size, int flags);
333 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
334 size_t size, int flags);
335 void tcp_release_cb(struct sock *sk);
336 void tcp_wfree(struct sk_buff *skb);
337 void tcp_write_timer_handler(struct sock *sk);
338 void tcp_delack_timer_handler(struct sock *sk);
339 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
340 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
341 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
342 void tcp_rcv_space_adjust(struct sock *sk);
343 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
344 void tcp_twsk_destructor(struct sock *sk);
345 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
346 struct pipe_inode_info *pipe, size_t len,
347 unsigned int flags);
348
tcp_dec_quickack_mode(struct sock * sk)349 static inline void tcp_dec_quickack_mode(struct sock *sk)
350 {
351 struct inet_connection_sock *icsk = inet_csk(sk);
352
353 if (icsk->icsk_ack.quick) {
354 /* How many ACKs S/ACKing new data have we sent? */
355 const unsigned int pkts = inet_csk_ack_scheduled(sk) ? 1 : 0;
356
357 if (pkts >= icsk->icsk_ack.quick) {
358 icsk->icsk_ack.quick = 0;
359 /* Leaving quickack mode we deflate ATO. */
360 icsk->icsk_ack.ato = TCP_ATO_MIN;
361 } else
362 icsk->icsk_ack.quick -= pkts;
363 }
364 }
365
366 #define TCP_ECN_OK 1
367 #define TCP_ECN_QUEUE_CWR 2
368 #define TCP_ECN_DEMAND_CWR 4
369 #define TCP_ECN_SEEN 8
370
371 enum tcp_tw_status {
372 TCP_TW_SUCCESS = 0,
373 TCP_TW_RST = 1,
374 TCP_TW_ACK = 2,
375 TCP_TW_SYN = 3
376 };
377
378
379 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
380 struct sk_buff *skb,
381 const struct tcphdr *th);
382 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
383 struct request_sock *req, bool fastopen,
384 bool *lost_race);
385 int tcp_child_process(struct sock *parent, struct sock *child,
386 struct sk_buff *skb);
387 void tcp_enter_loss(struct sock *sk);
388 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
389 void tcp_clear_retrans(struct tcp_sock *tp);
390 void tcp_update_metrics(struct sock *sk);
391 void tcp_init_metrics(struct sock *sk);
392 void tcp_metrics_init(void);
393 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
394 void __tcp_close(struct sock *sk, long timeout);
395 void tcp_close(struct sock *sk, long timeout);
396 void tcp_init_sock(struct sock *sk);
397 void tcp_init_transfer(struct sock *sk, int bpf_op);
398 __poll_t tcp_poll(struct file *file, struct socket *sock,
399 struct poll_table_struct *wait);
400 int tcp_getsockopt(struct sock *sk, int level, int optname,
401 char __user *optval, int __user *optlen);
402 int tcp_setsockopt(struct sock *sk, int level, int optname,
403 char __user *optval, unsigned int optlen);
404 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
405 char __user *optval, int __user *optlen);
406 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
407 char __user *optval, unsigned int optlen);
408 void tcp_set_keepalive(struct sock *sk, int val);
409 void tcp_syn_ack_timeout(const struct request_sock *req);
410 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
411 int flags, int *addr_len);
412 int tcp_set_rcvlowat(struct sock *sk, int val);
413 void tcp_data_ready(struct sock *sk);
414 int tcp_mmap(struct file *file, struct socket *sock,
415 struct vm_area_struct *vma);
416 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
417 struct tcp_options_received *opt_rx,
418 int estab, struct tcp_fastopen_cookie *foc);
419 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
420
421 /*
422 * TCP v4 functions exported for the inet6 API
423 */
424
425 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
426 void tcp_v4_mtu_reduced(struct sock *sk);
427 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
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 int tcp_disconnect(struct sock *sk, int flags);
451
452 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
453 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
454 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
455
456 /* From syncookies.c */
457 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
458 struct request_sock *req,
459 struct dst_entry *dst, u32 tsoff);
460 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
461 u32 cookie);
462 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
463 #ifdef CONFIG_SYN_COOKIES
464
465 /* Syncookies use a monotonic timer which increments every 60 seconds.
466 * This counter is used both as a hash input and partially encoded into
467 * the cookie value. A cookie is only validated further if the delta
468 * between the current counter value and the encoded one is less than this,
469 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
470 * the counter advances immediately after a cookie is generated).
471 */
472 #define MAX_SYNCOOKIE_AGE 2
473 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
474 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
475
476 /* syncookies: remember time of last synqueue overflow
477 * But do not dirty this field too often (once per second is enough)
478 * It is racy as we do not hold a lock, but race is very minor.
479 */
tcp_synq_overflow(const struct sock * sk)480 static inline void tcp_synq_overflow(const struct sock *sk)
481 {
482 unsigned int last_overflow;
483 unsigned int now = jiffies;
484
485 if (sk->sk_reuseport) {
486 struct sock_reuseport *reuse;
487
488 reuse = rcu_dereference(sk->sk_reuseport_cb);
489 if (likely(reuse)) {
490 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
491 if (!time_between32(now, last_overflow,
492 last_overflow + HZ))
493 WRITE_ONCE(reuse->synq_overflow_ts, now);
494 return;
495 }
496 }
497
498 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
499 if (!time_between32(now, last_overflow, last_overflow + HZ))
500 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
501 }
502
503 /* syncookies: no recent synqueue overflow on this listening socket? */
tcp_synq_no_recent_overflow(const struct sock * sk)504 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
505 {
506 unsigned int last_overflow;
507 unsigned int now = jiffies;
508
509 if (sk->sk_reuseport) {
510 struct sock_reuseport *reuse;
511
512 reuse = rcu_dereference(sk->sk_reuseport_cb);
513 if (likely(reuse)) {
514 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
515 return !time_between32(now, last_overflow - HZ,
516 last_overflow +
517 TCP_SYNCOOKIE_VALID);
518 }
519 }
520
521 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
522
523 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
524 * then we're under synflood. However, we have to use
525 * 'last_overflow - HZ' as lower bound. That's because a concurrent
526 * tcp_synq_overflow() could update .ts_recent_stamp after we read
527 * jiffies but before we store .ts_recent_stamp into last_overflow,
528 * which could lead to rejecting a valid syncookie.
529 */
530 return !time_between32(now, last_overflow - HZ,
531 last_overflow + TCP_SYNCOOKIE_VALID);
532 }
533
tcp_cookie_time(void)534 static inline u32 tcp_cookie_time(void)
535 {
536 u64 val = get_jiffies_64();
537
538 do_div(val, TCP_SYNCOOKIE_PERIOD);
539 return val;
540 }
541
542 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
543 u16 *mssp);
544 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
545 u64 cookie_init_timestamp(struct request_sock *req);
546 bool cookie_timestamp_decode(const struct net *net,
547 struct tcp_options_received *opt);
548 bool cookie_ecn_ok(const struct tcp_options_received *opt,
549 const struct net *net, const struct dst_entry *dst);
550
551 /* From net/ipv6/syncookies.c */
552 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
553 u32 cookie);
554 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
555
556 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
557 const struct tcphdr *th, u16 *mssp);
558 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
559 #endif
560 /* tcp_output.c */
561
562 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
563 int nonagle);
564 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
565 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
566 void tcp_retransmit_timer(struct sock *sk);
567 void tcp_xmit_retransmit_queue(struct sock *);
568 void tcp_simple_retransmit(struct sock *);
569 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
570 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
571 enum tcp_queue {
572 TCP_FRAG_IN_WRITE_QUEUE,
573 TCP_FRAG_IN_RTX_QUEUE,
574 };
575 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
576 struct sk_buff *skb, u32 len,
577 unsigned int mss_now, gfp_t gfp);
578
579 void tcp_send_probe0(struct sock *);
580 void tcp_send_partial(struct sock *);
581 int tcp_write_wakeup(struct sock *, int mib);
582 void tcp_send_fin(struct sock *sk);
583 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
584 int tcp_send_synack(struct sock *);
585 void tcp_push_one(struct sock *, unsigned int mss_now);
586 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
587 void tcp_send_ack(struct sock *sk);
588 void tcp_send_delayed_ack(struct sock *sk);
589 void tcp_send_loss_probe(struct sock *sk);
590 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
591 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
592 const struct sk_buff *next_skb);
593
594 /* tcp_input.c */
595 void tcp_rearm_rto(struct sock *sk);
596 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
597 void tcp_reset(struct sock *sk);
598 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
599 void tcp_fin(struct sock *sk);
600 void tcp_check_space(struct sock *sk);
601
602 /* tcp_timer.c */
603 void tcp_init_xmit_timers(struct sock *);
tcp_clear_xmit_timers(struct sock * sk)604 static inline void tcp_clear_xmit_timers(struct sock *sk)
605 {
606 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
607 __sock_put(sk);
608
609 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
610 __sock_put(sk);
611
612 inet_csk_clear_xmit_timers(sk);
613 }
614
615 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
616 unsigned int tcp_current_mss(struct sock *sk);
617
618 /* Bound MSS / TSO packet size with the half of the window */
tcp_bound_to_half_wnd(struct tcp_sock * tp,int pktsize)619 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
620 {
621 int cutoff;
622
623 /* When peer uses tiny windows, there is no use in packetizing
624 * to sub-MSS pieces for the sake of SWS or making sure there
625 * are enough packets in the pipe for fast recovery.
626 *
627 * On the other hand, for extremely large MSS devices, handling
628 * smaller than MSS windows in this way does make sense.
629 */
630 if (tp->max_window > TCP_MSS_DEFAULT)
631 cutoff = (tp->max_window >> 1);
632 else
633 cutoff = tp->max_window;
634
635 if (cutoff && pktsize > cutoff)
636 return max_t(int, cutoff, 68U - tp->tcp_header_len);
637 else
638 return pktsize;
639 }
640
641 /* tcp.c */
642 void tcp_get_info(struct sock *, struct tcp_info *);
643
644 /* Read 'sendfile()'-style from a TCP socket */
645 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
646 sk_read_actor_t recv_actor);
647
648 void tcp_initialize_rcv_mss(struct sock *sk);
649
650 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
651 int tcp_mss_to_mtu(struct sock *sk, int mss);
652 void tcp_mtup_init(struct sock *sk);
653 void tcp_init_buffer_space(struct sock *sk);
654
tcp_bound_rto(const struct sock * sk)655 static inline void tcp_bound_rto(const struct sock *sk)
656 {
657 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
658 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
659 }
660
__tcp_set_rto(const struct tcp_sock * tp)661 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
662 {
663 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
664 }
665
__tcp_fast_path_on(struct tcp_sock * tp,u32 snd_wnd)666 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
667 {
668 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
669 ntohl(TCP_FLAG_ACK) |
670 snd_wnd);
671 }
672
tcp_fast_path_on(struct tcp_sock * tp)673 static inline void tcp_fast_path_on(struct tcp_sock *tp)
674 {
675 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
676 }
677
tcp_fast_path_check(struct sock * sk)678 static inline void tcp_fast_path_check(struct sock *sk)
679 {
680 struct tcp_sock *tp = tcp_sk(sk);
681
682 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
683 tp->rcv_wnd &&
684 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
685 !tp->urg_data)
686 tcp_fast_path_on(tp);
687 }
688
689 /* Compute the actual rto_min value */
tcp_rto_min(struct sock * sk)690 static inline u32 tcp_rto_min(struct sock *sk)
691 {
692 const struct dst_entry *dst = __sk_dst_get(sk);
693 u32 rto_min = TCP_RTO_MIN;
694
695 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
696 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
697 return rto_min;
698 }
699
tcp_rto_min_us(struct sock * sk)700 static inline u32 tcp_rto_min_us(struct sock *sk)
701 {
702 return jiffies_to_usecs(tcp_rto_min(sk));
703 }
704
tcp_ca_dst_locked(const struct dst_entry * dst)705 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
706 {
707 return dst_metric_locked(dst, RTAX_CC_ALGO);
708 }
709
710 /* Minimum RTT in usec. ~0 means not available. */
tcp_min_rtt(const struct tcp_sock * tp)711 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
712 {
713 return minmax_get(&tp->rtt_min);
714 }
715
716 /* Compute the actual receive window we are currently advertising.
717 * Rcv_nxt can be after the window if our peer push more data
718 * than the offered window.
719 */
tcp_receive_window(const struct tcp_sock * tp)720 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
721 {
722 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
723
724 if (win < 0)
725 win = 0;
726 return (u32) win;
727 }
728
729 /* Choose a new window, without checks for shrinking, and without
730 * scaling applied to the result. The caller does these things
731 * if necessary. This is a "raw" window selection.
732 */
733 u32 __tcp_select_window(struct sock *sk);
734
735 void tcp_send_window_probe(struct sock *sk);
736
737 /* TCP uses 32bit jiffies to save some space.
738 * Note that this is different from tcp_time_stamp, which
739 * historically has been the same until linux-4.13.
740 */
741 #define tcp_jiffies32 ((u32)jiffies)
742
743 /*
744 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
745 * It is no longer tied to jiffies, but to 1 ms clock.
746 * Note: double check if you want to use tcp_jiffies32 instead of this.
747 */
748 #define TCP_TS_HZ 1000
749
tcp_clock_ns(void)750 static inline u64 tcp_clock_ns(void)
751 {
752 return local_clock();
753 }
754
tcp_clock_us(void)755 static inline u64 tcp_clock_us(void)
756 {
757 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
758 }
759
760 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
tcp_time_stamp(const struct tcp_sock * tp)761 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
762 {
763 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
764 }
765
766 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
tcp_time_stamp_raw(void)767 static inline u32 tcp_time_stamp_raw(void)
768 {
769 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ);
770 }
771
772
773 /* Refresh 1us clock of a TCP socket,
774 * ensuring monotically increasing values.
775 */
tcp_mstamp_refresh(struct tcp_sock * tp)776 static inline void tcp_mstamp_refresh(struct tcp_sock *tp)
777 {
778 u64 val = tcp_clock_us();
779
780 if (val > tp->tcp_mstamp)
781 tp->tcp_mstamp = val;
782 }
783
tcp_stamp_us_delta(u64 t1,u64 t0)784 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
785 {
786 return max_t(s64, t1 - t0, 0);
787 }
788
tcp_skb_timestamp(const struct sk_buff * skb)789 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
790 {
791 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ);
792 }
793
794
795 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
796
797 #define TCPHDR_FIN 0x01
798 #define TCPHDR_SYN 0x02
799 #define TCPHDR_RST 0x04
800 #define TCPHDR_PSH 0x08
801 #define TCPHDR_ACK 0x10
802 #define TCPHDR_URG 0x20
803 #define TCPHDR_ECE 0x40
804 #define TCPHDR_CWR 0x80
805
806 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
807
808 /* This is what the send packet queuing engine uses to pass
809 * TCP per-packet control information to the transmission code.
810 * We also store the host-order sequence numbers in here too.
811 * This is 44 bytes if IPV6 is enabled.
812 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
813 */
814 struct tcp_skb_cb {
815 __u32 seq; /* Starting sequence number */
816 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
817 union {
818 /* Note : tcp_tw_isn is used in input path only
819 * (isn chosen by tcp_timewait_state_process())
820 *
821 * tcp_gso_segs/size are used in write queue only,
822 * cf tcp_skb_pcount()/tcp_skb_mss()
823 */
824 __u32 tcp_tw_isn;
825 struct {
826 u16 tcp_gso_segs;
827 u16 tcp_gso_size;
828 };
829 };
830 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
831
832 __u8 sacked; /* State flags for SACK. */
833 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
834 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
835 #define TCPCB_LOST 0x04 /* SKB is lost */
836 #define TCPCB_TAGBITS 0x07 /* All tag bits */
837 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
838 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
839 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
840 TCPCB_REPAIRED)
841
842 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
843 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
844 eor:1, /* Is skb MSG_EOR marked? */
845 has_rxtstamp:1, /* SKB has a RX timestamp */
846 unused:5;
847 __u32 ack_seq; /* Sequence number ACK'd */
848 union {
849 struct {
850 /* There is space for up to 24 bytes */
851 __u32 in_flight:30,/* Bytes in flight at transmit */
852 is_app_limited:1, /* cwnd not fully used? */
853 unused:1;
854 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
855 __u32 delivered;
856 /* start of send pipeline phase */
857 u64 first_tx_mstamp;
858 /* when we reached the "delivered" count */
859 u64 delivered_mstamp;
860 } tx; /* only used for outgoing skbs */
861 union {
862 struct inet_skb_parm h4;
863 #if IS_ENABLED(CONFIG_IPV6)
864 struct inet6_skb_parm h6;
865 #endif
866 } header; /* For incoming skbs */
867 struct {
868 __u32 flags;
869 struct sock *sk_redir;
870 void *data_end;
871 } bpf;
872 };
873 };
874
875 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
876
bpf_compute_data_end_sk_skb(struct sk_buff * skb)877 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb)
878 {
879 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb);
880 }
881
882 #if IS_ENABLED(CONFIG_IPV6)
883 /* This is the variant of inet6_iif() that must be used by TCP,
884 * as TCP moves IP6CB into a different location in skb->cb[]
885 */
tcp_v6_iif(const struct sk_buff * skb)886 static inline int tcp_v6_iif(const struct sk_buff *skb)
887 {
888 return TCP_SKB_CB(skb)->header.h6.iif;
889 }
890
tcp_v6_iif_l3_slave(const struct sk_buff * skb)891 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
892 {
893 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
894
895 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
896 }
897
898 /* TCP_SKB_CB reference means this can not be used from early demux */
tcp_v6_sdif(const struct sk_buff * skb)899 static inline int tcp_v6_sdif(const struct sk_buff *skb)
900 {
901 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
902 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
903 return TCP_SKB_CB(skb)->header.h6.iif;
904 #endif
905 return 0;
906 }
907
908 void tcp_v6_early_demux(struct sk_buff *skb);
909 #endif
910
inet_exact_dif_match(struct net * net,struct sk_buff * skb)911 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
912 {
913 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
914 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
915 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
916 return true;
917 #endif
918 return false;
919 }
920
921 /* TCP_SKB_CB reference means this can not be used from early demux */
tcp_v4_sdif(struct sk_buff * skb)922 static inline int tcp_v4_sdif(struct sk_buff *skb)
923 {
924 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
925 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
926 return TCP_SKB_CB(skb)->header.h4.iif;
927 #endif
928 return 0;
929 }
930
931 /* Due to TSO, an SKB can be composed of multiple actual
932 * packets. To keep these tracked properly, we use this.
933 */
tcp_skb_pcount(const struct sk_buff * skb)934 static inline int tcp_skb_pcount(const struct sk_buff *skb)
935 {
936 return TCP_SKB_CB(skb)->tcp_gso_segs;
937 }
938
tcp_skb_pcount_set(struct sk_buff * skb,int segs)939 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
940 {
941 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
942 }
943
tcp_skb_pcount_add(struct sk_buff * skb,int segs)944 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
945 {
946 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
947 }
948
949 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
tcp_skb_mss(const struct sk_buff * skb)950 static inline int tcp_skb_mss(const struct sk_buff *skb)
951 {
952 return TCP_SKB_CB(skb)->tcp_gso_size;
953 }
954
tcp_skb_can_collapse_to(const struct sk_buff * skb)955 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
956 {
957 return likely(!TCP_SKB_CB(skb)->eor);
958 }
959
960 /* Events passed to congestion control interface */
961 enum tcp_ca_event {
962 CA_EVENT_TX_START, /* first transmit when no packets in flight */
963 CA_EVENT_CWND_RESTART, /* congestion window restart */
964 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
965 CA_EVENT_LOSS, /* loss timeout */
966 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
967 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
968 };
969
970 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
971 enum tcp_ca_ack_event_flags {
972 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
973 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
974 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
975 };
976
977 /*
978 * Interface for adding new TCP congestion control handlers
979 */
980 #define TCP_CA_NAME_MAX 16
981 #define TCP_CA_MAX 128
982 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
983
984 #define TCP_CA_UNSPEC 0
985
986 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
987 #define TCP_CONG_NON_RESTRICTED 0x1
988 /* Requires ECN/ECT set on all packets */
989 #define TCP_CONG_NEEDS_ECN 0x2
990
991 union tcp_cc_info;
992
993 struct ack_sample {
994 u32 pkts_acked;
995 s32 rtt_us;
996 u32 in_flight;
997 };
998
999 /* A rate sample measures the number of (original/retransmitted) data
1000 * packets delivered "delivered" over an interval of time "interval_us".
1001 * The tcp_rate.c code fills in the rate sample, and congestion
1002 * control modules that define a cong_control function to run at the end
1003 * of ACK processing can optionally chose to consult this sample when
1004 * setting cwnd and pacing rate.
1005 * A sample is invalid if "delivered" or "interval_us" is negative.
1006 */
1007 struct rate_sample {
1008 u64 prior_mstamp; /* starting timestamp for interval */
1009 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1010 s32 delivered; /* number of packets delivered over interval */
1011 long interval_us; /* time for tp->delivered to incr "delivered" */
1012 u32 snd_interval_us; /* snd interval for delivered packets */
1013 u32 rcv_interval_us; /* rcv interval for delivered packets */
1014 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1015 int losses; /* number of packets marked lost upon ACK */
1016 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1017 u32 prior_in_flight; /* in flight before this ACK */
1018 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1019 bool is_retrans; /* is sample from retransmission? */
1020 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1021 };
1022
1023 struct tcp_congestion_ops {
1024 struct list_head list;
1025 u32 key;
1026 u32 flags;
1027
1028 /* initialize private data (optional) */
1029 void (*init)(struct sock *sk);
1030 /* cleanup private data (optional) */
1031 void (*release)(struct sock *sk);
1032
1033 /* return slow start threshold (required) */
1034 u32 (*ssthresh)(struct sock *sk);
1035 /* do new cwnd calculation (required) */
1036 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1037 /* call before changing ca_state (optional) */
1038 void (*set_state)(struct sock *sk, u8 new_state);
1039 /* call when cwnd event occurs (optional) */
1040 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1041 /* call when ack arrives (optional) */
1042 void (*in_ack_event)(struct sock *sk, u32 flags);
1043 /* new value of cwnd after loss (required) */
1044 u32 (*undo_cwnd)(struct sock *sk);
1045 /* hook for packet ack accounting (optional) */
1046 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1047 /* override sysctl_tcp_min_tso_segs */
1048 u32 (*min_tso_segs)(struct sock *sk);
1049 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1050 u32 (*sndbuf_expand)(struct sock *sk);
1051 /* call when packets are delivered to update cwnd and pacing rate,
1052 * after all the ca_state processing. (optional)
1053 */
1054 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1055 /* get info for inet_diag (optional) */
1056 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1057 union tcp_cc_info *info);
1058
1059 char name[TCP_CA_NAME_MAX];
1060 struct module *owner;
1061 };
1062
1063 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1064 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1065
1066 void tcp_assign_congestion_control(struct sock *sk);
1067 void tcp_init_congestion_control(struct sock *sk);
1068 void tcp_cleanup_congestion_control(struct sock *sk);
1069 int tcp_set_default_congestion_control(struct net *net, const char *name);
1070 void tcp_get_default_congestion_control(struct net *net, char *name);
1071 void tcp_get_available_congestion_control(char *buf, size_t len);
1072 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1073 int tcp_set_allowed_congestion_control(char *allowed);
1074 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1075 bool reinit, bool cap_net_admin);
1076 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1077 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1078
1079 u32 tcp_reno_ssthresh(struct sock *sk);
1080 u32 tcp_reno_undo_cwnd(struct sock *sk);
1081 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1082 extern struct tcp_congestion_ops tcp_reno;
1083
1084 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1085 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1086 #ifdef CONFIG_INET
1087 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1088 #else
tcp_ca_get_name_by_key(u32 key,char * buffer)1089 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1090 {
1091 return NULL;
1092 }
1093 #endif
1094
tcp_ca_needs_ecn(const struct sock * sk)1095 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1096 {
1097 const struct inet_connection_sock *icsk = inet_csk(sk);
1098
1099 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1100 }
1101
tcp_set_ca_state(struct sock * sk,const u8 ca_state)1102 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1103 {
1104 struct inet_connection_sock *icsk = inet_csk(sk);
1105
1106 if (icsk->icsk_ca_ops->set_state)
1107 icsk->icsk_ca_ops->set_state(sk, ca_state);
1108 icsk->icsk_ca_state = ca_state;
1109 }
1110
tcp_ca_event(struct sock * sk,const enum tcp_ca_event event)1111 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1112 {
1113 const struct inet_connection_sock *icsk = inet_csk(sk);
1114
1115 if (icsk->icsk_ca_ops->cwnd_event)
1116 icsk->icsk_ca_ops->cwnd_event(sk, event);
1117 }
1118
1119 /* From tcp_rate.c */
1120 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1121 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1122 struct rate_sample *rs);
1123 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1124 bool is_sack_reneg, struct rate_sample *rs);
1125 void tcp_rate_check_app_limited(struct sock *sk);
1126
1127 /* These functions determine how the current flow behaves in respect of SACK
1128 * handling. SACK is negotiated with the peer, and therefore it can vary
1129 * between different flows.
1130 *
1131 * tcp_is_sack - SACK enabled
1132 * tcp_is_reno - No SACK
1133 */
tcp_is_sack(const struct tcp_sock * tp)1134 static inline int tcp_is_sack(const struct tcp_sock *tp)
1135 {
1136 return tp->rx_opt.sack_ok;
1137 }
1138
tcp_is_reno(const struct tcp_sock * tp)1139 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1140 {
1141 return !tcp_is_sack(tp);
1142 }
1143
tcp_left_out(const struct tcp_sock * tp)1144 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1145 {
1146 return tp->sacked_out + tp->lost_out;
1147 }
1148
1149 /* This determines how many packets are "in the network" to the best
1150 * of our knowledge. In many cases it is conservative, but where
1151 * detailed information is available from the receiver (via SACK
1152 * blocks etc.) we can make more aggressive calculations.
1153 *
1154 * Use this for decisions involving congestion control, use just
1155 * tp->packets_out to determine if the send queue is empty or not.
1156 *
1157 * Read this equation as:
1158 *
1159 * "Packets sent once on transmission queue" MINUS
1160 * "Packets left network, but not honestly ACKed yet" PLUS
1161 * "Packets fast retransmitted"
1162 */
tcp_packets_in_flight(const struct tcp_sock * tp)1163 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1164 {
1165 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1166 }
1167
1168 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1169
tcp_in_slow_start(const struct tcp_sock * tp)1170 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1171 {
1172 return tp->snd_cwnd < tp->snd_ssthresh;
1173 }
1174
tcp_in_initial_slowstart(const struct tcp_sock * tp)1175 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1176 {
1177 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1178 }
1179
tcp_in_cwnd_reduction(const struct sock * sk)1180 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1181 {
1182 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1183 (1 << inet_csk(sk)->icsk_ca_state);
1184 }
1185
1186 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1187 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1188 * ssthresh.
1189 */
tcp_current_ssthresh(const struct sock * sk)1190 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1191 {
1192 const struct tcp_sock *tp = tcp_sk(sk);
1193
1194 if (tcp_in_cwnd_reduction(sk))
1195 return tp->snd_ssthresh;
1196 else
1197 return max(tp->snd_ssthresh,
1198 ((tp->snd_cwnd >> 1) +
1199 (tp->snd_cwnd >> 2)));
1200 }
1201
1202 /* Use define here intentionally to get WARN_ON location shown at the caller */
1203 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1204
1205 void tcp_enter_cwr(struct sock *sk);
1206 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1207
1208 /* The maximum number of MSS of available cwnd for which TSO defers
1209 * sending if not using sysctl_tcp_tso_win_divisor.
1210 */
tcp_max_tso_deferred_mss(const struct tcp_sock * tp)1211 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1212 {
1213 return 3;
1214 }
1215
1216 /* Returns end sequence number of the receiver's advertised window */
tcp_wnd_end(const struct tcp_sock * tp)1217 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1218 {
1219 return tp->snd_una + tp->snd_wnd;
1220 }
1221
1222 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1223 * flexible approach. The RFC suggests cwnd should not be raised unless
1224 * it was fully used previously. And that's exactly what we do in
1225 * congestion avoidance mode. But in slow start we allow cwnd to grow
1226 * as long as the application has used half the cwnd.
1227 * Example :
1228 * cwnd is 10 (IW10), but application sends 9 frames.
1229 * We allow cwnd to reach 18 when all frames are ACKed.
1230 * This check is safe because it's as aggressive as slow start which already
1231 * risks 100% overshoot. The advantage is that we discourage application to
1232 * either send more filler packets or data to artificially blow up the cwnd
1233 * usage, and allow application-limited process to probe bw more aggressively.
1234 */
tcp_is_cwnd_limited(const struct sock * sk)1235 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1236 {
1237 const struct tcp_sock *tp = tcp_sk(sk);
1238
1239 if (tp->is_cwnd_limited)
1240 return true;
1241
1242 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1243 if (tcp_in_slow_start(tp))
1244 return tp->snd_cwnd < 2 * tp->max_packets_out;
1245
1246 return false;
1247 }
1248
1249 /* BBR congestion control needs pacing.
1250 * Same remark for SO_MAX_PACING_RATE.
1251 * sch_fq packet scheduler is efficiently handling pacing,
1252 * but is not always installed/used.
1253 * Return true if TCP stack should pace packets itself.
1254 */
tcp_needs_internal_pacing(const struct sock * sk)1255 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1256 {
1257 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1258 }
1259
1260 /* Something is really bad, we could not queue an additional packet,
1261 * because qdisc is full or receiver sent a 0 window.
1262 * We do not want to add fuel to the fire, or abort too early,
1263 * so make sure the timer we arm now is at least 200ms in the future,
1264 * regardless of current icsk_rto value (as it could be ~2ms)
1265 */
tcp_probe0_base(const struct sock * sk)1266 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1267 {
1268 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1269 }
1270
1271 /* Variant of inet_csk_rto_backoff() used for zero window probes */
tcp_probe0_when(const struct sock * sk,unsigned long max_when)1272 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1273 unsigned long max_when)
1274 {
1275 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1276
1277 return (unsigned long)min_t(u64, when, max_when);
1278 }
1279
tcp_check_probe_timer(struct sock * sk)1280 static inline void tcp_check_probe_timer(struct sock *sk)
1281 {
1282 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1283 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1284 tcp_probe0_base(sk), TCP_RTO_MAX);
1285 }
1286
tcp_init_wl(struct tcp_sock * tp,u32 seq)1287 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1288 {
1289 tp->snd_wl1 = seq;
1290 }
1291
tcp_update_wl(struct tcp_sock * tp,u32 seq)1292 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1293 {
1294 tp->snd_wl1 = seq;
1295 }
1296
1297 /*
1298 * Calculate(/check) TCP checksum
1299 */
tcp_v4_check(int len,__be32 saddr,__be32 daddr,__wsum base)1300 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1301 __be32 daddr, __wsum base)
1302 {
1303 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1304 }
1305
__tcp_checksum_complete(struct sk_buff * skb)1306 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1307 {
1308 return __skb_checksum_complete(skb);
1309 }
1310
tcp_checksum_complete(struct sk_buff * skb)1311 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1312 {
1313 return !skb_csum_unnecessary(skb) &&
1314 __tcp_checksum_complete(skb);
1315 }
1316
1317 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1318 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1319
1320 #undef STATE_TRACE
1321
1322 #ifdef STATE_TRACE
1323 static const char *statename[]={
1324 "Unused","Established","Syn Sent","Syn Recv",
1325 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1326 "Close Wait","Last ACK","Listen","Closing"
1327 };
1328 #endif
1329 void tcp_set_state(struct sock *sk, int state);
1330
1331 void tcp_done(struct sock *sk);
1332
1333 int tcp_abort(struct sock *sk, int err);
1334
tcp_sack_reset(struct tcp_options_received * rx_opt)1335 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1336 {
1337 rx_opt->dsack = 0;
1338 rx_opt->num_sacks = 0;
1339 }
1340
1341 u32 tcp_default_init_rwnd(u32 mss);
1342 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1343
tcp_slow_start_after_idle_check(struct sock * sk)1344 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1345 {
1346 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1347 struct tcp_sock *tp = tcp_sk(sk);
1348 s32 delta;
1349
1350 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) ||
1351 tp->packets_out || ca_ops->cong_control)
1352 return;
1353 delta = tcp_jiffies32 - tp->lsndtime;
1354 if (delta > inet_csk(sk)->icsk_rto)
1355 tcp_cwnd_restart(sk, delta);
1356 }
1357
1358 /* Determine a window scaling and initial window to offer. */
1359 void tcp_select_initial_window(const struct sock *sk, int __space,
1360 __u32 mss, __u32 *rcv_wnd,
1361 __u32 *window_clamp, int wscale_ok,
1362 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1363
tcp_win_from_space(const struct sock * sk,int space)1364 static inline int tcp_win_from_space(const struct sock *sk, int space)
1365 {
1366 int tcp_adv_win_scale = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale);
1367
1368 return tcp_adv_win_scale <= 0 ?
1369 (space>>(-tcp_adv_win_scale)) :
1370 space - (space>>tcp_adv_win_scale);
1371 }
1372
1373 /* Note: caller must be prepared to deal with negative returns */
tcp_space(const struct sock * sk)1374 static inline int tcp_space(const struct sock *sk)
1375 {
1376 return tcp_win_from_space(sk, sk->sk_rcvbuf - sk->sk_backlog.len -
1377 atomic_read(&sk->sk_rmem_alloc));
1378 }
1379
tcp_full_space(const struct sock * sk)1380 static inline int tcp_full_space(const struct sock *sk)
1381 {
1382 return tcp_win_from_space(sk, sk->sk_rcvbuf);
1383 }
1384
1385 /* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1386 * If 87.5 % (7/8) of the space has been consumed, we want to override
1387 * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1388 * len/truesize ratio.
1389 */
tcp_rmem_pressure(const struct sock * sk)1390 static inline bool tcp_rmem_pressure(const struct sock *sk)
1391 {
1392 int rcvbuf, threshold;
1393
1394 if (tcp_under_memory_pressure(sk))
1395 return true;
1396
1397 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1398 threshold = rcvbuf - (rcvbuf >> 3);
1399
1400 return atomic_read(&sk->sk_rmem_alloc) > threshold;
1401 }
1402
1403 extern void tcp_openreq_init_rwin(struct request_sock *req,
1404 const struct sock *sk_listener,
1405 const struct dst_entry *dst);
1406
1407 void tcp_enter_memory_pressure(struct sock *sk);
1408 void tcp_leave_memory_pressure(struct sock *sk);
1409
keepalive_intvl_when(const struct tcp_sock * tp)1410 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1411 {
1412 struct net *net = sock_net((struct sock *)tp);
1413
1414 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1415 }
1416
keepalive_time_when(const struct tcp_sock * tp)1417 static inline int keepalive_time_when(const struct tcp_sock *tp)
1418 {
1419 struct net *net = sock_net((struct sock *)tp);
1420
1421 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1422 }
1423
keepalive_probes(const struct tcp_sock * tp)1424 static inline int keepalive_probes(const struct tcp_sock *tp)
1425 {
1426 struct net *net = sock_net((struct sock *)tp);
1427
1428 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1429 }
1430
keepalive_time_elapsed(const struct tcp_sock * tp)1431 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1432 {
1433 const struct inet_connection_sock *icsk = &tp->inet_conn;
1434
1435 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1436 tcp_jiffies32 - tp->rcv_tstamp);
1437 }
1438
tcp_fin_time(const struct sock * sk)1439 static inline int tcp_fin_time(const struct sock *sk)
1440 {
1441 int fin_timeout = tcp_sk(sk)->linger2 ? :
1442 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout);
1443 const int rto = inet_csk(sk)->icsk_rto;
1444
1445 if (fin_timeout < (rto << 2) - (rto >> 1))
1446 fin_timeout = (rto << 2) - (rto >> 1);
1447
1448 return fin_timeout;
1449 }
1450
tcp_paws_check(const struct tcp_options_received * rx_opt,int paws_win)1451 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1452 int paws_win)
1453 {
1454 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1455 return true;
1456 if (unlikely(!time_before32(ktime_get_seconds(),
1457 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1458 return true;
1459 /*
1460 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1461 * then following tcp messages have valid values. Ignore 0 value,
1462 * or else 'negative' tsval might forbid us to accept their packets.
1463 */
1464 if (!rx_opt->ts_recent)
1465 return true;
1466 return false;
1467 }
1468
tcp_paws_reject(const struct tcp_options_received * rx_opt,int rst)1469 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1470 int rst)
1471 {
1472 if (tcp_paws_check(rx_opt, 0))
1473 return false;
1474
1475 /* RST segments are not recommended to carry timestamp,
1476 and, if they do, it is recommended to ignore PAWS because
1477 "their cleanup function should take precedence over timestamps."
1478 Certainly, it is mistake. It is necessary to understand the reasons
1479 of this constraint to relax it: if peer reboots, clock may go
1480 out-of-sync and half-open connections will not be reset.
1481 Actually, the problem would be not existing if all
1482 the implementations followed draft about maintaining clock
1483 via reboots. Linux-2.2 DOES NOT!
1484
1485 However, we can relax time bounds for RST segments to MSL.
1486 */
1487 if (rst && !time_before32(ktime_get_seconds(),
1488 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1489 return false;
1490 return true;
1491 }
1492
1493 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1494 int mib_idx, u32 *last_oow_ack_time);
1495
tcp_mib_init(struct net * net)1496 static inline void tcp_mib_init(struct net *net)
1497 {
1498 /* See RFC 2012 */
1499 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1500 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1501 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1502 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1503 }
1504
1505 /* from STCP */
tcp_clear_retrans_hints_partial(struct tcp_sock * tp)1506 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1507 {
1508 tp->lost_skb_hint = NULL;
1509 }
1510
tcp_clear_all_retrans_hints(struct tcp_sock * tp)1511 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1512 {
1513 tcp_clear_retrans_hints_partial(tp);
1514 tp->retransmit_skb_hint = NULL;
1515 }
1516
1517 union tcp_md5_addr {
1518 struct in_addr a4;
1519 #if IS_ENABLED(CONFIG_IPV6)
1520 struct in6_addr a6;
1521 #endif
1522 };
1523
1524 /* - key database */
1525 struct tcp_md5sig_key {
1526 struct hlist_node node;
1527 u8 keylen;
1528 u8 family; /* AF_INET or AF_INET6 */
1529 union tcp_md5_addr addr;
1530 u8 prefixlen;
1531 u8 key[TCP_MD5SIG_MAXKEYLEN];
1532 struct rcu_head rcu;
1533 };
1534
1535 /* - sock block */
1536 struct tcp_md5sig_info {
1537 struct hlist_head head;
1538 struct rcu_head rcu;
1539 };
1540
1541 /* - pseudo header */
1542 struct tcp4_pseudohdr {
1543 __be32 saddr;
1544 __be32 daddr;
1545 __u8 pad;
1546 __u8 protocol;
1547 __be16 len;
1548 };
1549
1550 struct tcp6_pseudohdr {
1551 struct in6_addr saddr;
1552 struct in6_addr daddr;
1553 __be32 len;
1554 __be32 protocol; /* including padding */
1555 };
1556
1557 union tcp_md5sum_block {
1558 struct tcp4_pseudohdr ip4;
1559 #if IS_ENABLED(CONFIG_IPV6)
1560 struct tcp6_pseudohdr ip6;
1561 #endif
1562 };
1563
1564 /* - pool: digest algorithm, hash description and scratch buffer */
1565 struct tcp_md5sig_pool {
1566 struct ahash_request *md5_req;
1567 void *scratch;
1568 };
1569
1570 /* - functions */
1571 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1572 const struct sock *sk, const struct sk_buff *skb);
1573 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1574 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1575 gfp_t gfp);
1576 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1577 int family, u8 prefixlen);
1578 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1579 const struct sock *addr_sk);
1580
1581 #ifdef CONFIG_TCP_MD5SIG
1582 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1583 const union tcp_md5_addr *addr,
1584 int family);
1585 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1586 #else
tcp_md5_do_lookup(const struct sock * sk,const union tcp_md5_addr * addr,int family)1587 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1588 const union tcp_md5_addr *addr,
1589 int family)
1590 {
1591 return NULL;
1592 }
1593 #define tcp_twsk_md5_key(twsk) NULL
1594 #endif
1595
1596 bool tcp_alloc_md5sig_pool(void);
1597
1598 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
tcp_put_md5sig_pool(void)1599 static inline void tcp_put_md5sig_pool(void)
1600 {
1601 local_bh_enable();
1602 }
1603
1604 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1605 unsigned int header_len);
1606 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1607 const struct tcp_md5sig_key *key);
1608
1609 /* From tcp_fastopen.c */
1610 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1611 struct tcp_fastopen_cookie *cookie);
1612 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1613 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1614 u16 try_exp);
1615 struct tcp_fastopen_request {
1616 /* Fast Open cookie. Size 0 means a cookie request */
1617 struct tcp_fastopen_cookie cookie;
1618 struct msghdr *data; /* data in MSG_FASTOPEN */
1619 size_t size;
1620 int copied; /* queued in tcp_connect() */
1621 };
1622 void tcp_free_fastopen_req(struct tcp_sock *tp);
1623 void tcp_fastopen_destroy_cipher(struct sock *sk);
1624 void tcp_fastopen_ctx_destroy(struct net *net);
1625 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1626 void *key, unsigned int len);
1627 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1628 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1629 struct request_sock *req,
1630 struct tcp_fastopen_cookie *foc,
1631 const struct dst_entry *dst);
1632 void tcp_fastopen_init_key_once(struct net *net);
1633 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1634 struct tcp_fastopen_cookie *cookie);
1635 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1636 #define TCP_FASTOPEN_KEY_LENGTH 16
1637
1638 /* Fastopen key context */
1639 struct tcp_fastopen_context {
1640 struct crypto_cipher *tfm;
1641 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1642 struct rcu_head rcu;
1643 };
1644
1645 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1646 void tcp_fastopen_active_disable(struct sock *sk);
1647 bool tcp_fastopen_active_should_disable(struct sock *sk);
1648 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1649 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1650
1651 /* Latencies incurred by various limits for a sender. They are
1652 * chronograph-like stats that are mutually exclusive.
1653 */
1654 enum tcp_chrono {
1655 TCP_CHRONO_UNSPEC,
1656 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1657 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1658 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1659 __TCP_CHRONO_MAX,
1660 };
1661
1662 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1663 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1664
1665 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1666 * the same memory storage than skb->destructor/_skb_refdst
1667 */
tcp_skb_tsorted_anchor_cleanup(struct sk_buff * skb)1668 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1669 {
1670 skb->destructor = NULL;
1671 skb->_skb_refdst = 0UL;
1672 }
1673
1674 #define tcp_skb_tsorted_save(skb) { \
1675 unsigned long _save = skb->_skb_refdst; \
1676 skb->_skb_refdst = 0UL;
1677
1678 #define tcp_skb_tsorted_restore(skb) \
1679 skb->_skb_refdst = _save; \
1680 }
1681
1682 void tcp_write_queue_purge(struct sock *sk);
1683
tcp_rtx_queue_head(const struct sock * sk)1684 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1685 {
1686 return skb_rb_first(&sk->tcp_rtx_queue);
1687 }
1688
tcp_rtx_queue_tail(const struct sock * sk)1689 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1690 {
1691 return skb_rb_last(&sk->tcp_rtx_queue);
1692 }
1693
tcp_write_queue_head(const struct sock * sk)1694 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1695 {
1696 return skb_peek(&sk->sk_write_queue);
1697 }
1698
tcp_write_queue_tail(const struct sock * sk)1699 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1700 {
1701 return skb_peek_tail(&sk->sk_write_queue);
1702 }
1703
1704 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1705 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1706
tcp_send_head(const struct sock * sk)1707 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1708 {
1709 return skb_peek(&sk->sk_write_queue);
1710 }
1711
tcp_skb_is_last(const struct sock * sk,const struct sk_buff * skb)1712 static inline bool tcp_skb_is_last(const struct sock *sk,
1713 const struct sk_buff *skb)
1714 {
1715 return skb_queue_is_last(&sk->sk_write_queue, skb);
1716 }
1717
tcp_write_queue_empty(const struct sock * sk)1718 static inline bool tcp_write_queue_empty(const struct sock *sk)
1719 {
1720 return skb_queue_empty(&sk->sk_write_queue);
1721 }
1722
tcp_rtx_queue_empty(const struct sock * sk)1723 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1724 {
1725 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1726 }
1727
tcp_rtx_and_write_queues_empty(const struct sock * sk)1728 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1729 {
1730 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1731 }
1732
tcp_check_send_head(struct sock * sk,struct sk_buff * skb_unlinked)1733 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1734 {
1735 if (tcp_write_queue_empty(sk))
1736 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1737 }
1738
__tcp_add_write_queue_tail(struct sock * sk,struct sk_buff * skb)1739 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1740 {
1741 __skb_queue_tail(&sk->sk_write_queue, skb);
1742 }
1743
tcp_add_write_queue_tail(struct sock * sk,struct sk_buff * skb)1744 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1745 {
1746 __tcp_add_write_queue_tail(sk, skb);
1747
1748 /* Queue it, remembering where we must start sending. */
1749 if (sk->sk_write_queue.next == skb)
1750 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1751 }
1752
1753 /* 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)1754 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1755 struct sk_buff *skb,
1756 struct sock *sk)
1757 {
1758 __skb_queue_before(&sk->sk_write_queue, skb, new);
1759 }
1760
tcp_unlink_write_queue(struct sk_buff * skb,struct sock * sk)1761 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1762 {
1763 tcp_skb_tsorted_anchor_cleanup(skb);
1764 __skb_unlink(skb, &sk->sk_write_queue);
1765 }
1766
1767 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1768
tcp_rtx_queue_unlink(struct sk_buff * skb,struct sock * sk)1769 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1770 {
1771 tcp_skb_tsorted_anchor_cleanup(skb);
1772 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1773 }
1774
tcp_rtx_queue_unlink_and_free(struct sk_buff * skb,struct sock * sk)1775 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1776 {
1777 list_del(&skb->tcp_tsorted_anchor);
1778 tcp_rtx_queue_unlink(skb, sk);
1779 sk_wmem_free_skb(sk, skb);
1780 }
1781
tcp_push_pending_frames(struct sock * sk)1782 static inline void tcp_push_pending_frames(struct sock *sk)
1783 {
1784 if (tcp_send_head(sk)) {
1785 struct tcp_sock *tp = tcp_sk(sk);
1786
1787 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1788 }
1789 }
1790
1791 /* Start sequence of the skb just after the highest skb with SACKed
1792 * bit, valid only if sacked_out > 0 or when the caller has ensured
1793 * validity by itself.
1794 */
tcp_highest_sack_seq(struct tcp_sock * tp)1795 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1796 {
1797 if (!tp->sacked_out)
1798 return tp->snd_una;
1799
1800 if (tp->highest_sack == NULL)
1801 return tp->snd_nxt;
1802
1803 return TCP_SKB_CB(tp->highest_sack)->seq;
1804 }
1805
tcp_advance_highest_sack(struct sock * sk,struct sk_buff * skb)1806 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1807 {
1808 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1809 }
1810
tcp_highest_sack(struct sock * sk)1811 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1812 {
1813 return tcp_sk(sk)->highest_sack;
1814 }
1815
tcp_highest_sack_reset(struct sock * sk)1816 static inline void tcp_highest_sack_reset(struct sock *sk)
1817 {
1818 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1819 }
1820
1821 /* 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)1822 static inline void tcp_highest_sack_replace(struct sock *sk,
1823 struct sk_buff *old,
1824 struct sk_buff *new)
1825 {
1826 if (old == tcp_highest_sack(sk))
1827 tcp_sk(sk)->highest_sack = new;
1828 }
1829
1830 /* This helper checks if socket has IP_TRANSPARENT set */
inet_sk_transparent(const struct sock * sk)1831 static inline bool inet_sk_transparent(const struct sock *sk)
1832 {
1833 switch (sk->sk_state) {
1834 case TCP_TIME_WAIT:
1835 return inet_twsk(sk)->tw_transparent;
1836 case TCP_NEW_SYN_RECV:
1837 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1838 }
1839 return inet_sk(sk)->transparent;
1840 }
1841
1842 /* Determines whether this is a thin stream (which may suffer from
1843 * increased latency). Used to trigger latency-reducing mechanisms.
1844 */
tcp_stream_is_thin(struct tcp_sock * tp)1845 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1846 {
1847 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1848 }
1849
1850 /* /proc */
1851 enum tcp_seq_states {
1852 TCP_SEQ_STATE_LISTENING,
1853 TCP_SEQ_STATE_ESTABLISHED,
1854 };
1855
1856 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1857 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1858 void tcp_seq_stop(struct seq_file *seq, void *v);
1859
1860 struct tcp_seq_afinfo {
1861 sa_family_t family;
1862 };
1863
1864 struct tcp_iter_state {
1865 struct seq_net_private p;
1866 enum tcp_seq_states state;
1867 struct sock *syn_wait_sk;
1868 int bucket, offset, sbucket, num;
1869 loff_t last_pos;
1870 };
1871
1872 extern struct request_sock_ops tcp_request_sock_ops;
1873 extern struct request_sock_ops tcp6_request_sock_ops;
1874
1875 void tcp_v4_destroy_sock(struct sock *sk);
1876
1877 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1878 netdev_features_t features);
1879 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
1880 int tcp_gro_complete(struct sk_buff *skb);
1881
1882 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1883
tcp_notsent_lowat(const struct tcp_sock * tp)1884 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1885 {
1886 struct net *net = sock_net((struct sock *)tp);
1887 u32 val;
1888
1889 val = READ_ONCE(tp->notsent_lowat);
1890
1891 return val ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
1892 }
1893
1894 /* @wake is one when sk_stream_write_space() calls us.
1895 * This sends EPOLLOUT only if notsent_bytes is half the limit.
1896 * This mimics the strategy used in sock_def_write_space().
1897 */
tcp_stream_memory_free(const struct sock * sk,int wake)1898 static inline bool tcp_stream_memory_free(const struct sock *sk, int wake)
1899 {
1900 const struct tcp_sock *tp = tcp_sk(sk);
1901 u32 notsent_bytes = READ_ONCE(tp->write_seq) - tp->snd_nxt;
1902
1903 return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
1904 }
1905
1906 #ifdef CONFIG_PROC_FS
1907 int tcp4_proc_init(void);
1908 void tcp4_proc_exit(void);
1909 #endif
1910
1911 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1912 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1913 const struct tcp_request_sock_ops *af_ops,
1914 struct sock *sk, struct sk_buff *skb);
1915
1916 /* TCP af-specific functions */
1917 struct tcp_sock_af_ops {
1918 #ifdef CONFIG_TCP_MD5SIG
1919 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1920 const struct sock *addr_sk);
1921 int (*calc_md5_hash)(char *location,
1922 const struct tcp_md5sig_key *md5,
1923 const struct sock *sk,
1924 const struct sk_buff *skb);
1925 int (*md5_parse)(struct sock *sk,
1926 int optname,
1927 char __user *optval,
1928 int optlen);
1929 #endif
1930 };
1931
1932 struct tcp_request_sock_ops {
1933 u16 mss_clamp;
1934 #ifdef CONFIG_TCP_MD5SIG
1935 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1936 const struct sock *addr_sk);
1937 int (*calc_md5_hash) (char *location,
1938 const struct tcp_md5sig_key *md5,
1939 const struct sock *sk,
1940 const struct sk_buff *skb);
1941 #endif
1942 void (*init_req)(struct request_sock *req,
1943 const struct sock *sk_listener,
1944 struct sk_buff *skb);
1945 #ifdef CONFIG_SYN_COOKIES
1946 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1947 __u16 *mss);
1948 #endif
1949 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1950 const struct request_sock *req);
1951 u32 (*init_seq)(const struct sk_buff *skb);
1952 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
1953 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1954 struct flowi *fl, struct request_sock *req,
1955 struct tcp_fastopen_cookie *foc,
1956 enum tcp_synack_type synack_type);
1957 };
1958
1959 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
1960 #if IS_ENABLED(CONFIG_IPV6)
1961 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
1962 #endif
1963
1964 #ifdef CONFIG_SYN_COOKIES
cookie_init_sequence(const struct tcp_request_sock_ops * ops,const struct sock * sk,struct sk_buff * skb,__u16 * mss)1965 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1966 const struct sock *sk, struct sk_buff *skb,
1967 __u16 *mss)
1968 {
1969 tcp_synq_overflow(sk);
1970 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1971 return ops->cookie_init_seq(skb, mss);
1972 }
1973 #else
cookie_init_sequence(const struct tcp_request_sock_ops * ops,const struct sock * sk,struct sk_buff * skb,__u16 * mss)1974 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1975 const struct sock *sk, struct sk_buff *skb,
1976 __u16 *mss)
1977 {
1978 return 0;
1979 }
1980 #endif
1981
1982 int tcpv4_offload_init(void);
1983
1984 void tcp_v4_init(void);
1985 void tcp_init(void);
1986
1987 /* tcp_recovery.c */
1988 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
1989 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
1990 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
1991 u32 reo_wnd);
1992 extern bool tcp_rack_mark_lost(struct sock *sk);
1993 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
1994 u64 xmit_time);
1995 extern void tcp_rack_reo_timeout(struct sock *sk);
1996 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
1997
1998 /* At how many usecs into the future should the RTO fire? */
tcp_rto_delta_us(const struct sock * sk)1999 static inline s64 tcp_rto_delta_us(const struct sock *sk)
2000 {
2001 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2002 u32 rto = inet_csk(sk)->icsk_rto;
2003 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto);
2004
2005 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2006 }
2007
2008 /*
2009 * Save and compile IPv4 options, return a pointer to it
2010 */
tcp_v4_save_options(struct net * net,struct sk_buff * skb)2011 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2012 struct sk_buff *skb)
2013 {
2014 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2015 struct ip_options_rcu *dopt = NULL;
2016
2017 if (opt->optlen) {
2018 int opt_size = sizeof(*dopt) + opt->optlen;
2019
2020 dopt = kmalloc(opt_size, GFP_ATOMIC);
2021 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2022 kfree(dopt);
2023 dopt = NULL;
2024 }
2025 }
2026 return dopt;
2027 }
2028
2029 /* locally generated TCP pure ACKs have skb->truesize == 2
2030 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2031 * This is much faster than dissecting the packet to find out.
2032 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2033 */
skb_is_tcp_pure_ack(const struct sk_buff * skb)2034 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2035 {
2036 return skb->truesize == 2;
2037 }
2038
skb_set_tcp_pure_ack(struct sk_buff * skb)2039 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2040 {
2041 skb->truesize = 2;
2042 }
2043
tcp_inq(struct sock * sk)2044 static inline int tcp_inq(struct sock *sk)
2045 {
2046 struct tcp_sock *tp = tcp_sk(sk);
2047 int answ;
2048
2049 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2050 answ = 0;
2051 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2052 !tp->urg_data ||
2053 before(tp->urg_seq, tp->copied_seq) ||
2054 !before(tp->urg_seq, tp->rcv_nxt)) {
2055
2056 answ = tp->rcv_nxt - tp->copied_seq;
2057
2058 /* Subtract 1, if FIN was received */
2059 if (answ && sock_flag(sk, SOCK_DONE))
2060 answ--;
2061 } else {
2062 answ = tp->urg_seq - tp->copied_seq;
2063 }
2064
2065 return answ;
2066 }
2067
2068 int tcp_peek_len(struct socket *sock);
2069
tcp_segs_in(struct tcp_sock * tp,const struct sk_buff * skb)2070 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2071 {
2072 u16 segs_in;
2073
2074 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2075 tp->segs_in += segs_in;
2076 if (skb->len > tcp_hdrlen(skb))
2077 tp->data_segs_in += segs_in;
2078 }
2079
2080 /*
2081 * TCP listen path runs lockless.
2082 * We forced "struct sock" to be const qualified to make sure
2083 * we don't modify one of its field by mistake.
2084 * Here, we increment sk_drops which is an atomic_t, so we can safely
2085 * make sock writable again.
2086 */
tcp_listendrop(const struct sock * sk)2087 static inline void tcp_listendrop(const struct sock *sk)
2088 {
2089 atomic_inc(&((struct sock *)sk)->sk_drops);
2090 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2091 }
2092
2093 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2094
2095 /*
2096 * Interface for adding Upper Level Protocols over TCP
2097 */
2098
2099 #define TCP_ULP_NAME_MAX 16
2100 #define TCP_ULP_MAX 128
2101 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2102
2103 enum {
2104 TCP_ULP_TLS,
2105 TCP_ULP_BPF,
2106 };
2107
2108 struct tcp_ulp_ops {
2109 struct list_head list;
2110
2111 /* initialize ulp */
2112 int (*init)(struct sock *sk);
2113 /* cleanup ulp */
2114 void (*release)(struct sock *sk);
2115
2116 int uid;
2117 char name[TCP_ULP_NAME_MAX];
2118 bool user_visible;
2119 struct module *owner;
2120 };
2121 int tcp_register_ulp(struct tcp_ulp_ops *type);
2122 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2123 int tcp_set_ulp(struct sock *sk, const char *name);
2124 int tcp_set_ulp_id(struct sock *sk, const int ulp);
2125 void tcp_get_available_ulp(char *buf, size_t len);
2126 void tcp_cleanup_ulp(struct sock *sk);
2127
2128 #define MODULE_ALIAS_TCP_ULP(name) \
2129 __MODULE_INFO(alias, alias_userspace, name); \
2130 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2131
2132 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2133 * is < 0, then the BPF op failed (for example if the loaded BPF
2134 * program does not support the chosen operation or there is no BPF
2135 * program loaded).
2136 */
2137 #ifdef CONFIG_BPF
tcp_call_bpf(struct sock * sk,int op,u32 nargs,u32 * args)2138 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2139 {
2140 struct bpf_sock_ops_kern sock_ops;
2141 int ret;
2142
2143 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2144 if (sk_fullsock(sk)) {
2145 sock_ops.is_fullsock = 1;
2146 sock_owned_by_me(sk);
2147 }
2148
2149 sock_ops.sk = sk;
2150 sock_ops.op = op;
2151 if (nargs > 0)
2152 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2153
2154 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2155 if (ret == 0)
2156 ret = sock_ops.reply;
2157 else
2158 ret = -1;
2159 return ret;
2160 }
2161
tcp_call_bpf_2arg(struct sock * sk,int op,u32 arg1,u32 arg2)2162 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2163 {
2164 u32 args[2] = {arg1, arg2};
2165
2166 return tcp_call_bpf(sk, op, 2, args);
2167 }
2168
tcp_call_bpf_3arg(struct sock * sk,int op,u32 arg1,u32 arg2,u32 arg3)2169 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2170 u32 arg3)
2171 {
2172 u32 args[3] = {arg1, arg2, arg3};
2173
2174 return tcp_call_bpf(sk, op, 3, args);
2175 }
2176
2177 #else
tcp_call_bpf(struct sock * sk,int op,u32 nargs,u32 * args)2178 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2179 {
2180 return -EPERM;
2181 }
2182
tcp_call_bpf_2arg(struct sock * sk,int op,u32 arg1,u32 arg2)2183 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2184 {
2185 return -EPERM;
2186 }
2187
tcp_call_bpf_3arg(struct sock * sk,int op,u32 arg1,u32 arg2,u32 arg3)2188 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2189 u32 arg3)
2190 {
2191 return -EPERM;
2192 }
2193
2194 #endif
2195
tcp_timeout_init(struct sock * sk)2196 static inline u32 tcp_timeout_init(struct sock *sk)
2197 {
2198 int timeout;
2199
2200 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2201
2202 if (timeout <= 0)
2203 timeout = TCP_TIMEOUT_INIT;
2204 return timeout;
2205 }
2206
tcp_rwnd_init_bpf(struct sock * sk)2207 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2208 {
2209 int rwnd;
2210
2211 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2212
2213 if (rwnd < 0)
2214 rwnd = 0;
2215 return rwnd;
2216 }
2217
tcp_bpf_ca_needs_ecn(struct sock * sk)2218 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2219 {
2220 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2221 }
2222
2223 #if IS_ENABLED(CONFIG_SMC)
2224 extern struct static_key_false tcp_have_smc;
2225 #endif
2226
2227 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2228 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2229 void (*cad)(struct sock *sk, u32 ack_seq));
2230 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2231
2232 #endif
2233
2234 #endif /* _TCP_H */
2235