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.
6 * Definitions for the TCP module.
8 * Version: @(#)tcp.h 1.0.5 05/23/93
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
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.
21 #define FASTRETRANS_DEBUG 1
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>
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>
42 #include <net/tcp_states.h>
43 #include <net/inet_ecn.h>
46 #include <linux/seq_file.h>
47 #include <linux/memcontrol.h>
49 #include <linux/bpf.h>
50 #include <linux/filter.h>
51 #include <linux/bpf-cgroup.h>
53 extern struct inet_hashinfo tcp_hashinfo;
55 extern struct percpu_counter tcp_orphan_count;
56 void tcp_time_wait(struct sock *sk, int state, int timeo);
58 #define MAX_TCP_HEADER (128 + MAX_HEADER)
59 #define MAX_TCP_OPTION_SPACE 40
62 * Never offer a window over 32767 without using window scaling. Some
63 * poor stacks do signed 16bit maths!
65 #define MAX_TCP_WINDOW 32767U
67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68 #define TCP_MIN_MSS 88U
70 /* The least MTU to use for probing */
71 #define TCP_BASE_MSS 1024
73 /* probing interval, default to 10 minutes as per RFC4821 */
74 #define TCP_PROBE_INTERVAL 600
76 /* Specify interval when tcp mtu probing will stop */
77 #define TCP_PROBE_THRESHOLD 8
79 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
80 #define TCP_FASTRETRANS_THRESH 3
82 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
83 #define TCP_MAX_QUICKACKS 16U
85 /* Maximal number of window scale according to RFC1323 */
86 #define TCP_MAX_WSCALE 14U
89 #define TCP_URG_VALID 0x0100
90 #define TCP_URG_NOTYET 0x0200
91 #define TCP_URG_READ 0x0400
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.
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.
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.
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
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
132 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
134 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
135 #define TCP_ATO_MIN ((unsigned)(HZ/25))
137 #define TCP_DELACK_MIN 4U
138 #define TCP_ATO_MIN 4U
140 #define TCP_RTO_MAX ((unsigned)(120*HZ))
141 #define TCP_RTO_MIN ((unsigned)(HZ/5))
142 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
143 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
144 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
145 * used as a fallback RTO for the
146 * initial data transmission if no
147 * valid RTT sample has been acquired,
148 * most likely due to retrans in 3WHS.
151 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
152 * for local resources.
154 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
155 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
156 #define TCP_KEEPALIVE_INTVL (75*HZ)
158 #define MAX_TCP_KEEPIDLE 32767
159 #define MAX_TCP_KEEPINTVL 32767
160 #define MAX_TCP_KEEPCNT 127
161 #define MAX_TCP_SYNCNT 127
163 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
165 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
166 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
167 * after this time. It should be equal
168 * (or greater than) TCP_TIMEWAIT_LEN
169 * to provide reliability equal to one
170 * provided by timewait state.
172 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
173 * timestamps. It must be less than
174 * minimal timewait lifetime.
180 #define TCPOPT_NOP 1 /* Padding */
181 #define TCPOPT_EOL 0 /* End of options */
182 #define TCPOPT_MSS 2 /* Segment size negotiating */
183 #define TCPOPT_WINDOW 3 /* Window scaling */
184 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
185 #define TCPOPT_SACK 5 /* SACK Block */
186 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
187 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
188 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
189 #define TCPOPT_EXP 254 /* Experimental */
190 /* Magic number to be after the option value for sharing TCP
191 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
193 #define TCPOPT_FASTOPEN_MAGIC 0xF989
199 #define TCPOLEN_MSS 4
200 #define TCPOLEN_WINDOW 3
201 #define TCPOLEN_SACK_PERM 2
202 #define TCPOLEN_TIMESTAMP 10
203 #define TCPOLEN_MD5SIG 18
204 #define TCPOLEN_FASTOPEN_BASE 2
205 #define TCPOLEN_EXP_FASTOPEN_BASE 4
207 /* But this is what stacks really send out. */
208 #define TCPOLEN_TSTAMP_ALIGNED 12
209 #define TCPOLEN_WSCALE_ALIGNED 4
210 #define TCPOLEN_SACKPERM_ALIGNED 4
211 #define TCPOLEN_SACK_BASE 2
212 #define TCPOLEN_SACK_BASE_ALIGNED 4
213 #define TCPOLEN_SACK_PERBLOCK 8
214 #define TCPOLEN_MD5SIG_ALIGNED 20
215 #define TCPOLEN_MSS_ALIGNED 4
217 /* Flags in tp->nonagle */
218 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
219 #define TCP_NAGLE_CORK 2 /* Socket is corked */
220 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
222 /* TCP thin-stream limits */
223 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
225 /* TCP initial congestion window as per rfc6928 */
226 #define TCP_INIT_CWND 10
228 /* Bit Flags for sysctl_tcp_fastopen */
229 #define TFO_CLIENT_ENABLE 1
230 #define TFO_SERVER_ENABLE 2
231 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
233 /* Accept SYN data w/o any cookie option */
234 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
236 /* Force enable TFO on all listeners, i.e., not requiring the
237 * TCP_FASTOPEN socket option.
239 #define TFO_SERVER_WO_SOCKOPT1 0x400
242 /* sysctl variables for tcp */
243 extern int sysctl_tcp_fastopen;
244 extern int sysctl_tcp_retrans_collapse;
245 extern int sysctl_tcp_stdurg;
246 extern int sysctl_tcp_rfc1337;
247 extern int sysctl_tcp_abort_on_overflow;
248 extern int sysctl_tcp_max_orphans;
249 extern int sysctl_tcp_fack;
250 extern int sysctl_tcp_reordering;
251 extern int sysctl_tcp_max_reordering;
252 extern int sysctl_tcp_dsack;
253 extern long sysctl_tcp_mem[3];
254 extern int sysctl_tcp_wmem[3];
255 extern int sysctl_tcp_rmem[3];
256 extern int sysctl_tcp_app_win;
257 extern int sysctl_tcp_adv_win_scale;
258 extern int sysctl_tcp_frto;
259 extern int sysctl_tcp_nometrics_save;
260 extern int sysctl_tcp_moderate_rcvbuf;
261 extern int sysctl_tcp_tso_win_divisor;
262 extern int sysctl_tcp_workaround_signed_windows;
263 extern int sysctl_tcp_slow_start_after_idle;
264 extern int sysctl_tcp_thin_linear_timeouts;
265 extern int sysctl_tcp_thin_dupack;
266 extern int sysctl_tcp_early_retrans;
267 extern int sysctl_tcp_recovery;
268 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
270 extern int sysctl_tcp_limit_output_bytes;
271 extern int sysctl_tcp_challenge_ack_limit;
272 extern int sysctl_tcp_min_tso_segs;
273 extern int sysctl_tcp_min_rtt_wlen;
274 extern int sysctl_tcp_autocorking;
275 extern int sysctl_tcp_invalid_ratelimit;
276 extern int sysctl_tcp_pacing_ss_ratio;
277 extern int sysctl_tcp_pacing_ca_ratio;
279 extern atomic_long_t tcp_memory_allocated;
280 extern struct percpu_counter tcp_sockets_allocated;
281 extern unsigned long tcp_memory_pressure;
283 /* optimized version of sk_under_memory_pressure() for TCP sockets */
284 static inline bool tcp_under_memory_pressure(const struct sock *sk)
286 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
287 mem_cgroup_under_socket_pressure(sk->sk_memcg))
290 return tcp_memory_pressure;
293 * The next routines deal with comparing 32 bit unsigned ints
294 * and worry about wraparound (automatic with unsigned arithmetic).
297 static inline bool before(__u32 seq1, __u32 seq2)
299 return (__s32)(seq1-seq2) < 0;
301 #define after(seq2, seq1) before(seq1, seq2)
303 /* is s2<=s1<=s3 ? */
304 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
306 return seq3 - seq2 >= seq1 - seq2;
309 static inline bool tcp_out_of_memory(struct sock *sk)
311 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
312 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
317 void sk_forced_mem_schedule(struct sock *sk, int size);
319 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
321 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
322 int orphans = percpu_counter_read_positive(ocp);
324 if (orphans << shift > sysctl_tcp_max_orphans) {
325 orphans = percpu_counter_sum_positive(ocp);
326 if (orphans << shift > sysctl_tcp_max_orphans)
332 bool tcp_check_oom(struct sock *sk, int shift);
335 extern struct proto tcp_prot;
337 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
338 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
339 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
340 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
342 void tcp_tasklet_init(void);
344 void tcp_v4_err(struct sk_buff *skb, u32);
346 void tcp_shutdown(struct sock *sk, int how);
348 int tcp_v4_early_demux(struct sk_buff *skb);
349 int tcp_v4_rcv(struct sk_buff *skb);
351 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
352 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
353 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
354 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
356 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
357 size_t size, int flags);
358 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
359 size_t size, int flags);
360 void tcp_release_cb(struct sock *sk);
361 void tcp_wfree(struct sk_buff *skb);
362 void tcp_write_timer_handler(struct sock *sk);
363 void tcp_delack_timer_handler(struct sock *sk);
364 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
365 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
366 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb,
367 const struct tcphdr *th);
368 void tcp_rcv_space_adjust(struct sock *sk);
369 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
370 void tcp_twsk_destructor(struct sock *sk);
371 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
372 struct pipe_inode_info *pipe, size_t len,
375 static inline void tcp_dec_quickack_mode(struct sock *sk,
376 const unsigned int pkts)
378 struct inet_connection_sock *icsk = inet_csk(sk);
380 if (icsk->icsk_ack.quick) {
381 if (pkts >= icsk->icsk_ack.quick) {
382 icsk->icsk_ack.quick = 0;
383 /* Leaving quickack mode we deflate ATO. */
384 icsk->icsk_ack.ato = TCP_ATO_MIN;
386 icsk->icsk_ack.quick -= pkts;
391 #define TCP_ECN_QUEUE_CWR 2
392 #define TCP_ECN_DEMAND_CWR 4
393 #define TCP_ECN_SEEN 8
403 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
405 const struct tcphdr *th);
406 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
407 struct request_sock *req, bool fastopen);
408 int tcp_child_process(struct sock *parent, struct sock *child,
409 struct sk_buff *skb);
410 void tcp_enter_loss(struct sock *sk);
411 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
412 void tcp_clear_retrans(struct tcp_sock *tp);
413 void tcp_update_metrics(struct sock *sk);
414 void tcp_init_metrics(struct sock *sk);
415 void tcp_metrics_init(void);
416 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
417 void tcp_disable_fack(struct tcp_sock *tp);
418 void tcp_close(struct sock *sk, long timeout);
419 void tcp_init_sock(struct sock *sk);
420 unsigned int tcp_poll(struct file *file, struct socket *sock,
421 struct poll_table_struct *wait);
422 int tcp_getsockopt(struct sock *sk, int level, int optname,
423 char __user *optval, int __user *optlen);
424 int tcp_setsockopt(struct sock *sk, int level, int optname,
425 char __user *optval, unsigned int optlen);
426 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
427 char __user *optval, int __user *optlen);
428 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
429 char __user *optval, unsigned int optlen);
430 void tcp_set_keepalive(struct sock *sk, int val);
431 void tcp_syn_ack_timeout(const struct request_sock *req);
432 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
433 int flags, int *addr_len);
434 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
435 struct tcp_options_received *opt_rx,
436 int estab, struct tcp_fastopen_cookie *foc);
437 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
440 * TCP v4 functions exported for the inet6 API
443 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
444 void tcp_v4_mtu_reduced(struct sock *sk);
445 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
446 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
447 struct sock *tcp_create_openreq_child(const struct sock *sk,
448 struct request_sock *req,
449 struct sk_buff *skb);
450 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
451 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
452 struct request_sock *req,
453 struct dst_entry *dst,
454 struct request_sock *req_unhash,
456 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
457 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
458 int tcp_connect(struct sock *sk);
459 enum tcp_synack_type {
464 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
465 struct request_sock *req,
466 struct tcp_fastopen_cookie *foc,
467 enum tcp_synack_type synack_type);
468 int tcp_disconnect(struct sock *sk, int flags);
470 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
471 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
472 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
474 /* From syncookies.c */
475 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
476 struct request_sock *req,
477 struct dst_entry *dst, u32 tsoff);
478 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
480 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
481 #ifdef CONFIG_SYN_COOKIES
483 /* Syncookies use a monotonic timer which increments every 60 seconds.
484 * This counter is used both as a hash input and partially encoded into
485 * the cookie value. A cookie is only validated further if the delta
486 * between the current counter value and the encoded one is less than this,
487 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
488 * the counter advances immediately after a cookie is generated).
490 #define MAX_SYNCOOKIE_AGE 2
491 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
492 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
494 /* syncookies: remember time of last synqueue overflow
495 * But do not dirty this field too often (once per second is enough)
496 * It is racy as we do not hold a lock, but race is very minor.
498 static inline void tcp_synq_overflow(const struct sock *sk)
500 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
501 unsigned long now = jiffies;
503 if (time_after(now, last_overflow + HZ))
504 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
507 /* syncookies: no recent synqueue overflow on this listening socket? */
508 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
510 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
512 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
515 static inline u32 tcp_cookie_time(void)
517 u64 val = get_jiffies_64();
519 do_div(val, TCP_SYNCOOKIE_PERIOD);
523 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
525 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
526 u64 cookie_init_timestamp(struct request_sock *req);
527 bool cookie_timestamp_decode(const struct net *net,
528 struct tcp_options_received *opt);
529 bool cookie_ecn_ok(const struct tcp_options_received *opt,
530 const struct net *net, const struct dst_entry *dst);
532 /* From net/ipv6/syncookies.c */
533 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
535 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
537 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
538 const struct tcphdr *th, u16 *mssp);
539 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
543 u32 tcp_tso_autosize(const struct sock *sk, unsigned int mss_now,
545 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
547 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
548 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
549 void tcp_retransmit_timer(struct sock *sk);
550 void tcp_xmit_retransmit_queue(struct sock *);
551 void tcp_simple_retransmit(struct sock *);
552 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
553 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
554 int tcp_fragment(struct sock *, struct sk_buff *, u32, unsigned int, gfp_t);
556 void tcp_send_probe0(struct sock *);
557 void tcp_send_partial(struct sock *);
558 int tcp_write_wakeup(struct sock *, int mib);
559 void tcp_send_fin(struct sock *sk);
560 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
561 int tcp_send_synack(struct sock *);
562 void tcp_push_one(struct sock *, unsigned int mss_now);
563 void tcp_send_ack(struct sock *sk);
564 void tcp_send_delayed_ack(struct sock *sk);
565 void tcp_send_loss_probe(struct sock *sk);
566 bool tcp_schedule_loss_probe(struct sock *sk);
567 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
568 const struct sk_buff *next_skb);
571 void tcp_rearm_rto(struct sock *sk);
572 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
573 void tcp_reset(struct sock *sk);
574 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
575 void tcp_fin(struct sock *sk);
578 void tcp_init_xmit_timers(struct sock *);
579 static inline void tcp_clear_xmit_timers(struct sock *sk)
581 hrtimer_cancel(&tcp_sk(sk)->pacing_timer);
582 inet_csk_clear_xmit_timers(sk);
585 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
586 unsigned int tcp_current_mss(struct sock *sk);
588 /* Bound MSS / TSO packet size with the half of the window */
589 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
593 /* When peer uses tiny windows, there is no use in packetizing
594 * to sub-MSS pieces for the sake of SWS or making sure there
595 * are enough packets in the pipe for fast recovery.
597 * On the other hand, for extremely large MSS devices, handling
598 * smaller than MSS windows in this way does make sense.
600 if (tp->max_window > TCP_MSS_DEFAULT)
601 cutoff = (tp->max_window >> 1);
603 cutoff = tp->max_window;
605 if (cutoff && pktsize > cutoff)
606 return max_t(int, cutoff, 68U - tp->tcp_header_len);
612 void tcp_get_info(struct sock *, struct tcp_info *);
614 /* Read 'sendfile()'-style from a TCP socket */
615 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
616 sk_read_actor_t recv_actor);
618 void tcp_initialize_rcv_mss(struct sock *sk);
620 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
621 int tcp_mss_to_mtu(struct sock *sk, int mss);
622 void tcp_mtup_init(struct sock *sk);
623 void tcp_init_buffer_space(struct sock *sk);
625 static inline void tcp_bound_rto(const struct sock *sk)
627 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
628 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
631 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
633 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
636 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
638 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
639 ntohl(TCP_FLAG_ACK) |
643 static inline void tcp_fast_path_on(struct tcp_sock *tp)
645 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
648 static inline void tcp_fast_path_check(struct sock *sk)
650 struct tcp_sock *tp = tcp_sk(sk);
652 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
654 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
656 tcp_fast_path_on(tp);
659 /* Compute the actual rto_min value */
660 static inline u32 tcp_rto_min(struct sock *sk)
662 const struct dst_entry *dst = __sk_dst_get(sk);
663 u32 rto_min = TCP_RTO_MIN;
665 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
666 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
670 static inline u32 tcp_rto_min_us(struct sock *sk)
672 return jiffies_to_usecs(tcp_rto_min(sk));
675 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
677 return dst_metric_locked(dst, RTAX_CC_ALGO);
680 /* Minimum RTT in usec. ~0 means not available. */
681 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
683 return minmax_get(&tp->rtt_min);
686 /* Compute the actual receive window we are currently advertising.
687 * Rcv_nxt can be after the window if our peer push more data
688 * than the offered window.
690 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
692 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
699 /* Choose a new window, without checks for shrinking, and without
700 * scaling applied to the result. The caller does these things
701 * if necessary. This is a "raw" window selection.
703 u32 __tcp_select_window(struct sock *sk);
705 void tcp_send_window_probe(struct sock *sk);
707 /* TCP uses 32bit jiffies to save some space.
708 * Note that this is different from tcp_time_stamp, which
709 * historically has been the same until linux-4.13.
711 #define tcp_jiffies32 ((u32)jiffies)
714 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
715 * It is no longer tied to jiffies, but to 1 ms clock.
716 * Note: double check if you want to use tcp_jiffies32 instead of this.
718 #define TCP_TS_HZ 1000
720 static inline u64 tcp_clock_ns(void)
722 return local_clock();
725 static inline u64 tcp_clock_us(void)
727 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
730 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
731 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
733 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
736 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
737 static inline u32 tcp_time_stamp_raw(void)
739 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ);
743 /* Refresh 1us clock of a TCP socket,
744 * ensuring monotically increasing values.
746 static inline void tcp_mstamp_refresh(struct tcp_sock *tp)
748 u64 val = tcp_clock_us();
750 if (val > tp->tcp_mstamp)
751 tp->tcp_mstamp = val;
754 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
756 return max_t(s64, t1 - t0, 0);
759 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
761 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ);
765 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
767 #define TCPHDR_FIN 0x01
768 #define TCPHDR_SYN 0x02
769 #define TCPHDR_RST 0x04
770 #define TCPHDR_PSH 0x08
771 #define TCPHDR_ACK 0x10
772 #define TCPHDR_URG 0x20
773 #define TCPHDR_ECE 0x40
774 #define TCPHDR_CWR 0x80
776 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
778 /* This is what the send packet queuing engine uses to pass
779 * TCP per-packet control information to the transmission code.
780 * We also store the host-order sequence numbers in here too.
781 * This is 44 bytes if IPV6 is enabled.
782 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
785 __u32 seq; /* Starting sequence number */
786 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
788 /* Note : tcp_tw_isn is used in input path only
789 * (isn chosen by tcp_timewait_state_process())
791 * tcp_gso_segs/size are used in write queue only,
792 * cf tcp_skb_pcount()/tcp_skb_mss()
800 /* Used to stash the receive timestamp while this skb is in the
801 * out of order queue, as skb->tstamp is overwritten by the
806 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
808 __u8 sacked; /* State flags for SACK/FACK. */
809 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
810 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
811 #define TCPCB_LOST 0x04 /* SKB is lost */
812 #define TCPCB_TAGBITS 0x07 /* All tag bits */
813 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
814 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
815 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
818 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
819 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
820 eor:1, /* Is skb MSG_EOR marked? */
821 has_rxtstamp:1, /* SKB has a RX timestamp */
823 __u32 ack_seq; /* Sequence number ACK'd */
826 /* There is space for up to 24 bytes */
827 __u32 in_flight:30,/* Bytes in flight at transmit */
828 is_app_limited:1, /* cwnd not fully used? */
830 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
832 /* start of send pipeline phase */
834 /* when we reached the "delivered" count */
835 u64 delivered_mstamp;
836 } tx; /* only used for outgoing skbs */
838 struct inet_skb_parm h4;
839 #if IS_ENABLED(CONFIG_IPV6)
840 struct inet6_skb_parm h6;
842 } header; /* For incoming skbs */
852 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
855 #if IS_ENABLED(CONFIG_IPV6)
856 /* This is the variant of inet6_iif() that must be used by TCP,
857 * as TCP moves IP6CB into a different location in skb->cb[]
859 static inline int tcp_v6_iif(const struct sk_buff *skb)
861 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
863 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
866 /* TCP_SKB_CB reference means this can not be used from early demux */
867 static inline int tcp_v6_sdif(const struct sk_buff *skb)
869 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
870 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
871 return TCP_SKB_CB(skb)->header.h6.iif;
877 /* TCP_SKB_CB reference means this can not be used from early demux */
878 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
880 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
881 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
882 skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
888 /* TCP_SKB_CB reference means this can not be used from early demux */
889 static inline int tcp_v4_sdif(struct sk_buff *skb)
891 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
892 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
893 return TCP_SKB_CB(skb)->header.h4.iif;
898 /* Due to TSO, an SKB can be composed of multiple actual
899 * packets. To keep these tracked properly, we use this.
901 static inline int tcp_skb_pcount(const struct sk_buff *skb)
903 return TCP_SKB_CB(skb)->tcp_gso_segs;
906 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
908 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
911 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
913 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
916 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
917 static inline int tcp_skb_mss(const struct sk_buff *skb)
919 return TCP_SKB_CB(skb)->tcp_gso_size;
922 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
924 return likely(!TCP_SKB_CB(skb)->eor);
927 /* Events passed to congestion control interface */
929 CA_EVENT_TX_START, /* first transmit when no packets in flight */
930 CA_EVENT_CWND_RESTART, /* congestion window restart */
931 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
932 CA_EVENT_LOSS, /* loss timeout */
933 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
934 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
935 CA_EVENT_DELAYED_ACK, /* Delayed ack is sent */
936 CA_EVENT_NON_DELAYED_ACK,
939 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
940 enum tcp_ca_ack_event_flags {
941 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
942 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
943 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
947 * Interface for adding new TCP congestion control handlers
949 #define TCP_CA_NAME_MAX 16
950 #define TCP_CA_MAX 128
951 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
953 #define TCP_CA_UNSPEC 0
955 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
956 #define TCP_CONG_NON_RESTRICTED 0x1
957 /* Requires ECN/ECT set on all packets */
958 #define TCP_CONG_NEEDS_ECN 0x2
968 /* A rate sample measures the number of (original/retransmitted) data
969 * packets delivered "delivered" over an interval of time "interval_us".
970 * The tcp_rate.c code fills in the rate sample, and congestion
971 * control modules that define a cong_control function to run at the end
972 * of ACK processing can optionally chose to consult this sample when
973 * setting cwnd and pacing rate.
974 * A sample is invalid if "delivered" or "interval_us" is negative.
977 u64 prior_mstamp; /* starting timestamp for interval */
978 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
979 s32 delivered; /* number of packets delivered over interval */
980 long interval_us; /* time for tp->delivered to incr "delivered" */
981 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
982 int losses; /* number of packets marked lost upon ACK */
983 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
984 u32 prior_in_flight; /* in flight before this ACK */
985 bool is_app_limited; /* is sample from packet with bubble in pipe? */
986 bool is_retrans; /* is sample from retransmission? */
989 struct tcp_congestion_ops {
990 struct list_head list;
994 /* initialize private data (optional) */
995 void (*init)(struct sock *sk);
996 /* cleanup private data (optional) */
997 void (*release)(struct sock *sk);
999 /* return slow start threshold (required) */
1000 u32 (*ssthresh)(struct sock *sk);
1001 /* do new cwnd calculation (required) */
1002 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1003 /* call before changing ca_state (optional) */
1004 void (*set_state)(struct sock *sk, u8 new_state);
1005 /* call when cwnd event occurs (optional) */
1006 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1007 /* call when ack arrives (optional) */
1008 void (*in_ack_event)(struct sock *sk, u32 flags);
1009 /* new value of cwnd after loss (required) */
1010 u32 (*undo_cwnd)(struct sock *sk);
1011 /* hook for packet ack accounting (optional) */
1012 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1013 /* suggest number of segments for each skb to transmit (optional) */
1014 u32 (*tso_segs_goal)(struct sock *sk);
1015 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1016 u32 (*sndbuf_expand)(struct sock *sk);
1017 /* call when packets are delivered to update cwnd and pacing rate,
1018 * after all the ca_state processing. (optional)
1020 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1021 /* get info for inet_diag (optional) */
1022 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1023 union tcp_cc_info *info);
1025 char name[TCP_CA_NAME_MAX];
1026 struct module *owner;
1029 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1030 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1032 void tcp_assign_congestion_control(struct sock *sk);
1033 void tcp_init_congestion_control(struct sock *sk);
1034 void tcp_cleanup_congestion_control(struct sock *sk);
1035 int tcp_set_default_congestion_control(const char *name);
1036 void tcp_get_default_congestion_control(char *name);
1037 void tcp_get_available_congestion_control(char *buf, size_t len);
1038 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1039 int tcp_set_allowed_congestion_control(char *allowed);
1040 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit);
1041 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1042 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1044 u32 tcp_reno_ssthresh(struct sock *sk);
1045 u32 tcp_reno_undo_cwnd(struct sock *sk);
1046 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1047 extern struct tcp_congestion_ops tcp_reno;
1049 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1050 u32 tcp_ca_get_key_by_name(const char *name, bool *ecn_ca);
1052 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1054 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1060 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1062 const struct inet_connection_sock *icsk = inet_csk(sk);
1064 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1067 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1069 struct inet_connection_sock *icsk = inet_csk(sk);
1071 if (icsk->icsk_ca_ops->set_state)
1072 icsk->icsk_ca_ops->set_state(sk, ca_state);
1073 icsk->icsk_ca_state = ca_state;
1076 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1078 const struct inet_connection_sock *icsk = inet_csk(sk);
1080 if (icsk->icsk_ca_ops->cwnd_event)
1081 icsk->icsk_ca_ops->cwnd_event(sk, event);
1084 /* From tcp_rate.c */
1085 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1086 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1087 struct rate_sample *rs);
1088 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1089 struct rate_sample *rs);
1090 void tcp_rate_check_app_limited(struct sock *sk);
1092 /* These functions determine how the current flow behaves in respect of SACK
1093 * handling. SACK is negotiated with the peer, and therefore it can vary
1094 * between different flows.
1096 * tcp_is_sack - SACK enabled
1097 * tcp_is_reno - No SACK
1098 * tcp_is_fack - FACK enabled, implies SACK enabled
1100 static inline int tcp_is_sack(const struct tcp_sock *tp)
1102 return tp->rx_opt.sack_ok;
1105 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1107 return !tcp_is_sack(tp);
1110 static inline bool tcp_is_fack(const struct tcp_sock *tp)
1112 return tp->rx_opt.sack_ok & TCP_FACK_ENABLED;
1115 static inline void tcp_enable_fack(struct tcp_sock *tp)
1117 tp->rx_opt.sack_ok |= TCP_FACK_ENABLED;
1120 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1122 return tp->sacked_out + tp->lost_out;
1125 /* This determines how many packets are "in the network" to the best
1126 * of our knowledge. In many cases it is conservative, but where
1127 * detailed information is available from the receiver (via SACK
1128 * blocks etc.) we can make more aggressive calculations.
1130 * Use this for decisions involving congestion control, use just
1131 * tp->packets_out to determine if the send queue is empty or not.
1133 * Read this equation as:
1135 * "Packets sent once on transmission queue" MINUS
1136 * "Packets left network, but not honestly ACKed yet" PLUS
1137 * "Packets fast retransmitted"
1139 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1141 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1144 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1146 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1148 return tp->snd_cwnd < tp->snd_ssthresh;
1151 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1153 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1156 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1158 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1159 (1 << inet_csk(sk)->icsk_ca_state);
1162 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1163 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1166 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1168 const struct tcp_sock *tp = tcp_sk(sk);
1170 if (tcp_in_cwnd_reduction(sk))
1171 return tp->snd_ssthresh;
1173 return max(tp->snd_ssthresh,
1174 ((tp->snd_cwnd >> 1) +
1175 (tp->snd_cwnd >> 2)));
1178 /* Use define here intentionally to get WARN_ON location shown at the caller */
1179 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1181 void tcp_enter_cwr(struct sock *sk);
1182 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1184 /* The maximum number of MSS of available cwnd for which TSO defers
1185 * sending if not using sysctl_tcp_tso_win_divisor.
1187 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1192 /* Returns end sequence number of the receiver's advertised window */
1193 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1195 return tp->snd_una + tp->snd_wnd;
1198 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1199 * flexible approach. The RFC suggests cwnd should not be raised unless
1200 * it was fully used previously. And that's exactly what we do in
1201 * congestion avoidance mode. But in slow start we allow cwnd to grow
1202 * as long as the application has used half the cwnd.
1204 * cwnd is 10 (IW10), but application sends 9 frames.
1205 * We allow cwnd to reach 18 when all frames are ACKed.
1206 * This check is safe because it's as aggressive as slow start which already
1207 * risks 100% overshoot. The advantage is that we discourage application to
1208 * either send more filler packets or data to artificially blow up the cwnd
1209 * usage, and allow application-limited process to probe bw more aggressively.
1211 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1213 const struct tcp_sock *tp = tcp_sk(sk);
1215 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1216 if (tcp_in_slow_start(tp))
1217 return tp->snd_cwnd < 2 * tp->max_packets_out;
1219 return tp->is_cwnd_limited;
1222 /* Something is really bad, we could not queue an additional packet,
1223 * because qdisc is full or receiver sent a 0 window.
1224 * We do not want to add fuel to the fire, or abort too early,
1225 * so make sure the timer we arm now is at least 200ms in the future,
1226 * regardless of current icsk_rto value (as it could be ~2ms)
1228 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1230 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1233 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1234 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1235 unsigned long max_when)
1237 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1239 return (unsigned long)min_t(u64, when, max_when);
1242 static inline void tcp_check_probe_timer(struct sock *sk)
1244 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1245 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1246 tcp_probe0_base(sk), TCP_RTO_MAX);
1249 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1254 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1260 * Calculate(/check) TCP checksum
1262 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1263 __be32 daddr, __wsum base)
1265 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1268 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1270 return __skb_checksum_complete(skb);
1273 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1275 return !skb_csum_unnecessary(skb) &&
1276 __tcp_checksum_complete(skb);
1279 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1280 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1285 static const char *statename[]={
1286 "Unused","Established","Syn Sent","Syn Recv",
1287 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1288 "Close Wait","Last ACK","Listen","Closing"
1291 void tcp_set_state(struct sock *sk, int state);
1293 void tcp_done(struct sock *sk);
1295 int tcp_abort(struct sock *sk, int err);
1297 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1300 rx_opt->num_sacks = 0;
1303 u32 tcp_default_init_rwnd(u32 mss);
1304 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1306 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1308 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1309 struct tcp_sock *tp = tcp_sk(sk);
1312 if (!sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1313 ca_ops->cong_control)
1315 delta = tcp_jiffies32 - tp->lsndtime;
1316 if (delta > inet_csk(sk)->icsk_rto)
1317 tcp_cwnd_restart(sk, delta);
1320 /* Determine a window scaling and initial window to offer. */
1321 void tcp_select_initial_window(int __space, __u32 mss, __u32 *rcv_wnd,
1322 __u32 *window_clamp, int wscale_ok,
1323 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1325 static inline int tcp_win_from_space(int space)
1327 int tcp_adv_win_scale = sysctl_tcp_adv_win_scale;
1329 return tcp_adv_win_scale <= 0 ?
1330 (space>>(-tcp_adv_win_scale)) :
1331 space - (space>>tcp_adv_win_scale);
1334 /* Note: caller must be prepared to deal with negative returns */
1335 static inline int tcp_space(const struct sock *sk)
1337 return tcp_win_from_space(sk->sk_rcvbuf -
1338 atomic_read(&sk->sk_rmem_alloc));
1341 static inline int tcp_full_space(const struct sock *sk)
1343 return tcp_win_from_space(sk->sk_rcvbuf);
1346 extern void tcp_openreq_init_rwin(struct request_sock *req,
1347 const struct sock *sk_listener,
1348 const struct dst_entry *dst);
1350 void tcp_enter_memory_pressure(struct sock *sk);
1351 void tcp_leave_memory_pressure(struct sock *sk);
1353 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1355 struct net *net = sock_net((struct sock *)tp);
1357 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1360 static inline int keepalive_time_when(const struct tcp_sock *tp)
1362 struct net *net = sock_net((struct sock *)tp);
1364 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1367 static inline int keepalive_probes(const struct tcp_sock *tp)
1369 struct net *net = sock_net((struct sock *)tp);
1371 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1374 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1376 const struct inet_connection_sock *icsk = &tp->inet_conn;
1378 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1379 tcp_jiffies32 - tp->rcv_tstamp);
1382 static inline int tcp_fin_time(const struct sock *sk)
1384 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1385 const int rto = inet_csk(sk)->icsk_rto;
1387 if (fin_timeout < (rto << 2) - (rto >> 1))
1388 fin_timeout = (rto << 2) - (rto >> 1);
1393 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1396 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1398 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1401 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1402 * then following tcp messages have valid values. Ignore 0 value,
1403 * or else 'negative' tsval might forbid us to accept their packets.
1405 if (!rx_opt->ts_recent)
1410 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1413 if (tcp_paws_check(rx_opt, 0))
1416 /* RST segments are not recommended to carry timestamp,
1417 and, if they do, it is recommended to ignore PAWS because
1418 "their cleanup function should take precedence over timestamps."
1419 Certainly, it is mistake. It is necessary to understand the reasons
1420 of this constraint to relax it: if peer reboots, clock may go
1421 out-of-sync and half-open connections will not be reset.
1422 Actually, the problem would be not existing if all
1423 the implementations followed draft about maintaining clock
1424 via reboots. Linux-2.2 DOES NOT!
1426 However, we can relax time bounds for RST segments to MSL.
1428 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1433 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1434 int mib_idx, u32 *last_oow_ack_time);
1436 static inline void tcp_mib_init(struct net *net)
1439 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1440 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1441 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1442 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1446 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1448 tp->lost_skb_hint = NULL;
1451 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1453 tcp_clear_retrans_hints_partial(tp);
1454 tp->retransmit_skb_hint = NULL;
1457 union tcp_md5_addr {
1459 #if IS_ENABLED(CONFIG_IPV6)
1464 /* - key database */
1465 struct tcp_md5sig_key {
1466 struct hlist_node node;
1468 u8 family; /* AF_INET or AF_INET6 */
1469 union tcp_md5_addr addr;
1471 u8 key[TCP_MD5SIG_MAXKEYLEN];
1472 struct rcu_head rcu;
1476 struct tcp_md5sig_info {
1477 struct hlist_head head;
1478 struct rcu_head rcu;
1481 /* - pseudo header */
1482 struct tcp4_pseudohdr {
1490 struct tcp6_pseudohdr {
1491 struct in6_addr saddr;
1492 struct in6_addr daddr;
1494 __be32 protocol; /* including padding */
1497 union tcp_md5sum_block {
1498 struct tcp4_pseudohdr ip4;
1499 #if IS_ENABLED(CONFIG_IPV6)
1500 struct tcp6_pseudohdr ip6;
1504 /* - pool: digest algorithm, hash description and scratch buffer */
1505 struct tcp_md5sig_pool {
1506 struct ahash_request *md5_req;
1511 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1512 const struct sock *sk, const struct sk_buff *skb);
1513 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1514 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1516 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1517 int family, u8 prefixlen);
1518 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1519 const struct sock *addr_sk);
1521 #ifdef CONFIG_TCP_MD5SIG
1522 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1523 const union tcp_md5_addr *addr,
1525 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1527 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1528 const union tcp_md5_addr *addr,
1533 #define tcp_twsk_md5_key(twsk) NULL
1536 bool tcp_alloc_md5sig_pool(void);
1538 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1539 static inline void tcp_put_md5sig_pool(void)
1544 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1545 unsigned int header_len);
1546 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1547 const struct tcp_md5sig_key *key);
1549 /* From tcp_fastopen.c */
1550 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1551 struct tcp_fastopen_cookie *cookie, int *syn_loss,
1552 unsigned long *last_syn_loss);
1553 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1554 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1556 struct tcp_fastopen_request {
1557 /* Fast Open cookie. Size 0 means a cookie request */
1558 struct tcp_fastopen_cookie cookie;
1559 struct msghdr *data; /* data in MSG_FASTOPEN */
1561 int copied; /* queued in tcp_connect() */
1563 void tcp_free_fastopen_req(struct tcp_sock *tp);
1565 extern struct tcp_fastopen_context __rcu *tcp_fastopen_ctx;
1566 int tcp_fastopen_reset_cipher(void *key, unsigned int len);
1567 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1568 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1569 struct request_sock *req,
1570 struct tcp_fastopen_cookie *foc);
1571 void tcp_fastopen_init_key_once(bool publish);
1572 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1573 struct tcp_fastopen_cookie *cookie);
1574 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1575 #define TCP_FASTOPEN_KEY_LENGTH 16
1577 /* Fastopen key context */
1578 struct tcp_fastopen_context {
1579 struct crypto_cipher *tfm;
1580 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1581 struct rcu_head rcu;
1584 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1585 void tcp_fastopen_active_disable(struct sock *sk);
1586 bool tcp_fastopen_active_should_disable(struct sock *sk);
1587 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1588 void tcp_fastopen_active_timeout_reset(void);
1590 /* Latencies incurred by various limits for a sender. They are
1591 * chronograph-like stats that are mutually exclusive.
1595 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1596 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1597 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1601 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1602 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1604 /* write queue abstraction */
1605 static inline void tcp_write_queue_purge(struct sock *sk)
1607 struct sk_buff *skb;
1609 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1610 while ((skb = __skb_dequeue(&sk->sk_write_queue)) != NULL)
1611 sk_wmem_free_skb(sk, skb);
1613 tcp_clear_all_retrans_hints(tcp_sk(sk));
1616 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1618 return skb_peek(&sk->sk_write_queue);
1621 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1623 return skb_peek_tail(&sk->sk_write_queue);
1626 static inline struct sk_buff *tcp_write_queue_next(const struct sock *sk,
1627 const struct sk_buff *skb)
1629 return skb_queue_next(&sk->sk_write_queue, skb);
1632 static inline struct sk_buff *tcp_write_queue_prev(const struct sock *sk,
1633 const struct sk_buff *skb)
1635 return skb_queue_prev(&sk->sk_write_queue, skb);
1638 #define tcp_for_write_queue(skb, sk) \
1639 skb_queue_walk(&(sk)->sk_write_queue, skb)
1641 #define tcp_for_write_queue_from(skb, sk) \
1642 skb_queue_walk_from(&(sk)->sk_write_queue, skb)
1644 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1645 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1647 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1649 return sk->sk_send_head;
1652 static inline bool tcp_skb_is_last(const struct sock *sk,
1653 const struct sk_buff *skb)
1655 return skb_queue_is_last(&sk->sk_write_queue, skb);
1658 static inline void tcp_advance_send_head(struct sock *sk, const struct sk_buff *skb)
1660 if (tcp_skb_is_last(sk, skb))
1661 sk->sk_send_head = NULL;
1663 sk->sk_send_head = tcp_write_queue_next(sk, skb);
1666 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1668 if (sk->sk_send_head == skb_unlinked) {
1669 sk->sk_send_head = NULL;
1670 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1672 if (tcp_sk(sk)->highest_sack == skb_unlinked)
1673 tcp_sk(sk)->highest_sack = NULL;
1676 static inline void tcp_init_send_head(struct sock *sk)
1678 sk->sk_send_head = NULL;
1681 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1683 __skb_queue_tail(&sk->sk_write_queue, skb);
1686 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1688 __tcp_add_write_queue_tail(sk, skb);
1690 /* Queue it, remembering where we must start sending. */
1691 if (sk->sk_send_head == NULL) {
1692 sk->sk_send_head = skb;
1693 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1695 if (tcp_sk(sk)->highest_sack == NULL)
1696 tcp_sk(sk)->highest_sack = skb;
1700 static inline void __tcp_add_write_queue_head(struct sock *sk, struct sk_buff *skb)
1702 __skb_queue_head(&sk->sk_write_queue, skb);
1705 /* Insert buff after skb on the write queue of sk. */
1706 static inline void tcp_insert_write_queue_after(struct sk_buff *skb,
1707 struct sk_buff *buff,
1710 __skb_queue_after(&sk->sk_write_queue, skb, buff);
1713 /* Insert new before skb on the write queue of sk. */
1714 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1715 struct sk_buff *skb,
1718 __skb_queue_before(&sk->sk_write_queue, skb, new);
1720 if (sk->sk_send_head == skb)
1721 sk->sk_send_head = new;
1724 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1726 __skb_unlink(skb, &sk->sk_write_queue);
1729 static inline bool tcp_write_queue_empty(struct sock *sk)
1731 return skb_queue_empty(&sk->sk_write_queue);
1734 static inline void tcp_push_pending_frames(struct sock *sk)
1736 if (tcp_send_head(sk)) {
1737 struct tcp_sock *tp = tcp_sk(sk);
1739 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1743 /* Start sequence of the skb just after the highest skb with SACKed
1744 * bit, valid only if sacked_out > 0 or when the caller has ensured
1745 * validity by itself.
1747 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1749 if (!tp->sacked_out)
1752 if (tp->highest_sack == NULL)
1755 return TCP_SKB_CB(tp->highest_sack)->seq;
1758 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1760 tcp_sk(sk)->highest_sack = tcp_skb_is_last(sk, skb) ? NULL :
1761 tcp_write_queue_next(sk, skb);
1764 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1766 return tcp_sk(sk)->highest_sack;
1769 static inline void tcp_highest_sack_reset(struct sock *sk)
1771 tcp_sk(sk)->highest_sack = tcp_write_queue_head(sk);
1774 /* Called when old skb is about to be deleted and replaced by new skb */
1775 static inline void tcp_highest_sack_replace(struct sock *sk,
1776 struct sk_buff *old,
1777 struct sk_buff *new)
1779 if (old == tcp_highest_sack(sk))
1780 tcp_sk(sk)->highest_sack = new;
1783 /* This helper checks if socket has IP_TRANSPARENT set */
1784 static inline bool inet_sk_transparent(const struct sock *sk)
1786 switch (sk->sk_state) {
1788 return inet_twsk(sk)->tw_transparent;
1789 case TCP_NEW_SYN_RECV:
1790 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1792 return inet_sk(sk)->transparent;
1795 /* Determines whether this is a thin stream (which may suffer from
1796 * increased latency). Used to trigger latency-reducing mechanisms.
1798 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1800 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1804 enum tcp_seq_states {
1805 TCP_SEQ_STATE_LISTENING,
1806 TCP_SEQ_STATE_ESTABLISHED,
1809 int tcp_seq_open(struct inode *inode, struct file *file);
1811 struct tcp_seq_afinfo {
1814 const struct file_operations *seq_fops;
1815 struct seq_operations seq_ops;
1818 struct tcp_iter_state {
1819 struct seq_net_private p;
1821 enum tcp_seq_states state;
1822 struct sock *syn_wait_sk;
1823 int bucket, offset, sbucket, num;
1827 int tcp_proc_register(struct net *net, struct tcp_seq_afinfo *afinfo);
1828 void tcp_proc_unregister(struct net *net, struct tcp_seq_afinfo *afinfo);
1830 extern struct request_sock_ops tcp_request_sock_ops;
1831 extern struct request_sock_ops tcp6_request_sock_ops;
1833 void tcp_v4_destroy_sock(struct sock *sk);
1835 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1836 netdev_features_t features);
1837 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1838 int tcp_gro_complete(struct sk_buff *skb);
1840 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1842 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1844 struct net *net = sock_net((struct sock *)tp);
1845 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1848 static inline bool tcp_stream_memory_free(const struct sock *sk)
1850 const struct tcp_sock *tp = tcp_sk(sk);
1851 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1853 return notsent_bytes < tcp_notsent_lowat(tp);
1856 #ifdef CONFIG_PROC_FS
1857 int tcp4_proc_init(void);
1858 void tcp4_proc_exit(void);
1861 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1862 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1863 const struct tcp_request_sock_ops *af_ops,
1864 struct sock *sk, struct sk_buff *skb);
1866 /* TCP af-specific functions */
1867 struct tcp_sock_af_ops {
1868 #ifdef CONFIG_TCP_MD5SIG
1869 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1870 const struct sock *addr_sk);
1871 int (*calc_md5_hash)(char *location,
1872 const struct tcp_md5sig_key *md5,
1873 const struct sock *sk,
1874 const struct sk_buff *skb);
1875 int (*md5_parse)(struct sock *sk,
1877 char __user *optval,
1882 struct tcp_request_sock_ops {
1884 #ifdef CONFIG_TCP_MD5SIG
1885 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1886 const struct sock *addr_sk);
1887 int (*calc_md5_hash) (char *location,
1888 const struct tcp_md5sig_key *md5,
1889 const struct sock *sk,
1890 const struct sk_buff *skb);
1892 void (*init_req)(struct request_sock *req,
1893 const struct sock *sk_listener,
1894 struct sk_buff *skb);
1895 #ifdef CONFIG_SYN_COOKIES
1896 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1899 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1900 const struct request_sock *req);
1901 u32 (*init_seq)(const struct sk_buff *skb);
1902 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
1903 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1904 struct flowi *fl, struct request_sock *req,
1905 struct tcp_fastopen_cookie *foc,
1906 enum tcp_synack_type synack_type);
1909 #ifdef CONFIG_SYN_COOKIES
1910 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1911 const struct sock *sk, struct sk_buff *skb,
1914 tcp_synq_overflow(sk);
1915 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1916 return ops->cookie_init_seq(skb, mss);
1919 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1920 const struct sock *sk, struct sk_buff *skb,
1927 int tcpv4_offload_init(void);
1929 void tcp_v4_init(void);
1930 void tcp_init(void);
1932 /* tcp_recovery.c */
1933 extern void tcp_rack_mark_lost(struct sock *sk);
1934 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
1936 extern void tcp_rack_reo_timeout(struct sock *sk);
1938 /* At how many usecs into the future should the RTO fire? */
1939 static inline s64 tcp_rto_delta_us(const struct sock *sk)
1941 const struct sk_buff *skb = tcp_write_queue_head(sk);
1942 u32 rto = inet_csk(sk)->icsk_rto;
1943 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto);
1945 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
1949 * Save and compile IPv4 options, return a pointer to it
1951 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
1952 struct sk_buff *skb)
1954 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1955 struct ip_options_rcu *dopt = NULL;
1958 int opt_size = sizeof(*dopt) + opt->optlen;
1960 dopt = kmalloc(opt_size, GFP_ATOMIC);
1961 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
1969 /* locally generated TCP pure ACKs have skb->truesize == 2
1970 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1971 * This is much faster than dissecting the packet to find out.
1972 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1974 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1976 return skb->truesize == 2;
1979 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1984 static inline int tcp_inq(struct sock *sk)
1986 struct tcp_sock *tp = tcp_sk(sk);
1989 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1991 } else if (sock_flag(sk, SOCK_URGINLINE) ||
1993 before(tp->urg_seq, tp->copied_seq) ||
1994 !before(tp->urg_seq, tp->rcv_nxt)) {
1996 answ = tp->rcv_nxt - tp->copied_seq;
1998 /* Subtract 1, if FIN was received */
1999 if (answ && sock_flag(sk, SOCK_DONE))
2002 answ = tp->urg_seq - tp->copied_seq;
2008 int tcp_peek_len(struct socket *sock);
2010 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2014 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2015 tp->segs_in += segs_in;
2016 if (skb->len > tcp_hdrlen(skb))
2017 tp->data_segs_in += segs_in;
2021 * TCP listen path runs lockless.
2022 * We forced "struct sock" to be const qualified to make sure
2023 * we don't modify one of its field by mistake.
2024 * Here, we increment sk_drops which is an atomic_t, so we can safely
2025 * make sock writable again.
2027 static inline void tcp_listendrop(const struct sock *sk)
2029 atomic_inc(&((struct sock *)sk)->sk_drops);
2030 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2033 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2036 * Interface for adding Upper Level Protocols over TCP
2039 #define TCP_ULP_NAME_MAX 16
2040 #define TCP_ULP_MAX 128
2041 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2043 struct tcp_ulp_ops {
2044 struct list_head list;
2046 /* initialize ulp */
2047 int (*init)(struct sock *sk);
2049 void (*release)(struct sock *sk);
2051 char name[TCP_ULP_NAME_MAX];
2052 struct module *owner;
2054 int tcp_register_ulp(struct tcp_ulp_ops *type);
2055 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2056 int tcp_set_ulp(struct sock *sk, const char *name);
2057 void tcp_get_available_ulp(char *buf, size_t len);
2058 void tcp_cleanup_ulp(struct sock *sk);
2060 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2061 * is < 0, then the BPF op failed (for example if the loaded BPF
2062 * program does not support the chosen operation or there is no BPF
2066 static inline int tcp_call_bpf(struct sock *sk, int op)
2068 struct bpf_sock_ops_kern sock_ops;
2071 if (sk_fullsock(sk))
2072 sock_owned_by_me(sk);
2074 memset(&sock_ops, 0, sizeof(sock_ops));
2078 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2080 ret = sock_ops.reply;
2086 static inline int tcp_call_bpf(struct sock *sk, int op)
2092 static inline u32 tcp_timeout_init(struct sock *sk)
2096 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT);
2099 timeout = TCP_TIMEOUT_INIT;
2103 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2107 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT);
2114 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2116 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN) == 1);