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>
48 #include <linux/bpf-cgroup.h>
50 extern struct inet_hashinfo tcp_hashinfo;
52 extern struct percpu_counter tcp_orphan_count;
53 void tcp_time_wait(struct sock *sk, int state, int timeo);
55 #define MAX_TCP_HEADER (128 + MAX_HEADER)
56 #define MAX_TCP_OPTION_SPACE 40
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
62 #define MAX_TCP_WINDOW 32767U
64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
65 #define TCP_MIN_MSS 88U
67 /* The least MTU to use for probing */
68 #define TCP_BASE_MSS 1024
70 /* probing interval, default to 10 minutes as per RFC4821 */
71 #define TCP_PROBE_INTERVAL 600
73 /* Specify interval when tcp mtu probing will stop */
74 #define TCP_PROBE_THRESHOLD 8
76 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
77 #define TCP_FASTRETRANS_THRESH 3
79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
80 #define TCP_MAX_QUICKACKS 16U
82 /* Maximal number of window scale according to RFC1323 */
83 #define TCP_MAX_WSCALE 14U
86 #define TCP_URG_VALID 0x0100
87 #define TCP_URG_NOTYET 0x0200
88 #define TCP_URG_READ 0x0400
90 #define TCP_RETR1 3 /*
91 * This is how many retries it does before it
92 * tries to figure out if the gateway is
93 * down. Minimal RFC value is 3; it corresponds
94 * to ~3sec-8min depending on RTO.
97 #define TCP_RETR2 15 /*
98 * This should take at least
99 * 90 minutes to time out.
100 * RFC1122 says that the limit is 100 sec.
101 * 15 is ~13-30min depending on RTO.
104 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
105 * when active opening a connection.
106 * RFC1122 says the minimum retry MUST
107 * be at least 180secs. Nevertheless
108 * this value is corresponding to
109 * 63secs of retransmission with the
110 * current initial RTO.
113 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
114 * when passive opening a connection.
115 * This is corresponding to 31secs of
116 * retransmission with the current
120 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
121 * state, about 60 seconds */
122 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
123 /* BSD style FIN_WAIT2 deadlock breaker.
124 * It used to be 3min, new value is 60sec,
125 * to combine FIN-WAIT-2 timeout with
129 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
131 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
132 #define TCP_ATO_MIN ((unsigned)(HZ/25))
134 #define TCP_DELACK_MIN 4U
135 #define TCP_ATO_MIN 4U
137 #define TCP_RTO_MAX ((unsigned)(120*HZ))
138 #define TCP_RTO_MIN ((unsigned)(HZ/5))
139 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
140 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
141 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
142 * used as a fallback RTO for the
143 * initial data transmission if no
144 * valid RTT sample has been acquired,
145 * most likely due to retrans in 3WHS.
148 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
149 * for local resources.
151 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
152 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
153 #define TCP_KEEPALIVE_INTVL (75*HZ)
155 #define MAX_TCP_KEEPIDLE 32767
156 #define MAX_TCP_KEEPINTVL 32767
157 #define MAX_TCP_KEEPCNT 127
158 #define MAX_TCP_SYNCNT 127
160 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
162 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
163 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
164 * after this time. It should be equal
165 * (or greater than) TCP_TIMEWAIT_LEN
166 * to provide reliability equal to one
167 * provided by timewait state.
169 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
170 * timestamps. It must be less than
171 * minimal timewait lifetime.
177 #define TCPOPT_NOP 1 /* Padding */
178 #define TCPOPT_EOL 0 /* End of options */
179 #define TCPOPT_MSS 2 /* Segment size negotiating */
180 #define TCPOPT_WINDOW 3 /* Window scaling */
181 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
182 #define TCPOPT_SACK 5 /* SACK Block */
183 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
184 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
185 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
186 #define TCPOPT_EXP 254 /* Experimental */
187 /* Magic number to be after the option value for sharing TCP
188 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
190 #define TCPOPT_FASTOPEN_MAGIC 0xF989
191 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9
197 #define TCPOLEN_MSS 4
198 #define TCPOLEN_WINDOW 3
199 #define TCPOLEN_SACK_PERM 2
200 #define TCPOLEN_TIMESTAMP 10
201 #define TCPOLEN_MD5SIG 18
202 #define TCPOLEN_FASTOPEN_BASE 2
203 #define TCPOLEN_EXP_FASTOPEN_BASE 4
204 #define TCPOLEN_EXP_SMC_BASE 6
206 /* But this is what stacks really send out. */
207 #define TCPOLEN_TSTAMP_ALIGNED 12
208 #define TCPOLEN_WSCALE_ALIGNED 4
209 #define TCPOLEN_SACKPERM_ALIGNED 4
210 #define TCPOLEN_SACK_BASE 2
211 #define TCPOLEN_SACK_BASE_ALIGNED 4
212 #define TCPOLEN_SACK_PERBLOCK 8
213 #define TCPOLEN_MD5SIG_ALIGNED 20
214 #define TCPOLEN_MSS_ALIGNED 4
215 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
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_max_orphans;
244 extern long sysctl_tcp_mem[3];
246 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
247 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
248 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */
250 extern atomic_long_t tcp_memory_allocated;
251 extern struct percpu_counter tcp_sockets_allocated;
252 extern unsigned long tcp_memory_pressure;
254 /* optimized version of sk_under_memory_pressure() for TCP sockets */
255 static inline bool tcp_under_memory_pressure(const struct sock *sk)
257 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
258 mem_cgroup_under_socket_pressure(sk->sk_memcg))
261 return tcp_memory_pressure;
264 * The next routines deal with comparing 32 bit unsigned ints
265 * and worry about wraparound (automatic with unsigned arithmetic).
268 static inline bool before(__u32 seq1, __u32 seq2)
270 return (__s32)(seq1-seq2) < 0;
272 #define after(seq2, seq1) before(seq1, seq2)
274 /* is s2<=s1<=s3 ? */
275 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
277 return seq3 - seq2 >= seq1 - seq2;
280 static inline bool tcp_out_of_memory(struct sock *sk)
282 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
283 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
288 void sk_forced_mem_schedule(struct sock *sk, int size);
290 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
292 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
293 int orphans = percpu_counter_read_positive(ocp);
295 if (orphans << shift > sysctl_tcp_max_orphans) {
296 orphans = percpu_counter_sum_positive(ocp);
297 if (orphans << shift > sysctl_tcp_max_orphans)
303 bool tcp_check_oom(struct sock *sk, int shift);
306 extern struct proto tcp_prot;
308 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
309 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
310 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
311 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
313 void tcp_tasklet_init(void);
315 void tcp_v4_err(struct sk_buff *skb, u32);
317 void tcp_shutdown(struct sock *sk, int how);
319 int tcp_v4_early_demux(struct sk_buff *skb);
320 int tcp_v4_rcv(struct sk_buff *skb);
322 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
323 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
324 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
325 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
327 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
328 size_t size, int flags);
329 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
330 size_t size, int flags);
331 void tcp_release_cb(struct sock *sk);
332 void tcp_wfree(struct sk_buff *skb);
333 void tcp_write_timer_handler(struct sock *sk);
334 void tcp_delack_timer_handler(struct sock *sk);
335 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
336 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
337 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
338 void tcp_rcv_space_adjust(struct sock *sk);
339 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
340 void tcp_twsk_destructor(struct sock *sk);
341 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
342 struct pipe_inode_info *pipe, size_t len,
345 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
346 static inline void tcp_dec_quickack_mode(struct sock *sk,
347 const unsigned int pkts)
349 struct inet_connection_sock *icsk = inet_csk(sk);
351 if (icsk->icsk_ack.quick) {
352 if (pkts >= icsk->icsk_ack.quick) {
353 icsk->icsk_ack.quick = 0;
354 /* Leaving quickack mode we deflate ATO. */
355 icsk->icsk_ack.ato = TCP_ATO_MIN;
357 icsk->icsk_ack.quick -= pkts;
362 #define TCP_ECN_QUEUE_CWR 2
363 #define TCP_ECN_DEMAND_CWR 4
364 #define TCP_ECN_SEEN 8
374 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
376 const struct tcphdr *th);
377 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
378 struct request_sock *req, bool fastopen,
380 int tcp_child_process(struct sock *parent, struct sock *child,
381 struct sk_buff *skb);
382 void tcp_enter_loss(struct sock *sk);
383 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
384 void tcp_clear_retrans(struct tcp_sock *tp);
385 void tcp_update_metrics(struct sock *sk);
386 void tcp_init_metrics(struct sock *sk);
387 void tcp_metrics_init(void);
388 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
389 void tcp_close(struct sock *sk, long timeout);
390 void tcp_init_sock(struct sock *sk);
391 void tcp_init_transfer(struct sock *sk, int bpf_op);
392 __poll_t tcp_poll(struct file *file, struct socket *sock,
393 struct poll_table_struct *wait);
394 int tcp_getsockopt(struct sock *sk, int level, int optname,
395 char __user *optval, int __user *optlen);
396 int tcp_setsockopt(struct sock *sk, int level, int optname,
397 char __user *optval, unsigned int optlen);
398 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
399 char __user *optval, int __user *optlen);
400 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
401 char __user *optval, unsigned int optlen);
402 void tcp_set_keepalive(struct sock *sk, int val);
403 void tcp_syn_ack_timeout(const struct request_sock *req);
404 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
405 int flags, int *addr_len);
406 int tcp_set_rcvlowat(struct sock *sk, int val);
407 void tcp_data_ready(struct sock *sk);
408 int tcp_mmap(struct file *file, struct socket *sock,
409 struct vm_area_struct *vma);
410 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
411 struct tcp_options_received *opt_rx,
412 int estab, struct tcp_fastopen_cookie *foc);
413 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
416 * TCP v4 functions exported for the inet6 API
419 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
420 void tcp_v4_mtu_reduced(struct sock *sk);
421 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
422 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
423 struct sock *tcp_create_openreq_child(const struct sock *sk,
424 struct request_sock *req,
425 struct sk_buff *skb);
426 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
427 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
428 struct request_sock *req,
429 struct dst_entry *dst,
430 struct request_sock *req_unhash,
432 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
433 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
434 int tcp_connect(struct sock *sk);
435 enum tcp_synack_type {
440 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
441 struct request_sock *req,
442 struct tcp_fastopen_cookie *foc,
443 enum tcp_synack_type synack_type);
444 int tcp_disconnect(struct sock *sk, int flags);
446 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
447 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
448 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
450 /* From syncookies.c */
451 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
452 struct request_sock *req,
453 struct dst_entry *dst, u32 tsoff);
454 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
456 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
457 #ifdef CONFIG_SYN_COOKIES
459 /* Syncookies use a monotonic timer which increments every 60 seconds.
460 * This counter is used both as a hash input and partially encoded into
461 * the cookie value. A cookie is only validated further if the delta
462 * between the current counter value and the encoded one is less than this,
463 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
464 * the counter advances immediately after a cookie is generated).
466 #define MAX_SYNCOOKIE_AGE 2
467 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
468 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
470 /* syncookies: remember time of last synqueue overflow
471 * But do not dirty this field too often (once per second is enough)
472 * It is racy as we do not hold a lock, but race is very minor.
474 static inline void tcp_synq_overflow(const struct sock *sk)
476 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
477 unsigned long now = jiffies;
479 if (time_after(now, last_overflow + HZ))
480 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
483 /* syncookies: no recent synqueue overflow on this listening socket? */
484 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
486 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
488 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
491 static inline u32 tcp_cookie_time(void)
493 u64 val = get_jiffies_64();
495 do_div(val, TCP_SYNCOOKIE_PERIOD);
499 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
501 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
502 u64 cookie_init_timestamp(struct request_sock *req);
503 bool cookie_timestamp_decode(const struct net *net,
504 struct tcp_options_received *opt);
505 bool cookie_ecn_ok(const struct tcp_options_received *opt,
506 const struct net *net, const struct dst_entry *dst);
508 /* From net/ipv6/syncookies.c */
509 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
511 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
513 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
514 const struct tcphdr *th, u16 *mssp);
515 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
519 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
521 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
522 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
523 void tcp_retransmit_timer(struct sock *sk);
524 void tcp_xmit_retransmit_queue(struct sock *);
525 void tcp_simple_retransmit(struct sock *);
526 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
527 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
529 TCP_FRAG_IN_WRITE_QUEUE,
530 TCP_FRAG_IN_RTX_QUEUE,
532 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
533 struct sk_buff *skb, u32 len,
534 unsigned int mss_now, gfp_t gfp);
536 void tcp_send_probe0(struct sock *);
537 void tcp_send_partial(struct sock *);
538 int tcp_write_wakeup(struct sock *, int mib);
539 void tcp_send_fin(struct sock *sk);
540 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
541 int tcp_send_synack(struct sock *);
542 void tcp_push_one(struct sock *, unsigned int mss_now);
543 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
544 void tcp_send_ack(struct sock *sk);
545 void tcp_send_delayed_ack(struct sock *sk);
546 void tcp_send_loss_probe(struct sock *sk);
547 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
548 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
549 const struct sk_buff *next_skb);
552 void tcp_rearm_rto(struct sock *sk);
553 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
554 void tcp_reset(struct sock *sk);
555 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
556 void tcp_fin(struct sock *sk);
559 void tcp_init_xmit_timers(struct sock *);
560 static inline void tcp_clear_xmit_timers(struct sock *sk)
562 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
565 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
568 inet_csk_clear_xmit_timers(sk);
571 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
572 unsigned int tcp_current_mss(struct sock *sk);
574 /* Bound MSS / TSO packet size with the half of the window */
575 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
579 /* When peer uses tiny windows, there is no use in packetizing
580 * to sub-MSS pieces for the sake of SWS or making sure there
581 * are enough packets in the pipe for fast recovery.
583 * On the other hand, for extremely large MSS devices, handling
584 * smaller than MSS windows in this way does make sense.
586 if (tp->max_window > TCP_MSS_DEFAULT)
587 cutoff = (tp->max_window >> 1);
589 cutoff = tp->max_window;
591 if (cutoff && pktsize > cutoff)
592 return max_t(int, cutoff, 68U - tp->tcp_header_len);
598 void tcp_get_info(struct sock *, struct tcp_info *);
600 /* Read 'sendfile()'-style from a TCP socket */
601 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
602 sk_read_actor_t recv_actor);
604 void tcp_initialize_rcv_mss(struct sock *sk);
606 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
607 int tcp_mss_to_mtu(struct sock *sk, int mss);
608 void tcp_mtup_init(struct sock *sk);
609 void tcp_init_buffer_space(struct sock *sk);
611 static inline void tcp_bound_rto(const struct sock *sk)
613 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
614 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
617 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
619 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
622 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
624 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
625 ntohl(TCP_FLAG_ACK) |
629 static inline void tcp_fast_path_on(struct tcp_sock *tp)
631 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
634 static inline void tcp_fast_path_check(struct sock *sk)
636 struct tcp_sock *tp = tcp_sk(sk);
638 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
640 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
642 tcp_fast_path_on(tp);
645 /* Compute the actual rto_min value */
646 static inline u32 tcp_rto_min(struct sock *sk)
648 const struct dst_entry *dst = __sk_dst_get(sk);
649 u32 rto_min = TCP_RTO_MIN;
651 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
652 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
656 static inline u32 tcp_rto_min_us(struct sock *sk)
658 return jiffies_to_usecs(tcp_rto_min(sk));
661 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
663 return dst_metric_locked(dst, RTAX_CC_ALGO);
666 /* Minimum RTT in usec. ~0 means not available. */
667 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
669 return minmax_get(&tp->rtt_min);
672 /* Compute the actual receive window we are currently advertising.
673 * Rcv_nxt can be after the window if our peer push more data
674 * than the offered window.
676 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
678 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
685 /* Choose a new window, without checks for shrinking, and without
686 * scaling applied to the result. The caller does these things
687 * if necessary. This is a "raw" window selection.
689 u32 __tcp_select_window(struct sock *sk);
691 void tcp_send_window_probe(struct sock *sk);
693 /* TCP uses 32bit jiffies to save some space.
694 * Note that this is different from tcp_time_stamp, which
695 * historically has been the same until linux-4.13.
697 #define tcp_jiffies32 ((u32)jiffies)
700 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
701 * It is no longer tied to jiffies, but to 1 ms clock.
702 * Note: double check if you want to use tcp_jiffies32 instead of this.
704 #define TCP_TS_HZ 1000
706 static inline u64 tcp_clock_ns(void)
708 return local_clock();
711 static inline u64 tcp_clock_us(void)
713 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
716 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
717 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
719 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
722 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
723 static inline u32 tcp_time_stamp_raw(void)
725 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ);
729 /* Refresh 1us clock of a TCP socket,
730 * ensuring monotically increasing values.
732 static inline void tcp_mstamp_refresh(struct tcp_sock *tp)
734 u64 val = tcp_clock_us();
736 if (val > tp->tcp_mstamp)
737 tp->tcp_mstamp = val;
740 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
742 return max_t(s64, t1 - t0, 0);
745 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
747 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ);
751 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
753 #define TCPHDR_FIN 0x01
754 #define TCPHDR_SYN 0x02
755 #define TCPHDR_RST 0x04
756 #define TCPHDR_PSH 0x08
757 #define TCPHDR_ACK 0x10
758 #define TCPHDR_URG 0x20
759 #define TCPHDR_ECE 0x40
760 #define TCPHDR_CWR 0x80
762 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
764 /* This is what the send packet queuing engine uses to pass
765 * TCP per-packet control information to the transmission code.
766 * We also store the host-order sequence numbers in here too.
767 * This is 44 bytes if IPV6 is enabled.
768 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
771 __u32 seq; /* Starting sequence number */
772 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
774 /* Note : tcp_tw_isn is used in input path only
775 * (isn chosen by tcp_timewait_state_process())
777 * tcp_gso_segs/size are used in write queue only,
778 * cf tcp_skb_pcount()/tcp_skb_mss()
786 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
788 __u8 sacked; /* State flags for SACK. */
789 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
790 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
791 #define TCPCB_LOST 0x04 /* SKB is lost */
792 #define TCPCB_TAGBITS 0x07 /* All tag bits */
793 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
794 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
795 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
798 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
799 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
800 eor:1, /* Is skb MSG_EOR marked? */
801 has_rxtstamp:1, /* SKB has a RX timestamp */
803 __u32 ack_seq; /* Sequence number ACK'd */
806 /* There is space for up to 24 bytes */
807 __u32 in_flight:30,/* Bytes in flight at transmit */
808 is_app_limited:1, /* cwnd not fully used? */
810 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
812 /* start of send pipeline phase */
814 /* when we reached the "delivered" count */
815 u64 delivered_mstamp;
816 } tx; /* only used for outgoing skbs */
818 struct inet_skb_parm h4;
819 #if IS_ENABLED(CONFIG_IPV6)
820 struct inet6_skb_parm h6;
822 } header; /* For incoming skbs */
825 struct sock *sk_redir;
831 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
833 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb)
835 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb);
838 #if IS_ENABLED(CONFIG_IPV6)
839 /* This is the variant of inet6_iif() that must be used by TCP,
840 * as TCP moves IP6CB into a different location in skb->cb[]
842 static inline int tcp_v6_iif(const struct sk_buff *skb)
844 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
846 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
849 /* TCP_SKB_CB reference means this can not be used from early demux */
850 static inline int tcp_v6_sdif(const struct sk_buff *skb)
852 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
853 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
854 return TCP_SKB_CB(skb)->header.h6.iif;
860 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
862 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
863 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
864 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
870 /* TCP_SKB_CB reference means this can not be used from early demux */
871 static inline int tcp_v4_sdif(struct sk_buff *skb)
873 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
874 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
875 return TCP_SKB_CB(skb)->header.h4.iif;
880 /* Due to TSO, an SKB can be composed of multiple actual
881 * packets. To keep these tracked properly, we use this.
883 static inline int tcp_skb_pcount(const struct sk_buff *skb)
885 return TCP_SKB_CB(skb)->tcp_gso_segs;
888 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
890 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
893 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
895 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
898 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
899 static inline int tcp_skb_mss(const struct sk_buff *skb)
901 return TCP_SKB_CB(skb)->tcp_gso_size;
904 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
906 return likely(!TCP_SKB_CB(skb)->eor);
909 /* Events passed to congestion control interface */
911 CA_EVENT_TX_START, /* first transmit when no packets in flight */
912 CA_EVENT_CWND_RESTART, /* congestion window restart */
913 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
914 CA_EVENT_LOSS, /* loss timeout */
915 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
916 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
919 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
920 enum tcp_ca_ack_event_flags {
921 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
922 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
923 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
927 * Interface for adding new TCP congestion control handlers
929 #define TCP_CA_NAME_MAX 16
930 #define TCP_CA_MAX 128
931 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
933 #define TCP_CA_UNSPEC 0
935 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
936 #define TCP_CONG_NON_RESTRICTED 0x1
937 /* Requires ECN/ECT set on all packets */
938 #define TCP_CONG_NEEDS_ECN 0x2
948 /* A rate sample measures the number of (original/retransmitted) data
949 * packets delivered "delivered" over an interval of time "interval_us".
950 * The tcp_rate.c code fills in the rate sample, and congestion
951 * control modules that define a cong_control function to run at the end
952 * of ACK processing can optionally chose to consult this sample when
953 * setting cwnd and pacing rate.
954 * A sample is invalid if "delivered" or "interval_us" is negative.
957 u64 prior_mstamp; /* starting timestamp for interval */
958 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
959 s32 delivered; /* number of packets delivered over interval */
960 long interval_us; /* time for tp->delivered to incr "delivered" */
961 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
962 int losses; /* number of packets marked lost upon ACK */
963 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
964 u32 prior_in_flight; /* in flight before this ACK */
965 bool is_app_limited; /* is sample from packet with bubble in pipe? */
966 bool is_retrans; /* is sample from retransmission? */
967 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
970 struct tcp_congestion_ops {
971 struct list_head list;
975 /* initialize private data (optional) */
976 void (*init)(struct sock *sk);
977 /* cleanup private data (optional) */
978 void (*release)(struct sock *sk);
980 /* return slow start threshold (required) */
981 u32 (*ssthresh)(struct sock *sk);
982 /* do new cwnd calculation (required) */
983 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
984 /* call before changing ca_state (optional) */
985 void (*set_state)(struct sock *sk, u8 new_state);
986 /* call when cwnd event occurs (optional) */
987 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
988 /* call when ack arrives (optional) */
989 void (*in_ack_event)(struct sock *sk, u32 flags);
990 /* new value of cwnd after loss (required) */
991 u32 (*undo_cwnd)(struct sock *sk);
992 /* hook for packet ack accounting (optional) */
993 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
994 /* override sysctl_tcp_min_tso_segs */
995 u32 (*min_tso_segs)(struct sock *sk);
996 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
997 u32 (*sndbuf_expand)(struct sock *sk);
998 /* call when packets are delivered to update cwnd and pacing rate,
999 * after all the ca_state processing. (optional)
1001 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1002 /* get info for inet_diag (optional) */
1003 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1004 union tcp_cc_info *info);
1006 char name[TCP_CA_NAME_MAX];
1007 struct module *owner;
1010 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1011 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1013 void tcp_assign_congestion_control(struct sock *sk);
1014 void tcp_init_congestion_control(struct sock *sk);
1015 void tcp_cleanup_congestion_control(struct sock *sk);
1016 int tcp_set_default_congestion_control(struct net *net, const char *name);
1017 void tcp_get_default_congestion_control(struct net *net, char *name);
1018 void tcp_get_available_congestion_control(char *buf, size_t len);
1019 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1020 int tcp_set_allowed_congestion_control(char *allowed);
1021 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit);
1022 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1023 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1025 u32 tcp_reno_ssthresh(struct sock *sk);
1026 u32 tcp_reno_undo_cwnd(struct sock *sk);
1027 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1028 extern struct tcp_congestion_ops tcp_reno;
1030 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1031 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1033 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1035 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1041 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1043 const struct inet_connection_sock *icsk = inet_csk(sk);
1045 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1048 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1050 struct inet_connection_sock *icsk = inet_csk(sk);
1052 if (icsk->icsk_ca_ops->set_state)
1053 icsk->icsk_ca_ops->set_state(sk, ca_state);
1054 icsk->icsk_ca_state = ca_state;
1057 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1059 const struct inet_connection_sock *icsk = inet_csk(sk);
1061 if (icsk->icsk_ca_ops->cwnd_event)
1062 icsk->icsk_ca_ops->cwnd_event(sk, event);
1065 /* From tcp_rate.c */
1066 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1067 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1068 struct rate_sample *rs);
1069 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1070 bool is_sack_reneg, struct rate_sample *rs);
1071 void tcp_rate_check_app_limited(struct sock *sk);
1073 /* These functions determine how the current flow behaves in respect of SACK
1074 * handling. SACK is negotiated with the peer, and therefore it can vary
1075 * between different flows.
1077 * tcp_is_sack - SACK enabled
1078 * tcp_is_reno - No SACK
1080 static inline int tcp_is_sack(const struct tcp_sock *tp)
1082 return tp->rx_opt.sack_ok;
1085 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1087 return !tcp_is_sack(tp);
1090 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1092 return tp->sacked_out + tp->lost_out;
1095 /* This determines how many packets are "in the network" to the best
1096 * of our knowledge. In many cases it is conservative, but where
1097 * detailed information is available from the receiver (via SACK
1098 * blocks etc.) we can make more aggressive calculations.
1100 * Use this for decisions involving congestion control, use just
1101 * tp->packets_out to determine if the send queue is empty or not.
1103 * Read this equation as:
1105 * "Packets sent once on transmission queue" MINUS
1106 * "Packets left network, but not honestly ACKed yet" PLUS
1107 * "Packets fast retransmitted"
1109 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1111 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1114 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1116 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1118 return tp->snd_cwnd < tp->snd_ssthresh;
1121 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1123 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1126 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1128 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1129 (1 << inet_csk(sk)->icsk_ca_state);
1132 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1133 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1136 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1138 const struct tcp_sock *tp = tcp_sk(sk);
1140 if (tcp_in_cwnd_reduction(sk))
1141 return tp->snd_ssthresh;
1143 return max(tp->snd_ssthresh,
1144 ((tp->snd_cwnd >> 1) +
1145 (tp->snd_cwnd >> 2)));
1148 /* Use define here intentionally to get WARN_ON location shown at the caller */
1149 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1151 void tcp_enter_cwr(struct sock *sk);
1152 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1154 /* The maximum number of MSS of available cwnd for which TSO defers
1155 * sending if not using sysctl_tcp_tso_win_divisor.
1157 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1162 /* Returns end sequence number of the receiver's advertised window */
1163 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1165 return tp->snd_una + tp->snd_wnd;
1168 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1169 * flexible approach. The RFC suggests cwnd should not be raised unless
1170 * it was fully used previously. And that's exactly what we do in
1171 * congestion avoidance mode. But in slow start we allow cwnd to grow
1172 * as long as the application has used half the cwnd.
1174 * cwnd is 10 (IW10), but application sends 9 frames.
1175 * We allow cwnd to reach 18 when all frames are ACKed.
1176 * This check is safe because it's as aggressive as slow start which already
1177 * risks 100% overshoot. The advantage is that we discourage application to
1178 * either send more filler packets or data to artificially blow up the cwnd
1179 * usage, and allow application-limited process to probe bw more aggressively.
1181 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1183 const struct tcp_sock *tp = tcp_sk(sk);
1185 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1186 if (tcp_in_slow_start(tp))
1187 return tp->snd_cwnd < 2 * tp->max_packets_out;
1189 return tp->is_cwnd_limited;
1192 /* Something is really bad, we could not queue an additional packet,
1193 * because qdisc is full or receiver sent a 0 window.
1194 * We do not want to add fuel to the fire, or abort too early,
1195 * so make sure the timer we arm now is at least 200ms in the future,
1196 * regardless of current icsk_rto value (as it could be ~2ms)
1198 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1200 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1203 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1204 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1205 unsigned long max_when)
1207 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1209 return (unsigned long)min_t(u64, when, max_when);
1212 static inline void tcp_check_probe_timer(struct sock *sk)
1214 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1215 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1216 tcp_probe0_base(sk), TCP_RTO_MAX);
1219 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1224 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1230 * Calculate(/check) TCP checksum
1232 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1233 __be32 daddr, __wsum base)
1235 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1238 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1240 return __skb_checksum_complete(skb);
1243 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1245 return !skb_csum_unnecessary(skb) &&
1246 __tcp_checksum_complete(skb);
1249 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1250 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1255 static const char *statename[]={
1256 "Unused","Established","Syn Sent","Syn Recv",
1257 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1258 "Close Wait","Last ACK","Listen","Closing"
1261 void tcp_set_state(struct sock *sk, int state);
1263 void tcp_done(struct sock *sk);
1265 int tcp_abort(struct sock *sk, int err);
1267 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1270 rx_opt->num_sacks = 0;
1273 u32 tcp_default_init_rwnd(u32 mss);
1274 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1276 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1278 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1279 struct tcp_sock *tp = tcp_sk(sk);
1282 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1283 ca_ops->cong_control)
1285 delta = tcp_jiffies32 - tp->lsndtime;
1286 if (delta > inet_csk(sk)->icsk_rto)
1287 tcp_cwnd_restart(sk, delta);
1290 /* Determine a window scaling and initial window to offer. */
1291 void tcp_select_initial_window(const struct sock *sk, int __space,
1292 __u32 mss, __u32 *rcv_wnd,
1293 __u32 *window_clamp, int wscale_ok,
1294 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1296 static inline int tcp_win_from_space(const struct sock *sk, int space)
1298 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1300 return tcp_adv_win_scale <= 0 ?
1301 (space>>(-tcp_adv_win_scale)) :
1302 space - (space>>tcp_adv_win_scale);
1305 /* Note: caller must be prepared to deal with negative returns */
1306 static inline int tcp_space(const struct sock *sk)
1308 return tcp_win_from_space(sk, sk->sk_rcvbuf -
1309 atomic_read(&sk->sk_rmem_alloc));
1312 static inline int tcp_full_space(const struct sock *sk)
1314 return tcp_win_from_space(sk, sk->sk_rcvbuf);
1317 extern void tcp_openreq_init_rwin(struct request_sock *req,
1318 const struct sock *sk_listener,
1319 const struct dst_entry *dst);
1321 void tcp_enter_memory_pressure(struct sock *sk);
1322 void tcp_leave_memory_pressure(struct sock *sk);
1324 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1326 struct net *net = sock_net((struct sock *)tp);
1328 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1331 static inline int keepalive_time_when(const struct tcp_sock *tp)
1333 struct net *net = sock_net((struct sock *)tp);
1335 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1338 static inline int keepalive_probes(const struct tcp_sock *tp)
1340 struct net *net = sock_net((struct sock *)tp);
1342 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1345 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1347 const struct inet_connection_sock *icsk = &tp->inet_conn;
1349 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1350 tcp_jiffies32 - tp->rcv_tstamp);
1353 static inline int tcp_fin_time(const struct sock *sk)
1355 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1356 const int rto = inet_csk(sk)->icsk_rto;
1358 if (fin_timeout < (rto << 2) - (rto >> 1))
1359 fin_timeout = (rto << 2) - (rto >> 1);
1364 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1367 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1369 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1372 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1373 * then following tcp messages have valid values. Ignore 0 value,
1374 * or else 'negative' tsval might forbid us to accept their packets.
1376 if (!rx_opt->ts_recent)
1381 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1384 if (tcp_paws_check(rx_opt, 0))
1387 /* RST segments are not recommended to carry timestamp,
1388 and, if they do, it is recommended to ignore PAWS because
1389 "their cleanup function should take precedence over timestamps."
1390 Certainly, it is mistake. It is necessary to understand the reasons
1391 of this constraint to relax it: if peer reboots, clock may go
1392 out-of-sync and half-open connections will not be reset.
1393 Actually, the problem would be not existing if all
1394 the implementations followed draft about maintaining clock
1395 via reboots. Linux-2.2 DOES NOT!
1397 However, we can relax time bounds for RST segments to MSL.
1399 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1404 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1405 int mib_idx, u32 *last_oow_ack_time);
1407 static inline void tcp_mib_init(struct net *net)
1410 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1411 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1412 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1413 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1417 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1419 tp->lost_skb_hint = NULL;
1422 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1424 tcp_clear_retrans_hints_partial(tp);
1425 tp->retransmit_skb_hint = NULL;
1428 union tcp_md5_addr {
1430 #if IS_ENABLED(CONFIG_IPV6)
1435 /* - key database */
1436 struct tcp_md5sig_key {
1437 struct hlist_node node;
1439 u8 family; /* AF_INET or AF_INET6 */
1440 union tcp_md5_addr addr;
1442 u8 key[TCP_MD5SIG_MAXKEYLEN];
1443 struct rcu_head rcu;
1447 struct tcp_md5sig_info {
1448 struct hlist_head head;
1449 struct rcu_head rcu;
1452 /* - pseudo header */
1453 struct tcp4_pseudohdr {
1461 struct tcp6_pseudohdr {
1462 struct in6_addr saddr;
1463 struct in6_addr daddr;
1465 __be32 protocol; /* including padding */
1468 union tcp_md5sum_block {
1469 struct tcp4_pseudohdr ip4;
1470 #if IS_ENABLED(CONFIG_IPV6)
1471 struct tcp6_pseudohdr ip6;
1475 /* - pool: digest algorithm, hash description and scratch buffer */
1476 struct tcp_md5sig_pool {
1477 struct ahash_request *md5_req;
1482 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1483 const struct sock *sk, const struct sk_buff *skb);
1484 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1485 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1487 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1488 int family, u8 prefixlen);
1489 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1490 const struct sock *addr_sk);
1492 #ifdef CONFIG_TCP_MD5SIG
1493 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1494 const union tcp_md5_addr *addr,
1496 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1498 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1499 const union tcp_md5_addr *addr,
1504 #define tcp_twsk_md5_key(twsk) NULL
1507 bool tcp_alloc_md5sig_pool(void);
1509 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1510 static inline void tcp_put_md5sig_pool(void)
1515 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1516 unsigned int header_len);
1517 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1518 const struct tcp_md5sig_key *key);
1520 /* From tcp_fastopen.c */
1521 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1522 struct tcp_fastopen_cookie *cookie);
1523 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1524 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1526 struct tcp_fastopen_request {
1527 /* Fast Open cookie. Size 0 means a cookie request */
1528 struct tcp_fastopen_cookie cookie;
1529 struct msghdr *data; /* data in MSG_FASTOPEN */
1531 int copied; /* queued in tcp_connect() */
1533 void tcp_free_fastopen_req(struct tcp_sock *tp);
1534 void tcp_fastopen_destroy_cipher(struct sock *sk);
1535 void tcp_fastopen_ctx_destroy(struct net *net);
1536 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1537 void *key, unsigned int len);
1538 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1539 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1540 struct request_sock *req,
1541 struct tcp_fastopen_cookie *foc,
1542 const struct dst_entry *dst);
1543 void tcp_fastopen_init_key_once(struct net *net);
1544 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1545 struct tcp_fastopen_cookie *cookie);
1546 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1547 #define TCP_FASTOPEN_KEY_LENGTH 16
1549 /* Fastopen key context */
1550 struct tcp_fastopen_context {
1551 struct crypto_cipher *tfm;
1552 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1553 struct rcu_head rcu;
1556 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1557 void tcp_fastopen_active_disable(struct sock *sk);
1558 bool tcp_fastopen_active_should_disable(struct sock *sk);
1559 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1560 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1562 /* Latencies incurred by various limits for a sender. They are
1563 * chronograph-like stats that are mutually exclusive.
1567 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1568 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1569 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1573 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1574 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1576 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1577 * the same memory storage than skb->destructor/_skb_refdst
1579 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1581 skb->destructor = NULL;
1582 skb->_skb_refdst = 0UL;
1585 #define tcp_skb_tsorted_save(skb) { \
1586 unsigned long _save = skb->_skb_refdst; \
1587 skb->_skb_refdst = 0UL;
1589 #define tcp_skb_tsorted_restore(skb) \
1590 skb->_skb_refdst = _save; \
1593 void tcp_write_queue_purge(struct sock *sk);
1595 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1597 return skb_rb_first(&sk->tcp_rtx_queue);
1600 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1602 return skb_peek(&sk->sk_write_queue);
1605 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1607 return skb_peek_tail(&sk->sk_write_queue);
1610 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1611 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1613 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1615 return skb_peek(&sk->sk_write_queue);
1618 static inline bool tcp_skb_is_last(const struct sock *sk,
1619 const struct sk_buff *skb)
1621 return skb_queue_is_last(&sk->sk_write_queue, skb);
1624 static inline bool tcp_write_queue_empty(const struct sock *sk)
1626 return skb_queue_empty(&sk->sk_write_queue);
1629 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1631 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1634 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1636 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1639 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1641 if (tcp_write_queue_empty(sk))
1642 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1645 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1647 __skb_queue_tail(&sk->sk_write_queue, skb);
1650 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1652 __tcp_add_write_queue_tail(sk, skb);
1654 /* Queue it, remembering where we must start sending. */
1655 if (sk->sk_write_queue.next == skb)
1656 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1659 /* Insert new before skb on the write queue of sk. */
1660 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1661 struct sk_buff *skb,
1664 __skb_queue_before(&sk->sk_write_queue, skb, new);
1667 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1669 tcp_skb_tsorted_anchor_cleanup(skb);
1670 __skb_unlink(skb, &sk->sk_write_queue);
1673 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1675 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1677 tcp_skb_tsorted_anchor_cleanup(skb);
1678 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1681 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1683 list_del(&skb->tcp_tsorted_anchor);
1684 tcp_rtx_queue_unlink(skb, sk);
1685 sk_wmem_free_skb(sk, skb);
1688 static inline void tcp_push_pending_frames(struct sock *sk)
1690 if (tcp_send_head(sk)) {
1691 struct tcp_sock *tp = tcp_sk(sk);
1693 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1697 /* Start sequence of the skb just after the highest skb with SACKed
1698 * bit, valid only if sacked_out > 0 or when the caller has ensured
1699 * validity by itself.
1701 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1703 if (!tp->sacked_out)
1706 if (tp->highest_sack == NULL)
1709 return TCP_SKB_CB(tp->highest_sack)->seq;
1712 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1714 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1717 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1719 return tcp_sk(sk)->highest_sack;
1722 static inline void tcp_highest_sack_reset(struct sock *sk)
1724 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1727 /* Called when old skb is about to be deleted and replaced by new skb */
1728 static inline void tcp_highest_sack_replace(struct sock *sk,
1729 struct sk_buff *old,
1730 struct sk_buff *new)
1732 if (old == tcp_highest_sack(sk))
1733 tcp_sk(sk)->highest_sack = new;
1736 /* This helper checks if socket has IP_TRANSPARENT set */
1737 static inline bool inet_sk_transparent(const struct sock *sk)
1739 switch (sk->sk_state) {
1741 return inet_twsk(sk)->tw_transparent;
1742 case TCP_NEW_SYN_RECV:
1743 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1745 return inet_sk(sk)->transparent;
1748 /* Determines whether this is a thin stream (which may suffer from
1749 * increased latency). Used to trigger latency-reducing mechanisms.
1751 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1753 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1757 enum tcp_seq_states {
1758 TCP_SEQ_STATE_LISTENING,
1759 TCP_SEQ_STATE_ESTABLISHED,
1762 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1763 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1764 void tcp_seq_stop(struct seq_file *seq, void *v);
1766 struct tcp_seq_afinfo {
1770 struct tcp_iter_state {
1771 struct seq_net_private p;
1772 enum tcp_seq_states state;
1773 struct sock *syn_wait_sk;
1774 int bucket, offset, sbucket, num;
1778 extern struct request_sock_ops tcp_request_sock_ops;
1779 extern struct request_sock_ops tcp6_request_sock_ops;
1781 void tcp_v4_destroy_sock(struct sock *sk);
1783 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1784 netdev_features_t features);
1785 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1786 int tcp_gro_complete(struct sk_buff *skb);
1788 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1790 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1792 struct net *net = sock_net((struct sock *)tp);
1793 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1796 static inline bool tcp_stream_memory_free(const struct sock *sk)
1798 const struct tcp_sock *tp = tcp_sk(sk);
1799 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1801 return notsent_bytes < tcp_notsent_lowat(tp);
1804 #ifdef CONFIG_PROC_FS
1805 int tcp4_proc_init(void);
1806 void tcp4_proc_exit(void);
1809 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1810 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1811 const struct tcp_request_sock_ops *af_ops,
1812 struct sock *sk, struct sk_buff *skb);
1814 /* TCP af-specific functions */
1815 struct tcp_sock_af_ops {
1816 #ifdef CONFIG_TCP_MD5SIG
1817 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1818 const struct sock *addr_sk);
1819 int (*calc_md5_hash)(char *location,
1820 const struct tcp_md5sig_key *md5,
1821 const struct sock *sk,
1822 const struct sk_buff *skb);
1823 int (*md5_parse)(struct sock *sk,
1825 char __user *optval,
1830 struct tcp_request_sock_ops {
1832 #ifdef CONFIG_TCP_MD5SIG
1833 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1834 const struct sock *addr_sk);
1835 int (*calc_md5_hash) (char *location,
1836 const struct tcp_md5sig_key *md5,
1837 const struct sock *sk,
1838 const struct sk_buff *skb);
1840 void (*init_req)(struct request_sock *req,
1841 const struct sock *sk_listener,
1842 struct sk_buff *skb);
1843 #ifdef CONFIG_SYN_COOKIES
1844 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1847 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1848 const struct request_sock *req);
1849 u32 (*init_seq)(const struct sk_buff *skb);
1850 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
1851 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1852 struct flowi *fl, struct request_sock *req,
1853 struct tcp_fastopen_cookie *foc,
1854 enum tcp_synack_type synack_type);
1857 #ifdef CONFIG_SYN_COOKIES
1858 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1859 const struct sock *sk, struct sk_buff *skb,
1862 tcp_synq_overflow(sk);
1863 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1864 return ops->cookie_init_seq(skb, mss);
1867 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1868 const struct sock *sk, struct sk_buff *skb,
1875 int tcpv4_offload_init(void);
1877 void tcp_v4_init(void);
1878 void tcp_init(void);
1880 /* tcp_recovery.c */
1881 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
1882 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
1883 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
1885 extern void tcp_rack_mark_lost(struct sock *sk);
1886 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
1888 extern void tcp_rack_reo_timeout(struct sock *sk);
1889 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
1891 /* At how many usecs into the future should the RTO fire? */
1892 static inline s64 tcp_rto_delta_us(const struct sock *sk)
1894 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
1895 u32 rto = inet_csk(sk)->icsk_rto;
1896 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto);
1898 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
1902 * Save and compile IPv4 options, return a pointer to it
1904 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
1905 struct sk_buff *skb)
1907 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1908 struct ip_options_rcu *dopt = NULL;
1911 int opt_size = sizeof(*dopt) + opt->optlen;
1913 dopt = kmalloc(opt_size, GFP_ATOMIC);
1914 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
1922 /* locally generated TCP pure ACKs have skb->truesize == 2
1923 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1924 * This is much faster than dissecting the packet to find out.
1925 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1927 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1929 return skb->truesize == 2;
1932 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1937 static inline int tcp_inq(struct sock *sk)
1939 struct tcp_sock *tp = tcp_sk(sk);
1942 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1944 } else if (sock_flag(sk, SOCK_URGINLINE) ||
1946 before(tp->urg_seq, tp->copied_seq) ||
1947 !before(tp->urg_seq, tp->rcv_nxt)) {
1949 answ = tp->rcv_nxt - tp->copied_seq;
1951 /* Subtract 1, if FIN was received */
1952 if (answ && sock_flag(sk, SOCK_DONE))
1955 answ = tp->urg_seq - tp->copied_seq;
1961 int tcp_peek_len(struct socket *sock);
1963 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1967 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1968 tp->segs_in += segs_in;
1969 if (skb->len > tcp_hdrlen(skb))
1970 tp->data_segs_in += segs_in;
1974 * TCP listen path runs lockless.
1975 * We forced "struct sock" to be const qualified to make sure
1976 * we don't modify one of its field by mistake.
1977 * Here, we increment sk_drops which is an atomic_t, so we can safely
1978 * make sock writable again.
1980 static inline void tcp_listendrop(const struct sock *sk)
1982 atomic_inc(&((struct sock *)sk)->sk_drops);
1983 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
1986 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
1989 * Interface for adding Upper Level Protocols over TCP
1992 #define TCP_ULP_NAME_MAX 16
1993 #define TCP_ULP_MAX 128
1994 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2001 struct tcp_ulp_ops {
2002 struct list_head list;
2004 /* initialize ulp */
2005 int (*init)(struct sock *sk);
2007 void (*release)(struct sock *sk);
2010 char name[TCP_ULP_NAME_MAX];
2012 struct module *owner;
2014 int tcp_register_ulp(struct tcp_ulp_ops *type);
2015 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2016 int tcp_set_ulp(struct sock *sk, const char *name);
2017 int tcp_set_ulp_id(struct sock *sk, const int ulp);
2018 void tcp_get_available_ulp(char *buf, size_t len);
2019 void tcp_cleanup_ulp(struct sock *sk);
2021 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2022 * is < 0, then the BPF op failed (for example if the loaded BPF
2023 * program does not support the chosen operation or there is no BPF
2027 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2029 struct bpf_sock_ops_kern sock_ops;
2032 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2033 if (sk_fullsock(sk)) {
2034 sock_ops.is_fullsock = 1;
2035 sock_owned_by_me(sk);
2041 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2043 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2045 ret = sock_ops.reply;
2051 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2053 u32 args[2] = {arg1, arg2};
2055 return tcp_call_bpf(sk, op, 2, args);
2058 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2061 u32 args[3] = {arg1, arg2, arg3};
2063 return tcp_call_bpf(sk, op, 3, args);
2067 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2072 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2077 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2085 static inline u32 tcp_timeout_init(struct sock *sk)
2089 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2092 timeout = TCP_TIMEOUT_INIT;
2096 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2100 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2107 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2109 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2112 #if IS_ENABLED(CONFIG_SMC)
2113 extern struct static_key_false tcp_have_smc;
2116 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2117 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2118 void (*cad)(struct sock *sk, u32 ack_seq));
2119 void clean_acked_data_disable(struct inet_connection_sock *icsk);