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 static inline void tcp_dec_quickack_mode(struct sock *sk,
346 const unsigned int pkts)
348 struct inet_connection_sock *icsk = inet_csk(sk);
350 if (icsk->icsk_ack.quick) {
351 if (pkts >= icsk->icsk_ack.quick) {
352 icsk->icsk_ack.quick = 0;
353 /* Leaving quickack mode we deflate ATO. */
354 icsk->icsk_ack.ato = TCP_ATO_MIN;
356 icsk->icsk_ack.quick -= pkts;
361 #define TCP_ECN_QUEUE_CWR 2
362 #define TCP_ECN_DEMAND_CWR 4
363 #define TCP_ECN_SEEN 8
373 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
375 const struct tcphdr *th);
376 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
377 struct request_sock *req, bool fastopen,
379 int tcp_child_process(struct sock *parent, struct sock *child,
380 struct sk_buff *skb);
381 void tcp_enter_loss(struct sock *sk);
382 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
383 void tcp_clear_retrans(struct tcp_sock *tp);
384 void tcp_update_metrics(struct sock *sk);
385 void tcp_init_metrics(struct sock *sk);
386 void tcp_metrics_init(void);
387 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
388 void tcp_close(struct sock *sk, long timeout);
389 void tcp_init_sock(struct sock *sk);
390 void tcp_init_transfer(struct sock *sk, int bpf_op);
391 __poll_t tcp_poll(struct file *file, struct socket *sock,
392 struct poll_table_struct *wait);
393 int tcp_getsockopt(struct sock *sk, int level, int optname,
394 char __user *optval, int __user *optlen);
395 int tcp_setsockopt(struct sock *sk, int level, int optname,
396 char __user *optval, unsigned int optlen);
397 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
398 char __user *optval, int __user *optlen);
399 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
400 char __user *optval, unsigned int optlen);
401 void tcp_set_keepalive(struct sock *sk, int val);
402 void tcp_syn_ack_timeout(const struct request_sock *req);
403 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
404 int flags, int *addr_len);
405 int tcp_set_rcvlowat(struct sock *sk, int val);
406 void tcp_data_ready(struct sock *sk);
407 int tcp_mmap(struct file *file, struct socket *sock,
408 struct vm_area_struct *vma);
409 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
410 struct tcp_options_received *opt_rx,
411 int estab, struct tcp_fastopen_cookie *foc);
412 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
415 * TCP v4 functions exported for the inet6 API
418 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
419 void tcp_v4_mtu_reduced(struct sock *sk);
420 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
421 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
422 struct sock *tcp_create_openreq_child(const struct sock *sk,
423 struct request_sock *req,
424 struct sk_buff *skb);
425 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
426 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
427 struct request_sock *req,
428 struct dst_entry *dst,
429 struct request_sock *req_unhash,
431 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
432 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
433 int tcp_connect(struct sock *sk);
434 enum tcp_synack_type {
439 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
440 struct request_sock *req,
441 struct tcp_fastopen_cookie *foc,
442 enum tcp_synack_type synack_type);
443 int tcp_disconnect(struct sock *sk, int flags);
445 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
446 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
447 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
449 /* From syncookies.c */
450 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
451 struct request_sock *req,
452 struct dst_entry *dst, u32 tsoff);
453 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
455 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
456 #ifdef CONFIG_SYN_COOKIES
458 /* Syncookies use a monotonic timer which increments every 60 seconds.
459 * This counter is used both as a hash input and partially encoded into
460 * the cookie value. A cookie is only validated further if the delta
461 * between the current counter value and the encoded one is less than this,
462 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
463 * the counter advances immediately after a cookie is generated).
465 #define MAX_SYNCOOKIE_AGE 2
466 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
467 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
469 /* syncookies: remember time of last synqueue overflow
470 * But do not dirty this field too often (once per second is enough)
471 * It is racy as we do not hold a lock, but race is very minor.
473 static inline void tcp_synq_overflow(const struct sock *sk)
475 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
476 unsigned long now = jiffies;
478 if (time_after(now, last_overflow + HZ))
479 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
482 /* syncookies: no recent synqueue overflow on this listening socket? */
483 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
485 unsigned long last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
487 return time_after(jiffies, last_overflow + TCP_SYNCOOKIE_VALID);
490 static inline u32 tcp_cookie_time(void)
492 u64 val = get_jiffies_64();
494 do_div(val, TCP_SYNCOOKIE_PERIOD);
498 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
500 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
501 u64 cookie_init_timestamp(struct request_sock *req);
502 bool cookie_timestamp_decode(const struct net *net,
503 struct tcp_options_received *opt);
504 bool cookie_ecn_ok(const struct tcp_options_received *opt,
505 const struct net *net, const struct dst_entry *dst);
507 /* From net/ipv6/syncookies.c */
508 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
510 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
512 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
513 const struct tcphdr *th, u16 *mssp);
514 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
518 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
520 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
521 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
522 void tcp_retransmit_timer(struct sock *sk);
523 void tcp_xmit_retransmit_queue(struct sock *);
524 void tcp_simple_retransmit(struct sock *);
525 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
526 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
528 TCP_FRAG_IN_WRITE_QUEUE,
529 TCP_FRAG_IN_RTX_QUEUE,
531 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
532 struct sk_buff *skb, u32 len,
533 unsigned int mss_now, gfp_t gfp);
535 void tcp_send_probe0(struct sock *);
536 void tcp_send_partial(struct sock *);
537 int tcp_write_wakeup(struct sock *, int mib);
538 void tcp_send_fin(struct sock *sk);
539 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
540 int tcp_send_synack(struct sock *);
541 void tcp_push_one(struct sock *, unsigned int mss_now);
542 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
543 void tcp_send_ack(struct sock *sk);
544 void tcp_send_delayed_ack(struct sock *sk);
545 void tcp_send_loss_probe(struct sock *sk);
546 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
547 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
548 const struct sk_buff *next_skb);
551 void tcp_rearm_rto(struct sock *sk);
552 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
553 void tcp_reset(struct sock *sk);
554 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
555 void tcp_fin(struct sock *sk);
558 void tcp_init_xmit_timers(struct sock *);
559 static inline void tcp_clear_xmit_timers(struct sock *sk)
561 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
564 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
567 inet_csk_clear_xmit_timers(sk);
570 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
571 unsigned int tcp_current_mss(struct sock *sk);
573 /* Bound MSS / TSO packet size with the half of the window */
574 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
578 /* When peer uses tiny windows, there is no use in packetizing
579 * to sub-MSS pieces for the sake of SWS or making sure there
580 * are enough packets in the pipe for fast recovery.
582 * On the other hand, for extremely large MSS devices, handling
583 * smaller than MSS windows in this way does make sense.
585 if (tp->max_window > TCP_MSS_DEFAULT)
586 cutoff = (tp->max_window >> 1);
588 cutoff = tp->max_window;
590 if (cutoff && pktsize > cutoff)
591 return max_t(int, cutoff, 68U - tp->tcp_header_len);
597 void tcp_get_info(struct sock *, struct tcp_info *);
599 /* Read 'sendfile()'-style from a TCP socket */
600 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
601 sk_read_actor_t recv_actor);
603 void tcp_initialize_rcv_mss(struct sock *sk);
605 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
606 int tcp_mss_to_mtu(struct sock *sk, int mss);
607 void tcp_mtup_init(struct sock *sk);
608 void tcp_init_buffer_space(struct sock *sk);
610 static inline void tcp_bound_rto(const struct sock *sk)
612 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
613 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
616 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
618 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
621 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
623 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
624 ntohl(TCP_FLAG_ACK) |
628 static inline void tcp_fast_path_on(struct tcp_sock *tp)
630 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
633 static inline void tcp_fast_path_check(struct sock *sk)
635 struct tcp_sock *tp = tcp_sk(sk);
637 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
639 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
641 tcp_fast_path_on(tp);
644 /* Compute the actual rto_min value */
645 static inline u32 tcp_rto_min(struct sock *sk)
647 const struct dst_entry *dst = __sk_dst_get(sk);
648 u32 rto_min = TCP_RTO_MIN;
650 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
651 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
655 static inline u32 tcp_rto_min_us(struct sock *sk)
657 return jiffies_to_usecs(tcp_rto_min(sk));
660 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
662 return dst_metric_locked(dst, RTAX_CC_ALGO);
665 /* Minimum RTT in usec. ~0 means not available. */
666 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
668 return minmax_get(&tp->rtt_min);
671 /* Compute the actual receive window we are currently advertising.
672 * Rcv_nxt can be after the window if our peer push more data
673 * than the offered window.
675 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
677 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
684 /* Choose a new window, without checks for shrinking, and without
685 * scaling applied to the result. The caller does these things
686 * if necessary. This is a "raw" window selection.
688 u32 __tcp_select_window(struct sock *sk);
690 void tcp_send_window_probe(struct sock *sk);
692 /* TCP uses 32bit jiffies to save some space.
693 * Note that this is different from tcp_time_stamp, which
694 * historically has been the same until linux-4.13.
696 #define tcp_jiffies32 ((u32)jiffies)
699 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
700 * It is no longer tied to jiffies, but to 1 ms clock.
701 * Note: double check if you want to use tcp_jiffies32 instead of this.
703 #define TCP_TS_HZ 1000
705 static inline u64 tcp_clock_ns(void)
707 return local_clock();
710 static inline u64 tcp_clock_us(void)
712 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
715 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
716 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
718 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
721 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
722 static inline u32 tcp_time_stamp_raw(void)
724 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ);
728 /* Refresh 1us clock of a TCP socket,
729 * ensuring monotically increasing values.
731 static inline void tcp_mstamp_refresh(struct tcp_sock *tp)
733 u64 val = tcp_clock_us();
735 if (val > tp->tcp_mstamp)
736 tp->tcp_mstamp = val;
739 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
741 return max_t(s64, t1 - t0, 0);
744 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
746 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ);
750 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
752 #define TCPHDR_FIN 0x01
753 #define TCPHDR_SYN 0x02
754 #define TCPHDR_RST 0x04
755 #define TCPHDR_PSH 0x08
756 #define TCPHDR_ACK 0x10
757 #define TCPHDR_URG 0x20
758 #define TCPHDR_ECE 0x40
759 #define TCPHDR_CWR 0x80
761 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
763 /* This is what the send packet queuing engine uses to pass
764 * TCP per-packet control information to the transmission code.
765 * We also store the host-order sequence numbers in here too.
766 * This is 44 bytes if IPV6 is enabled.
767 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
770 __u32 seq; /* Starting sequence number */
771 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
773 /* Note : tcp_tw_isn is used in input path only
774 * (isn chosen by tcp_timewait_state_process())
776 * tcp_gso_segs/size are used in write queue only,
777 * cf tcp_skb_pcount()/tcp_skb_mss()
785 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
787 __u8 sacked; /* State flags for SACK. */
788 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
789 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
790 #define TCPCB_LOST 0x04 /* SKB is lost */
791 #define TCPCB_TAGBITS 0x07 /* All tag bits */
792 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
793 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
794 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
797 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
798 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
799 eor:1, /* Is skb MSG_EOR marked? */
800 has_rxtstamp:1, /* SKB has a RX timestamp */
802 __u32 ack_seq; /* Sequence number ACK'd */
805 /* There is space for up to 24 bytes */
806 __u32 in_flight:30,/* Bytes in flight at transmit */
807 is_app_limited:1, /* cwnd not fully used? */
809 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
811 /* start of send pipeline phase */
813 /* when we reached the "delivered" count */
814 u64 delivered_mstamp;
815 } tx; /* only used for outgoing skbs */
817 struct inet_skb_parm h4;
818 #if IS_ENABLED(CONFIG_IPV6)
819 struct inet6_skb_parm h6;
821 } header; /* For incoming skbs */
824 struct sock *sk_redir;
830 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
832 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb)
834 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb);
837 #if IS_ENABLED(CONFIG_IPV6)
838 /* This is the variant of inet6_iif() that must be used by TCP,
839 * as TCP moves IP6CB into a different location in skb->cb[]
841 static inline int tcp_v6_iif(const struct sk_buff *skb)
843 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
845 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
848 /* TCP_SKB_CB reference means this can not be used from early demux */
849 static inline int tcp_v6_sdif(const struct sk_buff *skb)
851 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
852 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
853 return TCP_SKB_CB(skb)->header.h6.iif;
859 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
861 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
862 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
863 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
869 /* TCP_SKB_CB reference means this can not be used from early demux */
870 static inline int tcp_v4_sdif(struct sk_buff *skb)
872 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
873 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
874 return TCP_SKB_CB(skb)->header.h4.iif;
879 /* Due to TSO, an SKB can be composed of multiple actual
880 * packets. To keep these tracked properly, we use this.
882 static inline int tcp_skb_pcount(const struct sk_buff *skb)
884 return TCP_SKB_CB(skb)->tcp_gso_segs;
887 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
889 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
892 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
894 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
897 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
898 static inline int tcp_skb_mss(const struct sk_buff *skb)
900 return TCP_SKB_CB(skb)->tcp_gso_size;
903 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
905 return likely(!TCP_SKB_CB(skb)->eor);
908 /* Events passed to congestion control interface */
910 CA_EVENT_TX_START, /* first transmit when no packets in flight */
911 CA_EVENT_CWND_RESTART, /* congestion window restart */
912 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
913 CA_EVENT_LOSS, /* loss timeout */
914 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
915 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
918 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
919 enum tcp_ca_ack_event_flags {
920 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
921 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
922 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
926 * Interface for adding new TCP congestion control handlers
928 #define TCP_CA_NAME_MAX 16
929 #define TCP_CA_MAX 128
930 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
932 #define TCP_CA_UNSPEC 0
934 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
935 #define TCP_CONG_NON_RESTRICTED 0x1
936 /* Requires ECN/ECT set on all packets */
937 #define TCP_CONG_NEEDS_ECN 0x2
947 /* A rate sample measures the number of (original/retransmitted) data
948 * packets delivered "delivered" over an interval of time "interval_us".
949 * The tcp_rate.c code fills in the rate sample, and congestion
950 * control modules that define a cong_control function to run at the end
951 * of ACK processing can optionally chose to consult this sample when
952 * setting cwnd and pacing rate.
953 * A sample is invalid if "delivered" or "interval_us" is negative.
956 u64 prior_mstamp; /* starting timestamp for interval */
957 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
958 s32 delivered; /* number of packets delivered over interval */
959 long interval_us; /* time for tp->delivered to incr "delivered" */
960 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
961 int losses; /* number of packets marked lost upon ACK */
962 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
963 u32 prior_in_flight; /* in flight before this ACK */
964 bool is_app_limited; /* is sample from packet with bubble in pipe? */
965 bool is_retrans; /* is sample from retransmission? */
966 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
969 struct tcp_congestion_ops {
970 struct list_head list;
974 /* initialize private data (optional) */
975 void (*init)(struct sock *sk);
976 /* cleanup private data (optional) */
977 void (*release)(struct sock *sk);
979 /* return slow start threshold (required) */
980 u32 (*ssthresh)(struct sock *sk);
981 /* do new cwnd calculation (required) */
982 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
983 /* call before changing ca_state (optional) */
984 void (*set_state)(struct sock *sk, u8 new_state);
985 /* call when cwnd event occurs (optional) */
986 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
987 /* call when ack arrives (optional) */
988 void (*in_ack_event)(struct sock *sk, u32 flags);
989 /* new value of cwnd after loss (required) */
990 u32 (*undo_cwnd)(struct sock *sk);
991 /* hook for packet ack accounting (optional) */
992 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
993 /* override sysctl_tcp_min_tso_segs */
994 u32 (*min_tso_segs)(struct sock *sk);
995 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
996 u32 (*sndbuf_expand)(struct sock *sk);
997 /* call when packets are delivered to update cwnd and pacing rate,
998 * after all the ca_state processing. (optional)
1000 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1001 /* get info for inet_diag (optional) */
1002 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1003 union tcp_cc_info *info);
1005 char name[TCP_CA_NAME_MAX];
1006 struct module *owner;
1009 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1010 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1012 void tcp_assign_congestion_control(struct sock *sk);
1013 void tcp_init_congestion_control(struct sock *sk);
1014 void tcp_cleanup_congestion_control(struct sock *sk);
1015 int tcp_set_default_congestion_control(struct net *net, const char *name);
1016 void tcp_get_default_congestion_control(struct net *net, char *name);
1017 void tcp_get_available_congestion_control(char *buf, size_t len);
1018 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1019 int tcp_set_allowed_congestion_control(char *allowed);
1020 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit);
1021 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1022 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1024 u32 tcp_reno_ssthresh(struct sock *sk);
1025 u32 tcp_reno_undo_cwnd(struct sock *sk);
1026 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1027 extern struct tcp_congestion_ops tcp_reno;
1029 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1030 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1032 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1034 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1040 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1042 const struct inet_connection_sock *icsk = inet_csk(sk);
1044 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1047 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1049 struct inet_connection_sock *icsk = inet_csk(sk);
1051 if (icsk->icsk_ca_ops->set_state)
1052 icsk->icsk_ca_ops->set_state(sk, ca_state);
1053 icsk->icsk_ca_state = ca_state;
1056 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1058 const struct inet_connection_sock *icsk = inet_csk(sk);
1060 if (icsk->icsk_ca_ops->cwnd_event)
1061 icsk->icsk_ca_ops->cwnd_event(sk, event);
1064 /* From tcp_rate.c */
1065 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1066 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1067 struct rate_sample *rs);
1068 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1069 bool is_sack_reneg, struct rate_sample *rs);
1070 void tcp_rate_check_app_limited(struct sock *sk);
1072 /* These functions determine how the current flow behaves in respect of SACK
1073 * handling. SACK is negotiated with the peer, and therefore it can vary
1074 * between different flows.
1076 * tcp_is_sack - SACK enabled
1077 * tcp_is_reno - No SACK
1079 static inline int tcp_is_sack(const struct tcp_sock *tp)
1081 return tp->rx_opt.sack_ok;
1084 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1086 return !tcp_is_sack(tp);
1089 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1091 return tp->sacked_out + tp->lost_out;
1094 /* This determines how many packets are "in the network" to the best
1095 * of our knowledge. In many cases it is conservative, but where
1096 * detailed information is available from the receiver (via SACK
1097 * blocks etc.) we can make more aggressive calculations.
1099 * Use this for decisions involving congestion control, use just
1100 * tp->packets_out to determine if the send queue is empty or not.
1102 * Read this equation as:
1104 * "Packets sent once on transmission queue" MINUS
1105 * "Packets left network, but not honestly ACKed yet" PLUS
1106 * "Packets fast retransmitted"
1108 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1110 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1113 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1115 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1117 return tp->snd_cwnd < tp->snd_ssthresh;
1120 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1122 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1125 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1127 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1128 (1 << inet_csk(sk)->icsk_ca_state);
1131 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1132 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1135 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1137 const struct tcp_sock *tp = tcp_sk(sk);
1139 if (tcp_in_cwnd_reduction(sk))
1140 return tp->snd_ssthresh;
1142 return max(tp->snd_ssthresh,
1143 ((tp->snd_cwnd >> 1) +
1144 (tp->snd_cwnd >> 2)));
1147 /* Use define here intentionally to get WARN_ON location shown at the caller */
1148 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1150 void tcp_enter_cwr(struct sock *sk);
1151 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1153 /* The maximum number of MSS of available cwnd for which TSO defers
1154 * sending if not using sysctl_tcp_tso_win_divisor.
1156 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1161 /* Returns end sequence number of the receiver's advertised window */
1162 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1164 return tp->snd_una + tp->snd_wnd;
1167 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1168 * flexible approach. The RFC suggests cwnd should not be raised unless
1169 * it was fully used previously. And that's exactly what we do in
1170 * congestion avoidance mode. But in slow start we allow cwnd to grow
1171 * as long as the application has used half the cwnd.
1173 * cwnd is 10 (IW10), but application sends 9 frames.
1174 * We allow cwnd to reach 18 when all frames are ACKed.
1175 * This check is safe because it's as aggressive as slow start which already
1176 * risks 100% overshoot. The advantage is that we discourage application to
1177 * either send more filler packets or data to artificially blow up the cwnd
1178 * usage, and allow application-limited process to probe bw more aggressively.
1180 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1182 const struct tcp_sock *tp = tcp_sk(sk);
1184 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1185 if (tcp_in_slow_start(tp))
1186 return tp->snd_cwnd < 2 * tp->max_packets_out;
1188 return tp->is_cwnd_limited;
1191 /* Something is really bad, we could not queue an additional packet,
1192 * because qdisc is full or receiver sent a 0 window.
1193 * We do not want to add fuel to the fire, or abort too early,
1194 * so make sure the timer we arm now is at least 200ms in the future,
1195 * regardless of current icsk_rto value (as it could be ~2ms)
1197 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1199 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1202 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1203 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1204 unsigned long max_when)
1206 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1208 return (unsigned long)min_t(u64, when, max_when);
1211 static inline void tcp_check_probe_timer(struct sock *sk)
1213 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1214 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1215 tcp_probe0_base(sk), TCP_RTO_MAX);
1218 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1223 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1229 * Calculate(/check) TCP checksum
1231 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1232 __be32 daddr, __wsum base)
1234 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1237 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1239 return __skb_checksum_complete(skb);
1242 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1244 return !skb_csum_unnecessary(skb) &&
1245 __tcp_checksum_complete(skb);
1248 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1249 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1254 static const char *statename[]={
1255 "Unused","Established","Syn Sent","Syn Recv",
1256 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1257 "Close Wait","Last ACK","Listen","Closing"
1260 void tcp_set_state(struct sock *sk, int state);
1262 void tcp_done(struct sock *sk);
1264 int tcp_abort(struct sock *sk, int err);
1266 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1269 rx_opt->num_sacks = 0;
1272 u32 tcp_default_init_rwnd(u32 mss);
1273 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1275 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1277 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1278 struct tcp_sock *tp = tcp_sk(sk);
1281 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1282 ca_ops->cong_control)
1284 delta = tcp_jiffies32 - tp->lsndtime;
1285 if (delta > inet_csk(sk)->icsk_rto)
1286 tcp_cwnd_restart(sk, delta);
1289 /* Determine a window scaling and initial window to offer. */
1290 void tcp_select_initial_window(const struct sock *sk, int __space,
1291 __u32 mss, __u32 *rcv_wnd,
1292 __u32 *window_clamp, int wscale_ok,
1293 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1295 static inline int tcp_win_from_space(const struct sock *sk, int space)
1297 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1299 return tcp_adv_win_scale <= 0 ?
1300 (space>>(-tcp_adv_win_scale)) :
1301 space - (space>>tcp_adv_win_scale);
1304 /* Note: caller must be prepared to deal with negative returns */
1305 static inline int tcp_space(const struct sock *sk)
1307 return tcp_win_from_space(sk, sk->sk_rcvbuf -
1308 atomic_read(&sk->sk_rmem_alloc));
1311 static inline int tcp_full_space(const struct sock *sk)
1313 return tcp_win_from_space(sk, sk->sk_rcvbuf);
1316 extern void tcp_openreq_init_rwin(struct request_sock *req,
1317 const struct sock *sk_listener,
1318 const struct dst_entry *dst);
1320 void tcp_enter_memory_pressure(struct sock *sk);
1321 void tcp_leave_memory_pressure(struct sock *sk);
1323 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1325 struct net *net = sock_net((struct sock *)tp);
1327 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1330 static inline int keepalive_time_when(const struct tcp_sock *tp)
1332 struct net *net = sock_net((struct sock *)tp);
1334 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1337 static inline int keepalive_probes(const struct tcp_sock *tp)
1339 struct net *net = sock_net((struct sock *)tp);
1341 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1344 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1346 const struct inet_connection_sock *icsk = &tp->inet_conn;
1348 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1349 tcp_jiffies32 - tp->rcv_tstamp);
1352 static inline int tcp_fin_time(const struct sock *sk)
1354 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1355 const int rto = inet_csk(sk)->icsk_rto;
1357 if (fin_timeout < (rto << 2) - (rto >> 1))
1358 fin_timeout = (rto << 2) - (rto >> 1);
1363 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1366 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1368 if (unlikely(get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS))
1371 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1372 * then following tcp messages have valid values. Ignore 0 value,
1373 * or else 'negative' tsval might forbid us to accept their packets.
1375 if (!rx_opt->ts_recent)
1380 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1383 if (tcp_paws_check(rx_opt, 0))
1386 /* RST segments are not recommended to carry timestamp,
1387 and, if they do, it is recommended to ignore PAWS because
1388 "their cleanup function should take precedence over timestamps."
1389 Certainly, it is mistake. It is necessary to understand the reasons
1390 of this constraint to relax it: if peer reboots, clock may go
1391 out-of-sync and half-open connections will not be reset.
1392 Actually, the problem would be not existing if all
1393 the implementations followed draft about maintaining clock
1394 via reboots. Linux-2.2 DOES NOT!
1396 However, we can relax time bounds for RST segments to MSL.
1398 if (rst && get_seconds() >= rx_opt->ts_recent_stamp + TCP_PAWS_MSL)
1403 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1404 int mib_idx, u32 *last_oow_ack_time);
1406 static inline void tcp_mib_init(struct net *net)
1409 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1410 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1411 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1412 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1416 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1418 tp->lost_skb_hint = NULL;
1421 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1423 tcp_clear_retrans_hints_partial(tp);
1424 tp->retransmit_skb_hint = NULL;
1427 union tcp_md5_addr {
1429 #if IS_ENABLED(CONFIG_IPV6)
1434 /* - key database */
1435 struct tcp_md5sig_key {
1436 struct hlist_node node;
1438 u8 family; /* AF_INET or AF_INET6 */
1439 union tcp_md5_addr addr;
1441 u8 key[TCP_MD5SIG_MAXKEYLEN];
1442 struct rcu_head rcu;
1446 struct tcp_md5sig_info {
1447 struct hlist_head head;
1448 struct rcu_head rcu;
1451 /* - pseudo header */
1452 struct tcp4_pseudohdr {
1460 struct tcp6_pseudohdr {
1461 struct in6_addr saddr;
1462 struct in6_addr daddr;
1464 __be32 protocol; /* including padding */
1467 union tcp_md5sum_block {
1468 struct tcp4_pseudohdr ip4;
1469 #if IS_ENABLED(CONFIG_IPV6)
1470 struct tcp6_pseudohdr ip6;
1474 /* - pool: digest algorithm, hash description and scratch buffer */
1475 struct tcp_md5sig_pool {
1476 struct ahash_request *md5_req;
1481 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1482 const struct sock *sk, const struct sk_buff *skb);
1483 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1484 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1486 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1487 int family, u8 prefixlen);
1488 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1489 const struct sock *addr_sk);
1491 #ifdef CONFIG_TCP_MD5SIG
1492 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1493 const union tcp_md5_addr *addr,
1495 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1497 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1498 const union tcp_md5_addr *addr,
1503 #define tcp_twsk_md5_key(twsk) NULL
1506 bool tcp_alloc_md5sig_pool(void);
1508 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1509 static inline void tcp_put_md5sig_pool(void)
1514 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1515 unsigned int header_len);
1516 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1517 const struct tcp_md5sig_key *key);
1519 /* From tcp_fastopen.c */
1520 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1521 struct tcp_fastopen_cookie *cookie);
1522 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1523 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1525 struct tcp_fastopen_request {
1526 /* Fast Open cookie. Size 0 means a cookie request */
1527 struct tcp_fastopen_cookie cookie;
1528 struct msghdr *data; /* data in MSG_FASTOPEN */
1530 int copied; /* queued in tcp_connect() */
1532 void tcp_free_fastopen_req(struct tcp_sock *tp);
1533 void tcp_fastopen_destroy_cipher(struct sock *sk);
1534 void tcp_fastopen_ctx_destroy(struct net *net);
1535 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1536 void *key, unsigned int len);
1537 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1538 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1539 struct request_sock *req,
1540 struct tcp_fastopen_cookie *foc,
1541 const struct dst_entry *dst);
1542 void tcp_fastopen_init_key_once(struct net *net);
1543 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1544 struct tcp_fastopen_cookie *cookie);
1545 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1546 #define TCP_FASTOPEN_KEY_LENGTH 16
1548 /* Fastopen key context */
1549 struct tcp_fastopen_context {
1550 struct crypto_cipher *tfm;
1551 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1552 struct rcu_head rcu;
1555 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1556 void tcp_fastopen_active_disable(struct sock *sk);
1557 bool tcp_fastopen_active_should_disable(struct sock *sk);
1558 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1559 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1561 /* Latencies incurred by various limits for a sender. They are
1562 * chronograph-like stats that are mutually exclusive.
1566 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1567 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1568 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1572 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1573 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1575 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1576 * the same memory storage than skb->destructor/_skb_refdst
1578 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1580 skb->destructor = NULL;
1581 skb->_skb_refdst = 0UL;
1584 #define tcp_skb_tsorted_save(skb) { \
1585 unsigned long _save = skb->_skb_refdst; \
1586 skb->_skb_refdst = 0UL;
1588 #define tcp_skb_tsorted_restore(skb) \
1589 skb->_skb_refdst = _save; \
1592 void tcp_write_queue_purge(struct sock *sk);
1594 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1596 return skb_rb_first(&sk->tcp_rtx_queue);
1599 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1601 return skb_peek(&sk->sk_write_queue);
1604 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1606 return skb_peek_tail(&sk->sk_write_queue);
1609 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1610 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1612 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1614 return skb_peek(&sk->sk_write_queue);
1617 static inline bool tcp_skb_is_last(const struct sock *sk,
1618 const struct sk_buff *skb)
1620 return skb_queue_is_last(&sk->sk_write_queue, skb);
1623 static inline bool tcp_write_queue_empty(const struct sock *sk)
1625 return skb_queue_empty(&sk->sk_write_queue);
1628 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1630 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1633 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1635 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1638 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1640 if (tcp_write_queue_empty(sk))
1641 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1644 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1646 __skb_queue_tail(&sk->sk_write_queue, skb);
1649 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1651 __tcp_add_write_queue_tail(sk, skb);
1653 /* Queue it, remembering where we must start sending. */
1654 if (sk->sk_write_queue.next == skb)
1655 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1658 /* Insert new before skb on the write queue of sk. */
1659 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1660 struct sk_buff *skb,
1663 __skb_queue_before(&sk->sk_write_queue, skb, new);
1666 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1668 tcp_skb_tsorted_anchor_cleanup(skb);
1669 __skb_unlink(skb, &sk->sk_write_queue);
1672 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1674 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1676 tcp_skb_tsorted_anchor_cleanup(skb);
1677 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1680 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1682 list_del(&skb->tcp_tsorted_anchor);
1683 tcp_rtx_queue_unlink(skb, sk);
1684 sk_wmem_free_skb(sk, skb);
1687 static inline void tcp_push_pending_frames(struct sock *sk)
1689 if (tcp_send_head(sk)) {
1690 struct tcp_sock *tp = tcp_sk(sk);
1692 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1696 /* Start sequence of the skb just after the highest skb with SACKed
1697 * bit, valid only if sacked_out > 0 or when the caller has ensured
1698 * validity by itself.
1700 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1702 if (!tp->sacked_out)
1705 if (tp->highest_sack == NULL)
1708 return TCP_SKB_CB(tp->highest_sack)->seq;
1711 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1713 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1716 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1718 return tcp_sk(sk)->highest_sack;
1721 static inline void tcp_highest_sack_reset(struct sock *sk)
1723 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1726 /* Called when old skb is about to be deleted and replaced by new skb */
1727 static inline void tcp_highest_sack_replace(struct sock *sk,
1728 struct sk_buff *old,
1729 struct sk_buff *new)
1731 if (old == tcp_highest_sack(sk))
1732 tcp_sk(sk)->highest_sack = new;
1735 /* This helper checks if socket has IP_TRANSPARENT set */
1736 static inline bool inet_sk_transparent(const struct sock *sk)
1738 switch (sk->sk_state) {
1740 return inet_twsk(sk)->tw_transparent;
1741 case TCP_NEW_SYN_RECV:
1742 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1744 return inet_sk(sk)->transparent;
1747 /* Determines whether this is a thin stream (which may suffer from
1748 * increased latency). Used to trigger latency-reducing mechanisms.
1750 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1752 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1756 enum tcp_seq_states {
1757 TCP_SEQ_STATE_LISTENING,
1758 TCP_SEQ_STATE_ESTABLISHED,
1761 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1762 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1763 void tcp_seq_stop(struct seq_file *seq, void *v);
1765 struct tcp_seq_afinfo {
1769 struct tcp_iter_state {
1770 struct seq_net_private p;
1771 enum tcp_seq_states state;
1772 struct sock *syn_wait_sk;
1773 int bucket, offset, sbucket, num;
1777 extern struct request_sock_ops tcp_request_sock_ops;
1778 extern struct request_sock_ops tcp6_request_sock_ops;
1780 void tcp_v4_destroy_sock(struct sock *sk);
1782 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1783 netdev_features_t features);
1784 struct sk_buff **tcp_gro_receive(struct sk_buff **head, struct sk_buff *skb);
1785 int tcp_gro_complete(struct sk_buff *skb);
1787 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1789 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1791 struct net *net = sock_net((struct sock *)tp);
1792 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1795 static inline bool tcp_stream_memory_free(const struct sock *sk)
1797 const struct tcp_sock *tp = tcp_sk(sk);
1798 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1800 return notsent_bytes < tcp_notsent_lowat(tp);
1803 #ifdef CONFIG_PROC_FS
1804 int tcp4_proc_init(void);
1805 void tcp4_proc_exit(void);
1808 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1809 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1810 const struct tcp_request_sock_ops *af_ops,
1811 struct sock *sk, struct sk_buff *skb);
1813 /* TCP af-specific functions */
1814 struct tcp_sock_af_ops {
1815 #ifdef CONFIG_TCP_MD5SIG
1816 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1817 const struct sock *addr_sk);
1818 int (*calc_md5_hash)(char *location,
1819 const struct tcp_md5sig_key *md5,
1820 const struct sock *sk,
1821 const struct sk_buff *skb);
1822 int (*md5_parse)(struct sock *sk,
1824 char __user *optval,
1829 struct tcp_request_sock_ops {
1831 #ifdef CONFIG_TCP_MD5SIG
1832 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1833 const struct sock *addr_sk);
1834 int (*calc_md5_hash) (char *location,
1835 const struct tcp_md5sig_key *md5,
1836 const struct sock *sk,
1837 const struct sk_buff *skb);
1839 void (*init_req)(struct request_sock *req,
1840 const struct sock *sk_listener,
1841 struct sk_buff *skb);
1842 #ifdef CONFIG_SYN_COOKIES
1843 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1846 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1847 const struct request_sock *req);
1848 u32 (*init_seq)(const struct sk_buff *skb);
1849 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
1850 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1851 struct flowi *fl, struct request_sock *req,
1852 struct tcp_fastopen_cookie *foc,
1853 enum tcp_synack_type synack_type);
1856 #ifdef CONFIG_SYN_COOKIES
1857 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1858 const struct sock *sk, struct sk_buff *skb,
1861 tcp_synq_overflow(sk);
1862 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1863 return ops->cookie_init_seq(skb, mss);
1866 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1867 const struct sock *sk, struct sk_buff *skb,
1874 int tcpv4_offload_init(void);
1876 void tcp_v4_init(void);
1877 void tcp_init(void);
1879 /* tcp_recovery.c */
1880 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
1881 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
1882 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
1884 extern void tcp_rack_mark_lost(struct sock *sk);
1885 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
1887 extern void tcp_rack_reo_timeout(struct sock *sk);
1888 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
1890 /* At how many usecs into the future should the RTO fire? */
1891 static inline s64 tcp_rto_delta_us(const struct sock *sk)
1893 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
1894 u32 rto = inet_csk(sk)->icsk_rto;
1895 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto);
1897 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
1901 * Save and compile IPv4 options, return a pointer to it
1903 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
1904 struct sk_buff *skb)
1906 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1907 struct ip_options_rcu *dopt = NULL;
1910 int opt_size = sizeof(*dopt) + opt->optlen;
1912 dopt = kmalloc(opt_size, GFP_ATOMIC);
1913 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
1921 /* locally generated TCP pure ACKs have skb->truesize == 2
1922 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1923 * This is much faster than dissecting the packet to find out.
1924 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1926 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1928 return skb->truesize == 2;
1931 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
1936 static inline int tcp_inq(struct sock *sk)
1938 struct tcp_sock *tp = tcp_sk(sk);
1941 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1943 } else if (sock_flag(sk, SOCK_URGINLINE) ||
1945 before(tp->urg_seq, tp->copied_seq) ||
1946 !before(tp->urg_seq, tp->rcv_nxt)) {
1948 answ = tp->rcv_nxt - tp->copied_seq;
1950 /* Subtract 1, if FIN was received */
1951 if (answ && sock_flag(sk, SOCK_DONE))
1954 answ = tp->urg_seq - tp->copied_seq;
1960 int tcp_peek_len(struct socket *sock);
1962 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
1966 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
1967 tp->segs_in += segs_in;
1968 if (skb->len > tcp_hdrlen(skb))
1969 tp->data_segs_in += segs_in;
1973 * TCP listen path runs lockless.
1974 * We forced "struct sock" to be const qualified to make sure
1975 * we don't modify one of its field by mistake.
1976 * Here, we increment sk_drops which is an atomic_t, so we can safely
1977 * make sock writable again.
1979 static inline void tcp_listendrop(const struct sock *sk)
1981 atomic_inc(&((struct sock *)sk)->sk_drops);
1982 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
1985 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
1988 * Interface for adding Upper Level Protocols over TCP
1991 #define TCP_ULP_NAME_MAX 16
1992 #define TCP_ULP_MAX 128
1993 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2000 struct tcp_ulp_ops {
2001 struct list_head list;
2003 /* initialize ulp */
2004 int (*init)(struct sock *sk);
2006 void (*release)(struct sock *sk);
2009 char name[TCP_ULP_NAME_MAX];
2011 struct module *owner;
2013 int tcp_register_ulp(struct tcp_ulp_ops *type);
2014 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2015 int tcp_set_ulp(struct sock *sk, const char *name);
2016 int tcp_set_ulp_id(struct sock *sk, const int ulp);
2017 void tcp_get_available_ulp(char *buf, size_t len);
2018 void tcp_cleanup_ulp(struct sock *sk);
2020 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2021 * is < 0, then the BPF op failed (for example if the loaded BPF
2022 * program does not support the chosen operation or there is no BPF
2026 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2028 struct bpf_sock_ops_kern sock_ops;
2031 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2032 if (sk_fullsock(sk)) {
2033 sock_ops.is_fullsock = 1;
2034 sock_owned_by_me(sk);
2040 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2042 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2044 ret = sock_ops.reply;
2050 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2052 u32 args[2] = {arg1, arg2};
2054 return tcp_call_bpf(sk, op, 2, args);
2057 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2060 u32 args[3] = {arg1, arg2, arg3};
2062 return tcp_call_bpf(sk, op, 3, args);
2066 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2071 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2076 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2084 static inline u32 tcp_timeout_init(struct sock *sk)
2088 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2091 timeout = TCP_TIMEOUT_INIT;
2095 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2099 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2106 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2108 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2111 #if IS_ENABLED(CONFIG_SMC)
2112 extern struct static_key_false tcp_have_smc;
2115 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2116 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2117 void (*cad)(struct sock *sk, u32 ack_seq));
2118 void clean_acked_data_disable(struct inet_connection_sock *icsk);