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>
39 #include <net/sock_reuseport.h>
43 #include <net/tcp_states.h>
44 #include <net/inet_ecn.h>
47 #include <linux/seq_file.h>
48 #include <linux/memcontrol.h>
49 #include <linux/bpf-cgroup.h>
51 extern struct inet_hashinfo tcp_hashinfo;
53 extern struct percpu_counter tcp_orphan_count;
54 void tcp_time_wait(struct sock *sk, int state, int timeo);
56 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER)
57 #define MAX_TCP_OPTION_SPACE 40
58 #define TCP_MIN_SND_MSS 48
59 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
62 * Never offer a window over 32767 without using window scaling. Some
63 * poor stacks do signed 16bit maths!
65 #define MAX_TCP_WINDOW 32767U
67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68 #define TCP_MIN_MSS 88U
70 /* The least MTU to use for probing */
71 #define TCP_BASE_MSS 1024
73 /* probing interval, default to 10 minutes as per RFC4821 */
74 #define TCP_PROBE_INTERVAL 600
76 /* Specify interval when tcp mtu probing will stop */
77 #define TCP_PROBE_THRESHOLD 8
79 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
80 #define TCP_FASTRETRANS_THRESH 3
82 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
83 #define TCP_MAX_QUICKACKS 16U
85 /* Maximal number of window scale according to RFC1323 */
86 #define TCP_MAX_WSCALE 14U
89 #define TCP_URG_VALID 0x0100
90 #define TCP_URG_NOTYET 0x0200
91 #define TCP_URG_READ 0x0400
93 #define TCP_RETR1 3 /*
94 * This is how many retries it does before it
95 * tries to figure out if the gateway is
96 * down. Minimal RFC value is 3; it corresponds
97 * to ~3sec-8min depending on RTO.
100 #define TCP_RETR2 15 /*
101 * This should take at least
102 * 90 minutes to time out.
103 * RFC1122 says that the limit is 100 sec.
104 * 15 is ~13-30min depending on RTO.
107 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
108 * when active opening a connection.
109 * RFC1122 says the minimum retry MUST
110 * be at least 180secs. Nevertheless
111 * this value is corresponding to
112 * 63secs of retransmission with the
113 * current initial RTO.
116 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
117 * when passive opening a connection.
118 * This is corresponding to 31secs of
119 * retransmission with the current
123 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
124 * state, about 60 seconds */
125 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
126 /* BSD style FIN_WAIT2 deadlock breaker.
127 * It used to be 3min, new value is 60sec,
128 * to combine FIN-WAIT-2 timeout with
132 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
134 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
135 #define TCP_ATO_MIN ((unsigned)(HZ/25))
137 #define TCP_DELACK_MIN 4U
138 #define TCP_ATO_MIN 4U
140 #define TCP_RTO_MAX ((unsigned)(120*HZ))
141 #define TCP_RTO_MIN ((unsigned)(HZ/5))
142 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
143 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
144 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
145 * used as a fallback RTO for the
146 * initial data transmission if no
147 * valid RTT sample has been acquired,
148 * most likely due to retrans in 3WHS.
151 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
152 * for local resources.
154 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
155 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
156 #define TCP_KEEPALIVE_INTVL (75*HZ)
158 #define MAX_TCP_KEEPIDLE 32767
159 #define MAX_TCP_KEEPINTVL 32767
160 #define MAX_TCP_KEEPCNT 127
161 #define MAX_TCP_SYNCNT 127
163 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
165 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
166 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
167 * after this time. It should be equal
168 * (or greater than) TCP_TIMEWAIT_LEN
169 * to provide reliability equal to one
170 * provided by timewait state.
172 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
173 * timestamps. It must be less than
174 * minimal timewait lifetime.
180 #define TCPOPT_NOP 1 /* Padding */
181 #define TCPOPT_EOL 0 /* End of options */
182 #define TCPOPT_MSS 2 /* Segment size negotiating */
183 #define TCPOPT_WINDOW 3 /* Window scaling */
184 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
185 #define TCPOPT_SACK 5 /* SACK Block */
186 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
187 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
188 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
189 #define TCPOPT_EXP 254 /* Experimental */
190 /* Magic number to be after the option value for sharing TCP
191 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
193 #define TCPOPT_FASTOPEN_MAGIC 0xF989
194 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9
200 #define TCPOLEN_MSS 4
201 #define TCPOLEN_WINDOW 3
202 #define TCPOLEN_SACK_PERM 2
203 #define TCPOLEN_TIMESTAMP 10
204 #define TCPOLEN_MD5SIG 18
205 #define TCPOLEN_FASTOPEN_BASE 2
206 #define TCPOLEN_EXP_FASTOPEN_BASE 4
207 #define TCPOLEN_EXP_SMC_BASE 6
209 /* But this is what stacks really send out. */
210 #define TCPOLEN_TSTAMP_ALIGNED 12
211 #define TCPOLEN_WSCALE_ALIGNED 4
212 #define TCPOLEN_SACKPERM_ALIGNED 4
213 #define TCPOLEN_SACK_BASE 2
214 #define TCPOLEN_SACK_BASE_ALIGNED 4
215 #define TCPOLEN_SACK_PERBLOCK 8
216 #define TCPOLEN_MD5SIG_ALIGNED 20
217 #define TCPOLEN_MSS_ALIGNED 4
218 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
220 /* Flags in tp->nonagle */
221 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
222 #define TCP_NAGLE_CORK 2 /* Socket is corked */
223 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
225 /* TCP thin-stream limits */
226 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
228 /* TCP initial congestion window as per rfc6928 */
229 #define TCP_INIT_CWND 10
231 /* Bit Flags for sysctl_tcp_fastopen */
232 #define TFO_CLIENT_ENABLE 1
233 #define TFO_SERVER_ENABLE 2
234 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
236 /* Accept SYN data w/o any cookie option */
237 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
239 /* Force enable TFO on all listeners, i.e., not requiring the
240 * TCP_FASTOPEN socket option.
242 #define TFO_SERVER_WO_SOCKOPT1 0x400
245 /* sysctl variables for tcp */
246 extern int sysctl_tcp_max_orphans;
247 extern long sysctl_tcp_mem[3];
249 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
250 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
251 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */
253 extern atomic_long_t tcp_memory_allocated;
254 extern struct percpu_counter tcp_sockets_allocated;
255 extern unsigned long tcp_memory_pressure;
257 /* optimized version of sk_under_memory_pressure() for TCP sockets */
258 static inline bool tcp_under_memory_pressure(const struct sock *sk)
260 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
261 mem_cgroup_under_socket_pressure(sk->sk_memcg))
264 return READ_ONCE(tcp_memory_pressure);
267 * The next routines deal with comparing 32 bit unsigned ints
268 * and worry about wraparound (automatic with unsigned arithmetic).
271 static inline bool before(__u32 seq1, __u32 seq2)
273 return (__s32)(seq1-seq2) < 0;
275 #define after(seq2, seq1) before(seq1, seq2)
277 /* is s2<=s1<=s3 ? */
278 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
280 return seq3 - seq2 >= seq1 - seq2;
283 static inline bool tcp_out_of_memory(struct sock *sk)
285 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
286 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
291 void sk_forced_mem_schedule(struct sock *sk, int size);
293 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
295 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
296 int orphans = percpu_counter_read_positive(ocp);
298 if (orphans << shift > sysctl_tcp_max_orphans) {
299 orphans = percpu_counter_sum_positive(ocp);
300 if (orphans << shift > sysctl_tcp_max_orphans)
306 bool tcp_check_oom(struct sock *sk, int shift);
309 extern struct proto tcp_prot;
311 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
312 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
313 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
314 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
316 void tcp_tasklet_init(void);
318 void tcp_v4_err(struct sk_buff *skb, u32);
320 void tcp_shutdown(struct sock *sk, int how);
322 int tcp_v4_early_demux(struct sk_buff *skb);
323 int tcp_v4_rcv(struct sk_buff *skb);
325 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
326 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
327 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
328 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
330 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
331 size_t size, int flags);
332 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
333 size_t size, int flags);
334 void tcp_release_cb(struct sock *sk);
335 void tcp_wfree(struct sk_buff *skb);
336 void tcp_write_timer_handler(struct sock *sk);
337 void tcp_delack_timer_handler(struct sock *sk);
338 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
339 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
340 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
341 void tcp_rcv_space_adjust(struct sock *sk);
342 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
343 void tcp_twsk_destructor(struct sock *sk);
344 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
345 struct pipe_inode_info *pipe, size_t len,
348 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
349 static inline void tcp_dec_quickack_mode(struct sock *sk,
350 const unsigned int pkts)
352 struct inet_connection_sock *icsk = inet_csk(sk);
354 if (icsk->icsk_ack.quick) {
355 if (pkts >= icsk->icsk_ack.quick) {
356 icsk->icsk_ack.quick = 0;
357 /* Leaving quickack mode we deflate ATO. */
358 icsk->icsk_ack.ato = TCP_ATO_MIN;
360 icsk->icsk_ack.quick -= pkts;
365 #define TCP_ECN_QUEUE_CWR 2
366 #define TCP_ECN_DEMAND_CWR 4
367 #define TCP_ECN_SEEN 8
377 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
379 const struct tcphdr *th);
380 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
381 struct request_sock *req, bool fastopen,
383 int tcp_child_process(struct sock *parent, struct sock *child,
384 struct sk_buff *skb);
385 void tcp_enter_loss(struct sock *sk);
386 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
387 void tcp_clear_retrans(struct tcp_sock *tp);
388 void tcp_update_metrics(struct sock *sk);
389 void tcp_init_metrics(struct sock *sk);
390 void tcp_metrics_init(void);
391 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
392 void tcp_close(struct sock *sk, long timeout);
393 void tcp_init_sock(struct sock *sk);
394 void tcp_init_transfer(struct sock *sk, int bpf_op);
395 __poll_t tcp_poll(struct file *file, struct socket *sock,
396 struct poll_table_struct *wait);
397 int tcp_getsockopt(struct sock *sk, int level, int optname,
398 char __user *optval, int __user *optlen);
399 int tcp_setsockopt(struct sock *sk, int level, int optname,
400 char __user *optval, unsigned int optlen);
401 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
402 char __user *optval, int __user *optlen);
403 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
404 char __user *optval, unsigned int optlen);
405 void tcp_set_keepalive(struct sock *sk, int val);
406 void tcp_syn_ack_timeout(const struct request_sock *req);
407 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
408 int flags, int *addr_len);
409 int tcp_set_rcvlowat(struct sock *sk, int val);
410 void tcp_data_ready(struct sock *sk);
411 int tcp_mmap(struct file *file, struct socket *sock,
412 struct vm_area_struct *vma);
413 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
414 struct tcp_options_received *opt_rx,
415 int estab, struct tcp_fastopen_cookie *foc);
416 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
419 * TCP v4 functions exported for the inet6 API
422 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
423 void tcp_v4_mtu_reduced(struct sock *sk);
424 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
425 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
426 struct sock *tcp_create_openreq_child(const struct sock *sk,
427 struct request_sock *req,
428 struct sk_buff *skb);
429 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
430 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
431 struct request_sock *req,
432 struct dst_entry *dst,
433 struct request_sock *req_unhash,
435 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
436 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
437 int tcp_connect(struct sock *sk);
438 enum tcp_synack_type {
443 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
444 struct request_sock *req,
445 struct tcp_fastopen_cookie *foc,
446 enum tcp_synack_type synack_type);
447 int tcp_disconnect(struct sock *sk, int flags);
449 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
450 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
451 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
453 /* From syncookies.c */
454 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
455 struct request_sock *req,
456 struct dst_entry *dst, u32 tsoff);
457 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
459 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
460 #ifdef CONFIG_SYN_COOKIES
462 /* Syncookies use a monotonic timer which increments every 60 seconds.
463 * This counter is used both as a hash input and partially encoded into
464 * the cookie value. A cookie is only validated further if the delta
465 * between the current counter value and the encoded one is less than this,
466 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
467 * the counter advances immediately after a cookie is generated).
469 #define MAX_SYNCOOKIE_AGE 2
470 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
471 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
473 /* syncookies: remember time of last synqueue overflow
474 * But do not dirty this field too often (once per second is enough)
475 * It is racy as we do not hold a lock, but race is very minor.
477 static inline void tcp_synq_overflow(const struct sock *sk)
479 unsigned int last_overflow;
480 unsigned int now = jiffies;
482 if (sk->sk_reuseport) {
483 struct sock_reuseport *reuse;
485 reuse = rcu_dereference(sk->sk_reuseport_cb);
487 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
488 if (!time_between32(now, last_overflow,
490 WRITE_ONCE(reuse->synq_overflow_ts, now);
495 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
496 if (!time_between32(now, last_overflow, last_overflow + HZ))
497 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
500 /* syncookies: no recent synqueue overflow on this listening socket? */
501 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
503 unsigned int last_overflow;
504 unsigned int now = jiffies;
506 if (sk->sk_reuseport) {
507 struct sock_reuseport *reuse;
509 reuse = rcu_dereference(sk->sk_reuseport_cb);
511 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
512 return !time_between32(now, last_overflow - HZ,
514 TCP_SYNCOOKIE_VALID);
518 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
520 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
521 * then we're under synflood. However, we have to use
522 * 'last_overflow - HZ' as lower bound. That's because a concurrent
523 * tcp_synq_overflow() could update .ts_recent_stamp after we read
524 * jiffies but before we store .ts_recent_stamp into last_overflow,
525 * which could lead to rejecting a valid syncookie.
527 return !time_between32(now, last_overflow - HZ,
528 last_overflow + TCP_SYNCOOKIE_VALID);
531 static inline u32 tcp_cookie_time(void)
533 u64 val = get_jiffies_64();
535 do_div(val, TCP_SYNCOOKIE_PERIOD);
539 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
541 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
542 u64 cookie_init_timestamp(struct request_sock *req);
543 bool cookie_timestamp_decode(const struct net *net,
544 struct tcp_options_received *opt);
545 bool cookie_ecn_ok(const struct tcp_options_received *opt,
546 const struct net *net, const struct dst_entry *dst);
548 /* From net/ipv6/syncookies.c */
549 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
551 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
553 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
554 const struct tcphdr *th, u16 *mssp);
555 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
559 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
561 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
562 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
563 void tcp_retransmit_timer(struct sock *sk);
564 void tcp_xmit_retransmit_queue(struct sock *);
565 void tcp_simple_retransmit(struct sock *);
566 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
567 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
569 TCP_FRAG_IN_WRITE_QUEUE,
570 TCP_FRAG_IN_RTX_QUEUE,
572 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
573 struct sk_buff *skb, u32 len,
574 unsigned int mss_now, gfp_t gfp);
576 void tcp_send_probe0(struct sock *);
577 void tcp_send_partial(struct sock *);
578 int tcp_write_wakeup(struct sock *, int mib);
579 void tcp_send_fin(struct sock *sk);
580 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
581 int tcp_send_synack(struct sock *);
582 void tcp_push_one(struct sock *, unsigned int mss_now);
583 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
584 void tcp_send_ack(struct sock *sk);
585 void tcp_send_delayed_ack(struct sock *sk);
586 void tcp_send_loss_probe(struct sock *sk);
587 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
588 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
589 const struct sk_buff *next_skb);
592 void tcp_rearm_rto(struct sock *sk);
593 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
594 void tcp_reset(struct sock *sk);
595 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
596 void tcp_fin(struct sock *sk);
597 void tcp_check_space(struct sock *sk);
600 void tcp_init_xmit_timers(struct sock *);
601 static inline void tcp_clear_xmit_timers(struct sock *sk)
603 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
606 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
609 inet_csk_clear_xmit_timers(sk);
612 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
613 unsigned int tcp_current_mss(struct sock *sk);
615 /* Bound MSS / TSO packet size with the half of the window */
616 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
620 /* When peer uses tiny windows, there is no use in packetizing
621 * to sub-MSS pieces for the sake of SWS or making sure there
622 * are enough packets in the pipe for fast recovery.
624 * On the other hand, for extremely large MSS devices, handling
625 * smaller than MSS windows in this way does make sense.
627 if (tp->max_window > TCP_MSS_DEFAULT)
628 cutoff = (tp->max_window >> 1);
630 cutoff = tp->max_window;
632 if (cutoff && pktsize > cutoff)
633 return max_t(int, cutoff, 68U - tp->tcp_header_len);
639 void tcp_get_info(struct sock *, struct tcp_info *);
641 /* Read 'sendfile()'-style from a TCP socket */
642 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
643 sk_read_actor_t recv_actor);
645 void tcp_initialize_rcv_mss(struct sock *sk);
647 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
648 int tcp_mss_to_mtu(struct sock *sk, int mss);
649 void tcp_mtup_init(struct sock *sk);
650 void tcp_init_buffer_space(struct sock *sk);
652 static inline void tcp_bound_rto(const struct sock *sk)
654 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
655 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
658 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
660 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
663 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
665 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
666 ntohl(TCP_FLAG_ACK) |
670 static inline void tcp_fast_path_on(struct tcp_sock *tp)
672 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
675 static inline void tcp_fast_path_check(struct sock *sk)
677 struct tcp_sock *tp = tcp_sk(sk);
679 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
681 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
683 tcp_fast_path_on(tp);
686 /* Compute the actual rto_min value */
687 static inline u32 tcp_rto_min(struct sock *sk)
689 const struct dst_entry *dst = __sk_dst_get(sk);
690 u32 rto_min = TCP_RTO_MIN;
692 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
693 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
697 static inline u32 tcp_rto_min_us(struct sock *sk)
699 return jiffies_to_usecs(tcp_rto_min(sk));
702 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
704 return dst_metric_locked(dst, RTAX_CC_ALGO);
707 /* Minimum RTT in usec. ~0 means not available. */
708 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
710 return minmax_get(&tp->rtt_min);
713 /* Compute the actual receive window we are currently advertising.
714 * Rcv_nxt can be after the window if our peer push more data
715 * than the offered window.
717 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
719 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
726 /* Choose a new window, without checks for shrinking, and without
727 * scaling applied to the result. The caller does these things
728 * if necessary. This is a "raw" window selection.
730 u32 __tcp_select_window(struct sock *sk);
732 void tcp_send_window_probe(struct sock *sk);
734 /* TCP uses 32bit jiffies to save some space.
735 * Note that this is different from tcp_time_stamp, which
736 * historically has been the same until linux-4.13.
738 #define tcp_jiffies32 ((u32)jiffies)
741 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
742 * It is no longer tied to jiffies, but to 1 ms clock.
743 * Note: double check if you want to use tcp_jiffies32 instead of this.
745 #define TCP_TS_HZ 1000
747 static inline u64 tcp_clock_ns(void)
749 return local_clock();
752 static inline u64 tcp_clock_us(void)
754 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
757 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
758 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
760 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
763 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
764 static inline u32 tcp_time_stamp_raw(void)
766 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ);
770 /* Refresh 1us clock of a TCP socket,
771 * ensuring monotically increasing values.
773 static inline void tcp_mstamp_refresh(struct tcp_sock *tp)
775 u64 val = tcp_clock_us();
777 if (val > tp->tcp_mstamp)
778 tp->tcp_mstamp = val;
781 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
783 return max_t(s64, t1 - t0, 0);
786 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
788 return div_u64(skb->skb_mstamp, USEC_PER_SEC / TCP_TS_HZ);
792 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
794 #define TCPHDR_FIN 0x01
795 #define TCPHDR_SYN 0x02
796 #define TCPHDR_RST 0x04
797 #define TCPHDR_PSH 0x08
798 #define TCPHDR_ACK 0x10
799 #define TCPHDR_URG 0x20
800 #define TCPHDR_ECE 0x40
801 #define TCPHDR_CWR 0x80
803 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
805 /* This is what the send packet queuing engine uses to pass
806 * TCP per-packet control information to the transmission code.
807 * We also store the host-order sequence numbers in here too.
808 * This is 44 bytes if IPV6 is enabled.
809 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
812 __u32 seq; /* Starting sequence number */
813 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
815 /* Note : tcp_tw_isn is used in input path only
816 * (isn chosen by tcp_timewait_state_process())
818 * tcp_gso_segs/size are used in write queue only,
819 * cf tcp_skb_pcount()/tcp_skb_mss()
827 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
829 __u8 sacked; /* State flags for SACK. */
830 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
831 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
832 #define TCPCB_LOST 0x04 /* SKB is lost */
833 #define TCPCB_TAGBITS 0x07 /* All tag bits */
834 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp) */
835 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
836 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
839 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
840 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
841 eor:1, /* Is skb MSG_EOR marked? */
842 has_rxtstamp:1, /* SKB has a RX timestamp */
844 __u32 ack_seq; /* Sequence number ACK'd */
847 /* There is space for up to 24 bytes */
848 __u32 in_flight:30,/* Bytes in flight at transmit */
849 is_app_limited:1, /* cwnd not fully used? */
851 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
853 /* start of send pipeline phase */
855 /* when we reached the "delivered" count */
856 u64 delivered_mstamp;
857 } tx; /* only used for outgoing skbs */
859 struct inet_skb_parm h4;
860 #if IS_ENABLED(CONFIG_IPV6)
861 struct inet6_skb_parm h6;
863 } header; /* For incoming skbs */
866 struct sock *sk_redir;
872 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
874 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb)
876 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb);
879 #if IS_ENABLED(CONFIG_IPV6)
880 /* This is the variant of inet6_iif() that must be used by TCP,
881 * as TCP moves IP6CB into a different location in skb->cb[]
883 static inline int tcp_v6_iif(const struct sk_buff *skb)
885 return TCP_SKB_CB(skb)->header.h6.iif;
888 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
890 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
892 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
895 /* TCP_SKB_CB reference means this can not be used from early demux */
896 static inline int tcp_v6_sdif(const struct sk_buff *skb)
898 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
899 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
900 return TCP_SKB_CB(skb)->header.h6.iif;
906 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
908 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
909 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
910 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
916 /* TCP_SKB_CB reference means this can not be used from early demux */
917 static inline int tcp_v4_sdif(struct sk_buff *skb)
919 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
920 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
921 return TCP_SKB_CB(skb)->header.h4.iif;
926 /* Due to TSO, an SKB can be composed of multiple actual
927 * packets. To keep these tracked properly, we use this.
929 static inline int tcp_skb_pcount(const struct sk_buff *skb)
931 return TCP_SKB_CB(skb)->tcp_gso_segs;
934 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
936 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
939 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
941 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
944 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
945 static inline int tcp_skb_mss(const struct sk_buff *skb)
947 return TCP_SKB_CB(skb)->tcp_gso_size;
950 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
952 return likely(!TCP_SKB_CB(skb)->eor);
955 /* Events passed to congestion control interface */
957 CA_EVENT_TX_START, /* first transmit when no packets in flight */
958 CA_EVENT_CWND_RESTART, /* congestion window restart */
959 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
960 CA_EVENT_LOSS, /* loss timeout */
961 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
962 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
965 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
966 enum tcp_ca_ack_event_flags {
967 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
968 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
969 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
973 * Interface for adding new TCP congestion control handlers
975 #define TCP_CA_NAME_MAX 16
976 #define TCP_CA_MAX 128
977 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
979 #define TCP_CA_UNSPEC 0
981 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
982 #define TCP_CONG_NON_RESTRICTED 0x1
983 /* Requires ECN/ECT set on all packets */
984 #define TCP_CONG_NEEDS_ECN 0x2
994 /* A rate sample measures the number of (original/retransmitted) data
995 * packets delivered "delivered" over an interval of time "interval_us".
996 * The tcp_rate.c code fills in the rate sample, and congestion
997 * control modules that define a cong_control function to run at the end
998 * of ACK processing can optionally chose to consult this sample when
999 * setting cwnd and pacing rate.
1000 * A sample is invalid if "delivered" or "interval_us" is negative.
1002 struct rate_sample {
1003 u64 prior_mstamp; /* starting timestamp for interval */
1004 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1005 s32 delivered; /* number of packets delivered over interval */
1006 long interval_us; /* time for tp->delivered to incr "delivered" */
1007 u32 snd_interval_us; /* snd interval for delivered packets */
1008 u32 rcv_interval_us; /* rcv interval for delivered packets */
1009 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1010 int losses; /* number of packets marked lost upon ACK */
1011 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1012 u32 prior_in_flight; /* in flight before this ACK */
1013 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1014 bool is_retrans; /* is sample from retransmission? */
1015 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1018 struct tcp_congestion_ops {
1019 struct list_head list;
1023 /* initialize private data (optional) */
1024 void (*init)(struct sock *sk);
1025 /* cleanup private data (optional) */
1026 void (*release)(struct sock *sk);
1028 /* return slow start threshold (required) */
1029 u32 (*ssthresh)(struct sock *sk);
1030 /* do new cwnd calculation (required) */
1031 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1032 /* call before changing ca_state (optional) */
1033 void (*set_state)(struct sock *sk, u8 new_state);
1034 /* call when cwnd event occurs (optional) */
1035 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1036 /* call when ack arrives (optional) */
1037 void (*in_ack_event)(struct sock *sk, u32 flags);
1038 /* new value of cwnd after loss (required) */
1039 u32 (*undo_cwnd)(struct sock *sk);
1040 /* hook for packet ack accounting (optional) */
1041 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1042 /* override sysctl_tcp_min_tso_segs */
1043 u32 (*min_tso_segs)(struct sock *sk);
1044 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1045 u32 (*sndbuf_expand)(struct sock *sk);
1046 /* call when packets are delivered to update cwnd and pacing rate,
1047 * after all the ca_state processing. (optional)
1049 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1050 /* get info for inet_diag (optional) */
1051 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1052 union tcp_cc_info *info);
1054 char name[TCP_CA_NAME_MAX];
1055 struct module *owner;
1058 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1059 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1061 void tcp_assign_congestion_control(struct sock *sk);
1062 void tcp_init_congestion_control(struct sock *sk);
1063 void tcp_cleanup_congestion_control(struct sock *sk);
1064 int tcp_set_default_congestion_control(struct net *net, const char *name);
1065 void tcp_get_default_congestion_control(struct net *net, char *name);
1066 void tcp_get_available_congestion_control(char *buf, size_t len);
1067 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1068 int tcp_set_allowed_congestion_control(char *allowed);
1069 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1070 bool reinit, bool cap_net_admin);
1071 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1072 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1074 u32 tcp_reno_ssthresh(struct sock *sk);
1075 u32 tcp_reno_undo_cwnd(struct sock *sk);
1076 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1077 extern struct tcp_congestion_ops tcp_reno;
1079 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1080 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1082 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1084 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1090 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1092 const struct inet_connection_sock *icsk = inet_csk(sk);
1094 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1097 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1099 struct inet_connection_sock *icsk = inet_csk(sk);
1101 if (icsk->icsk_ca_ops->set_state)
1102 icsk->icsk_ca_ops->set_state(sk, ca_state);
1103 icsk->icsk_ca_state = ca_state;
1106 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1108 const struct inet_connection_sock *icsk = inet_csk(sk);
1110 if (icsk->icsk_ca_ops->cwnd_event)
1111 icsk->icsk_ca_ops->cwnd_event(sk, event);
1114 /* From tcp_rate.c */
1115 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1116 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1117 struct rate_sample *rs);
1118 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1119 bool is_sack_reneg, struct rate_sample *rs);
1120 void tcp_rate_check_app_limited(struct sock *sk);
1122 /* These functions determine how the current flow behaves in respect of SACK
1123 * handling. SACK is negotiated with the peer, and therefore it can vary
1124 * between different flows.
1126 * tcp_is_sack - SACK enabled
1127 * tcp_is_reno - No SACK
1129 static inline int tcp_is_sack(const struct tcp_sock *tp)
1131 return tp->rx_opt.sack_ok;
1134 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1136 return !tcp_is_sack(tp);
1139 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1141 return tp->sacked_out + tp->lost_out;
1144 /* This determines how many packets are "in the network" to the best
1145 * of our knowledge. In many cases it is conservative, but where
1146 * detailed information is available from the receiver (via SACK
1147 * blocks etc.) we can make more aggressive calculations.
1149 * Use this for decisions involving congestion control, use just
1150 * tp->packets_out to determine if the send queue is empty or not.
1152 * Read this equation as:
1154 * "Packets sent once on transmission queue" MINUS
1155 * "Packets left network, but not honestly ACKed yet" PLUS
1156 * "Packets fast retransmitted"
1158 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1160 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1163 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1165 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1167 return tp->snd_cwnd < tp->snd_ssthresh;
1170 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1172 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1175 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1177 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1178 (1 << inet_csk(sk)->icsk_ca_state);
1181 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1182 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1185 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1187 const struct tcp_sock *tp = tcp_sk(sk);
1189 if (tcp_in_cwnd_reduction(sk))
1190 return tp->snd_ssthresh;
1192 return max(tp->snd_ssthresh,
1193 ((tp->snd_cwnd >> 1) +
1194 (tp->snd_cwnd >> 2)));
1197 /* Use define here intentionally to get WARN_ON location shown at the caller */
1198 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1200 void tcp_enter_cwr(struct sock *sk);
1201 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1203 /* The maximum number of MSS of available cwnd for which TSO defers
1204 * sending if not using sysctl_tcp_tso_win_divisor.
1206 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1211 /* Returns end sequence number of the receiver's advertised window */
1212 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1214 return tp->snd_una + tp->snd_wnd;
1217 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1218 * flexible approach. The RFC suggests cwnd should not be raised unless
1219 * it was fully used previously. And that's exactly what we do in
1220 * congestion avoidance mode. But in slow start we allow cwnd to grow
1221 * as long as the application has used half the cwnd.
1223 * cwnd is 10 (IW10), but application sends 9 frames.
1224 * We allow cwnd to reach 18 when all frames are ACKed.
1225 * This check is safe because it's as aggressive as slow start which already
1226 * risks 100% overshoot. The advantage is that we discourage application to
1227 * either send more filler packets or data to artificially blow up the cwnd
1228 * usage, and allow application-limited process to probe bw more aggressively.
1230 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1232 const struct tcp_sock *tp = tcp_sk(sk);
1234 if (tp->is_cwnd_limited)
1237 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1238 if (tcp_in_slow_start(tp))
1239 return tp->snd_cwnd < 2 * tp->max_packets_out;
1244 /* BBR congestion control needs pacing.
1245 * Same remark for SO_MAX_PACING_RATE.
1246 * sch_fq packet scheduler is efficiently handling pacing,
1247 * but is not always installed/used.
1248 * Return true if TCP stack should pace packets itself.
1250 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1252 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1255 /* Something is really bad, we could not queue an additional packet,
1256 * because qdisc is full or receiver sent a 0 window.
1257 * We do not want to add fuel to the fire, or abort too early,
1258 * so make sure the timer we arm now is at least 200ms in the future,
1259 * regardless of current icsk_rto value (as it could be ~2ms)
1261 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1263 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1266 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1267 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1268 unsigned long max_when)
1270 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1272 return (unsigned long)min_t(u64, when, max_when);
1275 static inline void tcp_check_probe_timer(struct sock *sk)
1277 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1278 inet_csk_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1279 tcp_probe0_base(sk), TCP_RTO_MAX);
1282 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1287 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1293 * Calculate(/check) TCP checksum
1295 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1296 __be32 daddr, __wsum base)
1298 return csum_tcpudp_magic(saddr,daddr,len,IPPROTO_TCP,base);
1301 static inline __sum16 __tcp_checksum_complete(struct sk_buff *skb)
1303 return __skb_checksum_complete(skb);
1306 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1308 return !skb_csum_unnecessary(skb) &&
1309 __tcp_checksum_complete(skb);
1312 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1313 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1318 static const char *statename[]={
1319 "Unused","Established","Syn Sent","Syn Recv",
1320 "Fin Wait 1","Fin Wait 2","Time Wait", "Close",
1321 "Close Wait","Last ACK","Listen","Closing"
1324 void tcp_set_state(struct sock *sk, int state);
1326 void tcp_done(struct sock *sk);
1328 int tcp_abort(struct sock *sk, int err);
1330 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1333 rx_opt->num_sacks = 0;
1336 u32 tcp_default_init_rwnd(u32 mss);
1337 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1339 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1341 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1342 struct tcp_sock *tp = tcp_sk(sk);
1345 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) ||
1346 tp->packets_out || ca_ops->cong_control)
1348 delta = tcp_jiffies32 - tp->lsndtime;
1349 if (delta > inet_csk(sk)->icsk_rto)
1350 tcp_cwnd_restart(sk, delta);
1353 /* Determine a window scaling and initial window to offer. */
1354 void tcp_select_initial_window(const struct sock *sk, int __space,
1355 __u32 mss, __u32 *rcv_wnd,
1356 __u32 *window_clamp, int wscale_ok,
1357 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1359 static inline int tcp_win_from_space(const struct sock *sk, int space)
1361 int tcp_adv_win_scale = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale);
1363 return tcp_adv_win_scale <= 0 ?
1364 (space>>(-tcp_adv_win_scale)) :
1365 space - (space>>tcp_adv_win_scale);
1368 /* Note: caller must be prepared to deal with negative returns */
1369 static inline int tcp_space(const struct sock *sk)
1371 return tcp_win_from_space(sk, sk->sk_rcvbuf - sk->sk_backlog.len -
1372 atomic_read(&sk->sk_rmem_alloc));
1375 static inline int tcp_full_space(const struct sock *sk)
1377 return tcp_win_from_space(sk, sk->sk_rcvbuf);
1380 /* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1381 * If 87.5 % (7/8) of the space has been consumed, we want to override
1382 * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1383 * len/truesize ratio.
1385 static inline bool tcp_rmem_pressure(const struct sock *sk)
1387 int rcvbuf, threshold;
1389 if (tcp_under_memory_pressure(sk))
1392 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1393 threshold = rcvbuf - (rcvbuf >> 3);
1395 return atomic_read(&sk->sk_rmem_alloc) > threshold;
1398 extern void tcp_openreq_init_rwin(struct request_sock *req,
1399 const struct sock *sk_listener,
1400 const struct dst_entry *dst);
1402 void tcp_enter_memory_pressure(struct sock *sk);
1403 void tcp_leave_memory_pressure(struct sock *sk);
1405 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1407 struct net *net = sock_net((struct sock *)tp);
1409 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1412 static inline int keepalive_time_when(const struct tcp_sock *tp)
1414 struct net *net = sock_net((struct sock *)tp);
1416 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1419 static inline int keepalive_probes(const struct tcp_sock *tp)
1421 struct net *net = sock_net((struct sock *)tp);
1423 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1426 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1428 const struct inet_connection_sock *icsk = &tp->inet_conn;
1430 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1431 tcp_jiffies32 - tp->rcv_tstamp);
1434 static inline int tcp_fin_time(const struct sock *sk)
1436 int fin_timeout = tcp_sk(sk)->linger2 ? :
1437 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout);
1438 const int rto = inet_csk(sk)->icsk_rto;
1440 if (fin_timeout < (rto << 2) - (rto >> 1))
1441 fin_timeout = (rto << 2) - (rto >> 1);
1446 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1449 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1451 if (unlikely(!time_before32(ktime_get_seconds(),
1452 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1455 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1456 * then following tcp messages have valid values. Ignore 0 value,
1457 * or else 'negative' tsval might forbid us to accept their packets.
1459 if (!rx_opt->ts_recent)
1464 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1467 if (tcp_paws_check(rx_opt, 0))
1470 /* RST segments are not recommended to carry timestamp,
1471 and, if they do, it is recommended to ignore PAWS because
1472 "their cleanup function should take precedence over timestamps."
1473 Certainly, it is mistake. It is necessary to understand the reasons
1474 of this constraint to relax it: if peer reboots, clock may go
1475 out-of-sync and half-open connections will not be reset.
1476 Actually, the problem would be not existing if all
1477 the implementations followed draft about maintaining clock
1478 via reboots. Linux-2.2 DOES NOT!
1480 However, we can relax time bounds for RST segments to MSL.
1482 if (rst && !time_before32(ktime_get_seconds(),
1483 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1488 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1489 int mib_idx, u32 *last_oow_ack_time);
1491 static inline void tcp_mib_init(struct net *net)
1494 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1495 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1496 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1497 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1501 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1503 tp->lost_skb_hint = NULL;
1506 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1508 tcp_clear_retrans_hints_partial(tp);
1509 tp->retransmit_skb_hint = NULL;
1512 union tcp_md5_addr {
1514 #if IS_ENABLED(CONFIG_IPV6)
1519 /* - key database */
1520 struct tcp_md5sig_key {
1521 struct hlist_node node;
1523 u8 family; /* AF_INET or AF_INET6 */
1524 union tcp_md5_addr addr;
1526 u8 key[TCP_MD5SIG_MAXKEYLEN];
1527 struct rcu_head rcu;
1531 struct tcp_md5sig_info {
1532 struct hlist_head head;
1533 struct rcu_head rcu;
1536 /* - pseudo header */
1537 struct tcp4_pseudohdr {
1545 struct tcp6_pseudohdr {
1546 struct in6_addr saddr;
1547 struct in6_addr daddr;
1549 __be32 protocol; /* including padding */
1552 union tcp_md5sum_block {
1553 struct tcp4_pseudohdr ip4;
1554 #if IS_ENABLED(CONFIG_IPV6)
1555 struct tcp6_pseudohdr ip6;
1559 /* - pool: digest algorithm, hash description and scratch buffer */
1560 struct tcp_md5sig_pool {
1561 struct ahash_request *md5_req;
1566 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1567 const struct sock *sk, const struct sk_buff *skb);
1568 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1569 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1571 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1572 int family, u8 prefixlen);
1573 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1574 const struct sock *addr_sk);
1576 #ifdef CONFIG_TCP_MD5SIG
1577 struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1578 const union tcp_md5_addr *addr,
1580 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1582 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1583 const union tcp_md5_addr *addr,
1588 #define tcp_twsk_md5_key(twsk) NULL
1591 bool tcp_alloc_md5sig_pool(void);
1593 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1594 static inline void tcp_put_md5sig_pool(void)
1599 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1600 unsigned int header_len);
1601 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1602 const struct tcp_md5sig_key *key);
1604 /* From tcp_fastopen.c */
1605 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1606 struct tcp_fastopen_cookie *cookie);
1607 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1608 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1610 struct tcp_fastopen_request {
1611 /* Fast Open cookie. Size 0 means a cookie request */
1612 struct tcp_fastopen_cookie cookie;
1613 struct msghdr *data; /* data in MSG_FASTOPEN */
1615 int copied; /* queued in tcp_connect() */
1617 void tcp_free_fastopen_req(struct tcp_sock *tp);
1618 void tcp_fastopen_destroy_cipher(struct sock *sk);
1619 void tcp_fastopen_ctx_destroy(struct net *net);
1620 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1621 void *key, unsigned int len);
1622 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1623 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1624 struct request_sock *req,
1625 struct tcp_fastopen_cookie *foc,
1626 const struct dst_entry *dst);
1627 void tcp_fastopen_init_key_once(struct net *net);
1628 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1629 struct tcp_fastopen_cookie *cookie);
1630 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1631 #define TCP_FASTOPEN_KEY_LENGTH 16
1633 /* Fastopen key context */
1634 struct tcp_fastopen_context {
1635 struct crypto_cipher *tfm;
1636 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1637 struct rcu_head rcu;
1640 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1641 void tcp_fastopen_active_disable(struct sock *sk);
1642 bool tcp_fastopen_active_should_disable(struct sock *sk);
1643 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1644 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1646 /* Latencies incurred by various limits for a sender. They are
1647 * chronograph-like stats that are mutually exclusive.
1651 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1652 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1653 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1657 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1658 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1660 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1661 * the same memory storage than skb->destructor/_skb_refdst
1663 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1665 skb->destructor = NULL;
1666 skb->_skb_refdst = 0UL;
1669 #define tcp_skb_tsorted_save(skb) { \
1670 unsigned long _save = skb->_skb_refdst; \
1671 skb->_skb_refdst = 0UL;
1673 #define tcp_skb_tsorted_restore(skb) \
1674 skb->_skb_refdst = _save; \
1677 void tcp_write_queue_purge(struct sock *sk);
1679 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1681 return skb_rb_first(&sk->tcp_rtx_queue);
1684 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1686 return skb_rb_last(&sk->tcp_rtx_queue);
1689 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1691 return skb_peek(&sk->sk_write_queue);
1694 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1696 return skb_peek_tail(&sk->sk_write_queue);
1699 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1700 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1702 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1704 return skb_peek(&sk->sk_write_queue);
1707 static inline bool tcp_skb_is_last(const struct sock *sk,
1708 const struct sk_buff *skb)
1710 return skb_queue_is_last(&sk->sk_write_queue, skb);
1713 static inline bool tcp_write_queue_empty(const struct sock *sk)
1715 return skb_queue_empty(&sk->sk_write_queue);
1718 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1720 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1723 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1725 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1728 static inline void tcp_check_send_head(struct sock *sk, struct sk_buff *skb_unlinked)
1730 if (tcp_write_queue_empty(sk))
1731 tcp_chrono_stop(sk, TCP_CHRONO_BUSY);
1734 static inline void __tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1736 __skb_queue_tail(&sk->sk_write_queue, skb);
1739 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1741 __tcp_add_write_queue_tail(sk, skb);
1743 /* Queue it, remembering where we must start sending. */
1744 if (sk->sk_write_queue.next == skb)
1745 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1748 /* Insert new before skb on the write queue of sk. */
1749 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1750 struct sk_buff *skb,
1753 __skb_queue_before(&sk->sk_write_queue, skb, new);
1756 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1758 tcp_skb_tsorted_anchor_cleanup(skb);
1759 __skb_unlink(skb, &sk->sk_write_queue);
1762 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1764 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1766 tcp_skb_tsorted_anchor_cleanup(skb);
1767 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1770 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1772 list_del(&skb->tcp_tsorted_anchor);
1773 tcp_rtx_queue_unlink(skb, sk);
1774 sk_wmem_free_skb(sk, skb);
1777 static inline void tcp_push_pending_frames(struct sock *sk)
1779 if (tcp_send_head(sk)) {
1780 struct tcp_sock *tp = tcp_sk(sk);
1782 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1786 /* Start sequence of the skb just after the highest skb with SACKed
1787 * bit, valid only if sacked_out > 0 or when the caller has ensured
1788 * validity by itself.
1790 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1792 if (!tp->sacked_out)
1795 if (tp->highest_sack == NULL)
1798 return TCP_SKB_CB(tp->highest_sack)->seq;
1801 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1803 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1806 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1808 return tcp_sk(sk)->highest_sack;
1811 static inline void tcp_highest_sack_reset(struct sock *sk)
1813 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1816 /* Called when old skb is about to be deleted and replaced by new skb */
1817 static inline void tcp_highest_sack_replace(struct sock *sk,
1818 struct sk_buff *old,
1819 struct sk_buff *new)
1821 if (old == tcp_highest_sack(sk))
1822 tcp_sk(sk)->highest_sack = new;
1825 /* This helper checks if socket has IP_TRANSPARENT set */
1826 static inline bool inet_sk_transparent(const struct sock *sk)
1828 switch (sk->sk_state) {
1830 return inet_twsk(sk)->tw_transparent;
1831 case TCP_NEW_SYN_RECV:
1832 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1834 return inet_sk(sk)->transparent;
1837 /* Determines whether this is a thin stream (which may suffer from
1838 * increased latency). Used to trigger latency-reducing mechanisms.
1840 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1842 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1846 enum tcp_seq_states {
1847 TCP_SEQ_STATE_LISTENING,
1848 TCP_SEQ_STATE_ESTABLISHED,
1851 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1852 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1853 void tcp_seq_stop(struct seq_file *seq, void *v);
1855 struct tcp_seq_afinfo {
1859 struct tcp_iter_state {
1860 struct seq_net_private p;
1861 enum tcp_seq_states state;
1862 struct sock *syn_wait_sk;
1863 int bucket, offset, sbucket, num;
1867 extern struct request_sock_ops tcp_request_sock_ops;
1868 extern struct request_sock_ops tcp6_request_sock_ops;
1870 void tcp_v4_destroy_sock(struct sock *sk);
1872 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1873 netdev_features_t features);
1874 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
1875 int tcp_gro_complete(struct sk_buff *skb);
1877 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1879 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1881 struct net *net = sock_net((struct sock *)tp);
1882 return tp->notsent_lowat ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
1885 static inline bool tcp_stream_memory_free(const struct sock *sk)
1887 const struct tcp_sock *tp = tcp_sk(sk);
1888 u32 notsent_bytes = READ_ONCE(tp->write_seq) - tp->snd_nxt;
1890 return notsent_bytes < tcp_notsent_lowat(tp);
1893 #ifdef CONFIG_PROC_FS
1894 int tcp4_proc_init(void);
1895 void tcp4_proc_exit(void);
1898 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1899 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1900 const struct tcp_request_sock_ops *af_ops,
1901 struct sock *sk, struct sk_buff *skb);
1903 /* TCP af-specific functions */
1904 struct tcp_sock_af_ops {
1905 #ifdef CONFIG_TCP_MD5SIG
1906 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1907 const struct sock *addr_sk);
1908 int (*calc_md5_hash)(char *location,
1909 const struct tcp_md5sig_key *md5,
1910 const struct sock *sk,
1911 const struct sk_buff *skb);
1912 int (*md5_parse)(struct sock *sk,
1914 char __user *optval,
1919 struct tcp_request_sock_ops {
1921 #ifdef CONFIG_TCP_MD5SIG
1922 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1923 const struct sock *addr_sk);
1924 int (*calc_md5_hash) (char *location,
1925 const struct tcp_md5sig_key *md5,
1926 const struct sock *sk,
1927 const struct sk_buff *skb);
1929 void (*init_req)(struct request_sock *req,
1930 const struct sock *sk_listener,
1931 struct sk_buff *skb);
1932 #ifdef CONFIG_SYN_COOKIES
1933 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1936 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1937 const struct request_sock *req);
1938 u32 (*init_seq)(const struct sk_buff *skb);
1939 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
1940 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1941 struct flowi *fl, struct request_sock *req,
1942 struct tcp_fastopen_cookie *foc,
1943 enum tcp_synack_type synack_type);
1946 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
1947 #if IS_ENABLED(CONFIG_IPV6)
1948 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
1951 #ifdef CONFIG_SYN_COOKIES
1952 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1953 const struct sock *sk, struct sk_buff *skb,
1956 tcp_synq_overflow(sk);
1957 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1958 return ops->cookie_init_seq(skb, mss);
1961 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1962 const struct sock *sk, struct sk_buff *skb,
1969 int tcpv4_offload_init(void);
1971 void tcp_v4_init(void);
1972 void tcp_init(void);
1974 /* tcp_recovery.c */
1975 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
1976 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
1977 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
1979 extern bool tcp_rack_mark_lost(struct sock *sk);
1980 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
1982 extern void tcp_rack_reo_timeout(struct sock *sk);
1983 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
1985 /* At how many usecs into the future should the RTO fire? */
1986 static inline s64 tcp_rto_delta_us(const struct sock *sk)
1988 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
1989 u32 rto = inet_csk(sk)->icsk_rto;
1990 u64 rto_time_stamp_us = skb->skb_mstamp + jiffies_to_usecs(rto);
1992 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
1996 * Save and compile IPv4 options, return a pointer to it
1998 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
1999 struct sk_buff *skb)
2001 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2002 struct ip_options_rcu *dopt = NULL;
2005 int opt_size = sizeof(*dopt) + opt->optlen;
2007 dopt = kmalloc(opt_size, GFP_ATOMIC);
2008 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2016 /* locally generated TCP pure ACKs have skb->truesize == 2
2017 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2018 * This is much faster than dissecting the packet to find out.
2019 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2021 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2023 return skb->truesize == 2;
2026 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2031 static inline int tcp_inq(struct sock *sk)
2033 struct tcp_sock *tp = tcp_sk(sk);
2036 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2038 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2040 before(tp->urg_seq, tp->copied_seq) ||
2041 !before(tp->urg_seq, tp->rcv_nxt)) {
2043 answ = tp->rcv_nxt - tp->copied_seq;
2045 /* Subtract 1, if FIN was received */
2046 if (answ && sock_flag(sk, SOCK_DONE))
2049 answ = tp->urg_seq - tp->copied_seq;
2055 int tcp_peek_len(struct socket *sock);
2057 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2061 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2062 tp->segs_in += segs_in;
2063 if (skb->len > tcp_hdrlen(skb))
2064 tp->data_segs_in += segs_in;
2068 * TCP listen path runs lockless.
2069 * We forced "struct sock" to be const qualified to make sure
2070 * we don't modify one of its field by mistake.
2071 * Here, we increment sk_drops which is an atomic_t, so we can safely
2072 * make sock writable again.
2074 static inline void tcp_listendrop(const struct sock *sk)
2076 atomic_inc(&((struct sock *)sk)->sk_drops);
2077 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2080 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2083 * Interface for adding Upper Level Protocols over TCP
2086 #define TCP_ULP_NAME_MAX 16
2087 #define TCP_ULP_MAX 128
2088 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2095 struct tcp_ulp_ops {
2096 struct list_head list;
2098 /* initialize ulp */
2099 int (*init)(struct sock *sk);
2101 void (*release)(struct sock *sk);
2104 char name[TCP_ULP_NAME_MAX];
2106 struct module *owner;
2108 int tcp_register_ulp(struct tcp_ulp_ops *type);
2109 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2110 int tcp_set_ulp(struct sock *sk, const char *name);
2111 int tcp_set_ulp_id(struct sock *sk, const int ulp);
2112 void tcp_get_available_ulp(char *buf, size_t len);
2113 void tcp_cleanup_ulp(struct sock *sk);
2115 #define MODULE_ALIAS_TCP_ULP(name) \
2116 __MODULE_INFO(alias, alias_userspace, name); \
2117 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2119 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2120 * is < 0, then the BPF op failed (for example if the loaded BPF
2121 * program does not support the chosen operation or there is no BPF
2125 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2127 struct bpf_sock_ops_kern sock_ops;
2130 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2131 if (sk_fullsock(sk)) {
2132 sock_ops.is_fullsock = 1;
2133 sock_owned_by_me(sk);
2139 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2141 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2143 ret = sock_ops.reply;
2149 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2151 u32 args[2] = {arg1, arg2};
2153 return tcp_call_bpf(sk, op, 2, args);
2156 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2159 u32 args[3] = {arg1, arg2, arg3};
2161 return tcp_call_bpf(sk, op, 3, args);
2165 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2170 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2175 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2183 static inline u32 tcp_timeout_init(struct sock *sk)
2187 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2190 timeout = TCP_TIMEOUT_INIT;
2194 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2198 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2205 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2207 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2210 #if IS_ENABLED(CONFIG_SMC)
2211 extern struct static_key_false tcp_have_smc;
2214 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2215 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2216 void (*cad)(struct sock *sk, u32 ack_seq));
2217 void clean_acked_data_disable(struct inet_connection_sock *icsk);