1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Definitions for the TCP module.
9 * Version: @(#)tcp.h 1.0.5 05/23/93
12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
17 #define FASTRETRANS_DEBUG 1
19 #include <linux/list.h>
20 #include <linux/tcp.h>
21 #include <linux/bug.h>
22 #include <linux/slab.h>
23 #include <linux/cache.h>
24 #include <linux/percpu.h>
25 #include <linux/skbuff.h>
26 #include <linux/kref.h>
27 #include <linux/ktime.h>
28 #include <linux/indirect_call_wrapper.h>
30 #include <net/inet_connection_sock.h>
31 #include <net/inet_timewait_sock.h>
32 #include <net/inet_hashtables.h>
33 #include <net/checksum.h>
34 #include <net/request_sock.h>
35 #include <net/sock_reuseport.h>
39 #include <net/tcp_states.h>
40 #include <net/inet_ecn.h>
42 #include <net/mptcp.h>
44 #include <linux/seq_file.h>
45 #include <linux/memcontrol.h>
46 #include <linux/bpf-cgroup.h>
47 #include <linux/siphash.h>
49 extern struct inet_hashinfo tcp_hashinfo;
51 extern struct percpu_counter tcp_orphan_count;
52 void tcp_time_wait(struct sock *sk, int state, int timeo);
54 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER)
55 #define MAX_TCP_OPTION_SPACE 40
56 #define TCP_MIN_SND_MSS 48
57 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
60 * Never offer a window over 32767 without using window scaling. Some
61 * poor stacks do signed 16bit maths!
63 #define MAX_TCP_WINDOW 32767U
65 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
66 #define TCP_MIN_MSS 88U
68 /* The initial MTU to use for probing */
69 #define TCP_BASE_MSS 1024
71 /* probing interval, default to 10 minutes as per RFC4821 */
72 #define TCP_PROBE_INTERVAL 600
74 /* Specify interval when tcp mtu probing will stop */
75 #define TCP_PROBE_THRESHOLD 8
77 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
78 #define TCP_FASTRETRANS_THRESH 3
80 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
81 #define TCP_MAX_QUICKACKS 16U
83 /* Maximal number of window scale according to RFC1323 */
84 #define TCP_MAX_WSCALE 14U
87 #define TCP_URG_VALID 0x0100
88 #define TCP_URG_NOTYET 0x0200
89 #define TCP_URG_READ 0x0400
91 #define TCP_RETR1 3 /*
92 * This is how many retries it does before it
93 * tries to figure out if the gateway is
94 * down. Minimal RFC value is 3; it corresponds
95 * to ~3sec-8min depending on RTO.
98 #define TCP_RETR2 15 /*
99 * This should take at least
100 * 90 minutes to time out.
101 * RFC1122 says that the limit is 100 sec.
102 * 15 is ~13-30min depending on RTO.
105 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
106 * when active opening a connection.
107 * RFC1122 says the minimum retry MUST
108 * be at least 180secs. Nevertheless
109 * this value is corresponding to
110 * 63secs of retransmission with the
111 * current initial RTO.
114 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
115 * when passive opening a connection.
116 * This is corresponding to 31secs of
117 * retransmission with the current
121 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
122 * state, about 60 seconds */
123 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
124 /* BSD style FIN_WAIT2 deadlock breaker.
125 * It used to be 3min, new value is 60sec,
126 * to combine FIN-WAIT-2 timeout with
129 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */
131 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
133 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
134 #define TCP_ATO_MIN ((unsigned)(HZ/25))
136 #define TCP_DELACK_MIN 4U
137 #define TCP_ATO_MIN 4U
139 #define TCP_RTO_MAX ((unsigned)(120*HZ))
140 #define TCP_RTO_MIN ((unsigned)(HZ/5))
141 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
142 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
143 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
144 * used as a fallback RTO for the
145 * initial data transmission if no
146 * valid RTT sample has been acquired,
147 * most likely due to retrans in 3WHS.
150 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
151 * for local resources.
153 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
154 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
155 #define TCP_KEEPALIVE_INTVL (75*HZ)
157 #define MAX_TCP_KEEPIDLE 32767
158 #define MAX_TCP_KEEPINTVL 32767
159 #define MAX_TCP_KEEPCNT 127
160 #define MAX_TCP_SYNCNT 127
162 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
164 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
165 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
166 * after this time. It should be equal
167 * (or greater than) TCP_TIMEWAIT_LEN
168 * to provide reliability equal to one
169 * provided by timewait state.
171 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
172 * timestamps. It must be less than
173 * minimal timewait lifetime.
179 #define TCPOPT_NOP 1 /* Padding */
180 #define TCPOPT_EOL 0 /* End of options */
181 #define TCPOPT_MSS 2 /* Segment size negotiating */
182 #define TCPOPT_WINDOW 3 /* Window scaling */
183 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
184 #define TCPOPT_SACK 5 /* SACK Block */
185 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
186 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
187 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */
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 int 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 int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
335 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
337 void tcp_release_cb(struct sock *sk);
338 void tcp_wfree(struct sk_buff *skb);
339 void tcp_write_timer_handler(struct sock *sk);
340 void tcp_delack_timer_handler(struct sock *sk);
341 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
342 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
343 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
344 void tcp_rcv_space_adjust(struct sock *sk);
345 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
346 void tcp_twsk_destructor(struct sock *sk);
347 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
348 struct pipe_inode_info *pipe, size_t len,
351 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
352 static inline void tcp_dec_quickack_mode(struct sock *sk,
353 const unsigned int pkts)
355 struct inet_connection_sock *icsk = inet_csk(sk);
357 if (icsk->icsk_ack.quick) {
358 if (pkts >= icsk->icsk_ack.quick) {
359 icsk->icsk_ack.quick = 0;
360 /* Leaving quickack mode we deflate ATO. */
361 icsk->icsk_ack.ato = TCP_ATO_MIN;
363 icsk->icsk_ack.quick -= pkts;
368 #define TCP_ECN_QUEUE_CWR 2
369 #define TCP_ECN_DEMAND_CWR 4
370 #define TCP_ECN_SEEN 8
380 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
382 const struct tcphdr *th);
383 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
384 struct request_sock *req, bool fastopen,
386 int tcp_child_process(struct sock *parent, struct sock *child,
387 struct sk_buff *skb);
388 void tcp_enter_loss(struct sock *sk);
389 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
390 void tcp_clear_retrans(struct tcp_sock *tp);
391 void tcp_update_metrics(struct sock *sk);
392 void tcp_init_metrics(struct sock *sk);
393 void tcp_metrics_init(void);
394 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
395 void tcp_close(struct sock *sk, long timeout);
396 void tcp_init_sock(struct sock *sk);
397 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb);
398 __poll_t tcp_poll(struct file *file, struct socket *sock,
399 struct poll_table_struct *wait);
400 int tcp_getsockopt(struct sock *sk, int level, int optname,
401 char __user *optval, int __user *optlen);
402 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
403 unsigned int optlen);
404 void tcp_set_keepalive(struct sock *sk, int val);
405 void tcp_syn_ack_timeout(const struct request_sock *req);
406 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
407 int flags, int *addr_len);
408 int tcp_set_rcvlowat(struct sock *sk, int val);
409 void tcp_data_ready(struct sock *sk);
411 int tcp_mmap(struct file *file, struct socket *sock,
412 struct vm_area_struct *vma);
414 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
415 struct tcp_options_received *opt_rx,
416 int estab, struct tcp_fastopen_cookie *foc);
417 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
420 * BPF SKB-less helpers
422 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
423 struct tcphdr *th, u32 *cookie);
424 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
425 struct tcphdr *th, u32 *cookie);
426 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
427 const struct tcp_request_sock_ops *af_ops,
428 struct sock *sk, struct tcphdr *th);
430 * TCP v4 functions exported for the inet6 API
433 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
434 void tcp_v4_mtu_reduced(struct sock *sk);
435 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
436 void tcp_ld_RTO_revert(struct sock *sk, u32 seq);
437 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
438 struct sock *tcp_create_openreq_child(const struct sock *sk,
439 struct request_sock *req,
440 struct sk_buff *skb);
441 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
442 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
443 struct request_sock *req,
444 struct dst_entry *dst,
445 struct request_sock *req_unhash,
447 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
448 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
449 int tcp_connect(struct sock *sk);
450 enum tcp_synack_type {
455 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
456 struct request_sock *req,
457 struct tcp_fastopen_cookie *foc,
458 enum tcp_synack_type synack_type,
459 struct sk_buff *syn_skb);
460 int tcp_disconnect(struct sock *sk, int flags);
462 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
463 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
464 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
466 /* From syncookies.c */
467 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
468 struct request_sock *req,
469 struct dst_entry *dst, u32 tsoff);
470 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
472 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
473 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
474 struct sock *sk, struct sk_buff *skb);
475 #ifdef CONFIG_SYN_COOKIES
477 /* Syncookies use a monotonic timer which increments every 60 seconds.
478 * This counter is used both as a hash input and partially encoded into
479 * the cookie value. A cookie is only validated further if the delta
480 * between the current counter value and the encoded one is less than this,
481 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
482 * the counter advances immediately after a cookie is generated).
484 #define MAX_SYNCOOKIE_AGE 2
485 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
486 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
488 /* syncookies: remember time of last synqueue overflow
489 * But do not dirty this field too often (once per second is enough)
490 * It is racy as we do not hold a lock, but race is very minor.
492 static inline void tcp_synq_overflow(const struct sock *sk)
494 unsigned int last_overflow;
495 unsigned int now = jiffies;
497 if (sk->sk_reuseport) {
498 struct sock_reuseport *reuse;
500 reuse = rcu_dereference(sk->sk_reuseport_cb);
502 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
503 if (!time_between32(now, last_overflow,
505 WRITE_ONCE(reuse->synq_overflow_ts, now);
510 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
511 if (!time_between32(now, last_overflow, last_overflow + HZ))
512 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
515 /* syncookies: no recent synqueue overflow on this listening socket? */
516 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
518 unsigned int last_overflow;
519 unsigned int now = jiffies;
521 if (sk->sk_reuseport) {
522 struct sock_reuseport *reuse;
524 reuse = rcu_dereference(sk->sk_reuseport_cb);
526 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
527 return !time_between32(now, last_overflow - HZ,
529 TCP_SYNCOOKIE_VALID);
533 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
535 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
536 * then we're under synflood. However, we have to use
537 * 'last_overflow - HZ' as lower bound. That's because a concurrent
538 * tcp_synq_overflow() could update .ts_recent_stamp after we read
539 * jiffies but before we store .ts_recent_stamp into last_overflow,
540 * which could lead to rejecting a valid syncookie.
542 return !time_between32(now, last_overflow - HZ,
543 last_overflow + TCP_SYNCOOKIE_VALID);
546 static inline u32 tcp_cookie_time(void)
548 u64 val = get_jiffies_64();
550 do_div(val, TCP_SYNCOOKIE_PERIOD);
554 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
556 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
557 u64 cookie_init_timestamp(struct request_sock *req, u64 now);
558 bool cookie_timestamp_decode(const struct net *net,
559 struct tcp_options_received *opt);
560 bool cookie_ecn_ok(const struct tcp_options_received *opt,
561 const struct net *net, const struct dst_entry *dst);
563 /* From net/ipv6/syncookies.c */
564 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
566 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
568 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
569 const struct tcphdr *th, u16 *mssp);
570 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
574 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
576 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
577 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
578 void tcp_retransmit_timer(struct sock *sk);
579 void tcp_xmit_retransmit_queue(struct sock *);
580 void tcp_simple_retransmit(struct sock *);
581 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
582 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
584 TCP_FRAG_IN_WRITE_QUEUE,
585 TCP_FRAG_IN_RTX_QUEUE,
587 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
588 struct sk_buff *skb, u32 len,
589 unsigned int mss_now, gfp_t gfp);
591 void tcp_send_probe0(struct sock *);
592 void tcp_send_partial(struct sock *);
593 int tcp_write_wakeup(struct sock *, int mib);
594 void tcp_send_fin(struct sock *sk);
595 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
596 int tcp_send_synack(struct sock *);
597 void tcp_push_one(struct sock *, unsigned int mss_now);
598 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
599 void tcp_send_ack(struct sock *sk);
600 void tcp_send_delayed_ack(struct sock *sk);
601 void tcp_send_loss_probe(struct sock *sk);
602 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
603 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
604 const struct sk_buff *next_skb);
607 void tcp_rearm_rto(struct sock *sk);
608 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
609 void tcp_reset(struct sock *sk);
610 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
611 void tcp_fin(struct sock *sk);
614 void tcp_init_xmit_timers(struct sock *);
615 static inline void tcp_clear_xmit_timers(struct sock *sk)
617 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
620 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
623 inet_csk_clear_xmit_timers(sk);
626 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
627 unsigned int tcp_current_mss(struct sock *sk);
629 /* Bound MSS / TSO packet size with the half of the window */
630 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
634 /* When peer uses tiny windows, there is no use in packetizing
635 * to sub-MSS pieces for the sake of SWS or making sure there
636 * are enough packets in the pipe for fast recovery.
638 * On the other hand, for extremely large MSS devices, handling
639 * smaller than MSS windows in this way does make sense.
641 if (tp->max_window > TCP_MSS_DEFAULT)
642 cutoff = (tp->max_window >> 1);
644 cutoff = tp->max_window;
646 if (cutoff && pktsize > cutoff)
647 return max_t(int, cutoff, 68U - tp->tcp_header_len);
653 void tcp_get_info(struct sock *, struct tcp_info *);
655 /* Read 'sendfile()'-style from a TCP socket */
656 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
657 sk_read_actor_t recv_actor);
659 void tcp_initialize_rcv_mss(struct sock *sk);
661 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
662 int tcp_mss_to_mtu(struct sock *sk, int mss);
663 void tcp_mtup_init(struct sock *sk);
665 static inline void tcp_bound_rto(const struct sock *sk)
667 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
668 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
671 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
673 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
676 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
678 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
679 ntohl(TCP_FLAG_ACK) |
683 static inline void tcp_fast_path_on(struct tcp_sock *tp)
685 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
688 static inline void tcp_fast_path_check(struct sock *sk)
690 struct tcp_sock *tp = tcp_sk(sk);
692 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
694 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
696 tcp_fast_path_on(tp);
699 /* Compute the actual rto_min value */
700 static inline u32 tcp_rto_min(struct sock *sk)
702 const struct dst_entry *dst = __sk_dst_get(sk);
703 u32 rto_min = inet_csk(sk)->icsk_rto_min;
705 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
706 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
710 static inline u32 tcp_rto_min_us(struct sock *sk)
712 return jiffies_to_usecs(tcp_rto_min(sk));
715 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
717 return dst_metric_locked(dst, RTAX_CC_ALGO);
720 /* Minimum RTT in usec. ~0 means not available. */
721 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
723 return minmax_get(&tp->rtt_min);
726 /* Compute the actual receive window we are currently advertising.
727 * Rcv_nxt can be after the window if our peer push more data
728 * than the offered window.
730 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
732 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
739 /* Choose a new window, without checks for shrinking, and without
740 * scaling applied to the result. The caller does these things
741 * if necessary. This is a "raw" window selection.
743 u32 __tcp_select_window(struct sock *sk);
745 void tcp_send_window_probe(struct sock *sk);
747 /* TCP uses 32bit jiffies to save some space.
748 * Note that this is different from tcp_time_stamp, which
749 * historically has been the same until linux-4.13.
751 #define tcp_jiffies32 ((u32)jiffies)
754 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
755 * It is no longer tied to jiffies, but to 1 ms clock.
756 * Note: double check if you want to use tcp_jiffies32 instead of this.
758 #define TCP_TS_HZ 1000
760 static inline u64 tcp_clock_ns(void)
762 return ktime_get_ns();
765 static inline u64 tcp_clock_us(void)
767 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
770 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
771 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
773 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
776 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
777 static inline u32 tcp_ns_to_ts(u64 ns)
779 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
782 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
783 static inline u32 tcp_time_stamp_raw(void)
785 return tcp_ns_to_ts(tcp_clock_ns());
788 void tcp_mstamp_refresh(struct tcp_sock *tp);
790 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
792 return max_t(s64, t1 - t0, 0);
795 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
797 return tcp_ns_to_ts(skb->skb_mstamp_ns);
800 /* provide the departure time in us unit */
801 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
803 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
807 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
809 #define TCPHDR_FIN 0x01
810 #define TCPHDR_SYN 0x02
811 #define TCPHDR_RST 0x04
812 #define TCPHDR_PSH 0x08
813 #define TCPHDR_ACK 0x10
814 #define TCPHDR_URG 0x20
815 #define TCPHDR_ECE 0x40
816 #define TCPHDR_CWR 0x80
818 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
820 /* This is what the send packet queuing engine uses to pass
821 * TCP per-packet control information to the transmission code.
822 * We also store the host-order sequence numbers in here too.
823 * This is 44 bytes if IPV6 is enabled.
824 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
827 __u32 seq; /* Starting sequence number */
828 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
830 /* Note : tcp_tw_isn is used in input path only
831 * (isn chosen by tcp_timewait_state_process())
833 * tcp_gso_segs/size are used in write queue only,
834 * cf tcp_skb_pcount()/tcp_skb_mss()
842 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
844 __u8 sacked; /* State flags for SACK. */
845 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
846 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
847 #define TCPCB_LOST 0x04 /* SKB is lost */
848 #define TCPCB_TAGBITS 0x07 /* All tag bits */
849 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
850 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
851 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
854 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
855 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
856 eor:1, /* Is skb MSG_EOR marked? */
857 has_rxtstamp:1, /* SKB has a RX timestamp */
859 __u32 ack_seq; /* Sequence number ACK'd */
862 /* There is space for up to 24 bytes */
863 __u32 in_flight:30,/* Bytes in flight at transmit */
864 is_app_limited:1, /* cwnd not fully used? */
866 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
868 /* start of send pipeline phase */
870 /* when we reached the "delivered" count */
871 u64 delivered_mstamp;
872 } tx; /* only used for outgoing skbs */
874 struct inet_skb_parm h4;
875 #if IS_ENABLED(CONFIG_IPV6)
876 struct inet6_skb_parm h6;
878 } header; /* For incoming skbs */
881 struct sock *sk_redir;
887 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
889 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb)
891 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb);
894 static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb)
896 return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS;
899 static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb)
901 return TCP_SKB_CB(skb)->bpf.sk_redir;
904 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb)
906 TCP_SKB_CB(skb)->bpf.sk_redir = NULL;
909 extern const struct inet_connection_sock_af_ops ipv4_specific;
911 #if IS_ENABLED(CONFIG_IPV6)
912 /* This is the variant of inet6_iif() that must be used by TCP,
913 * as TCP moves IP6CB into a different location in skb->cb[]
915 static inline int tcp_v6_iif(const struct sk_buff *skb)
917 return TCP_SKB_CB(skb)->header.h6.iif;
920 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
922 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
924 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
927 /* TCP_SKB_CB reference means this can not be used from early demux */
928 static inline int tcp_v6_sdif(const struct sk_buff *skb)
930 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
931 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
932 return TCP_SKB_CB(skb)->header.h6.iif;
937 extern const struct inet_connection_sock_af_ops ipv6_specific;
939 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
940 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
941 INDIRECT_CALLABLE_DECLARE(void tcp_v6_early_demux(struct sk_buff *skb));
945 /* TCP_SKB_CB reference means this can not be used from early demux */
946 static inline int tcp_v4_sdif(struct sk_buff *skb)
948 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
949 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
950 return TCP_SKB_CB(skb)->header.h4.iif;
955 /* Due to TSO, an SKB can be composed of multiple actual
956 * packets. To keep these tracked properly, we use this.
958 static inline int tcp_skb_pcount(const struct sk_buff *skb)
960 return TCP_SKB_CB(skb)->tcp_gso_segs;
963 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
965 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
968 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
970 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
973 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
974 static inline int tcp_skb_mss(const struct sk_buff *skb)
976 return TCP_SKB_CB(skb)->tcp_gso_size;
979 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
981 return likely(!TCP_SKB_CB(skb)->eor);
984 static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
985 const struct sk_buff *from)
987 return likely(tcp_skb_can_collapse_to(to) &&
988 mptcp_skb_can_collapse(to, from));
991 /* Events passed to congestion control interface */
993 CA_EVENT_TX_START, /* first transmit when no packets in flight */
994 CA_EVENT_CWND_RESTART, /* congestion window restart */
995 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
996 CA_EVENT_LOSS, /* loss timeout */
997 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
998 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
1001 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
1002 enum tcp_ca_ack_event_flags {
1003 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
1004 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
1005 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
1009 * Interface for adding new TCP congestion control handlers
1011 #define TCP_CA_NAME_MAX 16
1012 #define TCP_CA_MAX 128
1013 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
1015 #define TCP_CA_UNSPEC 0
1017 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1018 #define TCP_CONG_NON_RESTRICTED 0x1
1019 /* Requires ECN/ECT set on all packets */
1020 #define TCP_CONG_NEEDS_ECN 0x2
1021 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1031 /* A rate sample measures the number of (original/retransmitted) data
1032 * packets delivered "delivered" over an interval of time "interval_us".
1033 * The tcp_rate.c code fills in the rate sample, and congestion
1034 * control modules that define a cong_control function to run at the end
1035 * of ACK processing can optionally chose to consult this sample when
1036 * setting cwnd and pacing rate.
1037 * A sample is invalid if "delivered" or "interval_us" is negative.
1039 struct rate_sample {
1040 u64 prior_mstamp; /* starting timestamp for interval */
1041 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1042 s32 delivered; /* number of packets delivered over interval */
1043 long interval_us; /* time for tp->delivered to incr "delivered" */
1044 u32 snd_interval_us; /* snd interval for delivered packets */
1045 u32 rcv_interval_us; /* rcv interval for delivered packets */
1046 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1047 int losses; /* number of packets marked lost upon ACK */
1048 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1049 u32 prior_in_flight; /* in flight before this ACK */
1050 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1051 bool is_retrans; /* is sample from retransmission? */
1052 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1055 struct tcp_congestion_ops {
1056 struct list_head list;
1060 /* initialize private data (optional) */
1061 void (*init)(struct sock *sk);
1062 /* cleanup private data (optional) */
1063 void (*release)(struct sock *sk);
1065 /* return slow start threshold (required) */
1066 u32 (*ssthresh)(struct sock *sk);
1067 /* do new cwnd calculation (required) */
1068 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1069 /* call before changing ca_state (optional) */
1070 void (*set_state)(struct sock *sk, u8 new_state);
1071 /* call when cwnd event occurs (optional) */
1072 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1073 /* call when ack arrives (optional) */
1074 void (*in_ack_event)(struct sock *sk, u32 flags);
1075 /* new value of cwnd after loss (required) */
1076 u32 (*undo_cwnd)(struct sock *sk);
1077 /* hook for packet ack accounting (optional) */
1078 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1079 /* override sysctl_tcp_min_tso_segs */
1080 u32 (*min_tso_segs)(struct sock *sk);
1081 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1082 u32 (*sndbuf_expand)(struct sock *sk);
1083 /* call when packets are delivered to update cwnd and pacing rate,
1084 * after all the ca_state processing. (optional)
1086 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1087 /* get info for inet_diag (optional) */
1088 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1089 union tcp_cc_info *info);
1091 char name[TCP_CA_NAME_MAX];
1092 struct module *owner;
1095 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1096 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1098 void tcp_assign_congestion_control(struct sock *sk);
1099 void tcp_init_congestion_control(struct sock *sk);
1100 void tcp_cleanup_congestion_control(struct sock *sk);
1101 int tcp_set_default_congestion_control(struct net *net, const char *name);
1102 void tcp_get_default_congestion_control(struct net *net, char *name);
1103 void tcp_get_available_congestion_control(char *buf, size_t len);
1104 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1105 int tcp_set_allowed_congestion_control(char *allowed);
1106 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1107 bool cap_net_admin);
1108 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1109 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1111 u32 tcp_reno_ssthresh(struct sock *sk);
1112 u32 tcp_reno_undo_cwnd(struct sock *sk);
1113 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1114 extern struct tcp_congestion_ops tcp_reno;
1116 struct tcp_congestion_ops *tcp_ca_find(const char *name);
1117 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1118 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1120 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1122 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1128 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1130 const struct inet_connection_sock *icsk = inet_csk(sk);
1132 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1135 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1137 struct inet_connection_sock *icsk = inet_csk(sk);
1139 if (icsk->icsk_ca_ops->set_state)
1140 icsk->icsk_ca_ops->set_state(sk, ca_state);
1141 icsk->icsk_ca_state = ca_state;
1144 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1146 const struct inet_connection_sock *icsk = inet_csk(sk);
1148 if (icsk->icsk_ca_ops->cwnd_event)
1149 icsk->icsk_ca_ops->cwnd_event(sk, event);
1152 /* From tcp_rate.c */
1153 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1154 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1155 struct rate_sample *rs);
1156 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1157 bool is_sack_reneg, struct rate_sample *rs);
1158 void tcp_rate_check_app_limited(struct sock *sk);
1160 /* These functions determine how the current flow behaves in respect of SACK
1161 * handling. SACK is negotiated with the peer, and therefore it can vary
1162 * between different flows.
1164 * tcp_is_sack - SACK enabled
1165 * tcp_is_reno - No SACK
1167 static inline int tcp_is_sack(const struct tcp_sock *tp)
1169 return likely(tp->rx_opt.sack_ok);
1172 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1174 return !tcp_is_sack(tp);
1177 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1179 return tp->sacked_out + tp->lost_out;
1182 /* This determines how many packets are "in the network" to the best
1183 * of our knowledge. In many cases it is conservative, but where
1184 * detailed information is available from the receiver (via SACK
1185 * blocks etc.) we can make more aggressive calculations.
1187 * Use this for decisions involving congestion control, use just
1188 * tp->packets_out to determine if the send queue is empty or not.
1190 * Read this equation as:
1192 * "Packets sent once on transmission queue" MINUS
1193 * "Packets left network, but not honestly ACKed yet" PLUS
1194 * "Packets fast retransmitted"
1196 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1198 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1201 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1203 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1205 return tp->snd_cwnd < tp->snd_ssthresh;
1208 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1210 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1213 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1215 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1216 (1 << inet_csk(sk)->icsk_ca_state);
1219 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1220 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1223 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1225 const struct tcp_sock *tp = tcp_sk(sk);
1227 if (tcp_in_cwnd_reduction(sk))
1228 return tp->snd_ssthresh;
1230 return max(tp->snd_ssthresh,
1231 ((tp->snd_cwnd >> 1) +
1232 (tp->snd_cwnd >> 2)));
1235 /* Use define here intentionally to get WARN_ON location shown at the caller */
1236 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1238 void tcp_enter_cwr(struct sock *sk);
1239 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1241 /* The maximum number of MSS of available cwnd for which TSO defers
1242 * sending if not using sysctl_tcp_tso_win_divisor.
1244 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1249 /* Returns end sequence number of the receiver's advertised window */
1250 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1252 return tp->snd_una + tp->snd_wnd;
1255 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1256 * flexible approach. The RFC suggests cwnd should not be raised unless
1257 * it was fully used previously. And that's exactly what we do in
1258 * congestion avoidance mode. But in slow start we allow cwnd to grow
1259 * as long as the application has used half the cwnd.
1261 * cwnd is 10 (IW10), but application sends 9 frames.
1262 * We allow cwnd to reach 18 when all frames are ACKed.
1263 * This check is safe because it's as aggressive as slow start which already
1264 * risks 100% overshoot. The advantage is that we discourage application to
1265 * either send more filler packets or data to artificially blow up the cwnd
1266 * usage, and allow application-limited process to probe bw more aggressively.
1268 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1270 const struct tcp_sock *tp = tcp_sk(sk);
1272 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1273 if (tcp_in_slow_start(tp))
1274 return tp->snd_cwnd < 2 * tp->max_packets_out;
1276 return tp->is_cwnd_limited;
1279 /* BBR congestion control needs pacing.
1280 * Same remark for SO_MAX_PACING_RATE.
1281 * sch_fq packet scheduler is efficiently handling pacing,
1282 * but is not always installed/used.
1283 * Return true if TCP stack should pace packets itself.
1285 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1287 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1290 /* Estimates in how many jiffies next packet for this flow can be sent.
1291 * Scheduling a retransmit timer too early would be silly.
1293 static inline unsigned long tcp_pacing_delay(const struct sock *sk)
1295 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
1297 return delay > 0 ? nsecs_to_jiffies(delay) : 0;
1300 static inline void tcp_reset_xmit_timer(struct sock *sk,
1303 const unsigned long max_when)
1305 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk),
1309 /* Something is really bad, we could not queue an additional packet,
1310 * because qdisc is full or receiver sent a 0 window, or we are paced.
1311 * We do not want to add fuel to the fire, or abort too early,
1312 * so make sure the timer we arm now is at least 200ms in the future,
1313 * regardless of current icsk_rto value (as it could be ~2ms)
1315 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1317 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1320 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1321 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1322 unsigned long max_when)
1324 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1326 return (unsigned long)min_t(u64, when, max_when);
1329 static inline void tcp_check_probe_timer(struct sock *sk)
1331 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1332 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1333 tcp_probe0_base(sk), TCP_RTO_MAX);
1336 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1341 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1347 * Calculate(/check) TCP checksum
1349 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1350 __be32 daddr, __wsum base)
1352 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1355 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1357 return !skb_csum_unnecessary(skb) &&
1358 __skb_checksum_complete(skb);
1361 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1362 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1363 void tcp_set_state(struct sock *sk, int state);
1364 void tcp_done(struct sock *sk);
1365 int tcp_abort(struct sock *sk, int err);
1367 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1370 rx_opt->num_sacks = 0;
1373 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1375 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1377 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1378 struct tcp_sock *tp = tcp_sk(sk);
1381 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1382 ca_ops->cong_control)
1384 delta = tcp_jiffies32 - tp->lsndtime;
1385 if (delta > inet_csk(sk)->icsk_rto)
1386 tcp_cwnd_restart(sk, delta);
1389 /* Determine a window scaling and initial window to offer. */
1390 void tcp_select_initial_window(const struct sock *sk, int __space,
1391 __u32 mss, __u32 *rcv_wnd,
1392 __u32 *window_clamp, int wscale_ok,
1393 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1395 static inline int tcp_win_from_space(const struct sock *sk, int space)
1397 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1399 return tcp_adv_win_scale <= 0 ?
1400 (space>>(-tcp_adv_win_scale)) :
1401 space - (space>>tcp_adv_win_scale);
1404 /* Note: caller must be prepared to deal with negative returns */
1405 static inline int tcp_space(const struct sock *sk)
1407 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1408 READ_ONCE(sk->sk_backlog.len) -
1409 atomic_read(&sk->sk_rmem_alloc));
1412 static inline int tcp_full_space(const struct sock *sk)
1414 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1417 void tcp_cleanup_rbuf(struct sock *sk, int copied);
1419 /* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1420 * If 87.5 % (7/8) of the space has been consumed, we want to override
1421 * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1422 * len/truesize ratio.
1424 static inline bool tcp_rmem_pressure(const struct sock *sk)
1426 int rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1427 int threshold = rcvbuf - (rcvbuf >> 3);
1429 return atomic_read(&sk->sk_rmem_alloc) > threshold;
1432 extern void tcp_openreq_init_rwin(struct request_sock *req,
1433 const struct sock *sk_listener,
1434 const struct dst_entry *dst);
1436 void tcp_enter_memory_pressure(struct sock *sk);
1437 void tcp_leave_memory_pressure(struct sock *sk);
1439 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1441 struct net *net = sock_net((struct sock *)tp);
1443 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1446 static inline int keepalive_time_when(const struct tcp_sock *tp)
1448 struct net *net = sock_net((struct sock *)tp);
1450 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1453 static inline int keepalive_probes(const struct tcp_sock *tp)
1455 struct net *net = sock_net((struct sock *)tp);
1457 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1460 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1462 const struct inet_connection_sock *icsk = &tp->inet_conn;
1464 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1465 tcp_jiffies32 - tp->rcv_tstamp);
1468 static inline int tcp_fin_time(const struct sock *sk)
1470 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1471 const int rto = inet_csk(sk)->icsk_rto;
1473 if (fin_timeout < (rto << 2) - (rto >> 1))
1474 fin_timeout = (rto << 2) - (rto >> 1);
1479 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1482 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1484 if (unlikely(!time_before32(ktime_get_seconds(),
1485 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1488 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1489 * then following tcp messages have valid values. Ignore 0 value,
1490 * or else 'negative' tsval might forbid us to accept their packets.
1492 if (!rx_opt->ts_recent)
1497 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1500 if (tcp_paws_check(rx_opt, 0))
1503 /* RST segments are not recommended to carry timestamp,
1504 and, if they do, it is recommended to ignore PAWS because
1505 "their cleanup function should take precedence over timestamps."
1506 Certainly, it is mistake. It is necessary to understand the reasons
1507 of this constraint to relax it: if peer reboots, clock may go
1508 out-of-sync and half-open connections will not be reset.
1509 Actually, the problem would be not existing if all
1510 the implementations followed draft about maintaining clock
1511 via reboots. Linux-2.2 DOES NOT!
1513 However, we can relax time bounds for RST segments to MSL.
1515 if (rst && !time_before32(ktime_get_seconds(),
1516 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1521 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1522 int mib_idx, u32 *last_oow_ack_time);
1524 static inline void tcp_mib_init(struct net *net)
1527 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1528 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1529 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1530 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1534 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1536 tp->lost_skb_hint = NULL;
1539 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1541 tcp_clear_retrans_hints_partial(tp);
1542 tp->retransmit_skb_hint = NULL;
1545 union tcp_md5_addr {
1547 #if IS_ENABLED(CONFIG_IPV6)
1552 /* - key database */
1553 struct tcp_md5sig_key {
1554 struct hlist_node node;
1556 u8 family; /* AF_INET or AF_INET6 */
1558 union tcp_md5_addr addr;
1559 int l3index; /* set if key added with L3 scope */
1560 u8 key[TCP_MD5SIG_MAXKEYLEN];
1561 struct rcu_head rcu;
1565 struct tcp_md5sig_info {
1566 struct hlist_head head;
1567 struct rcu_head rcu;
1570 /* - pseudo header */
1571 struct tcp4_pseudohdr {
1579 struct tcp6_pseudohdr {
1580 struct in6_addr saddr;
1581 struct in6_addr daddr;
1583 __be32 protocol; /* including padding */
1586 union tcp_md5sum_block {
1587 struct tcp4_pseudohdr ip4;
1588 #if IS_ENABLED(CONFIG_IPV6)
1589 struct tcp6_pseudohdr ip6;
1593 /* - pool: digest algorithm, hash description and scratch buffer */
1594 struct tcp_md5sig_pool {
1595 struct ahash_request *md5_req;
1600 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1601 const struct sock *sk, const struct sk_buff *skb);
1602 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1603 int family, u8 prefixlen, int l3index,
1604 const u8 *newkey, u8 newkeylen, gfp_t gfp);
1605 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1606 int family, u8 prefixlen, int l3index);
1607 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1608 const struct sock *addr_sk);
1610 #ifdef CONFIG_TCP_MD5SIG
1611 #include <linux/jump_label.h>
1612 extern struct static_key_false tcp_md5_needed;
1613 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1614 const union tcp_md5_addr *addr,
1616 static inline struct tcp_md5sig_key *
1617 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1618 const union tcp_md5_addr *addr, int family)
1620 if (!static_branch_unlikely(&tcp_md5_needed))
1622 return __tcp_md5_do_lookup(sk, l3index, addr, family);
1625 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1627 static inline struct tcp_md5sig_key *
1628 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1629 const union tcp_md5_addr *addr, int family)
1633 #define tcp_twsk_md5_key(twsk) NULL
1636 bool tcp_alloc_md5sig_pool(void);
1638 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1639 static inline void tcp_put_md5sig_pool(void)
1644 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1645 unsigned int header_len);
1646 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1647 const struct tcp_md5sig_key *key);
1649 /* From tcp_fastopen.c */
1650 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1651 struct tcp_fastopen_cookie *cookie);
1652 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1653 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1655 struct tcp_fastopen_request {
1656 /* Fast Open cookie. Size 0 means a cookie request */
1657 struct tcp_fastopen_cookie cookie;
1658 struct msghdr *data; /* data in MSG_FASTOPEN */
1660 int copied; /* queued in tcp_connect() */
1661 struct ubuf_info *uarg;
1663 void tcp_free_fastopen_req(struct tcp_sock *tp);
1664 void tcp_fastopen_destroy_cipher(struct sock *sk);
1665 void tcp_fastopen_ctx_destroy(struct net *net);
1666 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1667 void *primary_key, void *backup_key);
1668 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
1670 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1671 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1672 struct request_sock *req,
1673 struct tcp_fastopen_cookie *foc,
1674 const struct dst_entry *dst);
1675 void tcp_fastopen_init_key_once(struct net *net);
1676 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1677 struct tcp_fastopen_cookie *cookie);
1678 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1679 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1680 #define TCP_FASTOPEN_KEY_MAX 2
1681 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1682 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1684 /* Fastopen key context */
1685 struct tcp_fastopen_context {
1686 siphash_key_t key[TCP_FASTOPEN_KEY_MAX];
1688 struct rcu_head rcu;
1691 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1692 void tcp_fastopen_active_disable(struct sock *sk);
1693 bool tcp_fastopen_active_should_disable(struct sock *sk);
1694 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1695 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1697 /* Caller needs to wrap with rcu_read_(un)lock() */
1699 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1701 struct tcp_fastopen_context *ctx;
1703 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1705 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1710 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1711 const struct tcp_fastopen_cookie *orig)
1713 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1714 orig->len == foc->len &&
1715 !memcmp(orig->val, foc->val, foc->len))
1721 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1726 /* Latencies incurred by various limits for a sender. They are
1727 * chronograph-like stats that are mutually exclusive.
1731 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1732 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1733 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1737 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1738 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1740 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1741 * the same memory storage than skb->destructor/_skb_refdst
1743 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1745 skb->destructor = NULL;
1746 skb->_skb_refdst = 0UL;
1749 #define tcp_skb_tsorted_save(skb) { \
1750 unsigned long _save = skb->_skb_refdst; \
1751 skb->_skb_refdst = 0UL;
1753 #define tcp_skb_tsorted_restore(skb) \
1754 skb->_skb_refdst = _save; \
1757 void tcp_write_queue_purge(struct sock *sk);
1759 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1761 return skb_rb_first(&sk->tcp_rtx_queue);
1764 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1766 return skb_rb_last(&sk->tcp_rtx_queue);
1769 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1771 return skb_peek(&sk->sk_write_queue);
1774 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1776 return skb_peek_tail(&sk->sk_write_queue);
1779 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1780 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1782 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1784 return skb_peek(&sk->sk_write_queue);
1787 static inline bool tcp_skb_is_last(const struct sock *sk,
1788 const struct sk_buff *skb)
1790 return skb_queue_is_last(&sk->sk_write_queue, skb);
1794 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
1797 * Since the write queue can have a temporary empty skb in it,
1798 * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
1800 static inline bool tcp_write_queue_empty(const struct sock *sk)
1802 const struct tcp_sock *tp = tcp_sk(sk);
1804 return tp->write_seq == tp->snd_nxt;
1807 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1809 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1812 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1814 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1817 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1819 __skb_queue_tail(&sk->sk_write_queue, skb);
1821 /* Queue it, remembering where we must start sending. */
1822 if (sk->sk_write_queue.next == skb)
1823 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1826 /* Insert new before skb on the write queue of sk. */
1827 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1828 struct sk_buff *skb,
1831 __skb_queue_before(&sk->sk_write_queue, skb, new);
1834 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1836 tcp_skb_tsorted_anchor_cleanup(skb);
1837 __skb_unlink(skb, &sk->sk_write_queue);
1840 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1842 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1844 tcp_skb_tsorted_anchor_cleanup(skb);
1845 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1848 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1850 list_del(&skb->tcp_tsorted_anchor);
1851 tcp_rtx_queue_unlink(skb, sk);
1852 sk_wmem_free_skb(sk, skb);
1855 static inline void tcp_push_pending_frames(struct sock *sk)
1857 if (tcp_send_head(sk)) {
1858 struct tcp_sock *tp = tcp_sk(sk);
1860 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1864 /* Start sequence of the skb just after the highest skb with SACKed
1865 * bit, valid only if sacked_out > 0 or when the caller has ensured
1866 * validity by itself.
1868 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1870 if (!tp->sacked_out)
1873 if (tp->highest_sack == NULL)
1876 return TCP_SKB_CB(tp->highest_sack)->seq;
1879 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1881 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1884 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1886 return tcp_sk(sk)->highest_sack;
1889 static inline void tcp_highest_sack_reset(struct sock *sk)
1891 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1894 /* Called when old skb is about to be deleted and replaced by new skb */
1895 static inline void tcp_highest_sack_replace(struct sock *sk,
1896 struct sk_buff *old,
1897 struct sk_buff *new)
1899 if (old == tcp_highest_sack(sk))
1900 tcp_sk(sk)->highest_sack = new;
1903 /* This helper checks if socket has IP_TRANSPARENT set */
1904 static inline bool inet_sk_transparent(const struct sock *sk)
1906 switch (sk->sk_state) {
1908 return inet_twsk(sk)->tw_transparent;
1909 case TCP_NEW_SYN_RECV:
1910 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1912 return inet_sk(sk)->transparent;
1915 /* Determines whether this is a thin stream (which may suffer from
1916 * increased latency). Used to trigger latency-reducing mechanisms.
1918 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1920 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1924 enum tcp_seq_states {
1925 TCP_SEQ_STATE_LISTENING,
1926 TCP_SEQ_STATE_ESTABLISHED,
1929 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1930 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1931 void tcp_seq_stop(struct seq_file *seq, void *v);
1933 struct tcp_seq_afinfo {
1937 struct tcp_iter_state {
1938 struct seq_net_private p;
1939 enum tcp_seq_states state;
1940 struct sock *syn_wait_sk;
1941 struct tcp_seq_afinfo *bpf_seq_afinfo;
1942 int bucket, offset, sbucket, num;
1946 extern struct request_sock_ops tcp_request_sock_ops;
1947 extern struct request_sock_ops tcp6_request_sock_ops;
1949 void tcp_v4_destroy_sock(struct sock *sk);
1951 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1952 netdev_features_t features);
1953 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
1954 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
1955 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
1956 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff));
1957 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb));
1958 int tcp_gro_complete(struct sk_buff *skb);
1960 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1962 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1964 struct net *net = sock_net((struct sock *)tp);
1965 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1968 /* @wake is one when sk_stream_write_space() calls us.
1969 * This sends EPOLLOUT only if notsent_bytes is half the limit.
1970 * This mimics the strategy used in sock_def_write_space().
1972 static inline bool tcp_stream_memory_free(const struct sock *sk, int wake)
1974 const struct tcp_sock *tp = tcp_sk(sk);
1975 u32 notsent_bytes = READ_ONCE(tp->write_seq) -
1976 READ_ONCE(tp->snd_nxt);
1978 return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
1981 #ifdef CONFIG_PROC_FS
1982 int tcp4_proc_init(void);
1983 void tcp4_proc_exit(void);
1986 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1987 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1988 const struct tcp_request_sock_ops *af_ops,
1989 struct sock *sk, struct sk_buff *skb);
1991 /* TCP af-specific functions */
1992 struct tcp_sock_af_ops {
1993 #ifdef CONFIG_TCP_MD5SIG
1994 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1995 const struct sock *addr_sk);
1996 int (*calc_md5_hash)(char *location,
1997 const struct tcp_md5sig_key *md5,
1998 const struct sock *sk,
1999 const struct sk_buff *skb);
2000 int (*md5_parse)(struct sock *sk,
2007 struct tcp_request_sock_ops {
2009 #ifdef CONFIG_TCP_MD5SIG
2010 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
2011 const struct sock *addr_sk);
2012 int (*calc_md5_hash) (char *location,
2013 const struct tcp_md5sig_key *md5,
2014 const struct sock *sk,
2015 const struct sk_buff *skb);
2017 void (*init_req)(struct request_sock *req,
2018 const struct sock *sk_listener,
2019 struct sk_buff *skb);
2020 #ifdef CONFIG_SYN_COOKIES
2021 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
2024 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
2025 const struct request_sock *req);
2026 u32 (*init_seq)(const struct sk_buff *skb);
2027 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
2028 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2029 struct flowi *fl, struct request_sock *req,
2030 struct tcp_fastopen_cookie *foc,
2031 enum tcp_synack_type synack_type,
2032 struct sk_buff *syn_skb);
2035 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2036 #if IS_ENABLED(CONFIG_IPV6)
2037 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2040 #ifdef CONFIG_SYN_COOKIES
2041 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2042 const struct sock *sk, struct sk_buff *skb,
2045 tcp_synq_overflow(sk);
2046 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2047 return ops->cookie_init_seq(skb, mss);
2050 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2051 const struct sock *sk, struct sk_buff *skb,
2058 int tcpv4_offload_init(void);
2060 void tcp_v4_init(void);
2061 void tcp_init(void);
2063 /* tcp_recovery.c */
2064 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2065 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2066 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2068 extern void tcp_rack_mark_lost(struct sock *sk);
2069 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2071 extern void tcp_rack_reo_timeout(struct sock *sk);
2072 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2074 /* At how many usecs into the future should the RTO fire? */
2075 static inline s64 tcp_rto_delta_us(const struct sock *sk)
2077 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2078 u32 rto = inet_csk(sk)->icsk_rto;
2079 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2081 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2085 * Save and compile IPv4 options, return a pointer to it
2087 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2088 struct sk_buff *skb)
2090 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2091 struct ip_options_rcu *dopt = NULL;
2094 int opt_size = sizeof(*dopt) + opt->optlen;
2096 dopt = kmalloc(opt_size, GFP_ATOMIC);
2097 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2105 /* locally generated TCP pure ACKs have skb->truesize == 2
2106 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2107 * This is much faster than dissecting the packet to find out.
2108 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2110 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2112 return skb->truesize == 2;
2115 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2120 static inline int tcp_inq(struct sock *sk)
2122 struct tcp_sock *tp = tcp_sk(sk);
2125 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2127 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2129 before(tp->urg_seq, tp->copied_seq) ||
2130 !before(tp->urg_seq, tp->rcv_nxt)) {
2132 answ = tp->rcv_nxt - tp->copied_seq;
2134 /* Subtract 1, if FIN was received */
2135 if (answ && sock_flag(sk, SOCK_DONE))
2138 answ = tp->urg_seq - tp->copied_seq;
2144 int tcp_peek_len(struct socket *sock);
2146 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2150 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2151 tp->segs_in += segs_in;
2152 if (skb->len > tcp_hdrlen(skb))
2153 tp->data_segs_in += segs_in;
2157 * TCP listen path runs lockless.
2158 * We forced "struct sock" to be const qualified to make sure
2159 * we don't modify one of its field by mistake.
2160 * Here, we increment sk_drops which is an atomic_t, so we can safely
2161 * make sock writable again.
2163 static inline void tcp_listendrop(const struct sock *sk)
2165 atomic_inc(&((struct sock *)sk)->sk_drops);
2166 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2169 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2172 * Interface for adding Upper Level Protocols over TCP
2175 #define TCP_ULP_NAME_MAX 16
2176 #define TCP_ULP_MAX 128
2177 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2179 struct tcp_ulp_ops {
2180 struct list_head list;
2182 /* initialize ulp */
2183 int (*init)(struct sock *sk);
2185 void (*update)(struct sock *sk, struct proto *p,
2186 void (*write_space)(struct sock *sk));
2188 void (*release)(struct sock *sk);
2190 int (*get_info)(const struct sock *sk, struct sk_buff *skb);
2191 size_t (*get_info_size)(const struct sock *sk);
2193 void (*clone)(const struct request_sock *req, struct sock *newsk,
2194 const gfp_t priority);
2196 char name[TCP_ULP_NAME_MAX];
2197 struct module *owner;
2199 int tcp_register_ulp(struct tcp_ulp_ops *type);
2200 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2201 int tcp_set_ulp(struct sock *sk, const char *name);
2202 void tcp_get_available_ulp(char *buf, size_t len);
2203 void tcp_cleanup_ulp(struct sock *sk);
2204 void tcp_update_ulp(struct sock *sk, struct proto *p,
2205 void (*write_space)(struct sock *sk));
2207 #define MODULE_ALIAS_TCP_ULP(name) \
2208 __MODULE_INFO(alias, alias_userspace, name); \
2209 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2214 #ifdef CONFIG_BPF_STREAM_PARSER
2215 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
2216 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2218 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2221 #endif /* CONFIG_BPF_STREAM_PARSER */
2223 #ifdef CONFIG_NET_SOCK_MSG
2224 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes,
2226 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock,
2227 struct msghdr *msg, int len, int flags);
2228 #endif /* CONFIG_NET_SOCK_MSG */
2230 #ifdef CONFIG_CGROUP_BPF
2231 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2232 struct sk_buff *skb,
2233 unsigned int end_offset)
2236 skops->skb_data_end = skb->data + end_offset;
2239 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2240 struct sk_buff *skb,
2241 unsigned int end_offset)
2246 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2247 * is < 0, then the BPF op failed (for example if the loaded BPF
2248 * program does not support the chosen operation or there is no BPF
2252 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2254 struct bpf_sock_ops_kern sock_ops;
2257 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2258 if (sk_fullsock(sk)) {
2259 sock_ops.is_fullsock = 1;
2260 sock_owned_by_me(sk);
2266 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2268 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2270 ret = sock_ops.reply;
2276 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2278 u32 args[2] = {arg1, arg2};
2280 return tcp_call_bpf(sk, op, 2, args);
2283 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2286 u32 args[3] = {arg1, arg2, arg3};
2288 return tcp_call_bpf(sk, op, 3, args);
2292 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2297 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2302 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2310 static inline u32 tcp_timeout_init(struct sock *sk)
2314 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2317 timeout = TCP_TIMEOUT_INIT;
2321 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2325 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2332 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2334 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2337 static inline void tcp_bpf_rtt(struct sock *sk)
2339 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2340 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2343 #if IS_ENABLED(CONFIG_SMC)
2344 extern struct static_key_false tcp_have_smc;
2347 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2348 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2349 void (*cad)(struct sock *sk, u32 ack_seq));
2350 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2351 void clean_acked_data_flush(void);
2354 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2355 static inline void tcp_add_tx_delay(struct sk_buff *skb,
2356 const struct tcp_sock *tp)
2358 if (static_branch_unlikely(&tcp_tx_delay_enabled))
2359 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2362 /* Compute Earliest Departure Time for some control packets
2363 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2365 static inline u64 tcp_transmit_time(const struct sock *sk)
2367 if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2368 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2369 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2371 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;