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 void tcp_remove_empty_skb(struct sock *sk, struct sk_buff *skb);
326 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
327 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
328 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
329 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
331 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
332 size_t size, int flags);
333 struct sk_buff *tcp_build_frag(struct sock *sk, int size_goal, int flags,
334 struct page *page, int offset, size_t *size);
335 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
336 size_t size, int flags);
337 int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
338 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
340 void tcp_release_cb(struct sock *sk);
341 void tcp_wfree(struct sk_buff *skb);
342 void tcp_write_timer_handler(struct sock *sk);
343 void tcp_delack_timer_handler(struct sock *sk);
344 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
345 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
346 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
347 void tcp_rcv_space_adjust(struct sock *sk);
348 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
349 void tcp_twsk_destructor(struct sock *sk);
350 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
351 struct pipe_inode_info *pipe, size_t len,
354 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
355 static inline void tcp_dec_quickack_mode(struct sock *sk,
356 const unsigned int pkts)
358 struct inet_connection_sock *icsk = inet_csk(sk);
360 if (icsk->icsk_ack.quick) {
361 if (pkts >= icsk->icsk_ack.quick) {
362 icsk->icsk_ack.quick = 0;
363 /* Leaving quickack mode we deflate ATO. */
364 icsk->icsk_ack.ato = TCP_ATO_MIN;
366 icsk->icsk_ack.quick -= pkts;
371 #define TCP_ECN_QUEUE_CWR 2
372 #define TCP_ECN_DEMAND_CWR 4
373 #define TCP_ECN_SEEN 8
383 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
385 const struct tcphdr *th);
386 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
387 struct request_sock *req, bool fastopen,
389 int tcp_child_process(struct sock *parent, struct sock *child,
390 struct sk_buff *skb);
391 void tcp_enter_loss(struct sock *sk);
392 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag);
393 void tcp_clear_retrans(struct tcp_sock *tp);
394 void tcp_update_metrics(struct sock *sk);
395 void tcp_init_metrics(struct sock *sk);
396 void tcp_metrics_init(void);
397 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
398 void __tcp_close(struct sock *sk, long timeout);
399 void tcp_close(struct sock *sk, long timeout);
400 void tcp_init_sock(struct sock *sk);
401 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb);
402 __poll_t tcp_poll(struct file *file, struct socket *sock,
403 struct poll_table_struct *wait);
404 int tcp_getsockopt(struct sock *sk, int level, int optname,
405 char __user *optval, int __user *optlen);
406 bool tcp_bpf_bypass_getsockopt(int level, int optname);
407 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
408 unsigned int optlen);
409 void tcp_set_keepalive(struct sock *sk, int val);
410 void tcp_syn_ack_timeout(const struct request_sock *req);
411 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
412 int flags, int *addr_len);
413 int tcp_set_rcvlowat(struct sock *sk, int val);
414 int tcp_set_window_clamp(struct sock *sk, int val);
415 void tcp_data_ready(struct sock *sk);
417 int tcp_mmap(struct file *file, struct socket *sock,
418 struct vm_area_struct *vma);
420 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
421 struct tcp_options_received *opt_rx,
422 int estab, struct tcp_fastopen_cookie *foc);
423 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
426 * BPF SKB-less helpers
428 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
429 struct tcphdr *th, u32 *cookie);
430 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
431 struct tcphdr *th, u32 *cookie);
432 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
433 const struct tcp_request_sock_ops *af_ops,
434 struct sock *sk, struct tcphdr *th);
436 * TCP v4 functions exported for the inet6 API
439 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
440 void tcp_v4_mtu_reduced(struct sock *sk);
441 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
442 void tcp_ld_RTO_revert(struct sock *sk, u32 seq);
443 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
444 struct sock *tcp_create_openreq_child(const struct sock *sk,
445 struct request_sock *req,
446 struct sk_buff *skb);
447 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
448 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
449 struct request_sock *req,
450 struct dst_entry *dst,
451 struct request_sock *req_unhash,
453 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
454 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
455 int tcp_connect(struct sock *sk);
456 enum tcp_synack_type {
461 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
462 struct request_sock *req,
463 struct tcp_fastopen_cookie *foc,
464 enum tcp_synack_type synack_type,
465 struct sk_buff *syn_skb);
466 int tcp_disconnect(struct sock *sk, int flags);
468 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
469 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
470 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
472 /* From syncookies.c */
473 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
474 struct request_sock *req,
475 struct dst_entry *dst, u32 tsoff);
476 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
478 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
479 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
480 struct sock *sk, struct sk_buff *skb);
481 #ifdef CONFIG_SYN_COOKIES
483 /* Syncookies use a monotonic timer which increments every 60 seconds.
484 * This counter is used both as a hash input and partially encoded into
485 * the cookie value. A cookie is only validated further if the delta
486 * between the current counter value and the encoded one is less than this,
487 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
488 * the counter advances immediately after a cookie is generated).
490 #define MAX_SYNCOOKIE_AGE 2
491 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
492 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
494 /* syncookies: remember time of last synqueue overflow
495 * But do not dirty this field too often (once per second is enough)
496 * It is racy as we do not hold a lock, but race is very minor.
498 static inline void tcp_synq_overflow(const struct sock *sk)
500 unsigned int last_overflow;
501 unsigned int now = jiffies;
503 if (sk->sk_reuseport) {
504 struct sock_reuseport *reuse;
506 reuse = rcu_dereference(sk->sk_reuseport_cb);
508 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
509 if (!time_between32(now, last_overflow,
511 WRITE_ONCE(reuse->synq_overflow_ts, now);
516 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
517 if (!time_between32(now, last_overflow, last_overflow + HZ))
518 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
521 /* syncookies: no recent synqueue overflow on this listening socket? */
522 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
524 unsigned int last_overflow;
525 unsigned int now = jiffies;
527 if (sk->sk_reuseport) {
528 struct sock_reuseport *reuse;
530 reuse = rcu_dereference(sk->sk_reuseport_cb);
532 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
533 return !time_between32(now, last_overflow - HZ,
535 TCP_SYNCOOKIE_VALID);
539 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
541 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
542 * then we're under synflood. However, we have to use
543 * 'last_overflow - HZ' as lower bound. That's because a concurrent
544 * tcp_synq_overflow() could update .ts_recent_stamp after we read
545 * jiffies but before we store .ts_recent_stamp into last_overflow,
546 * which could lead to rejecting a valid syncookie.
548 return !time_between32(now, last_overflow - HZ,
549 last_overflow + TCP_SYNCOOKIE_VALID);
552 static inline u32 tcp_cookie_time(void)
554 u64 val = get_jiffies_64();
556 do_div(val, TCP_SYNCOOKIE_PERIOD);
560 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
562 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
563 u64 cookie_init_timestamp(struct request_sock *req, u64 now);
564 bool cookie_timestamp_decode(const struct net *net,
565 struct tcp_options_received *opt);
566 bool cookie_ecn_ok(const struct tcp_options_received *opt,
567 const struct net *net, const struct dst_entry *dst);
569 /* From net/ipv6/syncookies.c */
570 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
572 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
574 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
575 const struct tcphdr *th, u16 *mssp);
576 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
580 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
582 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
583 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
584 void tcp_retransmit_timer(struct sock *sk);
585 void tcp_xmit_retransmit_queue(struct sock *);
586 void tcp_simple_retransmit(struct sock *);
587 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
588 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
590 TCP_FRAG_IN_WRITE_QUEUE,
591 TCP_FRAG_IN_RTX_QUEUE,
593 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
594 struct sk_buff *skb, u32 len,
595 unsigned int mss_now, gfp_t gfp);
597 void tcp_send_probe0(struct sock *);
598 void tcp_send_partial(struct sock *);
599 int tcp_write_wakeup(struct sock *, int mib);
600 void tcp_send_fin(struct sock *sk);
601 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
602 int tcp_send_synack(struct sock *);
603 void tcp_push_one(struct sock *, unsigned int mss_now);
604 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
605 void tcp_send_ack(struct sock *sk);
606 void tcp_send_delayed_ack(struct sock *sk);
607 void tcp_send_loss_probe(struct sock *sk);
608 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
609 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
610 const struct sk_buff *next_skb);
613 void tcp_rearm_rto(struct sock *sk);
614 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
615 void tcp_reset(struct sock *sk, struct sk_buff *skb);
616 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
617 void tcp_fin(struct sock *sk);
620 void tcp_init_xmit_timers(struct sock *);
621 static inline void tcp_clear_xmit_timers(struct sock *sk)
623 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
626 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
629 inet_csk_clear_xmit_timers(sk);
632 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
633 unsigned int tcp_current_mss(struct sock *sk);
634 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when);
636 /* Bound MSS / TSO packet size with the half of the window */
637 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
641 /* When peer uses tiny windows, there is no use in packetizing
642 * to sub-MSS pieces for the sake of SWS or making sure there
643 * are enough packets in the pipe for fast recovery.
645 * On the other hand, for extremely large MSS devices, handling
646 * smaller than MSS windows in this way does make sense.
648 if (tp->max_window > TCP_MSS_DEFAULT)
649 cutoff = (tp->max_window >> 1);
651 cutoff = tp->max_window;
653 if (cutoff && pktsize > cutoff)
654 return max_t(int, cutoff, 68U - tp->tcp_header_len);
660 void tcp_get_info(struct sock *, struct tcp_info *);
662 /* Read 'sendfile()'-style from a TCP socket */
663 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
664 sk_read_actor_t recv_actor);
666 void tcp_initialize_rcv_mss(struct sock *sk);
668 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
669 int tcp_mss_to_mtu(struct sock *sk, int mss);
670 void tcp_mtup_init(struct sock *sk);
672 static inline void tcp_bound_rto(const struct sock *sk)
674 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
675 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
678 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
680 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
683 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
685 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
686 ntohl(TCP_FLAG_ACK) |
690 static inline void tcp_fast_path_on(struct tcp_sock *tp)
692 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
695 static inline void tcp_fast_path_check(struct sock *sk)
697 struct tcp_sock *tp = tcp_sk(sk);
699 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
701 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
703 tcp_fast_path_on(tp);
706 /* Compute the actual rto_min value */
707 static inline u32 tcp_rto_min(struct sock *sk)
709 const struct dst_entry *dst = __sk_dst_get(sk);
710 u32 rto_min = inet_csk(sk)->icsk_rto_min;
712 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
713 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
717 static inline u32 tcp_rto_min_us(struct sock *sk)
719 return jiffies_to_usecs(tcp_rto_min(sk));
722 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
724 return dst_metric_locked(dst, RTAX_CC_ALGO);
727 /* Minimum RTT in usec. ~0 means not available. */
728 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
730 return minmax_get(&tp->rtt_min);
733 /* Compute the actual receive window we are currently advertising.
734 * Rcv_nxt can be after the window if our peer push more data
735 * than the offered window.
737 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
739 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
746 /* Choose a new window, without checks for shrinking, and without
747 * scaling applied to the result. The caller does these things
748 * if necessary. This is a "raw" window selection.
750 u32 __tcp_select_window(struct sock *sk);
752 void tcp_send_window_probe(struct sock *sk);
754 /* TCP uses 32bit jiffies to save some space.
755 * Note that this is different from tcp_time_stamp, which
756 * historically has been the same until linux-4.13.
758 #define tcp_jiffies32 ((u32)jiffies)
761 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
762 * It is no longer tied to jiffies, but to 1 ms clock.
763 * Note: double check if you want to use tcp_jiffies32 instead of this.
765 #define TCP_TS_HZ 1000
767 static inline u64 tcp_clock_ns(void)
769 return ktime_get_ns();
772 static inline u64 tcp_clock_us(void)
774 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
777 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
778 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
780 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
783 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
784 static inline u32 tcp_ns_to_ts(u64 ns)
786 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
789 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
790 static inline u32 tcp_time_stamp_raw(void)
792 return tcp_ns_to_ts(tcp_clock_ns());
795 void tcp_mstamp_refresh(struct tcp_sock *tp);
797 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
799 return max_t(s64, t1 - t0, 0);
802 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
804 return tcp_ns_to_ts(skb->skb_mstamp_ns);
807 /* provide the departure time in us unit */
808 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
810 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
814 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
816 #define TCPHDR_FIN 0x01
817 #define TCPHDR_SYN 0x02
818 #define TCPHDR_RST 0x04
819 #define TCPHDR_PSH 0x08
820 #define TCPHDR_ACK 0x10
821 #define TCPHDR_URG 0x20
822 #define TCPHDR_ECE 0x40
823 #define TCPHDR_CWR 0x80
825 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
827 /* This is what the send packet queuing engine uses to pass
828 * TCP per-packet control information to the transmission code.
829 * We also store the host-order sequence numbers in here too.
830 * This is 44 bytes if IPV6 is enabled.
831 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
834 __u32 seq; /* Starting sequence number */
835 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
837 /* Note : tcp_tw_isn is used in input path only
838 * (isn chosen by tcp_timewait_state_process())
840 * tcp_gso_segs/size are used in write queue only,
841 * cf tcp_skb_pcount()/tcp_skb_mss()
849 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
851 __u8 sacked; /* State flags for SACK. */
852 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
853 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
854 #define TCPCB_LOST 0x04 /* SKB is lost */
855 #define TCPCB_TAGBITS 0x07 /* All tag bits */
856 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
857 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
858 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
861 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
862 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
863 eor:1, /* Is skb MSG_EOR marked? */
864 has_rxtstamp:1, /* SKB has a RX timestamp */
866 __u32 ack_seq; /* Sequence number ACK'd */
869 /* There is space for up to 24 bytes */
870 __u32 in_flight:30,/* Bytes in flight at transmit */
871 is_app_limited:1, /* cwnd not fully used? */
873 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
875 /* start of send pipeline phase */
877 /* when we reached the "delivered" count */
878 u64 delivered_mstamp;
879 } tx; /* only used for outgoing skbs */
881 struct inet_skb_parm h4;
882 #if IS_ENABLED(CONFIG_IPV6)
883 struct inet6_skb_parm h6;
885 } header; /* For incoming skbs */
889 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
891 extern const struct inet_connection_sock_af_ops ipv4_specific;
893 #if IS_ENABLED(CONFIG_IPV6)
894 /* This is the variant of inet6_iif() that must be used by TCP,
895 * as TCP moves IP6CB into a different location in skb->cb[]
897 static inline int tcp_v6_iif(const struct sk_buff *skb)
899 return TCP_SKB_CB(skb)->header.h6.iif;
902 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
904 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
906 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
909 /* TCP_SKB_CB reference means this can not be used from early demux */
910 static inline int tcp_v6_sdif(const struct sk_buff *skb)
912 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
913 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
914 return TCP_SKB_CB(skb)->header.h6.iif;
919 extern const struct inet_connection_sock_af_ops ipv6_specific;
921 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
922 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
923 INDIRECT_CALLABLE_DECLARE(void tcp_v6_early_demux(struct sk_buff *skb));
927 /* TCP_SKB_CB reference means this can not be used from early demux */
928 static inline int tcp_v4_sdif(struct sk_buff *skb)
930 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
931 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
932 return TCP_SKB_CB(skb)->header.h4.iif;
937 /* Due to TSO, an SKB can be composed of multiple actual
938 * packets. To keep these tracked properly, we use this.
940 static inline int tcp_skb_pcount(const struct sk_buff *skb)
942 return TCP_SKB_CB(skb)->tcp_gso_segs;
945 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
947 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
950 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
952 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
955 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
956 static inline int tcp_skb_mss(const struct sk_buff *skb)
958 return TCP_SKB_CB(skb)->tcp_gso_size;
961 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
963 return likely(!TCP_SKB_CB(skb)->eor);
966 static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
967 const struct sk_buff *from)
969 return likely(tcp_skb_can_collapse_to(to) &&
970 mptcp_skb_can_collapse(to, from));
973 /* Events passed to congestion control interface */
975 CA_EVENT_TX_START, /* first transmit when no packets in flight */
976 CA_EVENT_CWND_RESTART, /* congestion window restart */
977 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
978 CA_EVENT_LOSS, /* loss timeout */
979 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
980 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
983 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
984 enum tcp_ca_ack_event_flags {
985 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
986 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
987 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
991 * Interface for adding new TCP congestion control handlers
993 #define TCP_CA_NAME_MAX 16
994 #define TCP_CA_MAX 128
995 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
997 #define TCP_CA_UNSPEC 0
999 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1000 #define TCP_CONG_NON_RESTRICTED 0x1
1001 /* Requires ECN/ECT set on all packets */
1002 #define TCP_CONG_NEEDS_ECN 0x2
1003 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1013 /* A rate sample measures the number of (original/retransmitted) data
1014 * packets delivered "delivered" over an interval of time "interval_us".
1015 * The tcp_rate.c code fills in the rate sample, and congestion
1016 * control modules that define a cong_control function to run at the end
1017 * of ACK processing can optionally chose to consult this sample when
1018 * setting cwnd and pacing rate.
1019 * A sample is invalid if "delivered" or "interval_us" is negative.
1021 struct rate_sample {
1022 u64 prior_mstamp; /* starting timestamp for interval */
1023 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1024 s32 delivered; /* number of packets delivered over interval */
1025 long interval_us; /* time for tp->delivered to incr "delivered" */
1026 u32 snd_interval_us; /* snd interval for delivered packets */
1027 u32 rcv_interval_us; /* rcv interval for delivered packets */
1028 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1029 int losses; /* number of packets marked lost upon ACK */
1030 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1031 u32 prior_in_flight; /* in flight before this ACK */
1032 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1033 bool is_retrans; /* is sample from retransmission? */
1034 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1037 struct tcp_congestion_ops {
1038 struct list_head list;
1042 /* initialize private data (optional) */
1043 void (*init)(struct sock *sk);
1044 /* cleanup private data (optional) */
1045 void (*release)(struct sock *sk);
1047 /* return slow start threshold (required) */
1048 u32 (*ssthresh)(struct sock *sk);
1049 /* do new cwnd calculation (required) */
1050 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1051 /* call before changing ca_state (optional) */
1052 void (*set_state)(struct sock *sk, u8 new_state);
1053 /* call when cwnd event occurs (optional) */
1054 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1055 /* call when ack arrives (optional) */
1056 void (*in_ack_event)(struct sock *sk, u32 flags);
1057 /* new value of cwnd after loss (required) */
1058 u32 (*undo_cwnd)(struct sock *sk);
1059 /* hook for packet ack accounting (optional) */
1060 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1061 /* override sysctl_tcp_min_tso_segs */
1062 u32 (*min_tso_segs)(struct sock *sk);
1063 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1064 u32 (*sndbuf_expand)(struct sock *sk);
1065 /* call when packets are delivered to update cwnd and pacing rate,
1066 * after all the ca_state processing. (optional)
1068 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1069 /* get info for inet_diag (optional) */
1070 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1071 union tcp_cc_info *info);
1073 char name[TCP_CA_NAME_MAX];
1074 struct module *owner;
1077 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1078 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1080 void tcp_assign_congestion_control(struct sock *sk);
1081 void tcp_init_congestion_control(struct sock *sk);
1082 void tcp_cleanup_congestion_control(struct sock *sk);
1083 int tcp_set_default_congestion_control(struct net *net, const char *name);
1084 void tcp_get_default_congestion_control(struct net *net, char *name);
1085 void tcp_get_available_congestion_control(char *buf, size_t len);
1086 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1087 int tcp_set_allowed_congestion_control(char *allowed);
1088 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1089 bool cap_net_admin);
1090 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1091 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1093 u32 tcp_reno_ssthresh(struct sock *sk);
1094 u32 tcp_reno_undo_cwnd(struct sock *sk);
1095 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1096 extern struct tcp_congestion_ops tcp_reno;
1098 struct tcp_congestion_ops *tcp_ca_find(const char *name);
1099 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1100 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1102 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1104 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1110 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1112 const struct inet_connection_sock *icsk = inet_csk(sk);
1114 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1117 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1119 struct inet_connection_sock *icsk = inet_csk(sk);
1121 if (icsk->icsk_ca_ops->set_state)
1122 icsk->icsk_ca_ops->set_state(sk, ca_state);
1123 icsk->icsk_ca_state = ca_state;
1126 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1128 const struct inet_connection_sock *icsk = inet_csk(sk);
1130 if (icsk->icsk_ca_ops->cwnd_event)
1131 icsk->icsk_ca_ops->cwnd_event(sk, event);
1134 /* From tcp_rate.c */
1135 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1136 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1137 struct rate_sample *rs);
1138 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1139 bool is_sack_reneg, struct rate_sample *rs);
1140 void tcp_rate_check_app_limited(struct sock *sk);
1142 /* These functions determine how the current flow behaves in respect of SACK
1143 * handling. SACK is negotiated with the peer, and therefore it can vary
1144 * between different flows.
1146 * tcp_is_sack - SACK enabled
1147 * tcp_is_reno - No SACK
1149 static inline int tcp_is_sack(const struct tcp_sock *tp)
1151 return likely(tp->rx_opt.sack_ok);
1154 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1156 return !tcp_is_sack(tp);
1159 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1161 return tp->sacked_out + tp->lost_out;
1164 /* This determines how many packets are "in the network" to the best
1165 * of our knowledge. In many cases it is conservative, but where
1166 * detailed information is available from the receiver (via SACK
1167 * blocks etc.) we can make more aggressive calculations.
1169 * Use this for decisions involving congestion control, use just
1170 * tp->packets_out to determine if the send queue is empty or not.
1172 * Read this equation as:
1174 * "Packets sent once on transmission queue" MINUS
1175 * "Packets left network, but not honestly ACKed yet" PLUS
1176 * "Packets fast retransmitted"
1178 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1180 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1183 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1185 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1187 return tp->snd_cwnd < tp->snd_ssthresh;
1190 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1192 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1195 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1197 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1198 (1 << inet_csk(sk)->icsk_ca_state);
1201 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1202 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1205 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1207 const struct tcp_sock *tp = tcp_sk(sk);
1209 if (tcp_in_cwnd_reduction(sk))
1210 return tp->snd_ssthresh;
1212 return max(tp->snd_ssthresh,
1213 ((tp->snd_cwnd >> 1) +
1214 (tp->snd_cwnd >> 2)));
1217 /* Use define here intentionally to get WARN_ON location shown at the caller */
1218 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1220 void tcp_enter_cwr(struct sock *sk);
1221 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1223 /* The maximum number of MSS of available cwnd for which TSO defers
1224 * sending if not using sysctl_tcp_tso_win_divisor.
1226 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1231 /* Returns end sequence number of the receiver's advertised window */
1232 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1234 return tp->snd_una + tp->snd_wnd;
1237 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1238 * flexible approach. The RFC suggests cwnd should not be raised unless
1239 * it was fully used previously. And that's exactly what we do in
1240 * congestion avoidance mode. But in slow start we allow cwnd to grow
1241 * as long as the application has used half the cwnd.
1243 * cwnd is 10 (IW10), but application sends 9 frames.
1244 * We allow cwnd to reach 18 when all frames are ACKed.
1245 * This check is safe because it's as aggressive as slow start which already
1246 * risks 100% overshoot. The advantage is that we discourage application to
1247 * either send more filler packets or data to artificially blow up the cwnd
1248 * usage, and allow application-limited process to probe bw more aggressively.
1250 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1252 const struct tcp_sock *tp = tcp_sk(sk);
1254 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1255 if (tcp_in_slow_start(tp))
1256 return tp->snd_cwnd < 2 * tp->max_packets_out;
1258 return tp->is_cwnd_limited;
1261 /* BBR congestion control needs pacing.
1262 * Same remark for SO_MAX_PACING_RATE.
1263 * sch_fq packet scheduler is efficiently handling pacing,
1264 * but is not always installed/used.
1265 * Return true if TCP stack should pace packets itself.
1267 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1269 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1272 /* Estimates in how many jiffies next packet for this flow can be sent.
1273 * Scheduling a retransmit timer too early would be silly.
1275 static inline unsigned long tcp_pacing_delay(const struct sock *sk)
1277 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
1279 return delay > 0 ? nsecs_to_jiffies(delay) : 0;
1282 static inline void tcp_reset_xmit_timer(struct sock *sk,
1285 const unsigned long max_when)
1287 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk),
1291 /* Something is really bad, we could not queue an additional packet,
1292 * because qdisc is full or receiver sent a 0 window, or we are paced.
1293 * We do not want to add fuel to the fire, or abort too early,
1294 * so make sure the timer we arm now is at least 200ms in the future,
1295 * regardless of current icsk_rto value (as it could be ~2ms)
1297 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1299 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1302 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1303 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1304 unsigned long max_when)
1306 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1,
1307 inet_csk(sk)->icsk_backoff);
1308 u64 when = (u64)tcp_probe0_base(sk) << backoff;
1310 return (unsigned long)min_t(u64, when, max_when);
1313 static inline void tcp_check_probe_timer(struct sock *sk)
1315 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1316 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1317 tcp_probe0_base(sk), TCP_RTO_MAX);
1320 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1325 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1331 * Calculate(/check) TCP checksum
1333 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1334 __be32 daddr, __wsum base)
1336 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1339 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1341 return !skb_csum_unnecessary(skb) &&
1342 __skb_checksum_complete(skb);
1345 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1346 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1347 void tcp_set_state(struct sock *sk, int state);
1348 void tcp_done(struct sock *sk);
1349 int tcp_abort(struct sock *sk, int err);
1351 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1354 rx_opt->num_sacks = 0;
1357 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1359 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1361 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1362 struct tcp_sock *tp = tcp_sk(sk);
1365 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1366 ca_ops->cong_control)
1368 delta = tcp_jiffies32 - tp->lsndtime;
1369 if (delta > inet_csk(sk)->icsk_rto)
1370 tcp_cwnd_restart(sk, delta);
1373 /* Determine a window scaling and initial window to offer. */
1374 void tcp_select_initial_window(const struct sock *sk, int __space,
1375 __u32 mss, __u32 *rcv_wnd,
1376 __u32 *window_clamp, int wscale_ok,
1377 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1379 static inline int tcp_win_from_space(const struct sock *sk, int space)
1381 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1383 return tcp_adv_win_scale <= 0 ?
1384 (space>>(-tcp_adv_win_scale)) :
1385 space - (space>>tcp_adv_win_scale);
1388 /* Note: caller must be prepared to deal with negative returns */
1389 static inline int tcp_space(const struct sock *sk)
1391 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1392 READ_ONCE(sk->sk_backlog.len) -
1393 atomic_read(&sk->sk_rmem_alloc));
1396 static inline int tcp_full_space(const struct sock *sk)
1398 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1401 void tcp_cleanup_rbuf(struct sock *sk, int copied);
1403 /* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1404 * If 87.5 % (7/8) of the space has been consumed, we want to override
1405 * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1406 * len/truesize ratio.
1408 static inline bool tcp_rmem_pressure(const struct sock *sk)
1410 int rcvbuf, threshold;
1412 if (tcp_under_memory_pressure(sk))
1415 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1416 threshold = rcvbuf - (rcvbuf >> 3);
1418 return atomic_read(&sk->sk_rmem_alloc) > threshold;
1421 static inline bool tcp_epollin_ready(const struct sock *sk, int target)
1423 const struct tcp_sock *tp = tcp_sk(sk);
1424 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq);
1429 return (avail >= target) || tcp_rmem_pressure(sk) ||
1430 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss);
1433 extern void tcp_openreq_init_rwin(struct request_sock *req,
1434 const struct sock *sk_listener,
1435 const struct dst_entry *dst);
1437 void tcp_enter_memory_pressure(struct sock *sk);
1438 void tcp_leave_memory_pressure(struct sock *sk);
1440 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1442 struct net *net = sock_net((struct sock *)tp);
1444 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1447 static inline int keepalive_time_when(const struct tcp_sock *tp)
1449 struct net *net = sock_net((struct sock *)tp);
1451 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1454 static inline int keepalive_probes(const struct tcp_sock *tp)
1456 struct net *net = sock_net((struct sock *)tp);
1458 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1461 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1463 const struct inet_connection_sock *icsk = &tp->inet_conn;
1465 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1466 tcp_jiffies32 - tp->rcv_tstamp);
1469 static inline int tcp_fin_time(const struct sock *sk)
1471 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1472 const int rto = inet_csk(sk)->icsk_rto;
1474 if (fin_timeout < (rto << 2) - (rto >> 1))
1475 fin_timeout = (rto << 2) - (rto >> 1);
1480 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1483 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1485 if (unlikely(!time_before32(ktime_get_seconds(),
1486 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1489 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1490 * then following tcp messages have valid values. Ignore 0 value,
1491 * or else 'negative' tsval might forbid us to accept their packets.
1493 if (!rx_opt->ts_recent)
1498 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1501 if (tcp_paws_check(rx_opt, 0))
1504 /* RST segments are not recommended to carry timestamp,
1505 and, if they do, it is recommended to ignore PAWS because
1506 "their cleanup function should take precedence over timestamps."
1507 Certainly, it is mistake. It is necessary to understand the reasons
1508 of this constraint to relax it: if peer reboots, clock may go
1509 out-of-sync and half-open connections will not be reset.
1510 Actually, the problem would be not existing if all
1511 the implementations followed draft about maintaining clock
1512 via reboots. Linux-2.2 DOES NOT!
1514 However, we can relax time bounds for RST segments to MSL.
1516 if (rst && !time_before32(ktime_get_seconds(),
1517 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1522 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1523 int mib_idx, u32 *last_oow_ack_time);
1525 static inline void tcp_mib_init(struct net *net)
1528 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1529 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1530 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1531 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1535 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1537 tp->lost_skb_hint = NULL;
1540 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1542 tcp_clear_retrans_hints_partial(tp);
1543 tp->retransmit_skb_hint = NULL;
1546 union tcp_md5_addr {
1548 #if IS_ENABLED(CONFIG_IPV6)
1553 /* - key database */
1554 struct tcp_md5sig_key {
1555 struct hlist_node node;
1557 u8 family; /* AF_INET or AF_INET6 */
1559 union tcp_md5_addr addr;
1560 int l3index; /* set if key added with L3 scope */
1561 u8 key[TCP_MD5SIG_MAXKEYLEN];
1562 struct rcu_head rcu;
1566 struct tcp_md5sig_info {
1567 struct hlist_head head;
1568 struct rcu_head rcu;
1571 /* - pseudo header */
1572 struct tcp4_pseudohdr {
1580 struct tcp6_pseudohdr {
1581 struct in6_addr saddr;
1582 struct in6_addr daddr;
1584 __be32 protocol; /* including padding */
1587 union tcp_md5sum_block {
1588 struct tcp4_pseudohdr ip4;
1589 #if IS_ENABLED(CONFIG_IPV6)
1590 struct tcp6_pseudohdr ip6;
1594 /* - pool: digest algorithm, hash description and scratch buffer */
1595 struct tcp_md5sig_pool {
1596 struct ahash_request *md5_req;
1601 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1602 const struct sock *sk, const struct sk_buff *skb);
1603 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1604 int family, u8 prefixlen, int l3index,
1605 const u8 *newkey, u8 newkeylen, gfp_t gfp);
1606 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1607 int family, u8 prefixlen, int l3index);
1608 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1609 const struct sock *addr_sk);
1611 #ifdef CONFIG_TCP_MD5SIG
1612 #include <linux/jump_label.h>
1613 extern struct static_key_false tcp_md5_needed;
1614 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1615 const union tcp_md5_addr *addr,
1617 static inline struct tcp_md5sig_key *
1618 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1619 const union tcp_md5_addr *addr, int family)
1621 if (!static_branch_unlikely(&tcp_md5_needed))
1623 return __tcp_md5_do_lookup(sk, l3index, addr, family);
1626 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1628 static inline struct tcp_md5sig_key *
1629 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1630 const union tcp_md5_addr *addr, int family)
1634 #define tcp_twsk_md5_key(twsk) NULL
1637 bool tcp_alloc_md5sig_pool(void);
1639 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1640 static inline void tcp_put_md5sig_pool(void)
1645 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1646 unsigned int header_len);
1647 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1648 const struct tcp_md5sig_key *key);
1650 /* From tcp_fastopen.c */
1651 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1652 struct tcp_fastopen_cookie *cookie);
1653 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1654 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1656 struct tcp_fastopen_request {
1657 /* Fast Open cookie. Size 0 means a cookie request */
1658 struct tcp_fastopen_cookie cookie;
1659 struct msghdr *data; /* data in MSG_FASTOPEN */
1661 int copied; /* queued in tcp_connect() */
1662 struct ubuf_info *uarg;
1664 void tcp_free_fastopen_req(struct tcp_sock *tp);
1665 void tcp_fastopen_destroy_cipher(struct sock *sk);
1666 void tcp_fastopen_ctx_destroy(struct net *net);
1667 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1668 void *primary_key, void *backup_key);
1669 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
1671 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1672 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1673 struct request_sock *req,
1674 struct tcp_fastopen_cookie *foc,
1675 const struct dst_entry *dst);
1676 void tcp_fastopen_init_key_once(struct net *net);
1677 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1678 struct tcp_fastopen_cookie *cookie);
1679 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1680 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1681 #define TCP_FASTOPEN_KEY_MAX 2
1682 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1683 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1685 /* Fastopen key context */
1686 struct tcp_fastopen_context {
1687 siphash_key_t key[TCP_FASTOPEN_KEY_MAX];
1689 struct rcu_head rcu;
1692 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1693 void tcp_fastopen_active_disable(struct sock *sk);
1694 bool tcp_fastopen_active_should_disable(struct sock *sk);
1695 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1696 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1698 /* Caller needs to wrap with rcu_read_(un)lock() */
1700 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1702 struct tcp_fastopen_context *ctx;
1704 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1706 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1711 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1712 const struct tcp_fastopen_cookie *orig)
1714 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1715 orig->len == foc->len &&
1716 !memcmp(orig->val, foc->val, foc->len))
1722 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1727 /* Latencies incurred by various limits for a sender. They are
1728 * chronograph-like stats that are mutually exclusive.
1732 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1733 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1734 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1738 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1739 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1741 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1742 * the same memory storage than skb->destructor/_skb_refdst
1744 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1746 skb->destructor = NULL;
1747 skb->_skb_refdst = 0UL;
1750 #define tcp_skb_tsorted_save(skb) { \
1751 unsigned long _save = skb->_skb_refdst; \
1752 skb->_skb_refdst = 0UL;
1754 #define tcp_skb_tsorted_restore(skb) \
1755 skb->_skb_refdst = _save; \
1758 void tcp_write_queue_purge(struct sock *sk);
1760 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1762 return skb_rb_first(&sk->tcp_rtx_queue);
1765 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1767 return skb_rb_last(&sk->tcp_rtx_queue);
1770 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1772 return skb_peek(&sk->sk_write_queue);
1775 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1777 return skb_peek_tail(&sk->sk_write_queue);
1780 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1781 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1783 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1785 return skb_peek(&sk->sk_write_queue);
1788 static inline bool tcp_skb_is_last(const struct sock *sk,
1789 const struct sk_buff *skb)
1791 return skb_queue_is_last(&sk->sk_write_queue, skb);
1795 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
1798 * Since the write queue can have a temporary empty skb in it,
1799 * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
1801 static inline bool tcp_write_queue_empty(const struct sock *sk)
1803 const struct tcp_sock *tp = tcp_sk(sk);
1805 return tp->write_seq == tp->snd_nxt;
1808 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1810 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1813 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1815 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1818 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1820 __skb_queue_tail(&sk->sk_write_queue, skb);
1822 /* Queue it, remembering where we must start sending. */
1823 if (sk->sk_write_queue.next == skb)
1824 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1827 /* Insert new before skb on the write queue of sk. */
1828 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1829 struct sk_buff *skb,
1832 __skb_queue_before(&sk->sk_write_queue, skb, new);
1835 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1837 tcp_skb_tsorted_anchor_cleanup(skb);
1838 __skb_unlink(skb, &sk->sk_write_queue);
1841 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1843 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1845 tcp_skb_tsorted_anchor_cleanup(skb);
1846 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1849 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1851 list_del(&skb->tcp_tsorted_anchor);
1852 tcp_rtx_queue_unlink(skb, sk);
1853 sk_wmem_free_skb(sk, skb);
1856 static inline void tcp_push_pending_frames(struct sock *sk)
1858 if (tcp_send_head(sk)) {
1859 struct tcp_sock *tp = tcp_sk(sk);
1861 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1865 /* Start sequence of the skb just after the highest skb with SACKed
1866 * bit, valid only if sacked_out > 0 or when the caller has ensured
1867 * validity by itself.
1869 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1871 if (!tp->sacked_out)
1874 if (tp->highest_sack == NULL)
1877 return TCP_SKB_CB(tp->highest_sack)->seq;
1880 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1882 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1885 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1887 return tcp_sk(sk)->highest_sack;
1890 static inline void tcp_highest_sack_reset(struct sock *sk)
1892 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1895 /* Called when old skb is about to be deleted and replaced by new skb */
1896 static inline void tcp_highest_sack_replace(struct sock *sk,
1897 struct sk_buff *old,
1898 struct sk_buff *new)
1900 if (old == tcp_highest_sack(sk))
1901 tcp_sk(sk)->highest_sack = new;
1904 /* This helper checks if socket has IP_TRANSPARENT set */
1905 static inline bool inet_sk_transparent(const struct sock *sk)
1907 switch (sk->sk_state) {
1909 return inet_twsk(sk)->tw_transparent;
1910 case TCP_NEW_SYN_RECV:
1911 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1913 return inet_sk(sk)->transparent;
1916 /* Determines whether this is a thin stream (which may suffer from
1917 * increased latency). Used to trigger latency-reducing mechanisms.
1919 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1921 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1925 enum tcp_seq_states {
1926 TCP_SEQ_STATE_LISTENING,
1927 TCP_SEQ_STATE_ESTABLISHED,
1930 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1931 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1932 void tcp_seq_stop(struct seq_file *seq, void *v);
1934 struct tcp_seq_afinfo {
1938 struct tcp_iter_state {
1939 struct seq_net_private p;
1940 enum tcp_seq_states state;
1941 struct sock *syn_wait_sk;
1942 struct tcp_seq_afinfo *bpf_seq_afinfo;
1943 int bucket, offset, sbucket, num;
1947 extern struct request_sock_ops tcp_request_sock_ops;
1948 extern struct request_sock_ops tcp6_request_sock_ops;
1950 void tcp_v4_destroy_sock(struct sock *sk);
1952 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1953 netdev_features_t features);
1954 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
1955 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
1956 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
1957 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff));
1958 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb));
1959 int tcp_gro_complete(struct sk_buff *skb);
1961 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1963 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1965 struct net *net = sock_net((struct sock *)tp);
1966 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1969 bool tcp_stream_memory_free(const struct sock *sk, int wake);
1971 #ifdef CONFIG_PROC_FS
1972 int tcp4_proc_init(void);
1973 void tcp4_proc_exit(void);
1976 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1977 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1978 const struct tcp_request_sock_ops *af_ops,
1979 struct sock *sk, struct sk_buff *skb);
1981 /* TCP af-specific functions */
1982 struct tcp_sock_af_ops {
1983 #ifdef CONFIG_TCP_MD5SIG
1984 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1985 const struct sock *addr_sk);
1986 int (*calc_md5_hash)(char *location,
1987 const struct tcp_md5sig_key *md5,
1988 const struct sock *sk,
1989 const struct sk_buff *skb);
1990 int (*md5_parse)(struct sock *sk,
1997 struct tcp_request_sock_ops {
1999 #ifdef CONFIG_TCP_MD5SIG
2000 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
2001 const struct sock *addr_sk);
2002 int (*calc_md5_hash) (char *location,
2003 const struct tcp_md5sig_key *md5,
2004 const struct sock *sk,
2005 const struct sk_buff *skb);
2007 #ifdef CONFIG_SYN_COOKIES
2008 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
2011 struct dst_entry *(*route_req)(const struct sock *sk,
2012 struct sk_buff *skb,
2014 struct request_sock *req);
2015 u32 (*init_seq)(const struct sk_buff *skb);
2016 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
2017 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2018 struct flowi *fl, struct request_sock *req,
2019 struct tcp_fastopen_cookie *foc,
2020 enum tcp_synack_type synack_type,
2021 struct sk_buff *syn_skb);
2024 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2025 #if IS_ENABLED(CONFIG_IPV6)
2026 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2029 #ifdef CONFIG_SYN_COOKIES
2030 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2031 const struct sock *sk, struct sk_buff *skb,
2034 tcp_synq_overflow(sk);
2035 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2036 return ops->cookie_init_seq(skb, mss);
2039 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2040 const struct sock *sk, struct sk_buff *skb,
2047 int tcpv4_offload_init(void);
2049 void tcp_v4_init(void);
2050 void tcp_init(void);
2052 /* tcp_recovery.c */
2053 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2054 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2055 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2057 extern bool tcp_rack_mark_lost(struct sock *sk);
2058 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2060 extern void tcp_rack_reo_timeout(struct sock *sk);
2061 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2063 /* At how many usecs into the future should the RTO fire? */
2064 static inline s64 tcp_rto_delta_us(const struct sock *sk)
2066 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2067 u32 rto = inet_csk(sk)->icsk_rto;
2068 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2070 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2074 * Save and compile IPv4 options, return a pointer to it
2076 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2077 struct sk_buff *skb)
2079 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2080 struct ip_options_rcu *dopt = NULL;
2083 int opt_size = sizeof(*dopt) + opt->optlen;
2085 dopt = kmalloc(opt_size, GFP_ATOMIC);
2086 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2094 /* locally generated TCP pure ACKs have skb->truesize == 2
2095 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2096 * This is much faster than dissecting the packet to find out.
2097 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2099 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2101 return skb->truesize == 2;
2104 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2109 static inline int tcp_inq(struct sock *sk)
2111 struct tcp_sock *tp = tcp_sk(sk);
2114 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2116 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2118 before(tp->urg_seq, tp->copied_seq) ||
2119 !before(tp->urg_seq, tp->rcv_nxt)) {
2121 answ = tp->rcv_nxt - tp->copied_seq;
2123 /* Subtract 1, if FIN was received */
2124 if (answ && sock_flag(sk, SOCK_DONE))
2127 answ = tp->urg_seq - tp->copied_seq;
2133 int tcp_peek_len(struct socket *sock);
2135 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2139 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2140 tp->segs_in += segs_in;
2141 if (skb->len > tcp_hdrlen(skb))
2142 tp->data_segs_in += segs_in;
2146 * TCP listen path runs lockless.
2147 * We forced "struct sock" to be const qualified to make sure
2148 * we don't modify one of its field by mistake.
2149 * Here, we increment sk_drops which is an atomic_t, so we can safely
2150 * make sock writable again.
2152 static inline void tcp_listendrop(const struct sock *sk)
2154 atomic_inc(&((struct sock *)sk)->sk_drops);
2155 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2158 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2161 * Interface for adding Upper Level Protocols over TCP
2164 #define TCP_ULP_NAME_MAX 16
2165 #define TCP_ULP_MAX 128
2166 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2168 struct tcp_ulp_ops {
2169 struct list_head list;
2171 /* initialize ulp */
2172 int (*init)(struct sock *sk);
2174 void (*update)(struct sock *sk, struct proto *p,
2175 void (*write_space)(struct sock *sk));
2177 void (*release)(struct sock *sk);
2179 int (*get_info)(const struct sock *sk, struct sk_buff *skb);
2180 size_t (*get_info_size)(const struct sock *sk);
2182 void (*clone)(const struct request_sock *req, struct sock *newsk,
2183 const gfp_t priority);
2185 char name[TCP_ULP_NAME_MAX];
2186 struct module *owner;
2188 int tcp_register_ulp(struct tcp_ulp_ops *type);
2189 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2190 int tcp_set_ulp(struct sock *sk, const char *name);
2191 void tcp_get_available_ulp(char *buf, size_t len);
2192 void tcp_cleanup_ulp(struct sock *sk);
2193 void tcp_update_ulp(struct sock *sk, struct proto *p,
2194 void (*write_space)(struct sock *sk));
2196 #define MODULE_ALIAS_TCP_ULP(name) \
2197 __MODULE_INFO(alias, alias_userspace, name); \
2198 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2200 #ifdef CONFIG_NET_SOCK_MSG
2204 #ifdef CONFIG_BPF_SYSCALL
2205 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
2206 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2207 #endif /* CONFIG_BPF_SYSCALL */
2209 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes,
2211 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock,
2212 struct msghdr *msg, int len, int flags);
2213 #endif /* CONFIG_NET_SOCK_MSG */
2215 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG)
2216 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2221 #ifdef CONFIG_CGROUP_BPF
2222 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2223 struct sk_buff *skb,
2224 unsigned int end_offset)
2227 skops->skb_data_end = skb->data + end_offset;
2230 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2231 struct sk_buff *skb,
2232 unsigned int end_offset)
2237 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2238 * is < 0, then the BPF op failed (for example if the loaded BPF
2239 * program does not support the chosen operation or there is no BPF
2243 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2245 struct bpf_sock_ops_kern sock_ops;
2248 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2249 if (sk_fullsock(sk)) {
2250 sock_ops.is_fullsock = 1;
2251 sock_owned_by_me(sk);
2257 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2259 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2261 ret = sock_ops.reply;
2267 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2269 u32 args[2] = {arg1, arg2};
2271 return tcp_call_bpf(sk, op, 2, args);
2274 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2277 u32 args[3] = {arg1, arg2, arg3};
2279 return tcp_call_bpf(sk, op, 3, args);
2283 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2288 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2293 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2301 static inline u32 tcp_timeout_init(struct sock *sk)
2305 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2308 timeout = TCP_TIMEOUT_INIT;
2312 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2316 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2323 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2325 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2328 static inline void tcp_bpf_rtt(struct sock *sk)
2330 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2331 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2334 #if IS_ENABLED(CONFIG_SMC)
2335 extern struct static_key_false tcp_have_smc;
2338 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2339 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2340 void (*cad)(struct sock *sk, u32 ack_seq));
2341 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2342 void clean_acked_data_flush(void);
2345 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2346 static inline void tcp_add_tx_delay(struct sk_buff *skb,
2347 const struct tcp_sock *tp)
2349 if (static_branch_unlikely(&tcp_tx_delay_enabled))
2350 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2353 /* Compute Earliest Departure Time for some control packets
2354 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2356 static inline u64 tcp_transmit_time(const struct sock *sk)
2358 if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2359 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2360 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2362 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;