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/cryptohash.h>
27 #include <linux/kref.h>
28 #include <linux/ktime.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>
43 #include <linux/seq_file.h>
44 #include <linux/memcontrol.h>
45 #include <linux/bpf-cgroup.h>
46 #include <linux/siphash.h>
48 extern struct inet_hashinfo tcp_hashinfo;
50 extern struct percpu_counter tcp_orphan_count;
51 void tcp_time_wait(struct sock *sk, int state, int timeo);
53 #define MAX_TCP_HEADER (128 + MAX_HEADER)
54 #define MAX_TCP_OPTION_SPACE 40
55 #define TCP_MIN_SND_MSS 48
56 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
59 * Never offer a window over 32767 without using window scaling. Some
60 * poor stacks do signed 16bit maths!
62 #define MAX_TCP_WINDOW 32767U
64 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
65 #define TCP_MIN_MSS 88U
67 /* The initial MTU to use for probing */
68 #define TCP_BASE_MSS 1024
70 /* probing interval, default to 10 minutes as per RFC4821 */
71 #define TCP_PROBE_INTERVAL 600
73 /* Specify interval when tcp mtu probing will stop */
74 #define TCP_PROBE_THRESHOLD 8
76 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
77 #define TCP_FASTRETRANS_THRESH 3
79 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
80 #define TCP_MAX_QUICKACKS 16U
82 /* Maximal number of window scale according to RFC1323 */
83 #define TCP_MAX_WSCALE 14U
86 #define TCP_URG_VALID 0x0100
87 #define TCP_URG_NOTYET 0x0200
88 #define TCP_URG_READ 0x0400
90 #define TCP_RETR1 3 /*
91 * This is how many retries it does before it
92 * tries to figure out if the gateway is
93 * down. Minimal RFC value is 3; it corresponds
94 * to ~3sec-8min depending on RTO.
97 #define TCP_RETR2 15 /*
98 * This should take at least
99 * 90 minutes to time out.
100 * RFC1122 says that the limit is 100 sec.
101 * 15 is ~13-30min depending on RTO.
104 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
105 * when active opening a connection.
106 * RFC1122 says the minimum retry MUST
107 * be at least 180secs. Nevertheless
108 * this value is corresponding to
109 * 63secs of retransmission with the
110 * current initial RTO.
113 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
114 * when passive opening a connection.
115 * This is corresponding to 31secs of
116 * retransmission with the current
120 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
121 * state, about 60 seconds */
122 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
123 /* BSD style FIN_WAIT2 deadlock breaker.
124 * It used to be 3min, new value is 60sec,
125 * to combine FIN-WAIT-2 timeout with
129 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
131 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
132 #define TCP_ATO_MIN ((unsigned)(HZ/25))
134 #define TCP_DELACK_MIN 4U
135 #define TCP_ATO_MIN 4U
137 #define TCP_RTO_MAX ((unsigned)(120*HZ))
138 #define TCP_RTO_MIN ((unsigned)(HZ/5))
139 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
140 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
141 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
142 * used as a fallback RTO for the
143 * initial data transmission if no
144 * valid RTT sample has been acquired,
145 * most likely due to retrans in 3WHS.
148 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
149 * for local resources.
151 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
152 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
153 #define TCP_KEEPALIVE_INTVL (75*HZ)
155 #define MAX_TCP_KEEPIDLE 32767
156 #define MAX_TCP_KEEPINTVL 32767
157 #define MAX_TCP_KEEPCNT 127
158 #define MAX_TCP_SYNCNT 127
160 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
162 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
163 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
164 * after this time. It should be equal
165 * (or greater than) TCP_TIMEWAIT_LEN
166 * to provide reliability equal to one
167 * provided by timewait state.
169 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
170 * timestamps. It must be less than
171 * minimal timewait lifetime.
177 #define TCPOPT_NOP 1 /* Padding */
178 #define TCPOPT_EOL 0 /* End of options */
179 #define TCPOPT_MSS 2 /* Segment size negotiating */
180 #define TCPOPT_WINDOW 3 /* Window scaling */
181 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
182 #define TCPOPT_SACK 5 /* SACK Block */
183 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
184 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
185 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
186 #define TCPOPT_EXP 254 /* Experimental */
187 /* Magic number to be after the option value for sharing TCP
188 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
190 #define TCPOPT_FASTOPEN_MAGIC 0xF989
191 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9
197 #define TCPOLEN_MSS 4
198 #define TCPOLEN_WINDOW 3
199 #define TCPOLEN_SACK_PERM 2
200 #define TCPOLEN_TIMESTAMP 10
201 #define TCPOLEN_MD5SIG 18
202 #define TCPOLEN_FASTOPEN_BASE 2
203 #define TCPOLEN_EXP_FASTOPEN_BASE 4
204 #define TCPOLEN_EXP_SMC_BASE 6
206 /* But this is what stacks really send out. */
207 #define TCPOLEN_TSTAMP_ALIGNED 12
208 #define TCPOLEN_WSCALE_ALIGNED 4
209 #define TCPOLEN_SACKPERM_ALIGNED 4
210 #define TCPOLEN_SACK_BASE 2
211 #define TCPOLEN_SACK_BASE_ALIGNED 4
212 #define TCPOLEN_SACK_PERBLOCK 8
213 #define TCPOLEN_MD5SIG_ALIGNED 20
214 #define TCPOLEN_MSS_ALIGNED 4
215 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
217 /* Flags in tp->nonagle */
218 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
219 #define TCP_NAGLE_CORK 2 /* Socket is corked */
220 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
222 /* TCP thin-stream limits */
223 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
225 /* TCP initial congestion window as per rfc6928 */
226 #define TCP_INIT_CWND 10
228 /* Bit Flags for sysctl_tcp_fastopen */
229 #define TFO_CLIENT_ENABLE 1
230 #define TFO_SERVER_ENABLE 2
231 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
233 /* Accept SYN data w/o any cookie option */
234 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
236 /* Force enable TFO on all listeners, i.e., not requiring the
237 * TCP_FASTOPEN socket option.
239 #define TFO_SERVER_WO_SOCKOPT1 0x400
242 /* sysctl variables for tcp */
243 extern int sysctl_tcp_max_orphans;
244 extern long sysctl_tcp_mem[3];
246 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
247 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
248 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */
250 extern atomic_long_t tcp_memory_allocated;
251 extern struct percpu_counter tcp_sockets_allocated;
252 extern unsigned long tcp_memory_pressure;
254 /* optimized version of sk_under_memory_pressure() for TCP sockets */
255 static inline bool tcp_under_memory_pressure(const struct sock *sk)
257 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
258 mem_cgroup_under_socket_pressure(sk->sk_memcg))
261 return READ_ONCE(tcp_memory_pressure);
264 * The next routines deal with comparing 32 bit unsigned ints
265 * and worry about wraparound (automatic with unsigned arithmetic).
268 static inline bool before(__u32 seq1, __u32 seq2)
270 return (__s32)(seq1-seq2) < 0;
272 #define after(seq2, seq1) before(seq1, seq2)
274 /* is s2<=s1<=s3 ? */
275 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
277 return seq3 - seq2 >= seq1 - seq2;
280 static inline bool tcp_out_of_memory(struct sock *sk)
282 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
283 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
288 void sk_forced_mem_schedule(struct sock *sk, int size);
290 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
292 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
293 int orphans = percpu_counter_read_positive(ocp);
295 if (orphans << shift > sysctl_tcp_max_orphans) {
296 orphans = percpu_counter_sum_positive(ocp);
297 if (orphans << shift > sysctl_tcp_max_orphans)
303 bool tcp_check_oom(struct sock *sk, int shift);
306 extern struct proto tcp_prot;
308 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
309 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
310 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
311 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
313 void tcp_tasklet_init(void);
315 int tcp_v4_err(struct sk_buff *skb, u32);
317 void tcp_shutdown(struct sock *sk, int how);
319 int tcp_v4_early_demux(struct sk_buff *skb);
320 int tcp_v4_rcv(struct sk_buff *skb);
322 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
323 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
324 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
325 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
327 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
328 size_t size, int flags);
329 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
330 size_t size, int flags);
331 void tcp_release_cb(struct sock *sk);
332 void tcp_wfree(struct sk_buff *skb);
333 void tcp_write_timer_handler(struct sock *sk);
334 void tcp_delack_timer_handler(struct sock *sk);
335 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
336 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
337 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
338 void tcp_rcv_space_adjust(struct sock *sk);
339 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
340 void tcp_twsk_destructor(struct sock *sk);
341 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
342 struct pipe_inode_info *pipe, size_t len,
345 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
346 static inline void tcp_dec_quickack_mode(struct sock *sk,
347 const unsigned int pkts)
349 struct inet_connection_sock *icsk = inet_csk(sk);
351 if (icsk->icsk_ack.quick) {
352 if (pkts >= icsk->icsk_ack.quick) {
353 icsk->icsk_ack.quick = 0;
354 /* Leaving quickack mode we deflate ATO. */
355 icsk->icsk_ack.ato = TCP_ATO_MIN;
357 icsk->icsk_ack.quick -= pkts;
362 #define TCP_ECN_QUEUE_CWR 2
363 #define TCP_ECN_DEMAND_CWR 4
364 #define TCP_ECN_SEEN 8
374 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
376 const struct tcphdr *th);
377 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
378 struct request_sock *req, bool fastopen,
380 int tcp_child_process(struct sock *parent, struct sock *child,
381 struct sk_buff *skb);
382 void tcp_enter_loss(struct sock *sk);
383 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
384 void tcp_clear_retrans(struct tcp_sock *tp);
385 void tcp_update_metrics(struct sock *sk);
386 void tcp_init_metrics(struct sock *sk);
387 void tcp_metrics_init(void);
388 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
389 void tcp_close(struct sock *sk, long timeout);
390 void tcp_init_sock(struct sock *sk);
391 void tcp_init_transfer(struct sock *sk, int bpf_op);
392 __poll_t tcp_poll(struct file *file, struct socket *sock,
393 struct poll_table_struct *wait);
394 int tcp_getsockopt(struct sock *sk, int level, int optname,
395 char __user *optval, int __user *optlen);
396 int tcp_setsockopt(struct sock *sk, int level, int optname,
397 char __user *optval, unsigned int optlen);
398 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
399 char __user *optval, int __user *optlen);
400 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
401 char __user *optval, unsigned int optlen);
402 void tcp_set_keepalive(struct sock *sk, int val);
403 void tcp_syn_ack_timeout(const struct request_sock *req);
404 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
405 int flags, int *addr_len);
406 int tcp_set_rcvlowat(struct sock *sk, int val);
407 void tcp_data_ready(struct sock *sk);
409 int tcp_mmap(struct file *file, struct socket *sock,
410 struct vm_area_struct *vma);
412 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
413 struct tcp_options_received *opt_rx,
414 int estab, struct tcp_fastopen_cookie *foc);
415 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
418 * BPF SKB-less helpers
420 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
421 struct tcphdr *th, u32 *cookie);
422 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
423 struct tcphdr *th, u32 *cookie);
424 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
425 const struct tcp_request_sock_ops *af_ops,
426 struct sock *sk, struct tcphdr *th);
428 * TCP v4 functions exported for the inet6 API
431 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
432 void tcp_v4_mtu_reduced(struct sock *sk);
433 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
434 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
435 struct sock *tcp_create_openreq_child(const struct sock *sk,
436 struct request_sock *req,
437 struct sk_buff *skb);
438 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
439 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
440 struct request_sock *req,
441 struct dst_entry *dst,
442 struct request_sock *req_unhash,
444 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
445 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
446 int tcp_connect(struct sock *sk);
447 enum tcp_synack_type {
452 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
453 struct request_sock *req,
454 struct tcp_fastopen_cookie *foc,
455 enum tcp_synack_type synack_type);
456 int tcp_disconnect(struct sock *sk, int flags);
458 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
459 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
460 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
462 /* From syncookies.c */
463 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
464 struct request_sock *req,
465 struct dst_entry *dst, u32 tsoff);
466 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
468 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
469 #ifdef CONFIG_SYN_COOKIES
471 /* Syncookies use a monotonic timer which increments every 60 seconds.
472 * This counter is used both as a hash input and partially encoded into
473 * the cookie value. A cookie is only validated further if the delta
474 * between the current counter value and the encoded one is less than this,
475 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
476 * the counter advances immediately after a cookie is generated).
478 #define MAX_SYNCOOKIE_AGE 2
479 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
480 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
482 /* syncookies: remember time of last synqueue overflow
483 * But do not dirty this field too often (once per second is enough)
484 * It is racy as we do not hold a lock, but race is very minor.
486 static inline void tcp_synq_overflow(const struct sock *sk)
488 unsigned int last_overflow;
489 unsigned int now = jiffies;
491 if (sk->sk_reuseport) {
492 struct sock_reuseport *reuse;
494 reuse = rcu_dereference(sk->sk_reuseport_cb);
496 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
497 if (!time_between32(now, last_overflow,
499 WRITE_ONCE(reuse->synq_overflow_ts, now);
504 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
505 if (!time_between32(now, last_overflow, last_overflow + HZ))
506 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
509 /* syncookies: no recent synqueue overflow on this listening socket? */
510 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
512 unsigned int last_overflow;
513 unsigned int now = jiffies;
515 if (sk->sk_reuseport) {
516 struct sock_reuseport *reuse;
518 reuse = rcu_dereference(sk->sk_reuseport_cb);
520 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
521 return !time_between32(now, last_overflow - HZ,
523 TCP_SYNCOOKIE_VALID);
527 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
529 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
530 * then we're under synflood. However, we have to use
531 * 'last_overflow - HZ' as lower bound. That's because a concurrent
532 * tcp_synq_overflow() could update .ts_recent_stamp after we read
533 * jiffies but before we store .ts_recent_stamp into last_overflow,
534 * which could lead to rejecting a valid syncookie.
536 return !time_between32(now, last_overflow - HZ,
537 last_overflow + TCP_SYNCOOKIE_VALID);
540 static inline u32 tcp_cookie_time(void)
542 u64 val = get_jiffies_64();
544 do_div(val, TCP_SYNCOOKIE_PERIOD);
548 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
550 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
551 u64 cookie_init_timestamp(struct request_sock *req, u64 now);
552 bool cookie_timestamp_decode(const struct net *net,
553 struct tcp_options_received *opt);
554 bool cookie_ecn_ok(const struct tcp_options_received *opt,
555 const struct net *net, const struct dst_entry *dst);
557 /* From net/ipv6/syncookies.c */
558 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
560 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
562 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
563 const struct tcphdr *th, u16 *mssp);
564 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
568 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
570 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
571 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
572 void tcp_retransmit_timer(struct sock *sk);
573 void tcp_xmit_retransmit_queue(struct sock *);
574 void tcp_simple_retransmit(struct sock *);
575 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
576 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
578 TCP_FRAG_IN_WRITE_QUEUE,
579 TCP_FRAG_IN_RTX_QUEUE,
581 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
582 struct sk_buff *skb, u32 len,
583 unsigned int mss_now, gfp_t gfp);
585 void tcp_send_probe0(struct sock *);
586 void tcp_send_partial(struct sock *);
587 int tcp_write_wakeup(struct sock *, int mib);
588 void tcp_send_fin(struct sock *sk);
589 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
590 int tcp_send_synack(struct sock *);
591 void tcp_push_one(struct sock *, unsigned int mss_now);
592 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
593 void tcp_send_ack(struct sock *sk);
594 void tcp_send_delayed_ack(struct sock *sk);
595 void tcp_send_loss_probe(struct sock *sk);
596 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
597 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
598 const struct sk_buff *next_skb);
601 void tcp_rearm_rto(struct sock *sk);
602 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
603 void tcp_reset(struct sock *sk);
604 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
605 void tcp_fin(struct sock *sk);
608 void tcp_init_xmit_timers(struct sock *);
609 static inline void tcp_clear_xmit_timers(struct sock *sk)
611 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
614 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
617 inet_csk_clear_xmit_timers(sk);
620 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
621 unsigned int tcp_current_mss(struct sock *sk);
623 /* Bound MSS / TSO packet size with the half of the window */
624 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
628 /* When peer uses tiny windows, there is no use in packetizing
629 * to sub-MSS pieces for the sake of SWS or making sure there
630 * are enough packets in the pipe for fast recovery.
632 * On the other hand, for extremely large MSS devices, handling
633 * smaller than MSS windows in this way does make sense.
635 if (tp->max_window > TCP_MSS_DEFAULT)
636 cutoff = (tp->max_window >> 1);
638 cutoff = tp->max_window;
640 if (cutoff && pktsize > cutoff)
641 return max_t(int, cutoff, 68U - tp->tcp_header_len);
647 void tcp_get_info(struct sock *, struct tcp_info *);
649 /* Read 'sendfile()'-style from a TCP socket */
650 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
651 sk_read_actor_t recv_actor);
653 void tcp_initialize_rcv_mss(struct sock *sk);
655 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
656 int tcp_mss_to_mtu(struct sock *sk, int mss);
657 void tcp_mtup_init(struct sock *sk);
658 void tcp_init_buffer_space(struct sock *sk);
660 static inline void tcp_bound_rto(const struct sock *sk)
662 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
663 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
666 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
668 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
671 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
673 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
674 ntohl(TCP_FLAG_ACK) |
678 static inline void tcp_fast_path_on(struct tcp_sock *tp)
680 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
683 static inline void tcp_fast_path_check(struct sock *sk)
685 struct tcp_sock *tp = tcp_sk(sk);
687 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
689 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
691 tcp_fast_path_on(tp);
694 /* Compute the actual rto_min value */
695 static inline u32 tcp_rto_min(struct sock *sk)
697 const struct dst_entry *dst = __sk_dst_get(sk);
698 u32 rto_min = TCP_RTO_MIN;
700 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
701 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
705 static inline u32 tcp_rto_min_us(struct sock *sk)
707 return jiffies_to_usecs(tcp_rto_min(sk));
710 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
712 return dst_metric_locked(dst, RTAX_CC_ALGO);
715 /* Minimum RTT in usec. ~0 means not available. */
716 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
718 return minmax_get(&tp->rtt_min);
721 /* Compute the actual receive window we are currently advertising.
722 * Rcv_nxt can be after the window if our peer push more data
723 * than the offered window.
725 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
727 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
734 /* Choose a new window, without checks for shrinking, and without
735 * scaling applied to the result. The caller does these things
736 * if necessary. This is a "raw" window selection.
738 u32 __tcp_select_window(struct sock *sk);
740 void tcp_send_window_probe(struct sock *sk);
742 /* TCP uses 32bit jiffies to save some space.
743 * Note that this is different from tcp_time_stamp, which
744 * historically has been the same until linux-4.13.
746 #define tcp_jiffies32 ((u32)jiffies)
749 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
750 * It is no longer tied to jiffies, but to 1 ms clock.
751 * Note: double check if you want to use tcp_jiffies32 instead of this.
753 #define TCP_TS_HZ 1000
755 static inline u64 tcp_clock_ns(void)
757 return ktime_get_ns();
760 static inline u64 tcp_clock_us(void)
762 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
765 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
766 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
768 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
771 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
772 static inline u32 tcp_ns_to_ts(u64 ns)
774 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
777 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
778 static inline u32 tcp_time_stamp_raw(void)
780 return tcp_ns_to_ts(tcp_clock_ns());
783 void tcp_mstamp_refresh(struct tcp_sock *tp);
785 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
787 return max_t(s64, t1 - t0, 0);
790 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
792 return tcp_ns_to_ts(skb->skb_mstamp_ns);
795 /* provide the departure time in us unit */
796 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
798 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
802 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
804 #define TCPHDR_FIN 0x01
805 #define TCPHDR_SYN 0x02
806 #define TCPHDR_RST 0x04
807 #define TCPHDR_PSH 0x08
808 #define TCPHDR_ACK 0x10
809 #define TCPHDR_URG 0x20
810 #define TCPHDR_ECE 0x40
811 #define TCPHDR_CWR 0x80
813 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
815 /* This is what the send packet queuing engine uses to pass
816 * TCP per-packet control information to the transmission code.
817 * We also store the host-order sequence numbers in here too.
818 * This is 44 bytes if IPV6 is enabled.
819 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
822 __u32 seq; /* Starting sequence number */
823 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
825 /* Note : tcp_tw_isn is used in input path only
826 * (isn chosen by tcp_timewait_state_process())
828 * tcp_gso_segs/size are used in write queue only,
829 * cf tcp_skb_pcount()/tcp_skb_mss()
837 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
839 __u8 sacked; /* State flags for SACK. */
840 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
841 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
842 #define TCPCB_LOST 0x04 /* SKB is lost */
843 #define TCPCB_TAGBITS 0x07 /* All tag bits */
844 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
845 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
846 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
849 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
850 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
851 eor:1, /* Is skb MSG_EOR marked? */
852 has_rxtstamp:1, /* SKB has a RX timestamp */
854 __u32 ack_seq; /* Sequence number ACK'd */
857 /* There is space for up to 24 bytes */
858 __u32 in_flight:30,/* Bytes in flight at transmit */
859 is_app_limited:1, /* cwnd not fully used? */
861 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
863 /* start of send pipeline phase */
865 /* when we reached the "delivered" count */
866 u64 delivered_mstamp;
867 } tx; /* only used for outgoing skbs */
869 struct inet_skb_parm h4;
870 #if IS_ENABLED(CONFIG_IPV6)
871 struct inet6_skb_parm h6;
873 } header; /* For incoming skbs */
876 struct sock *sk_redir;
882 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
884 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb)
886 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb);
889 static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb)
891 return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS;
894 static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb)
896 return TCP_SKB_CB(skb)->bpf.sk_redir;
899 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb)
901 TCP_SKB_CB(skb)->bpf.sk_redir = NULL;
904 #if IS_ENABLED(CONFIG_IPV6)
905 /* This is the variant of inet6_iif() that must be used by TCP,
906 * as TCP moves IP6CB into a different location in skb->cb[]
908 static inline int tcp_v6_iif(const struct sk_buff *skb)
910 return TCP_SKB_CB(skb)->header.h6.iif;
913 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
915 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
917 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
920 /* TCP_SKB_CB reference means this can not be used from early demux */
921 static inline int tcp_v6_sdif(const struct sk_buff *skb)
923 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
924 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
925 return TCP_SKB_CB(skb)->header.h6.iif;
931 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
933 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
934 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
935 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
941 /* TCP_SKB_CB reference means this can not be used from early demux */
942 static inline int tcp_v4_sdif(struct sk_buff *skb)
944 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
945 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
946 return TCP_SKB_CB(skb)->header.h4.iif;
951 /* Due to TSO, an SKB can be composed of multiple actual
952 * packets. To keep these tracked properly, we use this.
954 static inline int tcp_skb_pcount(const struct sk_buff *skb)
956 return TCP_SKB_CB(skb)->tcp_gso_segs;
959 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
961 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
964 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
966 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
969 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
970 static inline int tcp_skb_mss(const struct sk_buff *skb)
972 return TCP_SKB_CB(skb)->tcp_gso_size;
975 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
977 return likely(!TCP_SKB_CB(skb)->eor);
980 /* Events passed to congestion control interface */
982 CA_EVENT_TX_START, /* first transmit when no packets in flight */
983 CA_EVENT_CWND_RESTART, /* congestion window restart */
984 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
985 CA_EVENT_LOSS, /* loss timeout */
986 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
987 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
990 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
991 enum tcp_ca_ack_event_flags {
992 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
993 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
994 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
998 * Interface for adding new TCP congestion control handlers
1000 #define TCP_CA_NAME_MAX 16
1001 #define TCP_CA_MAX 128
1002 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
1004 #define TCP_CA_UNSPEC 0
1006 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1007 #define TCP_CONG_NON_RESTRICTED 0x1
1008 /* Requires ECN/ECT set on all packets */
1009 #define TCP_CONG_NEEDS_ECN 0x2
1019 /* A rate sample measures the number of (original/retransmitted) data
1020 * packets delivered "delivered" over an interval of time "interval_us".
1021 * The tcp_rate.c code fills in the rate sample, and congestion
1022 * control modules that define a cong_control function to run at the end
1023 * of ACK processing can optionally chose to consult this sample when
1024 * setting cwnd and pacing rate.
1025 * A sample is invalid if "delivered" or "interval_us" is negative.
1027 struct rate_sample {
1028 u64 prior_mstamp; /* starting timestamp for interval */
1029 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1030 s32 delivered; /* number of packets delivered over interval */
1031 long interval_us; /* time for tp->delivered to incr "delivered" */
1032 u32 snd_interval_us; /* snd interval for delivered packets */
1033 u32 rcv_interval_us; /* rcv interval for delivered packets */
1034 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1035 int losses; /* number of packets marked lost upon ACK */
1036 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1037 u32 prior_in_flight; /* in flight before this ACK */
1038 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1039 bool is_retrans; /* is sample from retransmission? */
1040 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1043 struct tcp_congestion_ops {
1044 struct list_head list;
1048 /* initialize private data (optional) */
1049 void (*init)(struct sock *sk);
1050 /* cleanup private data (optional) */
1051 void (*release)(struct sock *sk);
1053 /* return slow start threshold (required) */
1054 u32 (*ssthresh)(struct sock *sk);
1055 /* do new cwnd calculation (required) */
1056 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1057 /* call before changing ca_state (optional) */
1058 void (*set_state)(struct sock *sk, u8 new_state);
1059 /* call when cwnd event occurs (optional) */
1060 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1061 /* call when ack arrives (optional) */
1062 void (*in_ack_event)(struct sock *sk, u32 flags);
1063 /* new value of cwnd after loss (required) */
1064 u32 (*undo_cwnd)(struct sock *sk);
1065 /* hook for packet ack accounting (optional) */
1066 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1067 /* override sysctl_tcp_min_tso_segs */
1068 u32 (*min_tso_segs)(struct sock *sk);
1069 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1070 u32 (*sndbuf_expand)(struct sock *sk);
1071 /* call when packets are delivered to update cwnd and pacing rate,
1072 * after all the ca_state processing. (optional)
1074 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1075 /* get info for inet_diag (optional) */
1076 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1077 union tcp_cc_info *info);
1079 char name[TCP_CA_NAME_MAX];
1080 struct module *owner;
1083 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1084 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1086 void tcp_assign_congestion_control(struct sock *sk);
1087 void tcp_init_congestion_control(struct sock *sk);
1088 void tcp_cleanup_congestion_control(struct sock *sk);
1089 int tcp_set_default_congestion_control(struct net *net, const char *name);
1090 void tcp_get_default_congestion_control(struct net *net, char *name);
1091 void tcp_get_available_congestion_control(char *buf, size_t len);
1092 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1093 int tcp_set_allowed_congestion_control(char *allowed);
1094 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1095 bool reinit, bool cap_net_admin);
1096 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1097 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1099 u32 tcp_reno_ssthresh(struct sock *sk);
1100 u32 tcp_reno_undo_cwnd(struct sock *sk);
1101 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1102 extern struct tcp_congestion_ops tcp_reno;
1104 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1105 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1107 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1109 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1115 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1117 const struct inet_connection_sock *icsk = inet_csk(sk);
1119 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1122 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1124 struct inet_connection_sock *icsk = inet_csk(sk);
1126 if (icsk->icsk_ca_ops->set_state)
1127 icsk->icsk_ca_ops->set_state(sk, ca_state);
1128 icsk->icsk_ca_state = ca_state;
1131 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1133 const struct inet_connection_sock *icsk = inet_csk(sk);
1135 if (icsk->icsk_ca_ops->cwnd_event)
1136 icsk->icsk_ca_ops->cwnd_event(sk, event);
1139 /* From tcp_rate.c */
1140 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1141 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1142 struct rate_sample *rs);
1143 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1144 bool is_sack_reneg, struct rate_sample *rs);
1145 void tcp_rate_check_app_limited(struct sock *sk);
1147 /* These functions determine how the current flow behaves in respect of SACK
1148 * handling. SACK is negotiated with the peer, and therefore it can vary
1149 * between different flows.
1151 * tcp_is_sack - SACK enabled
1152 * tcp_is_reno - No SACK
1154 static inline int tcp_is_sack(const struct tcp_sock *tp)
1156 return likely(tp->rx_opt.sack_ok);
1159 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1161 return !tcp_is_sack(tp);
1164 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1166 return tp->sacked_out + tp->lost_out;
1169 /* This determines how many packets are "in the network" to the best
1170 * of our knowledge. In many cases it is conservative, but where
1171 * detailed information is available from the receiver (via SACK
1172 * blocks etc.) we can make more aggressive calculations.
1174 * Use this for decisions involving congestion control, use just
1175 * tp->packets_out to determine if the send queue is empty or not.
1177 * Read this equation as:
1179 * "Packets sent once on transmission queue" MINUS
1180 * "Packets left network, but not honestly ACKed yet" PLUS
1181 * "Packets fast retransmitted"
1183 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1185 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1188 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1190 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1192 return tp->snd_cwnd < tp->snd_ssthresh;
1195 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1197 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1200 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1202 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1203 (1 << inet_csk(sk)->icsk_ca_state);
1206 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1207 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1210 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1212 const struct tcp_sock *tp = tcp_sk(sk);
1214 if (tcp_in_cwnd_reduction(sk))
1215 return tp->snd_ssthresh;
1217 return max(tp->snd_ssthresh,
1218 ((tp->snd_cwnd >> 1) +
1219 (tp->snd_cwnd >> 2)));
1222 /* Use define here intentionally to get WARN_ON location shown at the caller */
1223 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1225 void tcp_enter_cwr(struct sock *sk);
1226 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1228 /* The maximum number of MSS of available cwnd for which TSO defers
1229 * sending if not using sysctl_tcp_tso_win_divisor.
1231 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1236 /* Returns end sequence number of the receiver's advertised window */
1237 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1239 return tp->snd_una + tp->snd_wnd;
1242 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1243 * flexible approach. The RFC suggests cwnd should not be raised unless
1244 * it was fully used previously. And that's exactly what we do in
1245 * congestion avoidance mode. But in slow start we allow cwnd to grow
1246 * as long as the application has used half the cwnd.
1248 * cwnd is 10 (IW10), but application sends 9 frames.
1249 * We allow cwnd to reach 18 when all frames are ACKed.
1250 * This check is safe because it's as aggressive as slow start which already
1251 * risks 100% overshoot. The advantage is that we discourage application to
1252 * either send more filler packets or data to artificially blow up the cwnd
1253 * usage, and allow application-limited process to probe bw more aggressively.
1255 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1257 const struct tcp_sock *tp = tcp_sk(sk);
1259 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1260 if (tcp_in_slow_start(tp))
1261 return tp->snd_cwnd < 2 * tp->max_packets_out;
1263 return tp->is_cwnd_limited;
1266 /* BBR congestion control needs pacing.
1267 * Same remark for SO_MAX_PACING_RATE.
1268 * sch_fq packet scheduler is efficiently handling pacing,
1269 * but is not always installed/used.
1270 * Return true if TCP stack should pace packets itself.
1272 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1274 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1277 /* Return in jiffies the delay before one skb is sent.
1278 * If @skb is NULL, we look at EDT for next packet being sent on the socket.
1280 static inline unsigned long tcp_pacing_delay(const struct sock *sk,
1281 const struct sk_buff *skb)
1283 s64 pacing_delay = skb ? skb->tstamp : tcp_sk(sk)->tcp_wstamp_ns;
1285 pacing_delay -= tcp_sk(sk)->tcp_clock_cache;
1287 return pacing_delay > 0 ? nsecs_to_jiffies(pacing_delay) : 0;
1290 static inline void tcp_reset_xmit_timer(struct sock *sk,
1293 const unsigned long max_when,
1294 const struct sk_buff *skb)
1296 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk, skb),
1300 /* Something is really bad, we could not queue an additional packet,
1301 * because qdisc is full or receiver sent a 0 window, or we are paced.
1302 * We do not want to add fuel to the fire, or abort too early,
1303 * so make sure the timer we arm now is at least 200ms in the future,
1304 * regardless of current icsk_rto value (as it could be ~2ms)
1306 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1308 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1311 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1312 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1313 unsigned long max_when)
1315 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1317 return (unsigned long)min_t(u64, when, max_when);
1320 static inline void tcp_check_probe_timer(struct sock *sk)
1322 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1323 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1324 tcp_probe0_base(sk), TCP_RTO_MAX,
1328 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1333 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1339 * Calculate(/check) TCP checksum
1341 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1342 __be32 daddr, __wsum base)
1344 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1347 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1349 return !skb_csum_unnecessary(skb) &&
1350 __skb_checksum_complete(skb);
1353 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1354 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1355 void tcp_set_state(struct sock *sk, int state);
1356 void tcp_done(struct sock *sk);
1357 int tcp_abort(struct sock *sk, int err);
1359 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1362 rx_opt->num_sacks = 0;
1365 u32 tcp_default_init_rwnd(u32 mss);
1366 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1368 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1370 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1371 struct tcp_sock *tp = tcp_sk(sk);
1374 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1375 ca_ops->cong_control)
1377 delta = tcp_jiffies32 - tp->lsndtime;
1378 if (delta > inet_csk(sk)->icsk_rto)
1379 tcp_cwnd_restart(sk, delta);
1382 /* Determine a window scaling and initial window to offer. */
1383 void tcp_select_initial_window(const struct sock *sk, int __space,
1384 __u32 mss, __u32 *rcv_wnd,
1385 __u32 *window_clamp, int wscale_ok,
1386 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1388 static inline int tcp_win_from_space(const struct sock *sk, int space)
1390 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1392 return tcp_adv_win_scale <= 0 ?
1393 (space>>(-tcp_adv_win_scale)) :
1394 space - (space>>tcp_adv_win_scale);
1397 /* Note: caller must be prepared to deal with negative returns */
1398 static inline int tcp_space(const struct sock *sk)
1400 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1401 READ_ONCE(sk->sk_backlog.len) -
1402 atomic_read(&sk->sk_rmem_alloc));
1405 static inline int tcp_full_space(const struct sock *sk)
1407 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1410 extern void tcp_openreq_init_rwin(struct request_sock *req,
1411 const struct sock *sk_listener,
1412 const struct dst_entry *dst);
1414 void tcp_enter_memory_pressure(struct sock *sk);
1415 void tcp_leave_memory_pressure(struct sock *sk);
1417 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1419 struct net *net = sock_net((struct sock *)tp);
1421 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1424 static inline int keepalive_time_when(const struct tcp_sock *tp)
1426 struct net *net = sock_net((struct sock *)tp);
1428 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1431 static inline int keepalive_probes(const struct tcp_sock *tp)
1433 struct net *net = sock_net((struct sock *)tp);
1435 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1438 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1440 const struct inet_connection_sock *icsk = &tp->inet_conn;
1442 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1443 tcp_jiffies32 - tp->rcv_tstamp);
1446 static inline int tcp_fin_time(const struct sock *sk)
1448 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1449 const int rto = inet_csk(sk)->icsk_rto;
1451 if (fin_timeout < (rto << 2) - (rto >> 1))
1452 fin_timeout = (rto << 2) - (rto >> 1);
1457 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1460 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1462 if (unlikely(!time_before32(ktime_get_seconds(),
1463 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1466 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1467 * then following tcp messages have valid values. Ignore 0 value,
1468 * or else 'negative' tsval might forbid us to accept their packets.
1470 if (!rx_opt->ts_recent)
1475 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1478 if (tcp_paws_check(rx_opt, 0))
1481 /* RST segments are not recommended to carry timestamp,
1482 and, if they do, it is recommended to ignore PAWS because
1483 "their cleanup function should take precedence over timestamps."
1484 Certainly, it is mistake. It is necessary to understand the reasons
1485 of this constraint to relax it: if peer reboots, clock may go
1486 out-of-sync and half-open connections will not be reset.
1487 Actually, the problem would be not existing if all
1488 the implementations followed draft about maintaining clock
1489 via reboots. Linux-2.2 DOES NOT!
1491 However, we can relax time bounds for RST segments to MSL.
1493 if (rst && !time_before32(ktime_get_seconds(),
1494 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1499 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1500 int mib_idx, u32 *last_oow_ack_time);
1502 static inline void tcp_mib_init(struct net *net)
1505 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1506 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1507 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1508 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1512 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1514 tp->lost_skb_hint = NULL;
1517 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1519 tcp_clear_retrans_hints_partial(tp);
1520 tp->retransmit_skb_hint = NULL;
1523 union tcp_md5_addr {
1525 #if IS_ENABLED(CONFIG_IPV6)
1530 /* - key database */
1531 struct tcp_md5sig_key {
1532 struct hlist_node node;
1534 u8 family; /* AF_INET or AF_INET6 */
1535 union tcp_md5_addr addr;
1537 u8 key[TCP_MD5SIG_MAXKEYLEN];
1538 struct rcu_head rcu;
1542 struct tcp_md5sig_info {
1543 struct hlist_head head;
1544 struct rcu_head rcu;
1547 /* - pseudo header */
1548 struct tcp4_pseudohdr {
1556 struct tcp6_pseudohdr {
1557 struct in6_addr saddr;
1558 struct in6_addr daddr;
1560 __be32 protocol; /* including padding */
1563 union tcp_md5sum_block {
1564 struct tcp4_pseudohdr ip4;
1565 #if IS_ENABLED(CONFIG_IPV6)
1566 struct tcp6_pseudohdr ip6;
1570 /* - pool: digest algorithm, hash description and scratch buffer */
1571 struct tcp_md5sig_pool {
1572 struct ahash_request *md5_req;
1577 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1578 const struct sock *sk, const struct sk_buff *skb);
1579 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1580 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1582 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1583 int family, u8 prefixlen);
1584 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1585 const struct sock *addr_sk);
1587 #ifdef CONFIG_TCP_MD5SIG
1588 #include <linux/jump_label.h>
1589 extern struct static_key_false tcp_md5_needed;
1590 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk,
1591 const union tcp_md5_addr *addr,
1593 static inline struct tcp_md5sig_key *
1594 tcp_md5_do_lookup(const struct sock *sk,
1595 const union tcp_md5_addr *addr,
1598 if (!static_branch_unlikely(&tcp_md5_needed))
1600 return __tcp_md5_do_lookup(sk, addr, family);
1603 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1605 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1606 const union tcp_md5_addr *addr,
1611 #define tcp_twsk_md5_key(twsk) NULL
1614 bool tcp_alloc_md5sig_pool(void);
1616 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1617 static inline void tcp_put_md5sig_pool(void)
1622 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1623 unsigned int header_len);
1624 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1625 const struct tcp_md5sig_key *key);
1627 /* From tcp_fastopen.c */
1628 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1629 struct tcp_fastopen_cookie *cookie);
1630 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1631 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1633 struct tcp_fastopen_request {
1634 /* Fast Open cookie. Size 0 means a cookie request */
1635 struct tcp_fastopen_cookie cookie;
1636 struct msghdr *data; /* data in MSG_FASTOPEN */
1638 int copied; /* queued in tcp_connect() */
1639 struct ubuf_info *uarg;
1641 void tcp_free_fastopen_req(struct tcp_sock *tp);
1642 void tcp_fastopen_destroy_cipher(struct sock *sk);
1643 void tcp_fastopen_ctx_destroy(struct net *net);
1644 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1645 void *primary_key, void *backup_key);
1646 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1647 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1648 struct request_sock *req,
1649 struct tcp_fastopen_cookie *foc,
1650 const struct dst_entry *dst);
1651 void tcp_fastopen_init_key_once(struct net *net);
1652 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1653 struct tcp_fastopen_cookie *cookie);
1654 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1655 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1656 #define TCP_FASTOPEN_KEY_MAX 2
1657 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1658 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1660 /* Fastopen key context */
1661 struct tcp_fastopen_context {
1662 siphash_key_t key[TCP_FASTOPEN_KEY_MAX];
1664 struct rcu_head rcu;
1667 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1668 void tcp_fastopen_active_disable(struct sock *sk);
1669 bool tcp_fastopen_active_should_disable(struct sock *sk);
1670 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1671 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1673 /* Caller needs to wrap with rcu_read_(un)lock() */
1675 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1677 struct tcp_fastopen_context *ctx;
1679 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1681 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1686 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1687 const struct tcp_fastopen_cookie *orig)
1689 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1690 orig->len == foc->len &&
1691 !memcmp(orig->val, foc->val, foc->len))
1697 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1702 /* Latencies incurred by various limits for a sender. They are
1703 * chronograph-like stats that are mutually exclusive.
1707 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1708 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1709 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1713 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1714 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1716 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1717 * the same memory storage than skb->destructor/_skb_refdst
1719 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1721 skb->destructor = NULL;
1722 skb->_skb_refdst = 0UL;
1725 #define tcp_skb_tsorted_save(skb) { \
1726 unsigned long _save = skb->_skb_refdst; \
1727 skb->_skb_refdst = 0UL;
1729 #define tcp_skb_tsorted_restore(skb) \
1730 skb->_skb_refdst = _save; \
1733 void tcp_write_queue_purge(struct sock *sk);
1735 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1737 return skb_rb_first(&sk->tcp_rtx_queue);
1740 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1742 return skb_rb_last(&sk->tcp_rtx_queue);
1745 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1747 return skb_peek(&sk->sk_write_queue);
1750 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1752 return skb_peek_tail(&sk->sk_write_queue);
1755 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1756 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1758 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1760 return skb_peek(&sk->sk_write_queue);
1763 static inline bool tcp_skb_is_last(const struct sock *sk,
1764 const struct sk_buff *skb)
1766 return skb_queue_is_last(&sk->sk_write_queue, skb);
1769 static inline bool tcp_write_queue_empty(const struct sock *sk)
1771 return skb_queue_empty(&sk->sk_write_queue);
1774 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1776 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1779 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1781 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1784 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1786 __skb_queue_tail(&sk->sk_write_queue, skb);
1788 /* Queue it, remembering where we must start sending. */
1789 if (sk->sk_write_queue.next == skb)
1790 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1793 /* Insert new before skb on the write queue of sk. */
1794 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1795 struct sk_buff *skb,
1798 __skb_queue_before(&sk->sk_write_queue, skb, new);
1801 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1803 tcp_skb_tsorted_anchor_cleanup(skb);
1804 __skb_unlink(skb, &sk->sk_write_queue);
1807 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1809 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1811 tcp_skb_tsorted_anchor_cleanup(skb);
1812 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1815 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1817 list_del(&skb->tcp_tsorted_anchor);
1818 tcp_rtx_queue_unlink(skb, sk);
1819 sk_wmem_free_skb(sk, skb);
1822 static inline void tcp_push_pending_frames(struct sock *sk)
1824 if (tcp_send_head(sk)) {
1825 struct tcp_sock *tp = tcp_sk(sk);
1827 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1831 /* Start sequence of the skb just after the highest skb with SACKed
1832 * bit, valid only if sacked_out > 0 or when the caller has ensured
1833 * validity by itself.
1835 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1837 if (!tp->sacked_out)
1840 if (tp->highest_sack == NULL)
1843 return TCP_SKB_CB(tp->highest_sack)->seq;
1846 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1848 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1851 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1853 return tcp_sk(sk)->highest_sack;
1856 static inline void tcp_highest_sack_reset(struct sock *sk)
1858 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1861 /* Called when old skb is about to be deleted and replaced by new skb */
1862 static inline void tcp_highest_sack_replace(struct sock *sk,
1863 struct sk_buff *old,
1864 struct sk_buff *new)
1866 if (old == tcp_highest_sack(sk))
1867 tcp_sk(sk)->highest_sack = new;
1870 /* This helper checks if socket has IP_TRANSPARENT set */
1871 static inline bool inet_sk_transparent(const struct sock *sk)
1873 switch (sk->sk_state) {
1875 return inet_twsk(sk)->tw_transparent;
1876 case TCP_NEW_SYN_RECV:
1877 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1879 return inet_sk(sk)->transparent;
1882 /* Determines whether this is a thin stream (which may suffer from
1883 * increased latency). Used to trigger latency-reducing mechanisms.
1885 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1887 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1891 enum tcp_seq_states {
1892 TCP_SEQ_STATE_LISTENING,
1893 TCP_SEQ_STATE_ESTABLISHED,
1896 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1897 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1898 void tcp_seq_stop(struct seq_file *seq, void *v);
1900 struct tcp_seq_afinfo {
1904 struct tcp_iter_state {
1905 struct seq_net_private p;
1906 enum tcp_seq_states state;
1907 struct sock *syn_wait_sk;
1908 int bucket, offset, sbucket, num;
1912 extern struct request_sock_ops tcp_request_sock_ops;
1913 extern struct request_sock_ops tcp6_request_sock_ops;
1915 void tcp_v4_destroy_sock(struct sock *sk);
1917 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1918 netdev_features_t features);
1919 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
1920 int tcp_gro_complete(struct sk_buff *skb);
1922 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1924 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1926 struct net *net = sock_net((struct sock *)tp);
1927 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1930 /* @wake is one when sk_stream_write_space() calls us.
1931 * This sends EPOLLOUT only if notsent_bytes is half the limit.
1932 * This mimics the strategy used in sock_def_write_space().
1934 static inline bool tcp_stream_memory_free(const struct sock *sk, int wake)
1936 const struct tcp_sock *tp = tcp_sk(sk);
1937 u32 notsent_bytes = READ_ONCE(tp->write_seq) -
1938 READ_ONCE(tp->snd_nxt);
1940 return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
1943 #ifdef CONFIG_PROC_FS
1944 int tcp4_proc_init(void);
1945 void tcp4_proc_exit(void);
1948 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1949 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1950 const struct tcp_request_sock_ops *af_ops,
1951 struct sock *sk, struct sk_buff *skb);
1953 /* TCP af-specific functions */
1954 struct tcp_sock_af_ops {
1955 #ifdef CONFIG_TCP_MD5SIG
1956 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1957 const struct sock *addr_sk);
1958 int (*calc_md5_hash)(char *location,
1959 const struct tcp_md5sig_key *md5,
1960 const struct sock *sk,
1961 const struct sk_buff *skb);
1962 int (*md5_parse)(struct sock *sk,
1964 char __user *optval,
1969 struct tcp_request_sock_ops {
1971 #ifdef CONFIG_TCP_MD5SIG
1972 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1973 const struct sock *addr_sk);
1974 int (*calc_md5_hash) (char *location,
1975 const struct tcp_md5sig_key *md5,
1976 const struct sock *sk,
1977 const struct sk_buff *skb);
1979 void (*init_req)(struct request_sock *req,
1980 const struct sock *sk_listener,
1981 struct sk_buff *skb);
1982 #ifdef CONFIG_SYN_COOKIES
1983 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1986 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1987 const struct request_sock *req);
1988 u32 (*init_seq)(const struct sk_buff *skb);
1989 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
1990 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1991 struct flowi *fl, struct request_sock *req,
1992 struct tcp_fastopen_cookie *foc,
1993 enum tcp_synack_type synack_type);
1996 #ifdef CONFIG_SYN_COOKIES
1997 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1998 const struct sock *sk, struct sk_buff *skb,
2001 tcp_synq_overflow(sk);
2002 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2003 return ops->cookie_init_seq(skb, mss);
2006 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2007 const struct sock *sk, struct sk_buff *skb,
2014 int tcpv4_offload_init(void);
2016 void tcp_v4_init(void);
2017 void tcp_init(void);
2019 /* tcp_recovery.c */
2020 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2021 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2022 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2024 extern void tcp_rack_mark_lost(struct sock *sk);
2025 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2027 extern void tcp_rack_reo_timeout(struct sock *sk);
2028 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2030 /* At how many usecs into the future should the RTO fire? */
2031 static inline s64 tcp_rto_delta_us(const struct sock *sk)
2033 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2034 u32 rto = inet_csk(sk)->icsk_rto;
2035 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2037 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2041 * Save and compile IPv4 options, return a pointer to it
2043 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2044 struct sk_buff *skb)
2046 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2047 struct ip_options_rcu *dopt = NULL;
2050 int opt_size = sizeof(*dopt) + opt->optlen;
2052 dopt = kmalloc(opt_size, GFP_ATOMIC);
2053 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2061 /* locally generated TCP pure ACKs have skb->truesize == 2
2062 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2063 * This is much faster than dissecting the packet to find out.
2064 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2066 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2068 return skb->truesize == 2;
2071 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2076 static inline int tcp_inq(struct sock *sk)
2078 struct tcp_sock *tp = tcp_sk(sk);
2081 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2083 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2085 before(tp->urg_seq, tp->copied_seq) ||
2086 !before(tp->urg_seq, tp->rcv_nxt)) {
2088 answ = tp->rcv_nxt - tp->copied_seq;
2090 /* Subtract 1, if FIN was received */
2091 if (answ && sock_flag(sk, SOCK_DONE))
2094 answ = tp->urg_seq - tp->copied_seq;
2100 int tcp_peek_len(struct socket *sock);
2102 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2106 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2107 tp->segs_in += segs_in;
2108 if (skb->len > tcp_hdrlen(skb))
2109 tp->data_segs_in += segs_in;
2113 * TCP listen path runs lockless.
2114 * We forced "struct sock" to be const qualified to make sure
2115 * we don't modify one of its field by mistake.
2116 * Here, we increment sk_drops which is an atomic_t, so we can safely
2117 * make sock writable again.
2119 static inline void tcp_listendrop(const struct sock *sk)
2121 atomic_inc(&((struct sock *)sk)->sk_drops);
2122 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2125 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2128 * Interface for adding Upper Level Protocols over TCP
2131 #define TCP_ULP_NAME_MAX 16
2132 #define TCP_ULP_MAX 128
2133 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2135 struct tcp_ulp_ops {
2136 struct list_head list;
2138 /* initialize ulp */
2139 int (*init)(struct sock *sk);
2141 void (*update)(struct sock *sk, struct proto *p);
2143 void (*release)(struct sock *sk);
2145 int (*get_info)(const struct sock *sk, struct sk_buff *skb);
2146 size_t (*get_info_size)(const struct sock *sk);
2148 char name[TCP_ULP_NAME_MAX];
2149 struct module *owner;
2151 int tcp_register_ulp(struct tcp_ulp_ops *type);
2152 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2153 int tcp_set_ulp(struct sock *sk, const char *name);
2154 void tcp_get_available_ulp(char *buf, size_t len);
2155 void tcp_cleanup_ulp(struct sock *sk);
2156 void tcp_update_ulp(struct sock *sk, struct proto *p);
2158 #define MODULE_ALIAS_TCP_ULP(name) \
2159 __MODULE_INFO(alias, alias_userspace, name); \
2160 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2165 int tcp_bpf_init(struct sock *sk);
2166 void tcp_bpf_reinit(struct sock *sk);
2167 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes,
2169 int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
2170 int nonblock, int flags, int *addr_len);
2171 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock,
2172 struct msghdr *msg, int len, int flags);
2174 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2175 * is < 0, then the BPF op failed (for example if the loaded BPF
2176 * program does not support the chosen operation or there is no BPF
2180 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2182 struct bpf_sock_ops_kern sock_ops;
2185 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2186 if (sk_fullsock(sk)) {
2187 sock_ops.is_fullsock = 1;
2188 sock_owned_by_me(sk);
2194 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2196 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2198 ret = sock_ops.reply;
2204 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2206 u32 args[2] = {arg1, arg2};
2208 return tcp_call_bpf(sk, op, 2, args);
2211 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2214 u32 args[3] = {arg1, arg2, arg3};
2216 return tcp_call_bpf(sk, op, 3, args);
2220 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2225 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2230 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2238 static inline u32 tcp_timeout_init(struct sock *sk)
2242 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2245 timeout = TCP_TIMEOUT_INIT;
2249 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2253 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2260 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2262 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2265 static inline void tcp_bpf_rtt(struct sock *sk)
2267 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2268 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2271 #if IS_ENABLED(CONFIG_SMC)
2272 extern struct static_key_false tcp_have_smc;
2275 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2276 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2277 void (*cad)(struct sock *sk, u32 ack_seq));
2278 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2279 void clean_acked_data_flush(void);
2282 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2283 static inline void tcp_add_tx_delay(struct sk_buff *skb,
2284 const struct tcp_sock *tp)
2286 if (static_branch_unlikely(&tcp_tx_delay_enabled))
2287 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2290 /* Compute Earliest Departure Time for some control packets
2291 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2293 static inline u64 tcp_transmit_time(const struct sock *sk)
2295 if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2296 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2297 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2299 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;