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
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>
47 extern struct inet_hashinfo tcp_hashinfo;
49 extern struct percpu_counter tcp_orphan_count;
50 void tcp_time_wait(struct sock *sk, int state, int timeo);
52 #define MAX_TCP_HEADER (128 + MAX_HEADER)
53 #define MAX_TCP_OPTION_SPACE 40
56 * Never offer a window over 32767 without using window scaling. Some
57 * poor stacks do signed 16bit maths!
59 #define MAX_TCP_WINDOW 32767U
61 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
62 #define TCP_MIN_MSS 88U
64 /* The least MTU to use for probing */
65 #define TCP_BASE_MSS 1024
67 /* probing interval, default to 10 minutes as per RFC4821 */
68 #define TCP_PROBE_INTERVAL 600
70 /* Specify interval when tcp mtu probing will stop */
71 #define TCP_PROBE_THRESHOLD 8
73 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
74 #define TCP_FASTRETRANS_THRESH 3
76 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
77 #define TCP_MAX_QUICKACKS 16U
79 /* Maximal number of window scale according to RFC1323 */
80 #define TCP_MAX_WSCALE 14U
83 #define TCP_URG_VALID 0x0100
84 #define TCP_URG_NOTYET 0x0200
85 #define TCP_URG_READ 0x0400
87 #define TCP_RETR1 3 /*
88 * This is how many retries it does before it
89 * tries to figure out if the gateway is
90 * down. Minimal RFC value is 3; it corresponds
91 * to ~3sec-8min depending on RTO.
94 #define TCP_RETR2 15 /*
95 * This should take at least
96 * 90 minutes to time out.
97 * RFC1122 says that the limit is 100 sec.
98 * 15 is ~13-30min depending on RTO.
101 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
102 * when active opening a connection.
103 * RFC1122 says the minimum retry MUST
104 * be at least 180secs. Nevertheless
105 * this value is corresponding to
106 * 63secs of retransmission with the
107 * current initial RTO.
110 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
111 * when passive opening a connection.
112 * This is corresponding to 31secs of
113 * retransmission with the current
117 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
118 * state, about 60 seconds */
119 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
120 /* BSD style FIN_WAIT2 deadlock breaker.
121 * It used to be 3min, new value is 60sec,
122 * to combine FIN-WAIT-2 timeout with
126 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
128 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
129 #define TCP_ATO_MIN ((unsigned)(HZ/25))
131 #define TCP_DELACK_MIN 4U
132 #define TCP_ATO_MIN 4U
134 #define TCP_RTO_MAX ((unsigned)(120*HZ))
135 #define TCP_RTO_MIN ((unsigned)(HZ/5))
136 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
137 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
138 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
139 * used as a fallback RTO for the
140 * initial data transmission if no
141 * valid RTT sample has been acquired,
142 * most likely due to retrans in 3WHS.
145 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
146 * for local resources.
148 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
149 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
150 #define TCP_KEEPALIVE_INTVL (75*HZ)
152 #define MAX_TCP_KEEPIDLE 32767
153 #define MAX_TCP_KEEPINTVL 32767
154 #define MAX_TCP_KEEPCNT 127
155 #define MAX_TCP_SYNCNT 127
157 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
159 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
160 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
161 * after this time. It should be equal
162 * (or greater than) TCP_TIMEWAIT_LEN
163 * to provide reliability equal to one
164 * provided by timewait state.
166 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
167 * timestamps. It must be less than
168 * minimal timewait lifetime.
174 #define TCPOPT_NOP 1 /* Padding */
175 #define TCPOPT_EOL 0 /* End of options */
176 #define TCPOPT_MSS 2 /* Segment size negotiating */
177 #define TCPOPT_WINDOW 3 /* Window scaling */
178 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
179 #define TCPOPT_SACK 5 /* SACK Block */
180 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
181 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
182 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
183 #define TCPOPT_EXP 254 /* Experimental */
184 /* Magic number to be after the option value for sharing TCP
185 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
187 #define TCPOPT_FASTOPEN_MAGIC 0xF989
188 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9
194 #define TCPOLEN_MSS 4
195 #define TCPOLEN_WINDOW 3
196 #define TCPOLEN_SACK_PERM 2
197 #define TCPOLEN_TIMESTAMP 10
198 #define TCPOLEN_MD5SIG 18
199 #define TCPOLEN_FASTOPEN_BASE 2
200 #define TCPOLEN_EXP_FASTOPEN_BASE 4
201 #define TCPOLEN_EXP_SMC_BASE 6
203 /* But this is what stacks really send out. */
204 #define TCPOLEN_TSTAMP_ALIGNED 12
205 #define TCPOLEN_WSCALE_ALIGNED 4
206 #define TCPOLEN_SACKPERM_ALIGNED 4
207 #define TCPOLEN_SACK_BASE 2
208 #define TCPOLEN_SACK_BASE_ALIGNED 4
209 #define TCPOLEN_SACK_PERBLOCK 8
210 #define TCPOLEN_MD5SIG_ALIGNED 20
211 #define TCPOLEN_MSS_ALIGNED 4
212 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
214 /* Flags in tp->nonagle */
215 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
216 #define TCP_NAGLE_CORK 2 /* Socket is corked */
217 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
219 /* TCP thin-stream limits */
220 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
222 /* TCP initial congestion window as per rfc6928 */
223 #define TCP_INIT_CWND 10
225 /* Bit Flags for sysctl_tcp_fastopen */
226 #define TFO_CLIENT_ENABLE 1
227 #define TFO_SERVER_ENABLE 2
228 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
230 /* Accept SYN data w/o any cookie option */
231 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
233 /* Force enable TFO on all listeners, i.e., not requiring the
234 * TCP_FASTOPEN socket option.
236 #define TFO_SERVER_WO_SOCKOPT1 0x400
239 /* sysctl variables for tcp */
240 extern int sysctl_tcp_max_orphans;
241 extern long sysctl_tcp_mem[3];
243 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
244 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
245 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */
247 extern atomic_long_t tcp_memory_allocated;
248 extern struct percpu_counter tcp_sockets_allocated;
249 extern unsigned long tcp_memory_pressure;
251 /* optimized version of sk_under_memory_pressure() for TCP sockets */
252 static inline bool tcp_under_memory_pressure(const struct sock *sk)
254 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
255 mem_cgroup_under_socket_pressure(sk->sk_memcg))
258 return tcp_memory_pressure;
261 * The next routines deal with comparing 32 bit unsigned ints
262 * and worry about wraparound (automatic with unsigned arithmetic).
265 static inline bool before(__u32 seq1, __u32 seq2)
267 return (__s32)(seq1-seq2) < 0;
269 #define after(seq2, seq1) before(seq1, seq2)
271 /* is s2<=s1<=s3 ? */
272 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
274 return seq3 - seq2 >= seq1 - seq2;
277 static inline bool tcp_out_of_memory(struct sock *sk)
279 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
280 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
285 void sk_forced_mem_schedule(struct sock *sk, int size);
287 static inline bool tcp_too_many_orphans(struct sock *sk, int shift)
289 struct percpu_counter *ocp = sk->sk_prot->orphan_count;
290 int orphans = percpu_counter_read_positive(ocp);
292 if (orphans << shift > sysctl_tcp_max_orphans) {
293 orphans = percpu_counter_sum_positive(ocp);
294 if (orphans << shift > sysctl_tcp_max_orphans)
300 bool tcp_check_oom(struct sock *sk, int shift);
303 extern struct proto tcp_prot;
305 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
306 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
307 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
308 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
310 void tcp_tasklet_init(void);
312 int tcp_v4_err(struct sk_buff *skb, u32);
314 void tcp_shutdown(struct sock *sk, int how);
316 int tcp_v4_early_demux(struct sk_buff *skb);
317 int tcp_v4_rcv(struct sk_buff *skb);
319 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
320 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
321 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
322 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
324 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
325 size_t size, int flags);
326 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
327 size_t size, int flags);
328 void tcp_release_cb(struct sock *sk);
329 void tcp_wfree(struct sk_buff *skb);
330 void tcp_write_timer_handler(struct sock *sk);
331 void tcp_delack_timer_handler(struct sock *sk);
332 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
333 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
334 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
335 void tcp_rcv_space_adjust(struct sock *sk);
336 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
337 void tcp_twsk_destructor(struct sock *sk);
338 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
339 struct pipe_inode_info *pipe, size_t len,
342 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
343 static inline void tcp_dec_quickack_mode(struct sock *sk,
344 const unsigned int pkts)
346 struct inet_connection_sock *icsk = inet_csk(sk);
348 if (icsk->icsk_ack.quick) {
349 if (pkts >= icsk->icsk_ack.quick) {
350 icsk->icsk_ack.quick = 0;
351 /* Leaving quickack mode we deflate ATO. */
352 icsk->icsk_ack.ato = TCP_ATO_MIN;
354 icsk->icsk_ack.quick -= pkts;
359 #define TCP_ECN_QUEUE_CWR 2
360 #define TCP_ECN_DEMAND_CWR 4
361 #define TCP_ECN_SEEN 8
371 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
373 const struct tcphdr *th);
374 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
375 struct request_sock *req, bool fastopen,
377 int tcp_child_process(struct sock *parent, struct sock *child,
378 struct sk_buff *skb);
379 void tcp_enter_loss(struct sock *sk);
380 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int flag);
381 void tcp_clear_retrans(struct tcp_sock *tp);
382 void tcp_update_metrics(struct sock *sk);
383 void tcp_init_metrics(struct sock *sk);
384 void tcp_metrics_init(void);
385 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
386 void tcp_close(struct sock *sk, long timeout);
387 void tcp_init_sock(struct sock *sk);
388 void tcp_init_transfer(struct sock *sk, int bpf_op);
389 __poll_t tcp_poll(struct file *file, struct socket *sock,
390 struct poll_table_struct *wait);
391 int tcp_getsockopt(struct sock *sk, int level, int optname,
392 char __user *optval, int __user *optlen);
393 int tcp_setsockopt(struct sock *sk, int level, int optname,
394 char __user *optval, unsigned int optlen);
395 int compat_tcp_getsockopt(struct sock *sk, int level, int optname,
396 char __user *optval, int __user *optlen);
397 int compat_tcp_setsockopt(struct sock *sk, int level, int optname,
398 char __user *optval, unsigned int optlen);
399 void tcp_set_keepalive(struct sock *sk, int val);
400 void tcp_syn_ack_timeout(const struct request_sock *req);
401 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
402 int flags, int *addr_len);
403 int tcp_set_rcvlowat(struct sock *sk, int val);
404 void tcp_data_ready(struct sock *sk);
406 int tcp_mmap(struct file *file, struct socket *sock,
407 struct vm_area_struct *vma);
409 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
410 struct tcp_options_received *opt_rx,
411 int estab, struct tcp_fastopen_cookie *foc);
412 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
415 * TCP v4 functions exported for the inet6 API
418 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
419 void tcp_v4_mtu_reduced(struct sock *sk);
420 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
421 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
422 struct sock *tcp_create_openreq_child(const struct sock *sk,
423 struct request_sock *req,
424 struct sk_buff *skb);
425 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
426 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
427 struct request_sock *req,
428 struct dst_entry *dst,
429 struct request_sock *req_unhash,
431 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
432 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
433 int tcp_connect(struct sock *sk);
434 enum tcp_synack_type {
439 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
440 struct request_sock *req,
441 struct tcp_fastopen_cookie *foc,
442 enum tcp_synack_type synack_type);
443 int tcp_disconnect(struct sock *sk, int flags);
445 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
446 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
447 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
449 /* From syncookies.c */
450 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
451 struct request_sock *req,
452 struct dst_entry *dst, u32 tsoff);
453 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
455 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
456 #ifdef CONFIG_SYN_COOKIES
458 /* Syncookies use a monotonic timer which increments every 60 seconds.
459 * This counter is used both as a hash input and partially encoded into
460 * the cookie value. A cookie is only validated further if the delta
461 * between the current counter value and the encoded one is less than this,
462 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
463 * the counter advances immediately after a cookie is generated).
465 #define MAX_SYNCOOKIE_AGE 2
466 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
467 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
469 /* syncookies: remember time of last synqueue overflow
470 * But do not dirty this field too often (once per second is enough)
471 * It is racy as we do not hold a lock, but race is very minor.
473 static inline void tcp_synq_overflow(const struct sock *sk)
475 unsigned int last_overflow;
476 unsigned int now = jiffies;
478 if (sk->sk_reuseport) {
479 struct sock_reuseport *reuse;
481 reuse = rcu_dereference(sk->sk_reuseport_cb);
483 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
484 if (time_after32(now, last_overflow + HZ))
485 WRITE_ONCE(reuse->synq_overflow_ts, now);
490 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
491 if (time_after32(now, last_overflow + HZ))
492 tcp_sk(sk)->rx_opt.ts_recent_stamp = now;
495 /* syncookies: no recent synqueue overflow on this listening socket? */
496 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
498 unsigned int last_overflow;
499 unsigned int now = jiffies;
501 if (sk->sk_reuseport) {
502 struct sock_reuseport *reuse;
504 reuse = rcu_dereference(sk->sk_reuseport_cb);
506 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
507 return time_after32(now, last_overflow +
508 TCP_SYNCOOKIE_VALID);
512 last_overflow = tcp_sk(sk)->rx_opt.ts_recent_stamp;
513 return time_after32(now, last_overflow + TCP_SYNCOOKIE_VALID);
516 static inline u32 tcp_cookie_time(void)
518 u64 val = get_jiffies_64();
520 do_div(val, TCP_SYNCOOKIE_PERIOD);
524 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
526 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
527 u64 cookie_init_timestamp(struct request_sock *req);
528 bool cookie_timestamp_decode(const struct net *net,
529 struct tcp_options_received *opt);
530 bool cookie_ecn_ok(const struct tcp_options_received *opt,
531 const struct net *net, const struct dst_entry *dst);
533 /* From net/ipv6/syncookies.c */
534 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
536 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
538 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
539 const struct tcphdr *th, u16 *mssp);
540 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
544 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
546 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
547 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
548 void tcp_retransmit_timer(struct sock *sk);
549 void tcp_xmit_retransmit_queue(struct sock *);
550 void tcp_simple_retransmit(struct sock *);
551 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
552 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
554 TCP_FRAG_IN_WRITE_QUEUE,
555 TCP_FRAG_IN_RTX_QUEUE,
557 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
558 struct sk_buff *skb, u32 len,
559 unsigned int mss_now, gfp_t gfp);
561 void tcp_send_probe0(struct sock *);
562 void tcp_send_partial(struct sock *);
563 int tcp_write_wakeup(struct sock *, int mib);
564 void tcp_send_fin(struct sock *sk);
565 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
566 int tcp_send_synack(struct sock *);
567 void tcp_push_one(struct sock *, unsigned int mss_now);
568 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
569 void tcp_send_ack(struct sock *sk);
570 void tcp_send_delayed_ack(struct sock *sk);
571 void tcp_send_loss_probe(struct sock *sk);
572 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
573 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
574 const struct sk_buff *next_skb);
577 void tcp_rearm_rto(struct sock *sk);
578 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
579 void tcp_reset(struct sock *sk);
580 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
581 void tcp_fin(struct sock *sk);
584 void tcp_init_xmit_timers(struct sock *);
585 static inline void tcp_clear_xmit_timers(struct sock *sk)
587 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
590 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
593 inet_csk_clear_xmit_timers(sk);
596 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
597 unsigned int tcp_current_mss(struct sock *sk);
599 /* Bound MSS / TSO packet size with the half of the window */
600 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
604 /* When peer uses tiny windows, there is no use in packetizing
605 * to sub-MSS pieces for the sake of SWS or making sure there
606 * are enough packets in the pipe for fast recovery.
608 * On the other hand, for extremely large MSS devices, handling
609 * smaller than MSS windows in this way does make sense.
611 if (tp->max_window > TCP_MSS_DEFAULT)
612 cutoff = (tp->max_window >> 1);
614 cutoff = tp->max_window;
616 if (cutoff && pktsize > cutoff)
617 return max_t(int, cutoff, 68U - tp->tcp_header_len);
623 void tcp_get_info(struct sock *, struct tcp_info *);
625 /* Read 'sendfile()'-style from a TCP socket */
626 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
627 sk_read_actor_t recv_actor);
629 void tcp_initialize_rcv_mss(struct sock *sk);
631 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
632 int tcp_mss_to_mtu(struct sock *sk, int mss);
633 void tcp_mtup_init(struct sock *sk);
634 void tcp_init_buffer_space(struct sock *sk);
636 static inline void tcp_bound_rto(const struct sock *sk)
638 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
639 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
642 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
644 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
647 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
649 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
650 ntohl(TCP_FLAG_ACK) |
654 static inline void tcp_fast_path_on(struct tcp_sock *tp)
656 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
659 static inline void tcp_fast_path_check(struct sock *sk)
661 struct tcp_sock *tp = tcp_sk(sk);
663 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
665 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
667 tcp_fast_path_on(tp);
670 /* Compute the actual rto_min value */
671 static inline u32 tcp_rto_min(struct sock *sk)
673 const struct dst_entry *dst = __sk_dst_get(sk);
674 u32 rto_min = TCP_RTO_MIN;
676 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
677 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
681 static inline u32 tcp_rto_min_us(struct sock *sk)
683 return jiffies_to_usecs(tcp_rto_min(sk));
686 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
688 return dst_metric_locked(dst, RTAX_CC_ALGO);
691 /* Minimum RTT in usec. ~0 means not available. */
692 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
694 return minmax_get(&tp->rtt_min);
697 /* Compute the actual receive window we are currently advertising.
698 * Rcv_nxt can be after the window if our peer push more data
699 * than the offered window.
701 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
703 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
710 /* Choose a new window, without checks for shrinking, and without
711 * scaling applied to the result. The caller does these things
712 * if necessary. This is a "raw" window selection.
714 u32 __tcp_select_window(struct sock *sk);
716 void tcp_send_window_probe(struct sock *sk);
718 /* TCP uses 32bit jiffies to save some space.
719 * Note that this is different from tcp_time_stamp, which
720 * historically has been the same until linux-4.13.
722 #define tcp_jiffies32 ((u32)jiffies)
725 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
726 * It is no longer tied to jiffies, but to 1 ms clock.
727 * Note: double check if you want to use tcp_jiffies32 instead of this.
729 #define TCP_TS_HZ 1000
731 static inline u64 tcp_clock_ns(void)
733 return ktime_get_ns();
736 static inline u64 tcp_clock_us(void)
738 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
741 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
742 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
744 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
747 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
748 static inline u32 tcp_time_stamp_raw(void)
750 return div_u64(tcp_clock_ns(), NSEC_PER_SEC / TCP_TS_HZ);
753 void tcp_mstamp_refresh(struct tcp_sock *tp);
755 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
757 return max_t(s64, t1 - t0, 0);
760 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
762 return div_u64(skb->skb_mstamp_ns, NSEC_PER_SEC / TCP_TS_HZ);
765 /* provide the departure time in us unit */
766 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
768 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
772 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
774 #define TCPHDR_FIN 0x01
775 #define TCPHDR_SYN 0x02
776 #define TCPHDR_RST 0x04
777 #define TCPHDR_PSH 0x08
778 #define TCPHDR_ACK 0x10
779 #define TCPHDR_URG 0x20
780 #define TCPHDR_ECE 0x40
781 #define TCPHDR_CWR 0x80
783 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
785 /* This is what the send packet queuing engine uses to pass
786 * TCP per-packet control information to the transmission code.
787 * We also store the host-order sequence numbers in here too.
788 * This is 44 bytes if IPV6 is enabled.
789 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
792 __u32 seq; /* Starting sequence number */
793 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
795 /* Note : tcp_tw_isn is used in input path only
796 * (isn chosen by tcp_timewait_state_process())
798 * tcp_gso_segs/size are used in write queue only,
799 * cf tcp_skb_pcount()/tcp_skb_mss()
807 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
809 __u8 sacked; /* State flags for SACK. */
810 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
811 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
812 #define TCPCB_LOST 0x04 /* SKB is lost */
813 #define TCPCB_TAGBITS 0x07 /* All tag bits */
814 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
815 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
816 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
819 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
820 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
821 eor:1, /* Is skb MSG_EOR marked? */
822 has_rxtstamp:1, /* SKB has a RX timestamp */
824 __u32 ack_seq; /* Sequence number ACK'd */
827 /* There is space for up to 24 bytes */
828 __u32 in_flight:30,/* Bytes in flight at transmit */
829 is_app_limited:1, /* cwnd not fully used? */
831 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
833 /* start of send pipeline phase */
835 /* when we reached the "delivered" count */
836 u64 delivered_mstamp;
837 } tx; /* only used for outgoing skbs */
839 struct inet_skb_parm h4;
840 #if IS_ENABLED(CONFIG_IPV6)
841 struct inet6_skb_parm h6;
843 } header; /* For incoming skbs */
846 struct sock *sk_redir;
852 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
854 static inline void bpf_compute_data_end_sk_skb(struct sk_buff *skb)
856 TCP_SKB_CB(skb)->bpf.data_end = skb->data + skb_headlen(skb);
859 static inline bool tcp_skb_bpf_ingress(const struct sk_buff *skb)
861 return TCP_SKB_CB(skb)->bpf.flags & BPF_F_INGRESS;
864 static inline struct sock *tcp_skb_bpf_redirect_fetch(struct sk_buff *skb)
866 return TCP_SKB_CB(skb)->bpf.sk_redir;
869 static inline void tcp_skb_bpf_redirect_clear(struct sk_buff *skb)
871 TCP_SKB_CB(skb)->bpf.sk_redir = NULL;
874 #if IS_ENABLED(CONFIG_IPV6)
875 /* This is the variant of inet6_iif() that must be used by TCP,
876 * as TCP moves IP6CB into a different location in skb->cb[]
878 static inline int tcp_v6_iif(const struct sk_buff *skb)
880 return TCP_SKB_CB(skb)->header.h6.iif;
883 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
885 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
887 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
890 /* TCP_SKB_CB reference means this can not be used from early demux */
891 static inline int tcp_v6_sdif(const struct sk_buff *skb)
893 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
894 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
895 return TCP_SKB_CB(skb)->header.h6.iif;
901 static inline bool inet_exact_dif_match(struct net *net, struct sk_buff *skb)
903 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
904 if (!net->ipv4.sysctl_tcp_l3mdev_accept &&
905 skb && ipv4_l3mdev_skb(IPCB(skb)->flags))
911 /* TCP_SKB_CB reference means this can not be used from early demux */
912 static inline int tcp_v4_sdif(struct sk_buff *skb)
914 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
915 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
916 return TCP_SKB_CB(skb)->header.h4.iif;
921 /* Due to TSO, an SKB can be composed of multiple actual
922 * packets. To keep these tracked properly, we use this.
924 static inline int tcp_skb_pcount(const struct sk_buff *skb)
926 return TCP_SKB_CB(skb)->tcp_gso_segs;
929 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
931 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
934 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
936 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
939 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
940 static inline int tcp_skb_mss(const struct sk_buff *skb)
942 return TCP_SKB_CB(skb)->tcp_gso_size;
945 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
947 return likely(!TCP_SKB_CB(skb)->eor);
950 /* Events passed to congestion control interface */
952 CA_EVENT_TX_START, /* first transmit when no packets in flight */
953 CA_EVENT_CWND_RESTART, /* congestion window restart */
954 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
955 CA_EVENT_LOSS, /* loss timeout */
956 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
957 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
960 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
961 enum tcp_ca_ack_event_flags {
962 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
963 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
964 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
968 * Interface for adding new TCP congestion control handlers
970 #define TCP_CA_NAME_MAX 16
971 #define TCP_CA_MAX 128
972 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
974 #define TCP_CA_UNSPEC 0
976 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
977 #define TCP_CONG_NON_RESTRICTED 0x1
978 /* Requires ECN/ECT set on all packets */
979 #define TCP_CONG_NEEDS_ECN 0x2
989 /* A rate sample measures the number of (original/retransmitted) data
990 * packets delivered "delivered" over an interval of time "interval_us".
991 * The tcp_rate.c code fills in the rate sample, and congestion
992 * control modules that define a cong_control function to run at the end
993 * of ACK processing can optionally chose to consult this sample when
994 * setting cwnd and pacing rate.
995 * A sample is invalid if "delivered" or "interval_us" is negative.
998 u64 prior_mstamp; /* starting timestamp for interval */
999 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1000 s32 delivered; /* number of packets delivered over interval */
1001 long interval_us; /* time for tp->delivered to incr "delivered" */
1002 u32 snd_interval_us; /* snd interval for delivered packets */
1003 u32 rcv_interval_us; /* rcv interval for delivered packets */
1004 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1005 int losses; /* number of packets marked lost upon ACK */
1006 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1007 u32 prior_in_flight; /* in flight before this ACK */
1008 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1009 bool is_retrans; /* is sample from retransmission? */
1010 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1013 struct tcp_congestion_ops {
1014 struct list_head list;
1018 /* initialize private data (optional) */
1019 void (*init)(struct sock *sk);
1020 /* cleanup private data (optional) */
1021 void (*release)(struct sock *sk);
1023 /* return slow start threshold (required) */
1024 u32 (*ssthresh)(struct sock *sk);
1025 /* do new cwnd calculation (required) */
1026 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1027 /* call before changing ca_state (optional) */
1028 void (*set_state)(struct sock *sk, u8 new_state);
1029 /* call when cwnd event occurs (optional) */
1030 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1031 /* call when ack arrives (optional) */
1032 void (*in_ack_event)(struct sock *sk, u32 flags);
1033 /* new value of cwnd after loss (required) */
1034 u32 (*undo_cwnd)(struct sock *sk);
1035 /* hook for packet ack accounting (optional) */
1036 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1037 /* override sysctl_tcp_min_tso_segs */
1038 u32 (*min_tso_segs)(struct sock *sk);
1039 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1040 u32 (*sndbuf_expand)(struct sock *sk);
1041 /* call when packets are delivered to update cwnd and pacing rate,
1042 * after all the ca_state processing. (optional)
1044 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1045 /* get info for inet_diag (optional) */
1046 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1047 union tcp_cc_info *info);
1049 char name[TCP_CA_NAME_MAX];
1050 struct module *owner;
1053 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1054 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1056 void tcp_assign_congestion_control(struct sock *sk);
1057 void tcp_init_congestion_control(struct sock *sk);
1058 void tcp_cleanup_congestion_control(struct sock *sk);
1059 int tcp_set_default_congestion_control(struct net *net, const char *name);
1060 void tcp_get_default_congestion_control(struct net *net, char *name);
1061 void tcp_get_available_congestion_control(char *buf, size_t len);
1062 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1063 int tcp_set_allowed_congestion_control(char *allowed);
1064 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load, bool reinit);
1065 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1066 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1068 u32 tcp_reno_ssthresh(struct sock *sk);
1069 u32 tcp_reno_undo_cwnd(struct sock *sk);
1070 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1071 extern struct tcp_congestion_ops tcp_reno;
1073 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1074 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1076 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1078 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1084 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1086 const struct inet_connection_sock *icsk = inet_csk(sk);
1088 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1091 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1093 struct inet_connection_sock *icsk = inet_csk(sk);
1095 if (icsk->icsk_ca_ops->set_state)
1096 icsk->icsk_ca_ops->set_state(sk, ca_state);
1097 icsk->icsk_ca_state = ca_state;
1100 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1102 const struct inet_connection_sock *icsk = inet_csk(sk);
1104 if (icsk->icsk_ca_ops->cwnd_event)
1105 icsk->icsk_ca_ops->cwnd_event(sk, event);
1108 /* From tcp_rate.c */
1109 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1110 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1111 struct rate_sample *rs);
1112 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1113 bool is_sack_reneg, struct rate_sample *rs);
1114 void tcp_rate_check_app_limited(struct sock *sk);
1116 /* These functions determine how the current flow behaves in respect of SACK
1117 * handling. SACK is negotiated with the peer, and therefore it can vary
1118 * between different flows.
1120 * tcp_is_sack - SACK enabled
1121 * tcp_is_reno - No SACK
1123 static inline int tcp_is_sack(const struct tcp_sock *tp)
1125 return likely(tp->rx_opt.sack_ok);
1128 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1130 return !tcp_is_sack(tp);
1133 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1135 return tp->sacked_out + tp->lost_out;
1138 /* This determines how many packets are "in the network" to the best
1139 * of our knowledge. In many cases it is conservative, but where
1140 * detailed information is available from the receiver (via SACK
1141 * blocks etc.) we can make more aggressive calculations.
1143 * Use this for decisions involving congestion control, use just
1144 * tp->packets_out to determine if the send queue is empty or not.
1146 * Read this equation as:
1148 * "Packets sent once on transmission queue" MINUS
1149 * "Packets left network, but not honestly ACKed yet" PLUS
1150 * "Packets fast retransmitted"
1152 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1154 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1157 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1159 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1161 return tp->snd_cwnd < tp->snd_ssthresh;
1164 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1166 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1169 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1171 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1172 (1 << inet_csk(sk)->icsk_ca_state);
1175 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1176 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1179 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1181 const struct tcp_sock *tp = tcp_sk(sk);
1183 if (tcp_in_cwnd_reduction(sk))
1184 return tp->snd_ssthresh;
1186 return max(tp->snd_ssthresh,
1187 ((tp->snd_cwnd >> 1) +
1188 (tp->snd_cwnd >> 2)));
1191 /* Use define here intentionally to get WARN_ON location shown at the caller */
1192 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1194 void tcp_enter_cwr(struct sock *sk);
1195 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1197 /* The maximum number of MSS of available cwnd for which TSO defers
1198 * sending if not using sysctl_tcp_tso_win_divisor.
1200 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1205 /* Returns end sequence number of the receiver's advertised window */
1206 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1208 return tp->snd_una + tp->snd_wnd;
1211 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1212 * flexible approach. The RFC suggests cwnd should not be raised unless
1213 * it was fully used previously. And that's exactly what we do in
1214 * congestion avoidance mode. But in slow start we allow cwnd to grow
1215 * as long as the application has used half the cwnd.
1217 * cwnd is 10 (IW10), but application sends 9 frames.
1218 * We allow cwnd to reach 18 when all frames are ACKed.
1219 * This check is safe because it's as aggressive as slow start which already
1220 * risks 100% overshoot. The advantage is that we discourage application to
1221 * either send more filler packets or data to artificially blow up the cwnd
1222 * usage, and allow application-limited process to probe bw more aggressively.
1224 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1226 const struct tcp_sock *tp = tcp_sk(sk);
1228 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1229 if (tcp_in_slow_start(tp))
1230 return tp->snd_cwnd < 2 * tp->max_packets_out;
1232 return tp->is_cwnd_limited;
1235 /* BBR congestion control needs pacing.
1236 * Same remark for SO_MAX_PACING_RATE.
1237 * sch_fq packet scheduler is efficiently handling pacing,
1238 * but is not always installed/used.
1239 * Return true if TCP stack should pace packets itself.
1241 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1243 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1246 /* Return in jiffies the delay before one skb is sent.
1247 * If @skb is NULL, we look at EDT for next packet being sent on the socket.
1249 static inline unsigned long tcp_pacing_delay(const struct sock *sk,
1250 const struct sk_buff *skb)
1252 s64 pacing_delay = skb ? skb->tstamp : tcp_sk(sk)->tcp_wstamp_ns;
1254 pacing_delay -= tcp_sk(sk)->tcp_clock_cache;
1256 return pacing_delay > 0 ? nsecs_to_jiffies(pacing_delay) : 0;
1259 static inline void tcp_reset_xmit_timer(struct sock *sk,
1262 const unsigned long max_when,
1263 const struct sk_buff *skb)
1265 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk, skb),
1269 /* Something is really bad, we could not queue an additional packet,
1270 * because qdisc is full or receiver sent a 0 window, or we are paced.
1271 * We do not want to add fuel to the fire, or abort too early,
1272 * so make sure the timer we arm now is at least 200ms in the future,
1273 * regardless of current icsk_rto value (as it could be ~2ms)
1275 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1277 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1280 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1281 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1282 unsigned long max_when)
1284 u64 when = (u64)tcp_probe0_base(sk) << inet_csk(sk)->icsk_backoff;
1286 return (unsigned long)min_t(u64, when, max_when);
1289 static inline void tcp_check_probe_timer(struct sock *sk)
1291 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1292 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1293 tcp_probe0_base(sk), TCP_RTO_MAX,
1297 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1302 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1308 * Calculate(/check) TCP checksum
1310 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1311 __be32 daddr, __wsum base)
1313 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1316 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1318 return !skb_csum_unnecessary(skb) &&
1319 __skb_checksum_complete(skb);
1322 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1323 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1324 void tcp_set_state(struct sock *sk, int state);
1325 void tcp_done(struct sock *sk);
1326 int tcp_abort(struct sock *sk, int err);
1328 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1331 rx_opt->num_sacks = 0;
1334 u32 tcp_default_init_rwnd(u32 mss);
1335 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1337 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1339 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1340 struct tcp_sock *tp = tcp_sk(sk);
1343 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1344 ca_ops->cong_control)
1346 delta = tcp_jiffies32 - tp->lsndtime;
1347 if (delta > inet_csk(sk)->icsk_rto)
1348 tcp_cwnd_restart(sk, delta);
1351 /* Determine a window scaling and initial window to offer. */
1352 void tcp_select_initial_window(const struct sock *sk, int __space,
1353 __u32 mss, __u32 *rcv_wnd,
1354 __u32 *window_clamp, int wscale_ok,
1355 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1357 static inline int tcp_win_from_space(const struct sock *sk, int space)
1359 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1361 return tcp_adv_win_scale <= 0 ?
1362 (space>>(-tcp_adv_win_scale)) :
1363 space - (space>>tcp_adv_win_scale);
1366 /* Note: caller must be prepared to deal with negative returns */
1367 static inline int tcp_space(const struct sock *sk)
1369 return tcp_win_from_space(sk, sk->sk_rcvbuf - sk->sk_backlog.len -
1370 atomic_read(&sk->sk_rmem_alloc));
1373 static inline int tcp_full_space(const struct sock *sk)
1375 return tcp_win_from_space(sk, sk->sk_rcvbuf);
1378 extern void tcp_openreq_init_rwin(struct request_sock *req,
1379 const struct sock *sk_listener,
1380 const struct dst_entry *dst);
1382 void tcp_enter_memory_pressure(struct sock *sk);
1383 void tcp_leave_memory_pressure(struct sock *sk);
1385 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1387 struct net *net = sock_net((struct sock *)tp);
1389 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1392 static inline int keepalive_time_when(const struct tcp_sock *tp)
1394 struct net *net = sock_net((struct sock *)tp);
1396 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1399 static inline int keepalive_probes(const struct tcp_sock *tp)
1401 struct net *net = sock_net((struct sock *)tp);
1403 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1406 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1408 const struct inet_connection_sock *icsk = &tp->inet_conn;
1410 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1411 tcp_jiffies32 - tp->rcv_tstamp);
1414 static inline int tcp_fin_time(const struct sock *sk)
1416 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1417 const int rto = inet_csk(sk)->icsk_rto;
1419 if (fin_timeout < (rto << 2) - (rto >> 1))
1420 fin_timeout = (rto << 2) - (rto >> 1);
1425 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1428 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1430 if (unlikely(!time_before32(ktime_get_seconds(),
1431 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1434 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1435 * then following tcp messages have valid values. Ignore 0 value,
1436 * or else 'negative' tsval might forbid us to accept their packets.
1438 if (!rx_opt->ts_recent)
1443 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1446 if (tcp_paws_check(rx_opt, 0))
1449 /* RST segments are not recommended to carry timestamp,
1450 and, if they do, it is recommended to ignore PAWS because
1451 "their cleanup function should take precedence over timestamps."
1452 Certainly, it is mistake. It is necessary to understand the reasons
1453 of this constraint to relax it: if peer reboots, clock may go
1454 out-of-sync and half-open connections will not be reset.
1455 Actually, the problem would be not existing if all
1456 the implementations followed draft about maintaining clock
1457 via reboots. Linux-2.2 DOES NOT!
1459 However, we can relax time bounds for RST segments to MSL.
1461 if (rst && !time_before32(ktime_get_seconds(),
1462 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1467 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1468 int mib_idx, u32 *last_oow_ack_time);
1470 static inline void tcp_mib_init(struct net *net)
1473 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1474 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1475 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1476 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1480 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1482 tp->lost_skb_hint = NULL;
1485 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1487 tcp_clear_retrans_hints_partial(tp);
1488 tp->retransmit_skb_hint = NULL;
1491 union tcp_md5_addr {
1493 #if IS_ENABLED(CONFIG_IPV6)
1498 /* - key database */
1499 struct tcp_md5sig_key {
1500 struct hlist_node node;
1502 u8 family; /* AF_INET or AF_INET6 */
1503 union tcp_md5_addr addr;
1505 u8 key[TCP_MD5SIG_MAXKEYLEN];
1506 struct rcu_head rcu;
1510 struct tcp_md5sig_info {
1511 struct hlist_head head;
1512 struct rcu_head rcu;
1515 /* - pseudo header */
1516 struct tcp4_pseudohdr {
1524 struct tcp6_pseudohdr {
1525 struct in6_addr saddr;
1526 struct in6_addr daddr;
1528 __be32 protocol; /* including padding */
1531 union tcp_md5sum_block {
1532 struct tcp4_pseudohdr ip4;
1533 #if IS_ENABLED(CONFIG_IPV6)
1534 struct tcp6_pseudohdr ip6;
1538 /* - pool: digest algorithm, hash description and scratch buffer */
1539 struct tcp_md5sig_pool {
1540 struct ahash_request *md5_req;
1545 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1546 const struct sock *sk, const struct sk_buff *skb);
1547 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1548 int family, u8 prefixlen, const u8 *newkey, u8 newkeylen,
1550 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1551 int family, u8 prefixlen);
1552 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1553 const struct sock *addr_sk);
1555 #ifdef CONFIG_TCP_MD5SIG
1556 #include <linux/jump_label.h>
1557 extern struct static_key_false tcp_md5_needed;
1558 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk,
1559 const union tcp_md5_addr *addr,
1561 static inline struct tcp_md5sig_key *
1562 tcp_md5_do_lookup(const struct sock *sk,
1563 const union tcp_md5_addr *addr,
1566 if (!static_branch_unlikely(&tcp_md5_needed))
1568 return __tcp_md5_do_lookup(sk, addr, family);
1571 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1573 static inline struct tcp_md5sig_key *tcp_md5_do_lookup(const struct sock *sk,
1574 const union tcp_md5_addr *addr,
1579 #define tcp_twsk_md5_key(twsk) NULL
1582 bool tcp_alloc_md5sig_pool(void);
1584 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1585 static inline void tcp_put_md5sig_pool(void)
1590 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1591 unsigned int header_len);
1592 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1593 const struct tcp_md5sig_key *key);
1595 /* From tcp_fastopen.c */
1596 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1597 struct tcp_fastopen_cookie *cookie);
1598 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1599 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1601 struct tcp_fastopen_request {
1602 /* Fast Open cookie. Size 0 means a cookie request */
1603 struct tcp_fastopen_cookie cookie;
1604 struct msghdr *data; /* data in MSG_FASTOPEN */
1606 int copied; /* queued in tcp_connect() */
1607 struct ubuf_info *uarg;
1609 void tcp_free_fastopen_req(struct tcp_sock *tp);
1610 void tcp_fastopen_destroy_cipher(struct sock *sk);
1611 void tcp_fastopen_ctx_destroy(struct net *net);
1612 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1613 void *key, unsigned int len);
1614 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1615 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1616 struct request_sock *req,
1617 struct tcp_fastopen_cookie *foc,
1618 const struct dst_entry *dst);
1619 void tcp_fastopen_init_key_once(struct net *net);
1620 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1621 struct tcp_fastopen_cookie *cookie);
1622 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1623 #define TCP_FASTOPEN_KEY_LENGTH 16
1625 /* Fastopen key context */
1626 struct tcp_fastopen_context {
1627 struct crypto_cipher *tfm;
1628 __u8 key[TCP_FASTOPEN_KEY_LENGTH];
1629 struct rcu_head rcu;
1632 extern unsigned int sysctl_tcp_fastopen_blackhole_timeout;
1633 void tcp_fastopen_active_disable(struct sock *sk);
1634 bool tcp_fastopen_active_should_disable(struct sock *sk);
1635 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1636 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1638 /* Latencies incurred by various limits for a sender. They are
1639 * chronograph-like stats that are mutually exclusive.
1643 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1644 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1645 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1649 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1650 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1652 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1653 * the same memory storage than skb->destructor/_skb_refdst
1655 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1657 skb->destructor = NULL;
1658 skb->_skb_refdst = 0UL;
1661 #define tcp_skb_tsorted_save(skb) { \
1662 unsigned long _save = skb->_skb_refdst; \
1663 skb->_skb_refdst = 0UL;
1665 #define tcp_skb_tsorted_restore(skb) \
1666 skb->_skb_refdst = _save; \
1669 void tcp_write_queue_purge(struct sock *sk);
1671 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1673 return skb_rb_first(&sk->tcp_rtx_queue);
1676 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1678 return skb_peek(&sk->sk_write_queue);
1681 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1683 return skb_peek_tail(&sk->sk_write_queue);
1686 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1687 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1689 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1691 return skb_peek(&sk->sk_write_queue);
1694 static inline bool tcp_skb_is_last(const struct sock *sk,
1695 const struct sk_buff *skb)
1697 return skb_queue_is_last(&sk->sk_write_queue, skb);
1700 static inline bool tcp_write_queue_empty(const struct sock *sk)
1702 return skb_queue_empty(&sk->sk_write_queue);
1705 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1707 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1710 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1712 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1715 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1717 __skb_queue_tail(&sk->sk_write_queue, skb);
1719 /* Queue it, remembering where we must start sending. */
1720 if (sk->sk_write_queue.next == skb)
1721 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1724 /* Insert new before skb on the write queue of sk. */
1725 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1726 struct sk_buff *skb,
1729 __skb_queue_before(&sk->sk_write_queue, skb, new);
1732 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1734 tcp_skb_tsorted_anchor_cleanup(skb);
1735 __skb_unlink(skb, &sk->sk_write_queue);
1738 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1740 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1742 tcp_skb_tsorted_anchor_cleanup(skb);
1743 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1746 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1748 list_del(&skb->tcp_tsorted_anchor);
1749 tcp_rtx_queue_unlink(skb, sk);
1750 sk_wmem_free_skb(sk, skb);
1753 static inline void tcp_push_pending_frames(struct sock *sk)
1755 if (tcp_send_head(sk)) {
1756 struct tcp_sock *tp = tcp_sk(sk);
1758 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1762 /* Start sequence of the skb just after the highest skb with SACKed
1763 * bit, valid only if sacked_out > 0 or when the caller has ensured
1764 * validity by itself.
1766 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1768 if (!tp->sacked_out)
1771 if (tp->highest_sack == NULL)
1774 return TCP_SKB_CB(tp->highest_sack)->seq;
1777 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1779 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1782 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1784 return tcp_sk(sk)->highest_sack;
1787 static inline void tcp_highest_sack_reset(struct sock *sk)
1789 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1792 /* Called when old skb is about to be deleted and replaced by new skb */
1793 static inline void tcp_highest_sack_replace(struct sock *sk,
1794 struct sk_buff *old,
1795 struct sk_buff *new)
1797 if (old == tcp_highest_sack(sk))
1798 tcp_sk(sk)->highest_sack = new;
1801 /* This helper checks if socket has IP_TRANSPARENT set */
1802 static inline bool inet_sk_transparent(const struct sock *sk)
1804 switch (sk->sk_state) {
1806 return inet_twsk(sk)->tw_transparent;
1807 case TCP_NEW_SYN_RECV:
1808 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1810 return inet_sk(sk)->transparent;
1813 /* Determines whether this is a thin stream (which may suffer from
1814 * increased latency). Used to trigger latency-reducing mechanisms.
1816 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1818 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1822 enum tcp_seq_states {
1823 TCP_SEQ_STATE_LISTENING,
1824 TCP_SEQ_STATE_ESTABLISHED,
1827 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1828 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1829 void tcp_seq_stop(struct seq_file *seq, void *v);
1831 struct tcp_seq_afinfo {
1835 struct tcp_iter_state {
1836 struct seq_net_private p;
1837 enum tcp_seq_states state;
1838 struct sock *syn_wait_sk;
1839 int bucket, offset, sbucket, num;
1843 extern struct request_sock_ops tcp_request_sock_ops;
1844 extern struct request_sock_ops tcp6_request_sock_ops;
1846 void tcp_v4_destroy_sock(struct sock *sk);
1848 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1849 netdev_features_t features);
1850 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
1851 int tcp_gro_complete(struct sk_buff *skb);
1853 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1855 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
1857 struct net *net = sock_net((struct sock *)tp);
1858 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
1861 /* @wake is one when sk_stream_write_space() calls us.
1862 * This sends EPOLLOUT only if notsent_bytes is half the limit.
1863 * This mimics the strategy used in sock_def_write_space().
1865 static inline bool tcp_stream_memory_free(const struct sock *sk, int wake)
1867 const struct tcp_sock *tp = tcp_sk(sk);
1868 u32 notsent_bytes = tp->write_seq - tp->snd_nxt;
1870 return (notsent_bytes << wake) < tcp_notsent_lowat(tp);
1873 #ifdef CONFIG_PROC_FS
1874 int tcp4_proc_init(void);
1875 void tcp4_proc_exit(void);
1878 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
1879 int tcp_conn_request(struct request_sock_ops *rsk_ops,
1880 const struct tcp_request_sock_ops *af_ops,
1881 struct sock *sk, struct sk_buff *skb);
1883 /* TCP af-specific functions */
1884 struct tcp_sock_af_ops {
1885 #ifdef CONFIG_TCP_MD5SIG
1886 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
1887 const struct sock *addr_sk);
1888 int (*calc_md5_hash)(char *location,
1889 const struct tcp_md5sig_key *md5,
1890 const struct sock *sk,
1891 const struct sk_buff *skb);
1892 int (*md5_parse)(struct sock *sk,
1894 char __user *optval,
1899 struct tcp_request_sock_ops {
1901 #ifdef CONFIG_TCP_MD5SIG
1902 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
1903 const struct sock *addr_sk);
1904 int (*calc_md5_hash) (char *location,
1905 const struct tcp_md5sig_key *md5,
1906 const struct sock *sk,
1907 const struct sk_buff *skb);
1909 void (*init_req)(struct request_sock *req,
1910 const struct sock *sk_listener,
1911 struct sk_buff *skb);
1912 #ifdef CONFIG_SYN_COOKIES
1913 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
1916 struct dst_entry *(*route_req)(const struct sock *sk, struct flowi *fl,
1917 const struct request_sock *req);
1918 u32 (*init_seq)(const struct sk_buff *skb);
1919 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
1920 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
1921 struct flowi *fl, struct request_sock *req,
1922 struct tcp_fastopen_cookie *foc,
1923 enum tcp_synack_type synack_type);
1926 #ifdef CONFIG_SYN_COOKIES
1927 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1928 const struct sock *sk, struct sk_buff *skb,
1931 tcp_synq_overflow(sk);
1932 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
1933 return ops->cookie_init_seq(skb, mss);
1936 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
1937 const struct sock *sk, struct sk_buff *skb,
1944 int tcpv4_offload_init(void);
1946 void tcp_v4_init(void);
1947 void tcp_init(void);
1949 /* tcp_recovery.c */
1950 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
1951 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
1952 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
1954 extern void tcp_rack_mark_lost(struct sock *sk);
1955 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
1957 extern void tcp_rack_reo_timeout(struct sock *sk);
1958 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
1960 /* At how many usecs into the future should the RTO fire? */
1961 static inline s64 tcp_rto_delta_us(const struct sock *sk)
1963 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
1964 u32 rto = inet_csk(sk)->icsk_rto;
1965 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
1967 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
1971 * Save and compile IPv4 options, return a pointer to it
1973 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
1974 struct sk_buff *skb)
1976 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
1977 struct ip_options_rcu *dopt = NULL;
1980 int opt_size = sizeof(*dopt) + opt->optlen;
1982 dopt = kmalloc(opt_size, GFP_ATOMIC);
1983 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
1991 /* locally generated TCP pure ACKs have skb->truesize == 2
1992 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
1993 * This is much faster than dissecting the packet to find out.
1994 * (Think of GRE encapsulations, IPv4, IPv6, ...)
1996 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
1998 return skb->truesize == 2;
2001 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2006 static inline int tcp_inq(struct sock *sk)
2008 struct tcp_sock *tp = tcp_sk(sk);
2011 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2013 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2015 before(tp->urg_seq, tp->copied_seq) ||
2016 !before(tp->urg_seq, tp->rcv_nxt)) {
2018 answ = tp->rcv_nxt - tp->copied_seq;
2020 /* Subtract 1, if FIN was received */
2021 if (answ && sock_flag(sk, SOCK_DONE))
2024 answ = tp->urg_seq - tp->copied_seq;
2030 int tcp_peek_len(struct socket *sock);
2032 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2036 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2037 tp->segs_in += segs_in;
2038 if (skb->len > tcp_hdrlen(skb))
2039 tp->data_segs_in += segs_in;
2043 * TCP listen path runs lockless.
2044 * We forced "struct sock" to be const qualified to make sure
2045 * we don't modify one of its field by mistake.
2046 * Here, we increment sk_drops which is an atomic_t, so we can safely
2047 * make sock writable again.
2049 static inline void tcp_listendrop(const struct sock *sk)
2051 atomic_inc(&((struct sock *)sk)->sk_drops);
2052 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2055 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2058 * Interface for adding Upper Level Protocols over TCP
2061 #define TCP_ULP_NAME_MAX 16
2062 #define TCP_ULP_MAX 128
2063 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2065 struct tcp_ulp_ops {
2066 struct list_head list;
2068 /* initialize ulp */
2069 int (*init)(struct sock *sk);
2071 void (*release)(struct sock *sk);
2073 char name[TCP_ULP_NAME_MAX];
2074 struct module *owner;
2076 int tcp_register_ulp(struct tcp_ulp_ops *type);
2077 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2078 int tcp_set_ulp(struct sock *sk, const char *name);
2079 void tcp_get_available_ulp(char *buf, size_t len);
2080 void tcp_cleanup_ulp(struct sock *sk);
2082 #define MODULE_ALIAS_TCP_ULP(name) \
2083 __MODULE_INFO(alias, alias_userspace, name); \
2084 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2089 int tcp_bpf_init(struct sock *sk);
2090 void tcp_bpf_reinit(struct sock *sk);
2091 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes,
2093 int tcp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
2094 int nonblock, int flags, int *addr_len);
2095 int __tcp_bpf_recvmsg(struct sock *sk, struct sk_psock *psock,
2096 struct msghdr *msg, int len, int flags);
2098 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2099 * is < 0, then the BPF op failed (for example if the loaded BPF
2100 * program does not support the chosen operation or there is no BPF
2104 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2106 struct bpf_sock_ops_kern sock_ops;
2109 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2110 if (sk_fullsock(sk)) {
2111 sock_ops.is_fullsock = 1;
2112 sock_owned_by_me(sk);
2118 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2120 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2122 ret = sock_ops.reply;
2128 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2130 u32 args[2] = {arg1, arg2};
2132 return tcp_call_bpf(sk, op, 2, args);
2135 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2138 u32 args[3] = {arg1, arg2, arg3};
2140 return tcp_call_bpf(sk, op, 3, args);
2144 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2149 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2154 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2162 static inline u32 tcp_timeout_init(struct sock *sk)
2166 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2169 timeout = TCP_TIMEOUT_INIT;
2173 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2177 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2184 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2186 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2189 #if IS_ENABLED(CONFIG_SMC)
2190 extern struct static_key_false tcp_have_smc;
2193 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2194 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2195 void (*cad)(struct sock *sk, u32 ack_seq));
2196 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2197 void clean_acked_data_flush(void);