1 /* SPDX-License-Identifier: GPL-2.0-or-later */
3 * INET An implementation of the TCP/IP protocol suite for the LINUX
4 * operating system. INET is implemented using the BSD Socket
5 * interface as the means of communication with the user level.
7 * Definitions for the TCP module.
9 * Version: @(#)tcp.h 1.0.5 05/23/93
12 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
17 #define FASTRETRANS_DEBUG 1
19 #include <linux/list.h>
20 #include <linux/tcp.h>
21 #include <linux/bug.h>
22 #include <linux/slab.h>
23 #include <linux/cache.h>
24 #include <linux/percpu.h>
25 #include <linux/skbuff.h>
26 #include <linux/kref.h>
27 #include <linux/ktime.h>
28 #include <linux/indirect_call_wrapper.h>
30 #include <net/inet_connection_sock.h>
31 #include <net/inet_timewait_sock.h>
32 #include <net/inet_hashtables.h>
33 #include <net/checksum.h>
34 #include <net/request_sock.h>
35 #include <net/sock_reuseport.h>
39 #include <net/tcp_states.h>
40 #include <net/inet_ecn.h>
42 #include <net/mptcp.h>
44 #include <linux/seq_file.h>
45 #include <linux/memcontrol.h>
46 #include <linux/bpf-cgroup.h>
47 #include <linux/siphash.h>
49 extern struct inet_hashinfo tcp_hashinfo;
51 DECLARE_PER_CPU(unsigned int, tcp_orphan_count);
52 int tcp_orphan_count_sum(void);
54 void tcp_time_wait(struct sock *sk, int state, int timeo);
56 #define MAX_TCP_HEADER L1_CACHE_ALIGN(128 + MAX_HEADER)
57 #define MAX_TCP_OPTION_SPACE 40
58 #define TCP_MIN_SND_MSS 48
59 #define TCP_MIN_GSO_SIZE (TCP_MIN_SND_MSS - MAX_TCP_OPTION_SPACE)
62 * Never offer a window over 32767 without using window scaling. Some
63 * poor stacks do signed 16bit maths!
65 #define MAX_TCP_WINDOW 32767U
67 /* Minimal accepted MSS. It is (60+60+8) - (20+20). */
68 #define TCP_MIN_MSS 88U
70 /* The initial MTU to use for probing */
71 #define TCP_BASE_MSS 1024
73 /* probing interval, default to 10 minutes as per RFC4821 */
74 #define TCP_PROBE_INTERVAL 600
76 /* Specify interval when tcp mtu probing will stop */
77 #define TCP_PROBE_THRESHOLD 8
79 /* After receiving this amount of duplicate ACKs fast retransmit starts. */
80 #define TCP_FASTRETRANS_THRESH 3
82 /* Maximal number of ACKs sent quickly to accelerate slow-start. */
83 #define TCP_MAX_QUICKACKS 16U
85 /* Maximal number of window scale according to RFC1323 */
86 #define TCP_MAX_WSCALE 14U
89 #define TCP_URG_VALID 0x0100
90 #define TCP_URG_NOTYET 0x0200
91 #define TCP_URG_READ 0x0400
93 #define TCP_RETR1 3 /*
94 * This is how many retries it does before it
95 * tries to figure out if the gateway is
96 * down. Minimal RFC value is 3; it corresponds
97 * to ~3sec-8min depending on RTO.
100 #define TCP_RETR2 15 /*
101 * This should take at least
102 * 90 minutes to time out.
103 * RFC1122 says that the limit is 100 sec.
104 * 15 is ~13-30min depending on RTO.
107 #define TCP_SYN_RETRIES 6 /* This is how many retries are done
108 * when active opening a connection.
109 * RFC1122 says the minimum retry MUST
110 * be at least 180secs. Nevertheless
111 * this value is corresponding to
112 * 63secs of retransmission with the
113 * current initial RTO.
116 #define TCP_SYNACK_RETRIES 5 /* This is how may retries are done
117 * when passive opening a connection.
118 * This is corresponding to 31secs of
119 * retransmission with the current
123 #define TCP_TIMEWAIT_LEN (60*HZ) /* how long to wait to destroy TIME-WAIT
124 * state, about 60 seconds */
125 #define TCP_FIN_TIMEOUT TCP_TIMEWAIT_LEN
126 /* BSD style FIN_WAIT2 deadlock breaker.
127 * It used to be 3min, new value is 60sec,
128 * to combine FIN-WAIT-2 timeout with
131 #define TCP_FIN_TIMEOUT_MAX (120 * HZ) /* max TCP_LINGER2 value (two minutes) */
133 #define TCP_DELACK_MAX ((unsigned)(HZ/5)) /* maximal time to delay before sending an ACK */
135 #define TCP_DELACK_MIN ((unsigned)(HZ/25)) /* minimal time to delay before sending an ACK */
136 #define TCP_ATO_MIN ((unsigned)(HZ/25))
138 #define TCP_DELACK_MIN 4U
139 #define TCP_ATO_MIN 4U
141 #define TCP_RTO_MAX ((unsigned)(120*HZ))
142 #define TCP_RTO_MIN ((unsigned)(HZ/5))
143 #define TCP_TIMEOUT_MIN (2U) /* Min timeout for TCP timers in jiffies */
144 #define TCP_TIMEOUT_INIT ((unsigned)(1*HZ)) /* RFC6298 2.1 initial RTO value */
145 #define TCP_TIMEOUT_FALLBACK ((unsigned)(3*HZ)) /* RFC 1122 initial RTO value, now
146 * used as a fallback RTO for the
147 * initial data transmission if no
148 * valid RTT sample has been acquired,
149 * most likely due to retrans in 3WHS.
152 #define TCP_RESOURCE_PROBE_INTERVAL ((unsigned)(HZ/2U)) /* Maximal interval between probes
153 * for local resources.
155 #define TCP_KEEPALIVE_TIME (120*60*HZ) /* two hours */
156 #define TCP_KEEPALIVE_PROBES 9 /* Max of 9 keepalive probes */
157 #define TCP_KEEPALIVE_INTVL (75*HZ)
159 #define MAX_TCP_KEEPIDLE 32767
160 #define MAX_TCP_KEEPINTVL 32767
161 #define MAX_TCP_KEEPCNT 127
162 #define MAX_TCP_SYNCNT 127
164 #define TCP_SYNQ_INTERVAL (HZ/5) /* Period of SYNACK timer */
166 #define TCP_PAWS_24DAYS (60 * 60 * 24 * 24)
167 #define TCP_PAWS_MSL 60 /* Per-host timestamps are invalidated
168 * after this time. It should be equal
169 * (or greater than) TCP_TIMEWAIT_LEN
170 * to provide reliability equal to one
171 * provided by timewait state.
173 #define TCP_PAWS_WINDOW 1 /* Replay window for per-host
174 * timestamps. It must be less than
175 * minimal timewait lifetime.
181 #define TCPOPT_NOP 1 /* Padding */
182 #define TCPOPT_EOL 0 /* End of options */
183 #define TCPOPT_MSS 2 /* Segment size negotiating */
184 #define TCPOPT_WINDOW 3 /* Window scaling */
185 #define TCPOPT_SACK_PERM 4 /* SACK Permitted */
186 #define TCPOPT_SACK 5 /* SACK Block */
187 #define TCPOPT_TIMESTAMP 8 /* Better RTT estimations/PAWS */
188 #define TCPOPT_MD5SIG 19 /* MD5 Signature (RFC2385) */
189 #define TCPOPT_MPTCP 30 /* Multipath TCP (RFC6824) */
190 #define TCPOPT_FASTOPEN 34 /* Fast open (RFC7413) */
191 #define TCPOPT_EXP 254 /* Experimental */
192 /* Magic number to be after the option value for sharing TCP
193 * experimental options. See draft-ietf-tcpm-experimental-options-00.txt
195 #define TCPOPT_FASTOPEN_MAGIC 0xF989
196 #define TCPOPT_SMC_MAGIC 0xE2D4C3D9
202 #define TCPOLEN_MSS 4
203 #define TCPOLEN_WINDOW 3
204 #define TCPOLEN_SACK_PERM 2
205 #define TCPOLEN_TIMESTAMP 10
206 #define TCPOLEN_MD5SIG 18
207 #define TCPOLEN_FASTOPEN_BASE 2
208 #define TCPOLEN_EXP_FASTOPEN_BASE 4
209 #define TCPOLEN_EXP_SMC_BASE 6
211 /* But this is what stacks really send out. */
212 #define TCPOLEN_TSTAMP_ALIGNED 12
213 #define TCPOLEN_WSCALE_ALIGNED 4
214 #define TCPOLEN_SACKPERM_ALIGNED 4
215 #define TCPOLEN_SACK_BASE 2
216 #define TCPOLEN_SACK_BASE_ALIGNED 4
217 #define TCPOLEN_SACK_PERBLOCK 8
218 #define TCPOLEN_MD5SIG_ALIGNED 20
219 #define TCPOLEN_MSS_ALIGNED 4
220 #define TCPOLEN_EXP_SMC_BASE_ALIGNED 8
222 /* Flags in tp->nonagle */
223 #define TCP_NAGLE_OFF 1 /* Nagle's algo is disabled */
224 #define TCP_NAGLE_CORK 2 /* Socket is corked */
225 #define TCP_NAGLE_PUSH 4 /* Cork is overridden for already queued data */
227 /* TCP thin-stream limits */
228 #define TCP_THIN_LINEAR_RETRIES 6 /* After 6 linear retries, do exp. backoff */
230 /* TCP initial congestion window as per rfc6928 */
231 #define TCP_INIT_CWND 10
233 /* Bit Flags for sysctl_tcp_fastopen */
234 #define TFO_CLIENT_ENABLE 1
235 #define TFO_SERVER_ENABLE 2
236 #define TFO_CLIENT_NO_COOKIE 4 /* Data in SYN w/o cookie option */
238 /* Accept SYN data w/o any cookie option */
239 #define TFO_SERVER_COOKIE_NOT_REQD 0x200
241 /* Force enable TFO on all listeners, i.e., not requiring the
242 * TCP_FASTOPEN socket option.
244 #define TFO_SERVER_WO_SOCKOPT1 0x400
247 /* sysctl variables for tcp */
248 extern int sysctl_tcp_max_orphans;
249 extern long sysctl_tcp_mem[3];
251 #define TCP_RACK_LOSS_DETECTION 0x1 /* Use RACK to detect losses */
252 #define TCP_RACK_STATIC_REO_WND 0x2 /* Use static RACK reo wnd */
253 #define TCP_RACK_NO_DUPTHRESH 0x4 /* Do not use DUPACK threshold in RACK */
255 extern atomic_long_t tcp_memory_allocated;
256 DECLARE_PER_CPU(int, tcp_memory_per_cpu_fw_alloc);
258 extern struct percpu_counter tcp_sockets_allocated;
259 extern unsigned long tcp_memory_pressure;
261 /* optimized version of sk_under_memory_pressure() for TCP sockets */
262 static inline bool tcp_under_memory_pressure(const struct sock *sk)
264 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
265 mem_cgroup_under_socket_pressure(sk->sk_memcg))
268 return READ_ONCE(tcp_memory_pressure);
271 * The next routines deal with comparing 32 bit unsigned ints
272 * and worry about wraparound (automatic with unsigned arithmetic).
275 static inline bool before(__u32 seq1, __u32 seq2)
277 return (__s32)(seq1-seq2) < 0;
279 #define after(seq2, seq1) before(seq1, seq2)
281 /* is s2<=s1<=s3 ? */
282 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
284 return seq3 - seq2 >= seq1 - seq2;
287 static inline bool tcp_out_of_memory(struct sock *sk)
289 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
290 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
295 static inline void tcp_wmem_free_skb(struct sock *sk, struct sk_buff *skb)
297 sk_wmem_queued_add(sk, -skb->truesize);
298 if (!skb_zcopy_pure(skb))
299 sk_mem_uncharge(sk, skb->truesize);
301 sk_mem_uncharge(sk, SKB_TRUESIZE(skb_end_offset(skb)));
305 void sk_forced_mem_schedule(struct sock *sk, int size);
307 bool tcp_check_oom(struct sock *sk, int shift);
310 extern struct proto tcp_prot;
312 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
313 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
314 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
315 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
317 void tcp_tasklet_init(void);
319 int tcp_v4_err(struct sk_buff *skb, u32);
321 void tcp_shutdown(struct sock *sk, int how);
323 int tcp_v4_early_demux(struct sk_buff *skb);
324 int tcp_v4_rcv(struct sk_buff *skb);
326 void tcp_remove_empty_skb(struct sock *sk);
327 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
328 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
329 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
330 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
332 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
333 size_t size, int flags);
334 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
335 size_t size, int flags);
336 int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
337 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
339 void tcp_release_cb(struct sock *sk);
340 void tcp_wfree(struct sk_buff *skb);
341 void tcp_write_timer_handler(struct sock *sk);
342 void tcp_delack_timer_handler(struct sock *sk);
343 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
344 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
345 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
346 void tcp_rcv_space_adjust(struct sock *sk);
347 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
348 void tcp_twsk_destructor(struct sock *sk);
349 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
350 struct pipe_inode_info *pipe, size_t len,
352 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
353 bool force_schedule);
355 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
356 static inline void tcp_dec_quickack_mode(struct sock *sk,
357 const unsigned int pkts)
359 struct inet_connection_sock *icsk = inet_csk(sk);
361 if (icsk->icsk_ack.quick) {
362 if (pkts >= icsk->icsk_ack.quick) {
363 icsk->icsk_ack.quick = 0;
364 /* Leaving quickack mode we deflate ATO. */
365 icsk->icsk_ack.ato = TCP_ATO_MIN;
367 icsk->icsk_ack.quick -= pkts;
372 #define TCP_ECN_QUEUE_CWR 2
373 #define TCP_ECN_DEMAND_CWR 4
374 #define TCP_ECN_SEEN 8
384 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
386 const struct tcphdr *th);
387 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
388 struct request_sock *req, bool fastopen,
390 int tcp_child_process(struct sock *parent, struct sock *child,
391 struct sk_buff *skb);
392 void tcp_enter_loss(struct sock *sk);
393 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag);
394 void tcp_clear_retrans(struct tcp_sock *tp);
395 void tcp_update_metrics(struct sock *sk);
396 void tcp_init_metrics(struct sock *sk);
397 void tcp_metrics_init(void);
398 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
399 void __tcp_close(struct sock *sk, long timeout);
400 void tcp_close(struct sock *sk, long timeout);
401 void tcp_init_sock(struct sock *sk);
402 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb);
403 __poll_t tcp_poll(struct file *file, struct socket *sock,
404 struct poll_table_struct *wait);
405 int tcp_getsockopt(struct sock *sk, int level, int optname,
406 char __user *optval, int __user *optlen);
407 bool tcp_bpf_bypass_getsockopt(int level, int optname);
408 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
409 unsigned int optlen);
410 void tcp_set_keepalive(struct sock *sk, int val);
411 void tcp_syn_ack_timeout(const struct request_sock *req);
412 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
413 int flags, int *addr_len);
414 int tcp_set_rcvlowat(struct sock *sk, int val);
415 int tcp_set_window_clamp(struct sock *sk, int val);
416 void tcp_update_recv_tstamps(struct sk_buff *skb,
417 struct scm_timestamping_internal *tss);
418 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
419 struct scm_timestamping_internal *tss);
420 void tcp_data_ready(struct sock *sk);
422 int tcp_mmap(struct file *file, struct socket *sock,
423 struct vm_area_struct *vma);
425 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
426 struct tcp_options_received *opt_rx,
427 int estab, struct tcp_fastopen_cookie *foc);
428 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
431 * BPF SKB-less helpers
433 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
434 struct tcphdr *th, u32 *cookie);
435 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
436 struct tcphdr *th, u32 *cookie);
437 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss);
438 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
439 const struct tcp_request_sock_ops *af_ops,
440 struct sock *sk, struct tcphdr *th);
442 * TCP v4 functions exported for the inet6 API
445 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
446 void tcp_v4_mtu_reduced(struct sock *sk);
447 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
448 void tcp_ld_RTO_revert(struct sock *sk, u32 seq);
449 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
450 struct sock *tcp_create_openreq_child(const struct sock *sk,
451 struct request_sock *req,
452 struct sk_buff *skb);
453 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
454 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
455 struct request_sock *req,
456 struct dst_entry *dst,
457 struct request_sock *req_unhash,
459 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
460 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
461 int tcp_connect(struct sock *sk);
462 enum tcp_synack_type {
467 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
468 struct request_sock *req,
469 struct tcp_fastopen_cookie *foc,
470 enum tcp_synack_type synack_type,
471 struct sk_buff *syn_skb);
472 int tcp_disconnect(struct sock *sk, int flags);
474 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
475 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
476 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
478 /* From syncookies.c */
479 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
480 struct request_sock *req,
481 struct dst_entry *dst, u32 tsoff);
482 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
484 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
485 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
486 const struct tcp_request_sock_ops *af_ops,
487 struct sock *sk, struct sk_buff *skb);
488 #ifdef CONFIG_SYN_COOKIES
490 /* Syncookies use a monotonic timer which increments every 60 seconds.
491 * This counter is used both as a hash input and partially encoded into
492 * the cookie value. A cookie is only validated further if the delta
493 * between the current counter value and the encoded one is less than this,
494 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
495 * the counter advances immediately after a cookie is generated).
497 #define MAX_SYNCOOKIE_AGE 2
498 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
499 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
501 /* syncookies: remember time of last synqueue overflow
502 * But do not dirty this field too often (once per second is enough)
503 * It is racy as we do not hold a lock, but race is very minor.
505 static inline void tcp_synq_overflow(const struct sock *sk)
507 unsigned int last_overflow;
508 unsigned int now = jiffies;
510 if (sk->sk_reuseport) {
511 struct sock_reuseport *reuse;
513 reuse = rcu_dereference(sk->sk_reuseport_cb);
515 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
516 if (!time_between32(now, last_overflow,
518 WRITE_ONCE(reuse->synq_overflow_ts, now);
523 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
524 if (!time_between32(now, last_overflow, last_overflow + HZ))
525 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
528 /* syncookies: no recent synqueue overflow on this listening socket? */
529 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
531 unsigned int last_overflow;
532 unsigned int now = jiffies;
534 if (sk->sk_reuseport) {
535 struct sock_reuseport *reuse;
537 reuse = rcu_dereference(sk->sk_reuseport_cb);
539 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
540 return !time_between32(now, last_overflow - HZ,
542 TCP_SYNCOOKIE_VALID);
546 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
548 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
549 * then we're under synflood. However, we have to use
550 * 'last_overflow - HZ' as lower bound. That's because a concurrent
551 * tcp_synq_overflow() could update .ts_recent_stamp after we read
552 * jiffies but before we store .ts_recent_stamp into last_overflow,
553 * which could lead to rejecting a valid syncookie.
555 return !time_between32(now, last_overflow - HZ,
556 last_overflow + TCP_SYNCOOKIE_VALID);
559 static inline u32 tcp_cookie_time(void)
561 u64 val = get_jiffies_64();
563 do_div(val, TCP_SYNCOOKIE_PERIOD);
567 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
569 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
570 u64 cookie_init_timestamp(struct request_sock *req, u64 now);
571 bool cookie_timestamp_decode(const struct net *net,
572 struct tcp_options_received *opt);
573 bool cookie_ecn_ok(const struct tcp_options_received *opt,
574 const struct net *net, const struct dst_entry *dst);
576 /* From net/ipv6/syncookies.c */
577 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
579 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
581 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
582 const struct tcphdr *th, u16 *mssp);
583 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
587 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb);
588 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb);
589 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
591 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
592 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
593 void tcp_retransmit_timer(struct sock *sk);
594 void tcp_xmit_retransmit_queue(struct sock *);
595 void tcp_simple_retransmit(struct sock *);
596 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
597 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
599 TCP_FRAG_IN_WRITE_QUEUE,
600 TCP_FRAG_IN_RTX_QUEUE,
602 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
603 struct sk_buff *skb, u32 len,
604 unsigned int mss_now, gfp_t gfp);
606 void tcp_send_probe0(struct sock *);
607 void tcp_send_partial(struct sock *);
608 int tcp_write_wakeup(struct sock *, int mib);
609 void tcp_send_fin(struct sock *sk);
610 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
611 int tcp_send_synack(struct sock *);
612 void tcp_push_one(struct sock *, unsigned int mss_now);
613 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
614 void tcp_send_ack(struct sock *sk);
615 void tcp_send_delayed_ack(struct sock *sk);
616 void tcp_send_loss_probe(struct sock *sk);
617 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
618 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
619 const struct sk_buff *next_skb);
622 void tcp_rearm_rto(struct sock *sk);
623 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
624 void tcp_reset(struct sock *sk, struct sk_buff *skb);
625 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
626 void tcp_fin(struct sock *sk);
627 void tcp_check_space(struct sock *sk);
630 void tcp_init_xmit_timers(struct sock *);
631 static inline void tcp_clear_xmit_timers(struct sock *sk)
633 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
636 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
639 inet_csk_clear_xmit_timers(sk);
642 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
643 unsigned int tcp_current_mss(struct sock *sk);
644 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when);
646 /* Bound MSS / TSO packet size with the half of the window */
647 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
651 /* When peer uses tiny windows, there is no use in packetizing
652 * to sub-MSS pieces for the sake of SWS or making sure there
653 * are enough packets in the pipe for fast recovery.
655 * On the other hand, for extremely large MSS devices, handling
656 * smaller than MSS windows in this way does make sense.
658 if (tp->max_window > TCP_MSS_DEFAULT)
659 cutoff = (tp->max_window >> 1);
661 cutoff = tp->max_window;
663 if (cutoff && pktsize > cutoff)
664 return max_t(int, cutoff, 68U - tp->tcp_header_len);
670 void tcp_get_info(struct sock *, struct tcp_info *);
672 /* Read 'sendfile()'-style from a TCP socket */
673 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
674 sk_read_actor_t recv_actor);
676 void tcp_initialize_rcv_mss(struct sock *sk);
678 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
679 int tcp_mss_to_mtu(struct sock *sk, int mss);
680 void tcp_mtup_init(struct sock *sk);
682 static inline void tcp_bound_rto(const struct sock *sk)
684 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
685 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
688 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
690 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
693 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
695 /* mptcp hooks are only on the slow path */
696 if (sk_is_mptcp((struct sock *)tp))
699 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
700 ntohl(TCP_FLAG_ACK) |
704 static inline void tcp_fast_path_on(struct tcp_sock *tp)
706 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
709 static inline void tcp_fast_path_check(struct sock *sk)
711 struct tcp_sock *tp = tcp_sk(sk);
713 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
715 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
717 tcp_fast_path_on(tp);
720 /* Compute the actual rto_min value */
721 static inline u32 tcp_rto_min(struct sock *sk)
723 const struct dst_entry *dst = __sk_dst_get(sk);
724 u32 rto_min = inet_csk(sk)->icsk_rto_min;
726 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
727 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
731 static inline u32 tcp_rto_min_us(struct sock *sk)
733 return jiffies_to_usecs(tcp_rto_min(sk));
736 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
738 return dst_metric_locked(dst, RTAX_CC_ALGO);
741 /* Minimum RTT in usec. ~0 means not available. */
742 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
744 return minmax_get(&tp->rtt_min);
747 /* Compute the actual receive window we are currently advertising.
748 * Rcv_nxt can be after the window if our peer push more data
749 * than the offered window.
751 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
753 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
760 /* Choose a new window, without checks for shrinking, and without
761 * scaling applied to the result. The caller does these things
762 * if necessary. This is a "raw" window selection.
764 u32 __tcp_select_window(struct sock *sk);
766 void tcp_send_window_probe(struct sock *sk);
768 /* TCP uses 32bit jiffies to save some space.
769 * Note that this is different from tcp_time_stamp, which
770 * historically has been the same until linux-4.13.
772 #define tcp_jiffies32 ((u32)jiffies)
775 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
776 * It is no longer tied to jiffies, but to 1 ms clock.
777 * Note: double check if you want to use tcp_jiffies32 instead of this.
779 #define TCP_TS_HZ 1000
781 static inline u64 tcp_clock_ns(void)
783 return ktime_get_ns();
786 static inline u64 tcp_clock_us(void)
788 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
791 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
792 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
794 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
797 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
798 static inline u32 tcp_ns_to_ts(u64 ns)
800 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
803 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
804 static inline u32 tcp_time_stamp_raw(void)
806 return tcp_ns_to_ts(tcp_clock_ns());
809 void tcp_mstamp_refresh(struct tcp_sock *tp);
811 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
813 return max_t(s64, t1 - t0, 0);
816 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
818 return tcp_ns_to_ts(skb->skb_mstamp_ns);
821 /* provide the departure time in us unit */
822 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
824 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
828 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
830 #define TCPHDR_FIN 0x01
831 #define TCPHDR_SYN 0x02
832 #define TCPHDR_RST 0x04
833 #define TCPHDR_PSH 0x08
834 #define TCPHDR_ACK 0x10
835 #define TCPHDR_URG 0x20
836 #define TCPHDR_ECE 0x40
837 #define TCPHDR_CWR 0x80
839 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
841 /* This is what the send packet queuing engine uses to pass
842 * TCP per-packet control information to the transmission code.
843 * We also store the host-order sequence numbers in here too.
844 * This is 44 bytes if IPV6 is enabled.
845 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
848 __u32 seq; /* Starting sequence number */
849 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
851 /* Note : tcp_tw_isn is used in input path only
852 * (isn chosen by tcp_timewait_state_process())
854 * tcp_gso_segs/size are used in write queue only,
855 * cf tcp_skb_pcount()/tcp_skb_mss()
863 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
865 __u8 sacked; /* State flags for SACK. */
866 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
867 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
868 #define TCPCB_LOST 0x04 /* SKB is lost */
869 #define TCPCB_TAGBITS 0x07 /* All tag bits */
870 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
871 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
872 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
875 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
876 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
877 eor:1, /* Is skb MSG_EOR marked? */
878 has_rxtstamp:1, /* SKB has a RX timestamp */
880 __u32 ack_seq; /* Sequence number ACK'd */
883 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1)
884 /* There is space for up to 24 bytes */
885 __u32 is_app_limited:1, /* cwnd not fully used? */
888 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
890 /* start of send pipeline phase */
892 /* when we reached the "delivered" count */
893 u64 delivered_mstamp;
894 } tx; /* only used for outgoing skbs */
896 struct inet_skb_parm h4;
897 #if IS_ENABLED(CONFIG_IPV6)
898 struct inet6_skb_parm h6;
900 } header; /* For incoming skbs */
904 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
906 extern const struct inet_connection_sock_af_ops ipv4_specific;
908 #if IS_ENABLED(CONFIG_IPV6)
909 /* This is the variant of inet6_iif() that must be used by TCP,
910 * as TCP moves IP6CB into a different location in skb->cb[]
912 static inline int tcp_v6_iif(const struct sk_buff *skb)
914 return TCP_SKB_CB(skb)->header.h6.iif;
917 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
919 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
921 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
924 /* TCP_SKB_CB reference means this can not be used from early demux */
925 static inline int tcp_v6_sdif(const struct sk_buff *skb)
927 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
928 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
929 return TCP_SKB_CB(skb)->header.h6.iif;
934 extern const struct inet_connection_sock_af_ops ipv6_specific;
936 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
937 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
938 INDIRECT_CALLABLE_DECLARE(void tcp_v6_early_demux(struct sk_buff *skb));
942 /* TCP_SKB_CB reference means this can not be used from early demux */
943 static inline int tcp_v4_sdif(struct sk_buff *skb)
945 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
946 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
947 return TCP_SKB_CB(skb)->header.h4.iif;
952 /* Due to TSO, an SKB can be composed of multiple actual
953 * packets. To keep these tracked properly, we use this.
955 static inline int tcp_skb_pcount(const struct sk_buff *skb)
957 return TCP_SKB_CB(skb)->tcp_gso_segs;
960 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
962 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
965 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
967 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
970 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
971 static inline int tcp_skb_mss(const struct sk_buff *skb)
973 return TCP_SKB_CB(skb)->tcp_gso_size;
976 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
978 return likely(!TCP_SKB_CB(skb)->eor);
981 static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
982 const struct sk_buff *from)
984 return likely(tcp_skb_can_collapse_to(to) &&
985 mptcp_skb_can_collapse(to, from) &&
986 skb_pure_zcopy_same(to, from));
989 /* Events passed to congestion control interface */
991 CA_EVENT_TX_START, /* first transmit when no packets in flight */
992 CA_EVENT_CWND_RESTART, /* congestion window restart */
993 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
994 CA_EVENT_LOSS, /* loss timeout */
995 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
996 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
999 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
1000 enum tcp_ca_ack_event_flags {
1001 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
1002 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
1003 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
1007 * Interface for adding new TCP congestion control handlers
1009 #define TCP_CA_NAME_MAX 16
1010 #define TCP_CA_MAX 128
1011 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
1013 #define TCP_CA_UNSPEC 0
1015 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1016 #define TCP_CONG_NON_RESTRICTED 0x1
1017 /* Requires ECN/ECT set on all packets */
1018 #define TCP_CONG_NEEDS_ECN 0x2
1019 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1029 /* A rate sample measures the number of (original/retransmitted) data
1030 * packets delivered "delivered" over an interval of time "interval_us".
1031 * The tcp_rate.c code fills in the rate sample, and congestion
1032 * control modules that define a cong_control function to run at the end
1033 * of ACK processing can optionally chose to consult this sample when
1034 * setting cwnd and pacing rate.
1035 * A sample is invalid if "delivered" or "interval_us" is negative.
1037 struct rate_sample {
1038 u64 prior_mstamp; /* starting timestamp for interval */
1039 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1040 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */
1041 s32 delivered; /* number of packets delivered over interval */
1042 s32 delivered_ce; /* number of packets delivered w/ CE marks*/
1043 long interval_us; /* time for tp->delivered to incr "delivered" */
1044 u32 snd_interval_us; /* snd interval for delivered packets */
1045 u32 rcv_interval_us; /* rcv interval for delivered packets */
1046 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1047 int losses; /* number of packets marked lost upon ACK */
1048 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1049 u32 prior_in_flight; /* in flight before this ACK */
1050 u32 last_end_seq; /* end_seq of most recently ACKed packet */
1051 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1052 bool is_retrans; /* is sample from retransmission? */
1053 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1056 struct tcp_congestion_ops {
1057 /* fast path fields are put first to fill one cache line */
1059 /* return slow start threshold (required) */
1060 u32 (*ssthresh)(struct sock *sk);
1062 /* do new cwnd calculation (required) */
1063 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1065 /* call before changing ca_state (optional) */
1066 void (*set_state)(struct sock *sk, u8 new_state);
1068 /* call when cwnd event occurs (optional) */
1069 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1071 /* call when ack arrives (optional) */
1072 void (*in_ack_event)(struct sock *sk, u32 flags);
1074 /* hook for packet ack accounting (optional) */
1075 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1077 /* override sysctl_tcp_min_tso_segs */
1078 u32 (*min_tso_segs)(struct sock *sk);
1080 /* call when packets are delivered to update cwnd and pacing rate,
1081 * after all the ca_state processing. (optional)
1083 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1086 /* new value of cwnd after loss (required) */
1087 u32 (*undo_cwnd)(struct sock *sk);
1088 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1089 u32 (*sndbuf_expand)(struct sock *sk);
1091 /* control/slow paths put last */
1092 /* get info for inet_diag (optional) */
1093 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1094 union tcp_cc_info *info);
1096 char name[TCP_CA_NAME_MAX];
1097 struct module *owner;
1098 struct list_head list;
1102 /* initialize private data (optional) */
1103 void (*init)(struct sock *sk);
1104 /* cleanup private data (optional) */
1105 void (*release)(struct sock *sk);
1106 } ____cacheline_aligned_in_smp;
1108 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1109 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1111 void tcp_assign_congestion_control(struct sock *sk);
1112 void tcp_init_congestion_control(struct sock *sk);
1113 void tcp_cleanup_congestion_control(struct sock *sk);
1114 int tcp_set_default_congestion_control(struct net *net, const char *name);
1115 void tcp_get_default_congestion_control(struct net *net, char *name);
1116 void tcp_get_available_congestion_control(char *buf, size_t len);
1117 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1118 int tcp_set_allowed_congestion_control(char *allowed);
1119 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1120 bool cap_net_admin);
1121 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1122 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1124 u32 tcp_reno_ssthresh(struct sock *sk);
1125 u32 tcp_reno_undo_cwnd(struct sock *sk);
1126 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1127 extern struct tcp_congestion_ops tcp_reno;
1129 struct tcp_congestion_ops *tcp_ca_find(const char *name);
1130 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1131 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1133 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1135 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1141 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1143 const struct inet_connection_sock *icsk = inet_csk(sk);
1145 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1148 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1150 const struct inet_connection_sock *icsk = inet_csk(sk);
1152 if (icsk->icsk_ca_ops->cwnd_event)
1153 icsk->icsk_ca_ops->cwnd_event(sk, event);
1156 /* From tcp_cong.c */
1157 void tcp_set_ca_state(struct sock *sk, const u8 ca_state);
1159 /* From tcp_rate.c */
1160 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1161 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1162 struct rate_sample *rs);
1163 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1164 bool is_sack_reneg, struct rate_sample *rs);
1165 void tcp_rate_check_app_limited(struct sock *sk);
1167 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2)
1169 return t1 > t2 || (t1 == t2 && after(seq1, seq2));
1172 /* These functions determine how the current flow behaves in respect of SACK
1173 * handling. SACK is negotiated with the peer, and therefore it can vary
1174 * between different flows.
1176 * tcp_is_sack - SACK enabled
1177 * tcp_is_reno - No SACK
1179 static inline int tcp_is_sack(const struct tcp_sock *tp)
1181 return likely(tp->rx_opt.sack_ok);
1184 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1186 return !tcp_is_sack(tp);
1189 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1191 return tp->sacked_out + tp->lost_out;
1194 /* This determines how many packets are "in the network" to the best
1195 * of our knowledge. In many cases it is conservative, but where
1196 * detailed information is available from the receiver (via SACK
1197 * blocks etc.) we can make more aggressive calculations.
1199 * Use this for decisions involving congestion control, use just
1200 * tp->packets_out to determine if the send queue is empty or not.
1202 * Read this equation as:
1204 * "Packets sent once on transmission queue" MINUS
1205 * "Packets left network, but not honestly ACKed yet" PLUS
1206 * "Packets fast retransmitted"
1208 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1210 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1213 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1215 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp)
1217 return tp->snd_cwnd;
1220 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val)
1222 WARN_ON_ONCE((int)val <= 0);
1226 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1228 return tcp_snd_cwnd(tp) < tp->snd_ssthresh;
1231 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1233 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1236 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1238 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1239 (1 << inet_csk(sk)->icsk_ca_state);
1242 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1243 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1246 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1248 const struct tcp_sock *tp = tcp_sk(sk);
1250 if (tcp_in_cwnd_reduction(sk))
1251 return tp->snd_ssthresh;
1253 return max(tp->snd_ssthresh,
1254 ((tcp_snd_cwnd(tp) >> 1) +
1255 (tcp_snd_cwnd(tp) >> 2)));
1258 /* Use define here intentionally to get WARN_ON location shown at the caller */
1259 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1261 void tcp_enter_cwr(struct sock *sk);
1262 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1264 /* The maximum number of MSS of available cwnd for which TSO defers
1265 * sending if not using sysctl_tcp_tso_win_divisor.
1267 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1272 /* Returns end sequence number of the receiver's advertised window */
1273 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1275 return tp->snd_una + tp->snd_wnd;
1278 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1279 * flexible approach. The RFC suggests cwnd should not be raised unless
1280 * it was fully used previously. And that's exactly what we do in
1281 * congestion avoidance mode. But in slow start we allow cwnd to grow
1282 * as long as the application has used half the cwnd.
1284 * cwnd is 10 (IW10), but application sends 9 frames.
1285 * We allow cwnd to reach 18 when all frames are ACKed.
1286 * This check is safe because it's as aggressive as slow start which already
1287 * risks 100% overshoot. The advantage is that we discourage application to
1288 * either send more filler packets or data to artificially blow up the cwnd
1289 * usage, and allow application-limited process to probe bw more aggressively.
1291 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1293 const struct tcp_sock *tp = tcp_sk(sk);
1295 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1296 if (tcp_in_slow_start(tp))
1297 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out;
1299 return tp->is_cwnd_limited;
1302 /* BBR congestion control needs pacing.
1303 * Same remark for SO_MAX_PACING_RATE.
1304 * sch_fq packet scheduler is efficiently handling pacing,
1305 * but is not always installed/used.
1306 * Return true if TCP stack should pace packets itself.
1308 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1310 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1313 /* Estimates in how many jiffies next packet for this flow can be sent.
1314 * Scheduling a retransmit timer too early would be silly.
1316 static inline unsigned long tcp_pacing_delay(const struct sock *sk)
1318 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
1320 return delay > 0 ? nsecs_to_jiffies(delay) : 0;
1323 static inline void tcp_reset_xmit_timer(struct sock *sk,
1326 const unsigned long max_when)
1328 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk),
1332 /* Something is really bad, we could not queue an additional packet,
1333 * because qdisc is full or receiver sent a 0 window, or we are paced.
1334 * We do not want to add fuel to the fire, or abort too early,
1335 * so make sure the timer we arm now is at least 200ms in the future,
1336 * regardless of current icsk_rto value (as it could be ~2ms)
1338 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1340 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1343 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1344 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1345 unsigned long max_when)
1347 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1,
1348 inet_csk(sk)->icsk_backoff);
1349 u64 when = (u64)tcp_probe0_base(sk) << backoff;
1351 return (unsigned long)min_t(u64, when, max_when);
1354 static inline void tcp_check_probe_timer(struct sock *sk)
1356 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1357 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1358 tcp_probe0_base(sk), TCP_RTO_MAX);
1361 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1366 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1372 * Calculate(/check) TCP checksum
1374 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1375 __be32 daddr, __wsum base)
1377 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1380 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1382 return !skb_csum_unnecessary(skb) &&
1383 __skb_checksum_complete(skb);
1386 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb,
1387 enum skb_drop_reason *reason);
1390 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1391 void tcp_set_state(struct sock *sk, int state);
1392 void tcp_done(struct sock *sk);
1393 int tcp_abort(struct sock *sk, int err);
1395 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1398 rx_opt->num_sacks = 0;
1401 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1403 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1405 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1406 struct tcp_sock *tp = tcp_sk(sk);
1409 if (!sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle || tp->packets_out ||
1410 ca_ops->cong_control)
1412 delta = tcp_jiffies32 - tp->lsndtime;
1413 if (delta > inet_csk(sk)->icsk_rto)
1414 tcp_cwnd_restart(sk, delta);
1417 /* Determine a window scaling and initial window to offer. */
1418 void tcp_select_initial_window(const struct sock *sk, int __space,
1419 __u32 mss, __u32 *rcv_wnd,
1420 __u32 *window_clamp, int wscale_ok,
1421 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1423 static inline int tcp_win_from_space(const struct sock *sk, int space)
1425 int tcp_adv_win_scale = sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale;
1427 return tcp_adv_win_scale <= 0 ?
1428 (space>>(-tcp_adv_win_scale)) :
1429 space - (space>>tcp_adv_win_scale);
1432 /* Note: caller must be prepared to deal with negative returns */
1433 static inline int tcp_space(const struct sock *sk)
1435 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1436 READ_ONCE(sk->sk_backlog.len) -
1437 atomic_read(&sk->sk_rmem_alloc));
1440 static inline int tcp_full_space(const struct sock *sk)
1442 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1445 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk)
1447 int unused_mem = sk_unused_reserved_mem(sk);
1448 struct tcp_sock *tp = tcp_sk(sk);
1450 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss);
1452 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh,
1453 tcp_win_from_space(sk, unused_mem));
1456 void tcp_cleanup_rbuf(struct sock *sk, int copied);
1458 /* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1459 * If 87.5 % (7/8) of the space has been consumed, we want to override
1460 * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1461 * len/truesize ratio.
1463 static inline bool tcp_rmem_pressure(const struct sock *sk)
1465 int rcvbuf, threshold;
1467 if (tcp_under_memory_pressure(sk))
1470 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1471 threshold = rcvbuf - (rcvbuf >> 3);
1473 return atomic_read(&sk->sk_rmem_alloc) > threshold;
1476 static inline bool tcp_epollin_ready(const struct sock *sk, int target)
1478 const struct tcp_sock *tp = tcp_sk(sk);
1479 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq);
1484 return (avail >= target) || tcp_rmem_pressure(sk) ||
1485 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss);
1488 extern void tcp_openreq_init_rwin(struct request_sock *req,
1489 const struct sock *sk_listener,
1490 const struct dst_entry *dst);
1492 void tcp_enter_memory_pressure(struct sock *sk);
1493 void tcp_leave_memory_pressure(struct sock *sk);
1495 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1497 struct net *net = sock_net((struct sock *)tp);
1499 return tp->keepalive_intvl ? : net->ipv4.sysctl_tcp_keepalive_intvl;
1502 static inline int keepalive_time_when(const struct tcp_sock *tp)
1504 struct net *net = sock_net((struct sock *)tp);
1506 return tp->keepalive_time ? : net->ipv4.sysctl_tcp_keepalive_time;
1509 static inline int keepalive_probes(const struct tcp_sock *tp)
1511 struct net *net = sock_net((struct sock *)tp);
1513 return tp->keepalive_probes ? : net->ipv4.sysctl_tcp_keepalive_probes;
1516 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1518 const struct inet_connection_sock *icsk = &tp->inet_conn;
1520 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1521 tcp_jiffies32 - tp->rcv_tstamp);
1524 static inline int tcp_fin_time(const struct sock *sk)
1526 int fin_timeout = tcp_sk(sk)->linger2 ? : sock_net(sk)->ipv4.sysctl_tcp_fin_timeout;
1527 const int rto = inet_csk(sk)->icsk_rto;
1529 if (fin_timeout < (rto << 2) - (rto >> 1))
1530 fin_timeout = (rto << 2) - (rto >> 1);
1535 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1538 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1540 if (unlikely(!time_before32(ktime_get_seconds(),
1541 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1544 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1545 * then following tcp messages have valid values. Ignore 0 value,
1546 * or else 'negative' tsval might forbid us to accept their packets.
1548 if (!rx_opt->ts_recent)
1553 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1556 if (tcp_paws_check(rx_opt, 0))
1559 /* RST segments are not recommended to carry timestamp,
1560 and, if they do, it is recommended to ignore PAWS because
1561 "their cleanup function should take precedence over timestamps."
1562 Certainly, it is mistake. It is necessary to understand the reasons
1563 of this constraint to relax it: if peer reboots, clock may go
1564 out-of-sync and half-open connections will not be reset.
1565 Actually, the problem would be not existing if all
1566 the implementations followed draft about maintaining clock
1567 via reboots. Linux-2.2 DOES NOT!
1569 However, we can relax time bounds for RST segments to MSL.
1571 if (rst && !time_before32(ktime_get_seconds(),
1572 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1577 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1578 int mib_idx, u32 *last_oow_ack_time);
1580 static inline void tcp_mib_init(struct net *net)
1583 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1584 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1585 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1586 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1590 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1592 tp->lost_skb_hint = NULL;
1595 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1597 tcp_clear_retrans_hints_partial(tp);
1598 tp->retransmit_skb_hint = NULL;
1601 union tcp_md5_addr {
1603 #if IS_ENABLED(CONFIG_IPV6)
1608 /* - key database */
1609 struct tcp_md5sig_key {
1610 struct hlist_node node;
1612 u8 family; /* AF_INET or AF_INET6 */
1615 union tcp_md5_addr addr;
1616 int l3index; /* set if key added with L3 scope */
1617 u8 key[TCP_MD5SIG_MAXKEYLEN];
1618 struct rcu_head rcu;
1622 struct tcp_md5sig_info {
1623 struct hlist_head head;
1624 struct rcu_head rcu;
1627 /* - pseudo header */
1628 struct tcp4_pseudohdr {
1636 struct tcp6_pseudohdr {
1637 struct in6_addr saddr;
1638 struct in6_addr daddr;
1640 __be32 protocol; /* including padding */
1643 union tcp_md5sum_block {
1644 struct tcp4_pseudohdr ip4;
1645 #if IS_ENABLED(CONFIG_IPV6)
1646 struct tcp6_pseudohdr ip6;
1650 /* - pool: digest algorithm, hash description and scratch buffer */
1651 struct tcp_md5sig_pool {
1652 struct ahash_request *md5_req;
1657 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1658 const struct sock *sk, const struct sk_buff *skb);
1659 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1660 int family, u8 prefixlen, int l3index, u8 flags,
1661 const u8 *newkey, u8 newkeylen, gfp_t gfp);
1662 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1663 int family, u8 prefixlen, int l3index, u8 flags);
1664 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1665 const struct sock *addr_sk);
1667 #ifdef CONFIG_TCP_MD5SIG
1668 #include <linux/jump_label.h>
1669 extern struct static_key_false tcp_md5_needed;
1670 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1671 const union tcp_md5_addr *addr,
1673 static inline struct tcp_md5sig_key *
1674 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1675 const union tcp_md5_addr *addr, int family)
1677 if (!static_branch_unlikely(&tcp_md5_needed))
1679 return __tcp_md5_do_lookup(sk, l3index, addr, family);
1682 enum skb_drop_reason
1683 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
1684 const void *saddr, const void *daddr,
1685 int family, int dif, int sdif);
1688 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1690 static inline struct tcp_md5sig_key *
1691 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1692 const union tcp_md5_addr *addr, int family)
1697 static inline enum skb_drop_reason
1698 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
1699 const void *saddr, const void *daddr,
1700 int family, int dif, int sdif)
1702 return SKB_NOT_DROPPED_YET;
1704 #define tcp_twsk_md5_key(twsk) NULL
1707 bool tcp_alloc_md5sig_pool(void);
1709 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1710 static inline void tcp_put_md5sig_pool(void)
1715 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1716 unsigned int header_len);
1717 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1718 const struct tcp_md5sig_key *key);
1720 /* From tcp_fastopen.c */
1721 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1722 struct tcp_fastopen_cookie *cookie);
1723 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1724 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1726 struct tcp_fastopen_request {
1727 /* Fast Open cookie. Size 0 means a cookie request */
1728 struct tcp_fastopen_cookie cookie;
1729 struct msghdr *data; /* data in MSG_FASTOPEN */
1731 int copied; /* queued in tcp_connect() */
1732 struct ubuf_info *uarg;
1734 void tcp_free_fastopen_req(struct tcp_sock *tp);
1735 void tcp_fastopen_destroy_cipher(struct sock *sk);
1736 void tcp_fastopen_ctx_destroy(struct net *net);
1737 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1738 void *primary_key, void *backup_key);
1739 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
1741 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1742 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1743 struct request_sock *req,
1744 struct tcp_fastopen_cookie *foc,
1745 const struct dst_entry *dst);
1746 void tcp_fastopen_init_key_once(struct net *net);
1747 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1748 struct tcp_fastopen_cookie *cookie);
1749 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1750 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1751 #define TCP_FASTOPEN_KEY_MAX 2
1752 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1753 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1755 /* Fastopen key context */
1756 struct tcp_fastopen_context {
1757 siphash_key_t key[TCP_FASTOPEN_KEY_MAX];
1759 struct rcu_head rcu;
1762 void tcp_fastopen_active_disable(struct sock *sk);
1763 bool tcp_fastopen_active_should_disable(struct sock *sk);
1764 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1765 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1767 /* Caller needs to wrap with rcu_read_(un)lock() */
1769 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1771 struct tcp_fastopen_context *ctx;
1773 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1775 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1780 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1781 const struct tcp_fastopen_cookie *orig)
1783 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1784 orig->len == foc->len &&
1785 !memcmp(orig->val, foc->val, foc->len))
1791 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1796 /* Latencies incurred by various limits for a sender. They are
1797 * chronograph-like stats that are mutually exclusive.
1801 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1802 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1803 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1807 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1808 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1810 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1811 * the same memory storage than skb->destructor/_skb_refdst
1813 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1815 skb->destructor = NULL;
1816 skb->_skb_refdst = 0UL;
1819 #define tcp_skb_tsorted_save(skb) { \
1820 unsigned long _save = skb->_skb_refdst; \
1821 skb->_skb_refdst = 0UL;
1823 #define tcp_skb_tsorted_restore(skb) \
1824 skb->_skb_refdst = _save; \
1827 void tcp_write_queue_purge(struct sock *sk);
1829 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1831 return skb_rb_first(&sk->tcp_rtx_queue);
1834 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1836 return skb_rb_last(&sk->tcp_rtx_queue);
1839 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1841 return skb_peek_tail(&sk->sk_write_queue);
1844 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1845 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1847 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1849 return skb_peek(&sk->sk_write_queue);
1852 static inline bool tcp_skb_is_last(const struct sock *sk,
1853 const struct sk_buff *skb)
1855 return skb_queue_is_last(&sk->sk_write_queue, skb);
1859 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
1862 * Since the write queue can have a temporary empty skb in it,
1863 * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
1865 static inline bool tcp_write_queue_empty(const struct sock *sk)
1867 const struct tcp_sock *tp = tcp_sk(sk);
1869 return tp->write_seq == tp->snd_nxt;
1872 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1874 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1877 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1879 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1882 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1884 __skb_queue_tail(&sk->sk_write_queue, skb);
1886 /* Queue it, remembering where we must start sending. */
1887 if (sk->sk_write_queue.next == skb)
1888 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1891 /* Insert new before skb on the write queue of sk. */
1892 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1893 struct sk_buff *skb,
1896 __skb_queue_before(&sk->sk_write_queue, skb, new);
1899 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1901 tcp_skb_tsorted_anchor_cleanup(skb);
1902 __skb_unlink(skb, &sk->sk_write_queue);
1905 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1907 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1909 tcp_skb_tsorted_anchor_cleanup(skb);
1910 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1913 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1915 list_del(&skb->tcp_tsorted_anchor);
1916 tcp_rtx_queue_unlink(skb, sk);
1917 tcp_wmem_free_skb(sk, skb);
1920 static inline void tcp_push_pending_frames(struct sock *sk)
1922 if (tcp_send_head(sk)) {
1923 struct tcp_sock *tp = tcp_sk(sk);
1925 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1929 /* Start sequence of the skb just after the highest skb with SACKed
1930 * bit, valid only if sacked_out > 0 or when the caller has ensured
1931 * validity by itself.
1933 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1935 if (!tp->sacked_out)
1938 if (tp->highest_sack == NULL)
1941 return TCP_SKB_CB(tp->highest_sack)->seq;
1944 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1946 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1949 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1951 return tcp_sk(sk)->highest_sack;
1954 static inline void tcp_highest_sack_reset(struct sock *sk)
1956 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1959 /* Called when old skb is about to be deleted and replaced by new skb */
1960 static inline void tcp_highest_sack_replace(struct sock *sk,
1961 struct sk_buff *old,
1962 struct sk_buff *new)
1964 if (old == tcp_highest_sack(sk))
1965 tcp_sk(sk)->highest_sack = new;
1968 /* This helper checks if socket has IP_TRANSPARENT set */
1969 static inline bool inet_sk_transparent(const struct sock *sk)
1971 switch (sk->sk_state) {
1973 return inet_twsk(sk)->tw_transparent;
1974 case TCP_NEW_SYN_RECV:
1975 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1977 return inet_sk(sk)->transparent;
1980 /* Determines whether this is a thin stream (which may suffer from
1981 * increased latency). Used to trigger latency-reducing mechanisms.
1983 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1985 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1989 enum tcp_seq_states {
1990 TCP_SEQ_STATE_LISTENING,
1991 TCP_SEQ_STATE_ESTABLISHED,
1994 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1995 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1996 void tcp_seq_stop(struct seq_file *seq, void *v);
1998 struct tcp_seq_afinfo {
2002 struct tcp_iter_state {
2003 struct seq_net_private p;
2004 enum tcp_seq_states state;
2005 struct sock *syn_wait_sk;
2006 int bucket, offset, sbucket, num;
2010 extern struct request_sock_ops tcp_request_sock_ops;
2011 extern struct request_sock_ops tcp6_request_sock_ops;
2013 void tcp_v4_destroy_sock(struct sock *sk);
2015 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
2016 netdev_features_t features);
2017 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
2018 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
2019 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
2020 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff));
2021 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb));
2022 int tcp_gro_complete(struct sk_buff *skb);
2024 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
2026 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
2028 struct net *net = sock_net((struct sock *)tp);
2029 return tp->notsent_lowat ?: net->ipv4.sysctl_tcp_notsent_lowat;
2032 bool tcp_stream_memory_free(const struct sock *sk, int wake);
2034 #ifdef CONFIG_PROC_FS
2035 int tcp4_proc_init(void);
2036 void tcp4_proc_exit(void);
2039 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
2040 int tcp_conn_request(struct request_sock_ops *rsk_ops,
2041 const struct tcp_request_sock_ops *af_ops,
2042 struct sock *sk, struct sk_buff *skb);
2044 /* TCP af-specific functions */
2045 struct tcp_sock_af_ops {
2046 #ifdef CONFIG_TCP_MD5SIG
2047 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
2048 const struct sock *addr_sk);
2049 int (*calc_md5_hash)(char *location,
2050 const struct tcp_md5sig_key *md5,
2051 const struct sock *sk,
2052 const struct sk_buff *skb);
2053 int (*md5_parse)(struct sock *sk,
2060 struct tcp_request_sock_ops {
2062 #ifdef CONFIG_TCP_MD5SIG
2063 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
2064 const struct sock *addr_sk);
2065 int (*calc_md5_hash) (char *location,
2066 const struct tcp_md5sig_key *md5,
2067 const struct sock *sk,
2068 const struct sk_buff *skb);
2070 #ifdef CONFIG_SYN_COOKIES
2071 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
2074 struct dst_entry *(*route_req)(const struct sock *sk,
2075 struct sk_buff *skb,
2077 struct request_sock *req);
2078 u32 (*init_seq)(const struct sk_buff *skb);
2079 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
2080 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2081 struct flowi *fl, struct request_sock *req,
2082 struct tcp_fastopen_cookie *foc,
2083 enum tcp_synack_type synack_type,
2084 struct sk_buff *syn_skb);
2087 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2088 #if IS_ENABLED(CONFIG_IPV6)
2089 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2092 #ifdef CONFIG_SYN_COOKIES
2093 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2094 const struct sock *sk, struct sk_buff *skb,
2097 tcp_synq_overflow(sk);
2098 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2099 return ops->cookie_init_seq(skb, mss);
2102 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2103 const struct sock *sk, struct sk_buff *skb,
2110 int tcpv4_offload_init(void);
2112 void tcp_v4_init(void);
2113 void tcp_init(void);
2115 /* tcp_recovery.c */
2116 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2117 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2118 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2120 extern bool tcp_rack_mark_lost(struct sock *sk);
2121 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2123 extern void tcp_rack_reo_timeout(struct sock *sk);
2124 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2126 /* At how many usecs into the future should the RTO fire? */
2127 static inline s64 tcp_rto_delta_us(const struct sock *sk)
2129 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2130 u32 rto = inet_csk(sk)->icsk_rto;
2131 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2133 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2137 * Save and compile IPv4 options, return a pointer to it
2139 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2140 struct sk_buff *skb)
2142 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2143 struct ip_options_rcu *dopt = NULL;
2146 int opt_size = sizeof(*dopt) + opt->optlen;
2148 dopt = kmalloc(opt_size, GFP_ATOMIC);
2149 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2157 /* locally generated TCP pure ACKs have skb->truesize == 2
2158 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2159 * This is much faster than dissecting the packet to find out.
2160 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2162 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2164 return skb->truesize == 2;
2167 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2172 static inline int tcp_inq(struct sock *sk)
2174 struct tcp_sock *tp = tcp_sk(sk);
2177 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2179 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2181 before(tp->urg_seq, tp->copied_seq) ||
2182 !before(tp->urg_seq, tp->rcv_nxt)) {
2184 answ = tp->rcv_nxt - tp->copied_seq;
2186 /* Subtract 1, if FIN was received */
2187 if (answ && sock_flag(sk, SOCK_DONE))
2190 answ = tp->urg_seq - tp->copied_seq;
2196 int tcp_peek_len(struct socket *sock);
2198 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2202 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2204 /* We update these fields while other threads might
2205 * read them from tcp_get_info()
2207 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in);
2208 if (skb->len > tcp_hdrlen(skb))
2209 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in);
2213 * TCP listen path runs lockless.
2214 * We forced "struct sock" to be const qualified to make sure
2215 * we don't modify one of its field by mistake.
2216 * Here, we increment sk_drops which is an atomic_t, so we can safely
2217 * make sock writable again.
2219 static inline void tcp_listendrop(const struct sock *sk)
2221 atomic_inc(&((struct sock *)sk)->sk_drops);
2222 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2225 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2228 * Interface for adding Upper Level Protocols over TCP
2231 #define TCP_ULP_NAME_MAX 16
2232 #define TCP_ULP_MAX 128
2233 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2235 struct tcp_ulp_ops {
2236 struct list_head list;
2238 /* initialize ulp */
2239 int (*init)(struct sock *sk);
2241 void (*update)(struct sock *sk, struct proto *p,
2242 void (*write_space)(struct sock *sk));
2244 void (*release)(struct sock *sk);
2246 int (*get_info)(const struct sock *sk, struct sk_buff *skb);
2247 size_t (*get_info_size)(const struct sock *sk);
2249 void (*clone)(const struct request_sock *req, struct sock *newsk,
2250 const gfp_t priority);
2252 char name[TCP_ULP_NAME_MAX];
2253 struct module *owner;
2255 int tcp_register_ulp(struct tcp_ulp_ops *type);
2256 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2257 int tcp_set_ulp(struct sock *sk, const char *name);
2258 void tcp_get_available_ulp(char *buf, size_t len);
2259 void tcp_cleanup_ulp(struct sock *sk);
2260 void tcp_update_ulp(struct sock *sk, struct proto *p,
2261 void (*write_space)(struct sock *sk));
2263 #define MODULE_ALIAS_TCP_ULP(name) \
2264 __MODULE_INFO(alias, alias_userspace, name); \
2265 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2267 #ifdef CONFIG_NET_SOCK_MSG
2271 #ifdef CONFIG_BPF_SYSCALL
2272 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
2273 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
2274 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2275 #endif /* CONFIG_BPF_SYSCALL */
2277 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes,
2279 #endif /* CONFIG_NET_SOCK_MSG */
2281 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG)
2282 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2287 #ifdef CONFIG_CGROUP_BPF
2288 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2289 struct sk_buff *skb,
2290 unsigned int end_offset)
2293 skops->skb_data_end = skb->data + end_offset;
2296 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2297 struct sk_buff *skb,
2298 unsigned int end_offset)
2303 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2304 * is < 0, then the BPF op failed (for example if the loaded BPF
2305 * program does not support the chosen operation or there is no BPF
2309 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2311 struct bpf_sock_ops_kern sock_ops;
2314 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2315 if (sk_fullsock(sk)) {
2316 sock_ops.is_fullsock = 1;
2317 sock_owned_by_me(sk);
2323 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2325 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2327 ret = sock_ops.reply;
2333 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2335 u32 args[2] = {arg1, arg2};
2337 return tcp_call_bpf(sk, op, 2, args);
2340 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2343 u32 args[3] = {arg1, arg2, arg3};
2345 return tcp_call_bpf(sk, op, 3, args);
2349 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2354 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2359 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2367 static inline u32 tcp_timeout_init(struct sock *sk)
2371 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2374 timeout = TCP_TIMEOUT_INIT;
2375 return min_t(int, timeout, TCP_RTO_MAX);
2378 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2382 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2389 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2391 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2394 static inline void tcp_bpf_rtt(struct sock *sk)
2396 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2397 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2400 #if IS_ENABLED(CONFIG_SMC)
2401 extern struct static_key_false tcp_have_smc;
2404 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2405 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2406 void (*cad)(struct sock *sk, u32 ack_seq));
2407 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2408 void clean_acked_data_flush(void);
2411 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2412 static inline void tcp_add_tx_delay(struct sk_buff *skb,
2413 const struct tcp_sock *tp)
2415 if (static_branch_unlikely(&tcp_tx_delay_enabled))
2416 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2419 /* Compute Earliest Departure Time for some control packets
2420 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2422 static inline u64 tcp_transmit_time(const struct sock *sk)
2424 if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2425 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2426 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2428 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;