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);
675 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
676 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off);
677 void tcp_read_done(struct sock *sk, size_t len);
679 void tcp_initialize_rcv_mss(struct sock *sk);
681 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
682 int tcp_mss_to_mtu(struct sock *sk, int mss);
683 void tcp_mtup_init(struct sock *sk);
685 static inline void tcp_bound_rto(const struct sock *sk)
687 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
688 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
691 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
693 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
696 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
698 /* mptcp hooks are only on the slow path */
699 if (sk_is_mptcp((struct sock *)tp))
702 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
703 ntohl(TCP_FLAG_ACK) |
707 static inline void tcp_fast_path_on(struct tcp_sock *tp)
709 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
712 static inline void tcp_fast_path_check(struct sock *sk)
714 struct tcp_sock *tp = tcp_sk(sk);
716 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
718 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
720 tcp_fast_path_on(tp);
723 /* Compute the actual rto_min value */
724 static inline u32 tcp_rto_min(struct sock *sk)
726 const struct dst_entry *dst = __sk_dst_get(sk);
727 u32 rto_min = inet_csk(sk)->icsk_rto_min;
729 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
730 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
734 static inline u32 tcp_rto_min_us(struct sock *sk)
736 return jiffies_to_usecs(tcp_rto_min(sk));
739 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
741 return dst_metric_locked(dst, RTAX_CC_ALGO);
744 /* Minimum RTT in usec. ~0 means not available. */
745 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
747 return minmax_get(&tp->rtt_min);
750 /* Compute the actual receive window we are currently advertising.
751 * Rcv_nxt can be after the window if our peer push more data
752 * than the offered window.
754 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
756 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
763 /* Choose a new window, without checks for shrinking, and without
764 * scaling applied to the result. The caller does these things
765 * if necessary. This is a "raw" window selection.
767 u32 __tcp_select_window(struct sock *sk);
769 void tcp_send_window_probe(struct sock *sk);
771 /* TCP uses 32bit jiffies to save some space.
772 * Note that this is different from tcp_time_stamp, which
773 * historically has been the same until linux-4.13.
775 #define tcp_jiffies32 ((u32)jiffies)
778 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
779 * It is no longer tied to jiffies, but to 1 ms clock.
780 * Note: double check if you want to use tcp_jiffies32 instead of this.
782 #define TCP_TS_HZ 1000
784 static inline u64 tcp_clock_ns(void)
786 return ktime_get_ns();
789 static inline u64 tcp_clock_us(void)
791 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
794 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
795 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
797 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
800 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
801 static inline u32 tcp_ns_to_ts(u64 ns)
803 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
806 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
807 static inline u32 tcp_time_stamp_raw(void)
809 return tcp_ns_to_ts(tcp_clock_ns());
812 void tcp_mstamp_refresh(struct tcp_sock *tp);
814 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
816 return max_t(s64, t1 - t0, 0);
819 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
821 return tcp_ns_to_ts(skb->skb_mstamp_ns);
824 /* provide the departure time in us unit */
825 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
827 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
831 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
833 #define TCPHDR_FIN 0x01
834 #define TCPHDR_SYN 0x02
835 #define TCPHDR_RST 0x04
836 #define TCPHDR_PSH 0x08
837 #define TCPHDR_ACK 0x10
838 #define TCPHDR_URG 0x20
839 #define TCPHDR_ECE 0x40
840 #define TCPHDR_CWR 0x80
842 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
844 /* This is what the send packet queuing engine uses to pass
845 * TCP per-packet control information to the transmission code.
846 * We also store the host-order sequence numbers in here too.
847 * This is 44 bytes if IPV6 is enabled.
848 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
851 __u32 seq; /* Starting sequence number */
852 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
854 /* Note : tcp_tw_isn is used in input path only
855 * (isn chosen by tcp_timewait_state_process())
857 * tcp_gso_segs/size are used in write queue only,
858 * cf tcp_skb_pcount()/tcp_skb_mss()
866 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
868 __u8 sacked; /* State flags for SACK. */
869 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
870 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
871 #define TCPCB_LOST 0x04 /* SKB is lost */
872 #define TCPCB_TAGBITS 0x07 /* All tag bits */
873 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
874 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
875 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
878 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
879 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
880 eor:1, /* Is skb MSG_EOR marked? */
881 has_rxtstamp:1, /* SKB has a RX timestamp */
883 __u32 ack_seq; /* Sequence number ACK'd */
886 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1)
887 /* There is space for up to 24 bytes */
888 __u32 is_app_limited:1, /* cwnd not fully used? */
891 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
893 /* start of send pipeline phase */
895 /* when we reached the "delivered" count */
896 u64 delivered_mstamp;
897 } tx; /* only used for outgoing skbs */
899 struct inet_skb_parm h4;
900 #if IS_ENABLED(CONFIG_IPV6)
901 struct inet6_skb_parm h6;
903 } header; /* For incoming skbs */
907 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
909 extern const struct inet_connection_sock_af_ops ipv4_specific;
911 #if IS_ENABLED(CONFIG_IPV6)
912 /* This is the variant of inet6_iif() that must be used by TCP,
913 * as TCP moves IP6CB into a different location in skb->cb[]
915 static inline int tcp_v6_iif(const struct sk_buff *skb)
917 return TCP_SKB_CB(skb)->header.h6.iif;
920 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
922 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
924 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
927 /* TCP_SKB_CB reference means this can not be used from early demux */
928 static inline int tcp_v6_sdif(const struct sk_buff *skb)
930 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
931 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
932 return TCP_SKB_CB(skb)->header.h6.iif;
937 extern const struct inet_connection_sock_af_ops ipv6_specific;
939 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
940 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
941 void tcp_v6_early_demux(struct sk_buff *skb);
945 /* TCP_SKB_CB reference means this can not be used from early demux */
946 static inline int tcp_v4_sdif(struct sk_buff *skb)
948 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
949 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
950 return TCP_SKB_CB(skb)->header.h4.iif;
955 /* Due to TSO, an SKB can be composed of multiple actual
956 * packets. To keep these tracked properly, we use this.
958 static inline int tcp_skb_pcount(const struct sk_buff *skb)
960 return TCP_SKB_CB(skb)->tcp_gso_segs;
963 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
965 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
968 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
970 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
973 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
974 static inline int tcp_skb_mss(const struct sk_buff *skb)
976 return TCP_SKB_CB(skb)->tcp_gso_size;
979 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
981 return likely(!TCP_SKB_CB(skb)->eor);
984 static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
985 const struct sk_buff *from)
987 return likely(tcp_skb_can_collapse_to(to) &&
988 mptcp_skb_can_collapse(to, from) &&
989 skb_pure_zcopy_same(to, from));
992 /* Events passed to congestion control interface */
994 CA_EVENT_TX_START, /* first transmit when no packets in flight */
995 CA_EVENT_CWND_RESTART, /* congestion window restart */
996 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
997 CA_EVENT_LOSS, /* loss timeout */
998 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
999 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
1002 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
1003 enum tcp_ca_ack_event_flags {
1004 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
1005 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
1006 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
1010 * Interface for adding new TCP congestion control handlers
1012 #define TCP_CA_NAME_MAX 16
1013 #define TCP_CA_MAX 128
1014 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
1016 #define TCP_CA_UNSPEC 0
1018 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1019 #define TCP_CONG_NON_RESTRICTED 0x1
1020 /* Requires ECN/ECT set on all packets */
1021 #define TCP_CONG_NEEDS_ECN 0x2
1022 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1032 /* A rate sample measures the number of (original/retransmitted) data
1033 * packets delivered "delivered" over an interval of time "interval_us".
1034 * The tcp_rate.c code fills in the rate sample, and congestion
1035 * control modules that define a cong_control function to run at the end
1036 * of ACK processing can optionally chose to consult this sample when
1037 * setting cwnd and pacing rate.
1038 * A sample is invalid if "delivered" or "interval_us" is negative.
1040 struct rate_sample {
1041 u64 prior_mstamp; /* starting timestamp for interval */
1042 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1043 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */
1044 s32 delivered; /* number of packets delivered over interval */
1045 s32 delivered_ce; /* number of packets delivered w/ CE marks*/
1046 long interval_us; /* time for tp->delivered to incr "delivered" */
1047 u32 snd_interval_us; /* snd interval for delivered packets */
1048 u32 rcv_interval_us; /* rcv interval for delivered packets */
1049 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1050 int losses; /* number of packets marked lost upon ACK */
1051 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1052 u32 prior_in_flight; /* in flight before this ACK */
1053 u32 last_end_seq; /* end_seq of most recently ACKed packet */
1054 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1055 bool is_retrans; /* is sample from retransmission? */
1056 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1059 struct tcp_congestion_ops {
1060 /* fast path fields are put first to fill one cache line */
1062 /* return slow start threshold (required) */
1063 u32 (*ssthresh)(struct sock *sk);
1065 /* do new cwnd calculation (required) */
1066 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1068 /* call before changing ca_state (optional) */
1069 void (*set_state)(struct sock *sk, u8 new_state);
1071 /* call when cwnd event occurs (optional) */
1072 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1074 /* call when ack arrives (optional) */
1075 void (*in_ack_event)(struct sock *sk, u32 flags);
1077 /* hook for packet ack accounting (optional) */
1078 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1080 /* override sysctl_tcp_min_tso_segs */
1081 u32 (*min_tso_segs)(struct sock *sk);
1083 /* call when packets are delivered to update cwnd and pacing rate,
1084 * after all the ca_state processing. (optional)
1086 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1089 /* new value of cwnd after loss (required) */
1090 u32 (*undo_cwnd)(struct sock *sk);
1091 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1092 u32 (*sndbuf_expand)(struct sock *sk);
1094 /* control/slow paths put last */
1095 /* get info for inet_diag (optional) */
1096 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1097 union tcp_cc_info *info);
1099 char name[TCP_CA_NAME_MAX];
1100 struct module *owner;
1101 struct list_head list;
1105 /* initialize private data (optional) */
1106 void (*init)(struct sock *sk);
1107 /* cleanup private data (optional) */
1108 void (*release)(struct sock *sk);
1109 } ____cacheline_aligned_in_smp;
1111 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1112 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1114 void tcp_assign_congestion_control(struct sock *sk);
1115 void tcp_init_congestion_control(struct sock *sk);
1116 void tcp_cleanup_congestion_control(struct sock *sk);
1117 int tcp_set_default_congestion_control(struct net *net, const char *name);
1118 void tcp_get_default_congestion_control(struct net *net, char *name);
1119 void tcp_get_available_congestion_control(char *buf, size_t len);
1120 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1121 int tcp_set_allowed_congestion_control(char *allowed);
1122 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1123 bool cap_net_admin);
1124 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1125 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1127 u32 tcp_reno_ssthresh(struct sock *sk);
1128 u32 tcp_reno_undo_cwnd(struct sock *sk);
1129 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1130 extern struct tcp_congestion_ops tcp_reno;
1132 struct tcp_congestion_ops *tcp_ca_find(const char *name);
1133 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1134 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1136 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1138 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1144 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1146 const struct inet_connection_sock *icsk = inet_csk(sk);
1148 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1151 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1153 const struct inet_connection_sock *icsk = inet_csk(sk);
1155 if (icsk->icsk_ca_ops->cwnd_event)
1156 icsk->icsk_ca_ops->cwnd_event(sk, event);
1159 /* From tcp_cong.c */
1160 void tcp_set_ca_state(struct sock *sk, const u8 ca_state);
1162 /* From tcp_rate.c */
1163 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1164 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1165 struct rate_sample *rs);
1166 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1167 bool is_sack_reneg, struct rate_sample *rs);
1168 void tcp_rate_check_app_limited(struct sock *sk);
1170 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2)
1172 return t1 > t2 || (t1 == t2 && after(seq1, seq2));
1175 /* These functions determine how the current flow behaves in respect of SACK
1176 * handling. SACK is negotiated with the peer, and therefore it can vary
1177 * between different flows.
1179 * tcp_is_sack - SACK enabled
1180 * tcp_is_reno - No SACK
1182 static inline int tcp_is_sack(const struct tcp_sock *tp)
1184 return likely(tp->rx_opt.sack_ok);
1187 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1189 return !tcp_is_sack(tp);
1192 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1194 return tp->sacked_out + tp->lost_out;
1197 /* This determines how many packets are "in the network" to the best
1198 * of our knowledge. In many cases it is conservative, but where
1199 * detailed information is available from the receiver (via SACK
1200 * blocks etc.) we can make more aggressive calculations.
1202 * Use this for decisions involving congestion control, use just
1203 * tp->packets_out to determine if the send queue is empty or not.
1205 * Read this equation as:
1207 * "Packets sent once on transmission queue" MINUS
1208 * "Packets left network, but not honestly ACKed yet" PLUS
1209 * "Packets fast retransmitted"
1211 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1213 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1216 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1218 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp)
1220 return tp->snd_cwnd;
1223 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val)
1225 WARN_ON_ONCE((int)val <= 0);
1229 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1231 return tcp_snd_cwnd(tp) < tp->snd_ssthresh;
1234 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1236 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1239 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1241 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1242 (1 << inet_csk(sk)->icsk_ca_state);
1245 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1246 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1249 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1251 const struct tcp_sock *tp = tcp_sk(sk);
1253 if (tcp_in_cwnd_reduction(sk))
1254 return tp->snd_ssthresh;
1256 return max(tp->snd_ssthresh,
1257 ((tcp_snd_cwnd(tp) >> 1) +
1258 (tcp_snd_cwnd(tp) >> 2)));
1261 /* Use define here intentionally to get WARN_ON location shown at the caller */
1262 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1264 void tcp_enter_cwr(struct sock *sk);
1265 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1267 /* The maximum number of MSS of available cwnd for which TSO defers
1268 * sending if not using sysctl_tcp_tso_win_divisor.
1270 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1275 /* Returns end sequence number of the receiver's advertised window */
1276 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1278 return tp->snd_una + tp->snd_wnd;
1281 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1282 * flexible approach. The RFC suggests cwnd should not be raised unless
1283 * it was fully used previously. And that's exactly what we do in
1284 * congestion avoidance mode. But in slow start we allow cwnd to grow
1285 * as long as the application has used half the cwnd.
1287 * cwnd is 10 (IW10), but application sends 9 frames.
1288 * We allow cwnd to reach 18 when all frames are ACKed.
1289 * This check is safe because it's as aggressive as slow start which already
1290 * risks 100% overshoot. The advantage is that we discourage application to
1291 * either send more filler packets or data to artificially blow up the cwnd
1292 * usage, and allow application-limited process to probe bw more aggressively.
1294 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1296 const struct tcp_sock *tp = tcp_sk(sk);
1298 if (tp->is_cwnd_limited)
1301 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1302 if (tcp_in_slow_start(tp))
1303 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out;
1308 /* BBR congestion control needs pacing.
1309 * Same remark for SO_MAX_PACING_RATE.
1310 * sch_fq packet scheduler is efficiently handling pacing,
1311 * but is not always installed/used.
1312 * Return true if TCP stack should pace packets itself.
1314 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1316 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1319 /* Estimates in how many jiffies next packet for this flow can be sent.
1320 * Scheduling a retransmit timer too early would be silly.
1322 static inline unsigned long tcp_pacing_delay(const struct sock *sk)
1324 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
1326 return delay > 0 ? nsecs_to_jiffies(delay) : 0;
1329 static inline void tcp_reset_xmit_timer(struct sock *sk,
1332 const unsigned long max_when)
1334 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk),
1338 /* Something is really bad, we could not queue an additional packet,
1339 * because qdisc is full or receiver sent a 0 window, or we are paced.
1340 * We do not want to add fuel to the fire, or abort too early,
1341 * so make sure the timer we arm now is at least 200ms in the future,
1342 * regardless of current icsk_rto value (as it could be ~2ms)
1344 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1346 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1349 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1350 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1351 unsigned long max_when)
1353 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1,
1354 inet_csk(sk)->icsk_backoff);
1355 u64 when = (u64)tcp_probe0_base(sk) << backoff;
1357 return (unsigned long)min_t(u64, when, max_when);
1360 static inline void tcp_check_probe_timer(struct sock *sk)
1362 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1363 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1364 tcp_probe0_base(sk), TCP_RTO_MAX);
1367 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1372 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1378 * Calculate(/check) TCP checksum
1380 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1381 __be32 daddr, __wsum base)
1383 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1386 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1388 return !skb_csum_unnecessary(skb) &&
1389 __skb_checksum_complete(skb);
1392 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb,
1393 enum skb_drop_reason *reason);
1396 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1397 void tcp_set_state(struct sock *sk, int state);
1398 void tcp_done(struct sock *sk);
1399 int tcp_abort(struct sock *sk, int err);
1401 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1404 rx_opt->num_sacks = 0;
1407 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1409 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1411 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1412 struct tcp_sock *tp = tcp_sk(sk);
1415 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) ||
1416 tp->packets_out || ca_ops->cong_control)
1418 delta = tcp_jiffies32 - tp->lsndtime;
1419 if (delta > inet_csk(sk)->icsk_rto)
1420 tcp_cwnd_restart(sk, delta);
1423 /* Determine a window scaling and initial window to offer. */
1424 void tcp_select_initial_window(const struct sock *sk, int __space,
1425 __u32 mss, __u32 *rcv_wnd,
1426 __u32 *window_clamp, int wscale_ok,
1427 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1429 static inline int tcp_win_from_space(const struct sock *sk, int space)
1431 int tcp_adv_win_scale = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale);
1433 return tcp_adv_win_scale <= 0 ?
1434 (space>>(-tcp_adv_win_scale)) :
1435 space - (space>>tcp_adv_win_scale);
1438 /* Note: caller must be prepared to deal with negative returns */
1439 static inline int tcp_space(const struct sock *sk)
1441 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1442 READ_ONCE(sk->sk_backlog.len) -
1443 atomic_read(&sk->sk_rmem_alloc));
1446 static inline int tcp_full_space(const struct sock *sk)
1448 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1451 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk)
1453 int unused_mem = sk_unused_reserved_mem(sk);
1454 struct tcp_sock *tp = tcp_sk(sk);
1456 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss);
1458 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh,
1459 tcp_win_from_space(sk, unused_mem));
1462 void tcp_cleanup_rbuf(struct sock *sk, int copied);
1464 /* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1465 * If 87.5 % (7/8) of the space has been consumed, we want to override
1466 * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1467 * len/truesize ratio.
1469 static inline bool tcp_rmem_pressure(const struct sock *sk)
1471 int rcvbuf, threshold;
1473 if (tcp_under_memory_pressure(sk))
1476 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1477 threshold = rcvbuf - (rcvbuf >> 3);
1479 return atomic_read(&sk->sk_rmem_alloc) > threshold;
1482 static inline bool tcp_epollin_ready(const struct sock *sk, int target)
1484 const struct tcp_sock *tp = tcp_sk(sk);
1485 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq);
1490 return (avail >= target) || tcp_rmem_pressure(sk) ||
1491 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss);
1494 extern void tcp_openreq_init_rwin(struct request_sock *req,
1495 const struct sock *sk_listener,
1496 const struct dst_entry *dst);
1498 void tcp_enter_memory_pressure(struct sock *sk);
1499 void tcp_leave_memory_pressure(struct sock *sk);
1501 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1503 struct net *net = sock_net((struct sock *)tp);
1505 return tp->keepalive_intvl ? :
1506 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl);
1509 static inline int keepalive_time_when(const struct tcp_sock *tp)
1511 struct net *net = sock_net((struct sock *)tp);
1513 return tp->keepalive_time ? :
1514 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time);
1517 static inline int keepalive_probes(const struct tcp_sock *tp)
1519 struct net *net = sock_net((struct sock *)tp);
1521 return tp->keepalive_probes ? :
1522 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes);
1525 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1527 const struct inet_connection_sock *icsk = &tp->inet_conn;
1529 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1530 tcp_jiffies32 - tp->rcv_tstamp);
1533 static inline int tcp_fin_time(const struct sock *sk)
1535 int fin_timeout = tcp_sk(sk)->linger2 ? :
1536 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout);
1537 const int rto = inet_csk(sk)->icsk_rto;
1539 if (fin_timeout < (rto << 2) - (rto >> 1))
1540 fin_timeout = (rto << 2) - (rto >> 1);
1545 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1548 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1550 if (unlikely(!time_before32(ktime_get_seconds(),
1551 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1554 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1555 * then following tcp messages have valid values. Ignore 0 value,
1556 * or else 'negative' tsval might forbid us to accept their packets.
1558 if (!rx_opt->ts_recent)
1563 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1566 if (tcp_paws_check(rx_opt, 0))
1569 /* RST segments are not recommended to carry timestamp,
1570 and, if they do, it is recommended to ignore PAWS because
1571 "their cleanup function should take precedence over timestamps."
1572 Certainly, it is mistake. It is necessary to understand the reasons
1573 of this constraint to relax it: if peer reboots, clock may go
1574 out-of-sync and half-open connections will not be reset.
1575 Actually, the problem would be not existing if all
1576 the implementations followed draft about maintaining clock
1577 via reboots. Linux-2.2 DOES NOT!
1579 However, we can relax time bounds for RST segments to MSL.
1581 if (rst && !time_before32(ktime_get_seconds(),
1582 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1587 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1588 int mib_idx, u32 *last_oow_ack_time);
1590 static inline void tcp_mib_init(struct net *net)
1593 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1594 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1595 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1596 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1600 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1602 tp->lost_skb_hint = NULL;
1605 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1607 tcp_clear_retrans_hints_partial(tp);
1608 tp->retransmit_skb_hint = NULL;
1611 union tcp_md5_addr {
1613 #if IS_ENABLED(CONFIG_IPV6)
1618 /* - key database */
1619 struct tcp_md5sig_key {
1620 struct hlist_node node;
1622 u8 family; /* AF_INET or AF_INET6 */
1625 union tcp_md5_addr addr;
1626 int l3index; /* set if key added with L3 scope */
1627 u8 key[TCP_MD5SIG_MAXKEYLEN];
1628 struct rcu_head rcu;
1632 struct tcp_md5sig_info {
1633 struct hlist_head head;
1634 struct rcu_head rcu;
1637 /* - pseudo header */
1638 struct tcp4_pseudohdr {
1646 struct tcp6_pseudohdr {
1647 struct in6_addr saddr;
1648 struct in6_addr daddr;
1650 __be32 protocol; /* including padding */
1653 union tcp_md5sum_block {
1654 struct tcp4_pseudohdr ip4;
1655 #if IS_ENABLED(CONFIG_IPV6)
1656 struct tcp6_pseudohdr ip6;
1660 /* - pool: digest algorithm, hash description and scratch buffer */
1661 struct tcp_md5sig_pool {
1662 struct ahash_request *md5_req;
1667 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1668 const struct sock *sk, const struct sk_buff *skb);
1669 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1670 int family, u8 prefixlen, int l3index, u8 flags,
1671 const u8 *newkey, u8 newkeylen, gfp_t gfp);
1672 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1673 int family, u8 prefixlen, int l3index, u8 flags);
1674 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1675 const struct sock *addr_sk);
1677 #ifdef CONFIG_TCP_MD5SIG
1678 #include <linux/jump_label.h>
1679 extern struct static_key_false tcp_md5_needed;
1680 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1681 const union tcp_md5_addr *addr,
1683 static inline struct tcp_md5sig_key *
1684 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1685 const union tcp_md5_addr *addr, int family)
1687 if (!static_branch_unlikely(&tcp_md5_needed))
1689 return __tcp_md5_do_lookup(sk, l3index, addr, family);
1692 enum skb_drop_reason
1693 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
1694 const void *saddr, const void *daddr,
1695 int family, int dif, int sdif);
1698 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1700 static inline struct tcp_md5sig_key *
1701 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1702 const union tcp_md5_addr *addr, int family)
1707 static inline enum skb_drop_reason
1708 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
1709 const void *saddr, const void *daddr,
1710 int family, int dif, int sdif)
1712 return SKB_NOT_DROPPED_YET;
1714 #define tcp_twsk_md5_key(twsk) NULL
1717 bool tcp_alloc_md5sig_pool(void);
1719 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1720 static inline void tcp_put_md5sig_pool(void)
1725 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1726 unsigned int header_len);
1727 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1728 const struct tcp_md5sig_key *key);
1730 /* From tcp_fastopen.c */
1731 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1732 struct tcp_fastopen_cookie *cookie);
1733 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1734 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1736 struct tcp_fastopen_request {
1737 /* Fast Open cookie. Size 0 means a cookie request */
1738 struct tcp_fastopen_cookie cookie;
1739 struct msghdr *data; /* data in MSG_FASTOPEN */
1741 int copied; /* queued in tcp_connect() */
1742 struct ubuf_info *uarg;
1744 void tcp_free_fastopen_req(struct tcp_sock *tp);
1745 void tcp_fastopen_destroy_cipher(struct sock *sk);
1746 void tcp_fastopen_ctx_destroy(struct net *net);
1747 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1748 void *primary_key, void *backup_key);
1749 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
1751 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1752 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1753 struct request_sock *req,
1754 struct tcp_fastopen_cookie *foc,
1755 const struct dst_entry *dst);
1756 void tcp_fastopen_init_key_once(struct net *net);
1757 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1758 struct tcp_fastopen_cookie *cookie);
1759 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1760 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1761 #define TCP_FASTOPEN_KEY_MAX 2
1762 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1763 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1765 /* Fastopen key context */
1766 struct tcp_fastopen_context {
1767 siphash_key_t key[TCP_FASTOPEN_KEY_MAX];
1769 struct rcu_head rcu;
1772 void tcp_fastopen_active_disable(struct sock *sk);
1773 bool tcp_fastopen_active_should_disable(struct sock *sk);
1774 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1775 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1777 /* Caller needs to wrap with rcu_read_(un)lock() */
1779 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1781 struct tcp_fastopen_context *ctx;
1783 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1785 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1790 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1791 const struct tcp_fastopen_cookie *orig)
1793 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1794 orig->len == foc->len &&
1795 !memcmp(orig->val, foc->val, foc->len))
1801 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1806 /* Latencies incurred by various limits for a sender. They are
1807 * chronograph-like stats that are mutually exclusive.
1811 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1812 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1813 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1817 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1818 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1820 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1821 * the same memory storage than skb->destructor/_skb_refdst
1823 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1825 skb->destructor = NULL;
1826 skb->_skb_refdst = 0UL;
1829 #define tcp_skb_tsorted_save(skb) { \
1830 unsigned long _save = skb->_skb_refdst; \
1831 skb->_skb_refdst = 0UL;
1833 #define tcp_skb_tsorted_restore(skb) \
1834 skb->_skb_refdst = _save; \
1837 void tcp_write_queue_purge(struct sock *sk);
1839 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1841 return skb_rb_first(&sk->tcp_rtx_queue);
1844 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1846 return skb_rb_last(&sk->tcp_rtx_queue);
1849 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1851 return skb_peek_tail(&sk->sk_write_queue);
1854 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1855 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1857 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1859 return skb_peek(&sk->sk_write_queue);
1862 static inline bool tcp_skb_is_last(const struct sock *sk,
1863 const struct sk_buff *skb)
1865 return skb_queue_is_last(&sk->sk_write_queue, skb);
1869 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
1872 * Since the write queue can have a temporary empty skb in it,
1873 * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
1875 static inline bool tcp_write_queue_empty(const struct sock *sk)
1877 const struct tcp_sock *tp = tcp_sk(sk);
1879 return tp->write_seq == tp->snd_nxt;
1882 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1884 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1887 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1889 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1892 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1894 __skb_queue_tail(&sk->sk_write_queue, skb);
1896 /* Queue it, remembering where we must start sending. */
1897 if (sk->sk_write_queue.next == skb)
1898 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1901 /* Insert new before skb on the write queue of sk. */
1902 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1903 struct sk_buff *skb,
1906 __skb_queue_before(&sk->sk_write_queue, skb, new);
1909 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1911 tcp_skb_tsorted_anchor_cleanup(skb);
1912 __skb_unlink(skb, &sk->sk_write_queue);
1915 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1917 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1919 tcp_skb_tsorted_anchor_cleanup(skb);
1920 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1923 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1925 list_del(&skb->tcp_tsorted_anchor);
1926 tcp_rtx_queue_unlink(skb, sk);
1927 tcp_wmem_free_skb(sk, skb);
1930 static inline void tcp_push_pending_frames(struct sock *sk)
1932 if (tcp_send_head(sk)) {
1933 struct tcp_sock *tp = tcp_sk(sk);
1935 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1939 /* Start sequence of the skb just after the highest skb with SACKed
1940 * bit, valid only if sacked_out > 0 or when the caller has ensured
1941 * validity by itself.
1943 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1945 if (!tp->sacked_out)
1948 if (tp->highest_sack == NULL)
1951 return TCP_SKB_CB(tp->highest_sack)->seq;
1954 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1956 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1959 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1961 return tcp_sk(sk)->highest_sack;
1964 static inline void tcp_highest_sack_reset(struct sock *sk)
1966 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1969 /* Called when old skb is about to be deleted and replaced by new skb */
1970 static inline void tcp_highest_sack_replace(struct sock *sk,
1971 struct sk_buff *old,
1972 struct sk_buff *new)
1974 if (old == tcp_highest_sack(sk))
1975 tcp_sk(sk)->highest_sack = new;
1978 /* This helper checks if socket has IP_TRANSPARENT set */
1979 static inline bool inet_sk_transparent(const struct sock *sk)
1981 switch (sk->sk_state) {
1983 return inet_twsk(sk)->tw_transparent;
1984 case TCP_NEW_SYN_RECV:
1985 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1987 return inet_sk(sk)->transparent;
1990 /* Determines whether this is a thin stream (which may suffer from
1991 * increased latency). Used to trigger latency-reducing mechanisms.
1993 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1995 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1999 enum tcp_seq_states {
2000 TCP_SEQ_STATE_LISTENING,
2001 TCP_SEQ_STATE_ESTABLISHED,
2004 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
2005 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
2006 void tcp_seq_stop(struct seq_file *seq, void *v);
2008 struct tcp_seq_afinfo {
2012 struct tcp_iter_state {
2013 struct seq_net_private p;
2014 enum tcp_seq_states state;
2015 struct sock *syn_wait_sk;
2016 int bucket, offset, sbucket, num;
2020 extern struct request_sock_ops tcp_request_sock_ops;
2021 extern struct request_sock_ops tcp6_request_sock_ops;
2023 void tcp_v4_destroy_sock(struct sock *sk);
2025 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
2026 netdev_features_t features);
2027 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
2028 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
2029 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
2030 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff));
2031 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb));
2032 int tcp_gro_complete(struct sk_buff *skb);
2034 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
2036 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
2038 struct net *net = sock_net((struct sock *)tp);
2039 return tp->notsent_lowat ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
2042 bool tcp_stream_memory_free(const struct sock *sk, int wake);
2044 #ifdef CONFIG_PROC_FS
2045 int tcp4_proc_init(void);
2046 void tcp4_proc_exit(void);
2049 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
2050 int tcp_conn_request(struct request_sock_ops *rsk_ops,
2051 const struct tcp_request_sock_ops *af_ops,
2052 struct sock *sk, struct sk_buff *skb);
2054 /* TCP af-specific functions */
2055 struct tcp_sock_af_ops {
2056 #ifdef CONFIG_TCP_MD5SIG
2057 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
2058 const struct sock *addr_sk);
2059 int (*calc_md5_hash)(char *location,
2060 const struct tcp_md5sig_key *md5,
2061 const struct sock *sk,
2062 const struct sk_buff *skb);
2063 int (*md5_parse)(struct sock *sk,
2070 struct tcp_request_sock_ops {
2072 #ifdef CONFIG_TCP_MD5SIG
2073 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
2074 const struct sock *addr_sk);
2075 int (*calc_md5_hash) (char *location,
2076 const struct tcp_md5sig_key *md5,
2077 const struct sock *sk,
2078 const struct sk_buff *skb);
2080 #ifdef CONFIG_SYN_COOKIES
2081 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
2084 struct dst_entry *(*route_req)(const struct sock *sk,
2085 struct sk_buff *skb,
2087 struct request_sock *req);
2088 u32 (*init_seq)(const struct sk_buff *skb);
2089 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
2090 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2091 struct flowi *fl, struct request_sock *req,
2092 struct tcp_fastopen_cookie *foc,
2093 enum tcp_synack_type synack_type,
2094 struct sk_buff *syn_skb);
2097 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2098 #if IS_ENABLED(CONFIG_IPV6)
2099 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2102 #ifdef CONFIG_SYN_COOKIES
2103 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2104 const struct sock *sk, struct sk_buff *skb,
2107 tcp_synq_overflow(sk);
2108 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2109 return ops->cookie_init_seq(skb, mss);
2112 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2113 const struct sock *sk, struct sk_buff *skb,
2120 int tcpv4_offload_init(void);
2122 void tcp_v4_init(void);
2123 void tcp_init(void);
2125 /* tcp_recovery.c */
2126 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2127 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2128 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2130 extern bool tcp_rack_mark_lost(struct sock *sk);
2131 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2133 extern void tcp_rack_reo_timeout(struct sock *sk);
2134 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2136 /* At how many usecs into the future should the RTO fire? */
2137 static inline s64 tcp_rto_delta_us(const struct sock *sk)
2139 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2140 u32 rto = inet_csk(sk)->icsk_rto;
2141 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2143 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2147 * Save and compile IPv4 options, return a pointer to it
2149 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2150 struct sk_buff *skb)
2152 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2153 struct ip_options_rcu *dopt = NULL;
2156 int opt_size = sizeof(*dopt) + opt->optlen;
2158 dopt = kmalloc(opt_size, GFP_ATOMIC);
2159 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2167 /* locally generated TCP pure ACKs have skb->truesize == 2
2168 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2169 * This is much faster than dissecting the packet to find out.
2170 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2172 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2174 return skb->truesize == 2;
2177 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2182 static inline int tcp_inq(struct sock *sk)
2184 struct tcp_sock *tp = tcp_sk(sk);
2187 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2189 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2191 before(tp->urg_seq, tp->copied_seq) ||
2192 !before(tp->urg_seq, tp->rcv_nxt)) {
2194 answ = tp->rcv_nxt - tp->copied_seq;
2196 /* Subtract 1, if FIN was received */
2197 if (answ && sock_flag(sk, SOCK_DONE))
2200 answ = tp->urg_seq - tp->copied_seq;
2206 int tcp_peek_len(struct socket *sock);
2208 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2212 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2214 /* We update these fields while other threads might
2215 * read them from tcp_get_info()
2217 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in);
2218 if (skb->len > tcp_hdrlen(skb))
2219 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in);
2223 * TCP listen path runs lockless.
2224 * We forced "struct sock" to be const qualified to make sure
2225 * we don't modify one of its field by mistake.
2226 * Here, we increment sk_drops which is an atomic_t, so we can safely
2227 * make sock writable again.
2229 static inline void tcp_listendrop(const struct sock *sk)
2231 atomic_inc(&((struct sock *)sk)->sk_drops);
2232 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2235 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2238 * Interface for adding Upper Level Protocols over TCP
2241 #define TCP_ULP_NAME_MAX 16
2242 #define TCP_ULP_MAX 128
2243 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2245 struct tcp_ulp_ops {
2246 struct list_head list;
2248 /* initialize ulp */
2249 int (*init)(struct sock *sk);
2251 void (*update)(struct sock *sk, struct proto *p,
2252 void (*write_space)(struct sock *sk));
2254 void (*release)(struct sock *sk);
2256 int (*get_info)(const struct sock *sk, struct sk_buff *skb);
2257 size_t (*get_info_size)(const struct sock *sk);
2259 void (*clone)(const struct request_sock *req, struct sock *newsk,
2260 const gfp_t priority);
2262 char name[TCP_ULP_NAME_MAX];
2263 struct module *owner;
2265 int tcp_register_ulp(struct tcp_ulp_ops *type);
2266 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2267 int tcp_set_ulp(struct sock *sk, const char *name);
2268 void tcp_get_available_ulp(char *buf, size_t len);
2269 void tcp_cleanup_ulp(struct sock *sk);
2270 void tcp_update_ulp(struct sock *sk, struct proto *p,
2271 void (*write_space)(struct sock *sk));
2273 #define MODULE_ALIAS_TCP_ULP(name) \
2274 __MODULE_INFO(alias, alias_userspace, name); \
2275 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2277 #ifdef CONFIG_NET_SOCK_MSG
2281 #ifdef CONFIG_BPF_SYSCALL
2282 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
2283 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
2284 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2285 #endif /* CONFIG_BPF_SYSCALL */
2287 int tcp_bpf_sendmsg_redir(struct sock *sk, struct sk_msg *msg, u32 bytes,
2289 #endif /* CONFIG_NET_SOCK_MSG */
2291 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG)
2292 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2297 #ifdef CONFIG_CGROUP_BPF
2298 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2299 struct sk_buff *skb,
2300 unsigned int end_offset)
2303 skops->skb_data_end = skb->data + end_offset;
2306 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2307 struct sk_buff *skb,
2308 unsigned int end_offset)
2313 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2314 * is < 0, then the BPF op failed (for example if the loaded BPF
2315 * program does not support the chosen operation or there is no BPF
2319 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2321 struct bpf_sock_ops_kern sock_ops;
2324 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2325 if (sk_fullsock(sk)) {
2326 sock_ops.is_fullsock = 1;
2327 sock_owned_by_me(sk);
2333 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2335 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2337 ret = sock_ops.reply;
2343 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2345 u32 args[2] = {arg1, arg2};
2347 return tcp_call_bpf(sk, op, 2, args);
2350 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2353 u32 args[3] = {arg1, arg2, arg3};
2355 return tcp_call_bpf(sk, op, 3, args);
2359 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2364 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2369 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2377 static inline u32 tcp_timeout_init(struct sock *sk)
2381 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2384 timeout = TCP_TIMEOUT_INIT;
2385 return min_t(int, timeout, TCP_RTO_MAX);
2388 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2392 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2399 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2401 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2404 static inline void tcp_bpf_rtt(struct sock *sk)
2406 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2407 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2410 #if IS_ENABLED(CONFIG_SMC)
2411 extern struct static_key_false tcp_have_smc;
2414 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2415 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2416 void (*cad)(struct sock *sk, u32 ack_seq));
2417 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2418 void clean_acked_data_flush(void);
2421 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2422 static inline void tcp_add_tx_delay(struct sk_buff *skb,
2423 const struct tcp_sock *tp)
2425 if (static_branch_unlikely(&tcp_tx_delay_enabled))
2426 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2429 /* Compute Earliest Departure Time for some control packets
2430 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2432 static inline u64 tcp_transmit_time(const struct sock *sk)
2434 if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2435 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2436 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2438 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;