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_sendmsg_fastopen(struct sock *sk, struct msghdr *msg, int *copied,
331 size_t size, struct ubuf_info *uarg);
332 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
334 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
335 size_t size, int flags);
336 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
337 size_t size, int flags);
338 int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
339 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
341 void tcp_release_cb(struct sock *sk);
342 void tcp_wfree(struct sk_buff *skb);
343 void tcp_write_timer_handler(struct sock *sk);
344 void tcp_delack_timer_handler(struct sock *sk);
345 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
346 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
347 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
348 void tcp_rcv_space_adjust(struct sock *sk);
349 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
350 void tcp_twsk_destructor(struct sock *sk);
351 void tcp_twsk_purge(struct list_head *net_exit_list, int family);
352 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
353 struct pipe_inode_info *pipe, size_t len,
355 struct sk_buff *tcp_stream_alloc_skb(struct sock *sk, int size, gfp_t gfp,
356 bool force_schedule);
358 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
359 static inline void tcp_dec_quickack_mode(struct sock *sk,
360 const unsigned int pkts)
362 struct inet_connection_sock *icsk = inet_csk(sk);
364 if (icsk->icsk_ack.quick) {
365 if (pkts >= icsk->icsk_ack.quick) {
366 icsk->icsk_ack.quick = 0;
367 /* Leaving quickack mode we deflate ATO. */
368 icsk->icsk_ack.ato = TCP_ATO_MIN;
370 icsk->icsk_ack.quick -= pkts;
375 #define TCP_ECN_QUEUE_CWR 2
376 #define TCP_ECN_DEMAND_CWR 4
377 #define TCP_ECN_SEEN 8
387 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
389 const struct tcphdr *th);
390 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
391 struct request_sock *req, bool fastopen,
393 int tcp_child_process(struct sock *parent, struct sock *child,
394 struct sk_buff *skb);
395 void tcp_enter_loss(struct sock *sk);
396 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag);
397 void tcp_clear_retrans(struct tcp_sock *tp);
398 void tcp_update_metrics(struct sock *sk);
399 void tcp_init_metrics(struct sock *sk);
400 void tcp_metrics_init(void);
401 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
402 void __tcp_close(struct sock *sk, long timeout);
403 void tcp_close(struct sock *sk, long timeout);
404 void tcp_init_sock(struct sock *sk);
405 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb);
406 __poll_t tcp_poll(struct file *file, struct socket *sock,
407 struct poll_table_struct *wait);
408 int do_tcp_getsockopt(struct sock *sk, int level,
409 int optname, sockptr_t optval, sockptr_t optlen);
410 int tcp_getsockopt(struct sock *sk, int level, int optname,
411 char __user *optval, int __user *optlen);
412 bool tcp_bpf_bypass_getsockopt(int level, int optname);
413 int do_tcp_setsockopt(struct sock *sk, int level, int optname,
414 sockptr_t optval, unsigned int optlen);
415 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
416 unsigned int optlen);
417 void tcp_set_keepalive(struct sock *sk, int val);
418 void tcp_syn_ack_timeout(const struct request_sock *req);
419 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
420 int flags, int *addr_len);
421 int tcp_set_rcvlowat(struct sock *sk, int val);
422 int tcp_set_window_clamp(struct sock *sk, int val);
423 void tcp_update_recv_tstamps(struct sk_buff *skb,
424 struct scm_timestamping_internal *tss);
425 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
426 struct scm_timestamping_internal *tss);
427 void tcp_data_ready(struct sock *sk);
429 int tcp_mmap(struct file *file, struct socket *sock,
430 struct vm_area_struct *vma);
432 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
433 struct tcp_options_received *opt_rx,
434 int estab, struct tcp_fastopen_cookie *foc);
435 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
438 * BPF SKB-less helpers
440 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
441 struct tcphdr *th, u32 *cookie);
442 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
443 struct tcphdr *th, u32 *cookie);
444 u16 tcp_parse_mss_option(const struct tcphdr *th, u16 user_mss);
445 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
446 const struct tcp_request_sock_ops *af_ops,
447 struct sock *sk, struct tcphdr *th);
449 * TCP v4 functions exported for the inet6 API
452 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
453 void tcp_v4_mtu_reduced(struct sock *sk);
454 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
455 void tcp_ld_RTO_revert(struct sock *sk, u32 seq);
456 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
457 struct sock *tcp_create_openreq_child(const struct sock *sk,
458 struct request_sock *req,
459 struct sk_buff *skb);
460 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
461 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
462 struct request_sock *req,
463 struct dst_entry *dst,
464 struct request_sock *req_unhash,
466 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
467 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
468 int tcp_connect(struct sock *sk);
469 enum tcp_synack_type {
474 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
475 struct request_sock *req,
476 struct tcp_fastopen_cookie *foc,
477 enum tcp_synack_type synack_type,
478 struct sk_buff *syn_skb);
479 int tcp_disconnect(struct sock *sk, int flags);
481 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
482 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
483 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
485 /* From syncookies.c */
486 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
487 struct request_sock *req,
488 struct dst_entry *dst, u32 tsoff);
489 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
491 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
492 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
493 const struct tcp_request_sock_ops *af_ops,
494 struct sock *sk, struct sk_buff *skb);
495 #ifdef CONFIG_SYN_COOKIES
497 /* Syncookies use a monotonic timer which increments every 60 seconds.
498 * This counter is used both as a hash input and partially encoded into
499 * the cookie value. A cookie is only validated further if the delta
500 * between the current counter value and the encoded one is less than this,
501 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
502 * the counter advances immediately after a cookie is generated).
504 #define MAX_SYNCOOKIE_AGE 2
505 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
506 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
508 /* syncookies: remember time of last synqueue overflow
509 * But do not dirty this field too often (once per second is enough)
510 * It is racy as we do not hold a lock, but race is very minor.
512 static inline void tcp_synq_overflow(const struct sock *sk)
514 unsigned int last_overflow;
515 unsigned int now = jiffies;
517 if (sk->sk_reuseport) {
518 struct sock_reuseport *reuse;
520 reuse = rcu_dereference(sk->sk_reuseport_cb);
522 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
523 if (!time_between32(now, last_overflow,
525 WRITE_ONCE(reuse->synq_overflow_ts, now);
530 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
531 if (!time_between32(now, last_overflow, last_overflow + HZ))
532 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
535 /* syncookies: no recent synqueue overflow on this listening socket? */
536 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
538 unsigned int last_overflow;
539 unsigned int now = jiffies;
541 if (sk->sk_reuseport) {
542 struct sock_reuseport *reuse;
544 reuse = rcu_dereference(sk->sk_reuseport_cb);
546 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
547 return !time_between32(now, last_overflow - HZ,
549 TCP_SYNCOOKIE_VALID);
553 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
555 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
556 * then we're under synflood. However, we have to use
557 * 'last_overflow - HZ' as lower bound. That's because a concurrent
558 * tcp_synq_overflow() could update .ts_recent_stamp after we read
559 * jiffies but before we store .ts_recent_stamp into last_overflow,
560 * which could lead to rejecting a valid syncookie.
562 return !time_between32(now, last_overflow - HZ,
563 last_overflow + TCP_SYNCOOKIE_VALID);
566 static inline u32 tcp_cookie_time(void)
568 u64 val = get_jiffies_64();
570 do_div(val, TCP_SYNCOOKIE_PERIOD);
574 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
576 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
577 u64 cookie_init_timestamp(struct request_sock *req, u64 now);
578 bool cookie_timestamp_decode(const struct net *net,
579 struct tcp_options_received *opt);
580 bool cookie_ecn_ok(const struct tcp_options_received *opt,
581 const struct net *net, const struct dst_entry *dst);
583 /* From net/ipv6/syncookies.c */
584 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
586 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
588 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
589 const struct tcphdr *th, u16 *mssp);
590 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
594 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb);
595 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb);
596 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
598 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
599 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
600 void tcp_retransmit_timer(struct sock *sk);
601 void tcp_xmit_retransmit_queue(struct sock *);
602 void tcp_simple_retransmit(struct sock *);
603 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
604 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
606 TCP_FRAG_IN_WRITE_QUEUE,
607 TCP_FRAG_IN_RTX_QUEUE,
609 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
610 struct sk_buff *skb, u32 len,
611 unsigned int mss_now, gfp_t gfp);
613 void tcp_send_probe0(struct sock *);
614 void tcp_send_partial(struct sock *);
615 int tcp_write_wakeup(struct sock *, int mib);
616 void tcp_send_fin(struct sock *sk);
617 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
618 int tcp_send_synack(struct sock *);
619 void tcp_push_one(struct sock *, unsigned int mss_now);
620 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
621 void tcp_send_ack(struct sock *sk);
622 void tcp_send_delayed_ack(struct sock *sk);
623 void tcp_send_loss_probe(struct sock *sk);
624 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
625 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
626 const struct sk_buff *next_skb);
629 void tcp_rearm_rto(struct sock *sk);
630 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
631 void tcp_reset(struct sock *sk, struct sk_buff *skb);
632 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
633 void tcp_fin(struct sock *sk);
634 void tcp_check_space(struct sock *sk);
637 void tcp_init_xmit_timers(struct sock *);
638 static inline void tcp_clear_xmit_timers(struct sock *sk)
640 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
643 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
646 inet_csk_clear_xmit_timers(sk);
649 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
650 unsigned int tcp_current_mss(struct sock *sk);
651 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when);
653 /* Bound MSS / TSO packet size with the half of the window */
654 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
658 /* When peer uses tiny windows, there is no use in packetizing
659 * to sub-MSS pieces for the sake of SWS or making sure there
660 * are enough packets in the pipe for fast recovery.
662 * On the other hand, for extremely large MSS devices, handling
663 * smaller than MSS windows in this way does make sense.
665 if (tp->max_window > TCP_MSS_DEFAULT)
666 cutoff = (tp->max_window >> 1);
668 cutoff = tp->max_window;
670 if (cutoff && pktsize > cutoff)
671 return max_t(int, cutoff, 68U - tp->tcp_header_len);
677 void tcp_get_info(struct sock *, struct tcp_info *);
679 /* Read 'sendfile()'-style from a TCP socket */
680 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
681 sk_read_actor_t recv_actor);
682 int tcp_read_skb(struct sock *sk, skb_read_actor_t recv_actor);
683 struct sk_buff *tcp_recv_skb(struct sock *sk, u32 seq, u32 *off);
684 void tcp_read_done(struct sock *sk, size_t len);
686 void tcp_initialize_rcv_mss(struct sock *sk);
688 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
689 int tcp_mss_to_mtu(struct sock *sk, int mss);
690 void tcp_mtup_init(struct sock *sk);
692 static inline void tcp_bound_rto(const struct sock *sk)
694 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
695 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
698 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
700 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
703 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
705 /* mptcp hooks are only on the slow path */
706 if (sk_is_mptcp((struct sock *)tp))
709 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
710 ntohl(TCP_FLAG_ACK) |
714 static inline void tcp_fast_path_on(struct tcp_sock *tp)
716 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
719 static inline void tcp_fast_path_check(struct sock *sk)
721 struct tcp_sock *tp = tcp_sk(sk);
723 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
725 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
727 tcp_fast_path_on(tp);
730 /* Compute the actual rto_min value */
731 static inline u32 tcp_rto_min(struct sock *sk)
733 const struct dst_entry *dst = __sk_dst_get(sk);
734 u32 rto_min = inet_csk(sk)->icsk_rto_min;
736 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
737 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
741 static inline u32 tcp_rto_min_us(struct sock *sk)
743 return jiffies_to_usecs(tcp_rto_min(sk));
746 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
748 return dst_metric_locked(dst, RTAX_CC_ALGO);
751 /* Minimum RTT in usec. ~0 means not available. */
752 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
754 return minmax_get(&tp->rtt_min);
757 /* Compute the actual receive window we are currently advertising.
758 * Rcv_nxt can be after the window if our peer push more data
759 * than the offered window.
761 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
763 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
770 /* Choose a new window, without checks for shrinking, and without
771 * scaling applied to the result. The caller does these things
772 * if necessary. This is a "raw" window selection.
774 u32 __tcp_select_window(struct sock *sk);
776 void tcp_send_window_probe(struct sock *sk);
778 /* TCP uses 32bit jiffies to save some space.
779 * Note that this is different from tcp_time_stamp, which
780 * historically has been the same until linux-4.13.
782 #define tcp_jiffies32 ((u32)jiffies)
785 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
786 * It is no longer tied to jiffies, but to 1 ms clock.
787 * Note: double check if you want to use tcp_jiffies32 instead of this.
789 #define TCP_TS_HZ 1000
791 static inline u64 tcp_clock_ns(void)
793 return ktime_get_ns();
796 static inline u64 tcp_clock_us(void)
798 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
801 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
802 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
804 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
807 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
808 static inline u32 tcp_ns_to_ts(u64 ns)
810 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
813 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
814 static inline u32 tcp_time_stamp_raw(void)
816 return tcp_ns_to_ts(tcp_clock_ns());
819 void tcp_mstamp_refresh(struct tcp_sock *tp);
821 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
823 return max_t(s64, t1 - t0, 0);
826 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
828 return tcp_ns_to_ts(skb->skb_mstamp_ns);
831 /* provide the departure time in us unit */
832 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
834 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
838 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
840 #define TCPHDR_FIN 0x01
841 #define TCPHDR_SYN 0x02
842 #define TCPHDR_RST 0x04
843 #define TCPHDR_PSH 0x08
844 #define TCPHDR_ACK 0x10
845 #define TCPHDR_URG 0x20
846 #define TCPHDR_ECE 0x40
847 #define TCPHDR_CWR 0x80
849 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
851 /* This is what the send packet queuing engine uses to pass
852 * TCP per-packet control information to the transmission code.
853 * We also store the host-order sequence numbers in here too.
854 * This is 44 bytes if IPV6 is enabled.
855 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
858 __u32 seq; /* Starting sequence number */
859 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
861 /* Note : tcp_tw_isn is used in input path only
862 * (isn chosen by tcp_timewait_state_process())
864 * tcp_gso_segs/size are used in write queue only,
865 * cf tcp_skb_pcount()/tcp_skb_mss()
873 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
875 __u8 sacked; /* State flags for SACK. */
876 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
877 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
878 #define TCPCB_LOST 0x04 /* SKB is lost */
879 #define TCPCB_TAGBITS 0x07 /* All tag bits */
880 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
881 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
882 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
885 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
886 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
887 eor:1, /* Is skb MSG_EOR marked? */
888 has_rxtstamp:1, /* SKB has a RX timestamp */
890 __u32 ack_seq; /* Sequence number ACK'd */
893 #define TCPCB_DELIVERED_CE_MASK ((1U<<20) - 1)
894 /* There is space for up to 24 bytes */
895 __u32 is_app_limited:1, /* cwnd not fully used? */
898 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
900 /* start of send pipeline phase */
902 /* when we reached the "delivered" count */
903 u64 delivered_mstamp;
904 } tx; /* only used for outgoing skbs */
906 struct inet_skb_parm h4;
907 #if IS_ENABLED(CONFIG_IPV6)
908 struct inet6_skb_parm h6;
910 } header; /* For incoming skbs */
914 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
916 extern const struct inet_connection_sock_af_ops ipv4_specific;
918 #if IS_ENABLED(CONFIG_IPV6)
919 /* This is the variant of inet6_iif() that must be used by TCP,
920 * as TCP moves IP6CB into a different location in skb->cb[]
922 static inline int tcp_v6_iif(const struct sk_buff *skb)
924 return TCP_SKB_CB(skb)->header.h6.iif;
927 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
929 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
931 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
934 /* TCP_SKB_CB reference means this can not be used from early demux */
935 static inline int tcp_v6_sdif(const struct sk_buff *skb)
937 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
938 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
939 return TCP_SKB_CB(skb)->header.h6.iif;
944 extern const struct inet_connection_sock_af_ops ipv6_specific;
946 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
947 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
948 void tcp_v6_early_demux(struct sk_buff *skb);
952 /* TCP_SKB_CB reference means this can not be used from early demux */
953 static inline int tcp_v4_sdif(struct sk_buff *skb)
955 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
956 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
957 return TCP_SKB_CB(skb)->header.h4.iif;
962 /* Due to TSO, an SKB can be composed of multiple actual
963 * packets. To keep these tracked properly, we use this.
965 static inline int tcp_skb_pcount(const struct sk_buff *skb)
967 return TCP_SKB_CB(skb)->tcp_gso_segs;
970 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
972 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
975 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
977 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
980 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
981 static inline int tcp_skb_mss(const struct sk_buff *skb)
983 return TCP_SKB_CB(skb)->tcp_gso_size;
986 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
988 return likely(!TCP_SKB_CB(skb)->eor);
991 static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
992 const struct sk_buff *from)
994 return likely(tcp_skb_can_collapse_to(to) &&
995 mptcp_skb_can_collapse(to, from) &&
996 skb_pure_zcopy_same(to, from));
999 /* Events passed to congestion control interface */
1001 CA_EVENT_TX_START, /* first transmit when no packets in flight */
1002 CA_EVENT_CWND_RESTART, /* congestion window restart */
1003 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
1004 CA_EVENT_LOSS, /* loss timeout */
1005 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
1006 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
1009 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
1010 enum tcp_ca_ack_event_flags {
1011 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
1012 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
1013 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
1017 * Interface for adding new TCP congestion control handlers
1019 #define TCP_CA_NAME_MAX 16
1020 #define TCP_CA_MAX 128
1021 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
1023 #define TCP_CA_UNSPEC 0
1025 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1026 #define TCP_CONG_NON_RESTRICTED 0x1
1027 /* Requires ECN/ECT set on all packets */
1028 #define TCP_CONG_NEEDS_ECN 0x2
1029 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1039 /* A rate sample measures the number of (original/retransmitted) data
1040 * packets delivered "delivered" over an interval of time "interval_us".
1041 * The tcp_rate.c code fills in the rate sample, and congestion
1042 * control modules that define a cong_control function to run at the end
1043 * of ACK processing can optionally chose to consult this sample when
1044 * setting cwnd and pacing rate.
1045 * A sample is invalid if "delivered" or "interval_us" is negative.
1047 struct rate_sample {
1048 u64 prior_mstamp; /* starting timestamp for interval */
1049 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1050 u32 prior_delivered_ce;/* tp->delivered_ce at "prior_mstamp" */
1051 s32 delivered; /* number of packets delivered over interval */
1052 s32 delivered_ce; /* number of packets delivered w/ CE marks*/
1053 long interval_us; /* time for tp->delivered to incr "delivered" */
1054 u32 snd_interval_us; /* snd interval for delivered packets */
1055 u32 rcv_interval_us; /* rcv interval for delivered packets */
1056 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1057 int losses; /* number of packets marked lost upon ACK */
1058 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1059 u32 prior_in_flight; /* in flight before this ACK */
1060 u32 last_end_seq; /* end_seq of most recently ACKed packet */
1061 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1062 bool is_retrans; /* is sample from retransmission? */
1063 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1066 struct tcp_congestion_ops {
1067 /* fast path fields are put first to fill one cache line */
1069 /* return slow start threshold (required) */
1070 u32 (*ssthresh)(struct sock *sk);
1072 /* do new cwnd calculation (required) */
1073 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1075 /* call before changing ca_state (optional) */
1076 void (*set_state)(struct sock *sk, u8 new_state);
1078 /* call when cwnd event occurs (optional) */
1079 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1081 /* call when ack arrives (optional) */
1082 void (*in_ack_event)(struct sock *sk, u32 flags);
1084 /* hook for packet ack accounting (optional) */
1085 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1087 /* override sysctl_tcp_min_tso_segs */
1088 u32 (*min_tso_segs)(struct sock *sk);
1090 /* call when packets are delivered to update cwnd and pacing rate,
1091 * after all the ca_state processing. (optional)
1093 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1096 /* new value of cwnd after loss (required) */
1097 u32 (*undo_cwnd)(struct sock *sk);
1098 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1099 u32 (*sndbuf_expand)(struct sock *sk);
1101 /* control/slow paths put last */
1102 /* get info for inet_diag (optional) */
1103 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1104 union tcp_cc_info *info);
1106 char name[TCP_CA_NAME_MAX];
1107 struct module *owner;
1108 struct list_head list;
1112 /* initialize private data (optional) */
1113 void (*init)(struct sock *sk);
1114 /* cleanup private data (optional) */
1115 void (*release)(struct sock *sk);
1116 } ____cacheline_aligned_in_smp;
1118 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1119 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1121 void tcp_assign_congestion_control(struct sock *sk);
1122 void tcp_init_congestion_control(struct sock *sk);
1123 void tcp_cleanup_congestion_control(struct sock *sk);
1124 int tcp_set_default_congestion_control(struct net *net, const char *name);
1125 void tcp_get_default_congestion_control(struct net *net, char *name);
1126 void tcp_get_available_congestion_control(char *buf, size_t len);
1127 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1128 int tcp_set_allowed_congestion_control(char *allowed);
1129 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1130 bool cap_net_admin);
1131 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1132 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1134 u32 tcp_reno_ssthresh(struct sock *sk);
1135 u32 tcp_reno_undo_cwnd(struct sock *sk);
1136 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1137 extern struct tcp_congestion_ops tcp_reno;
1139 struct tcp_congestion_ops *tcp_ca_find(const char *name);
1140 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1141 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1143 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1145 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1151 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1153 const struct inet_connection_sock *icsk = inet_csk(sk);
1155 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1158 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1160 const struct inet_connection_sock *icsk = inet_csk(sk);
1162 if (icsk->icsk_ca_ops->cwnd_event)
1163 icsk->icsk_ca_ops->cwnd_event(sk, event);
1166 /* From tcp_cong.c */
1167 void tcp_set_ca_state(struct sock *sk, const u8 ca_state);
1169 /* From tcp_rate.c */
1170 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1171 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1172 struct rate_sample *rs);
1173 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1174 bool is_sack_reneg, struct rate_sample *rs);
1175 void tcp_rate_check_app_limited(struct sock *sk);
1177 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2)
1179 return t1 > t2 || (t1 == t2 && after(seq1, seq2));
1182 /* These functions determine how the current flow behaves in respect of SACK
1183 * handling. SACK is negotiated with the peer, and therefore it can vary
1184 * between different flows.
1186 * tcp_is_sack - SACK enabled
1187 * tcp_is_reno - No SACK
1189 static inline int tcp_is_sack(const struct tcp_sock *tp)
1191 return likely(tp->rx_opt.sack_ok);
1194 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1196 return !tcp_is_sack(tp);
1199 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1201 return tp->sacked_out + tp->lost_out;
1204 /* This determines how many packets are "in the network" to the best
1205 * of our knowledge. In many cases it is conservative, but where
1206 * detailed information is available from the receiver (via SACK
1207 * blocks etc.) we can make more aggressive calculations.
1209 * Use this for decisions involving congestion control, use just
1210 * tp->packets_out to determine if the send queue is empty or not.
1212 * Read this equation as:
1214 * "Packets sent once on transmission queue" MINUS
1215 * "Packets left network, but not honestly ACKed yet" PLUS
1216 * "Packets fast retransmitted"
1218 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1220 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1223 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1225 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp)
1227 return tp->snd_cwnd;
1230 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val)
1232 WARN_ON_ONCE((int)val <= 0);
1236 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1238 return tcp_snd_cwnd(tp) < tp->snd_ssthresh;
1241 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1243 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1246 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1248 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1249 (1 << inet_csk(sk)->icsk_ca_state);
1252 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1253 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1256 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1258 const struct tcp_sock *tp = tcp_sk(sk);
1260 if (tcp_in_cwnd_reduction(sk))
1261 return tp->snd_ssthresh;
1263 return max(tp->snd_ssthresh,
1264 ((tcp_snd_cwnd(tp) >> 1) +
1265 (tcp_snd_cwnd(tp) >> 2)));
1268 /* Use define here intentionally to get WARN_ON location shown at the caller */
1269 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1271 void tcp_enter_cwr(struct sock *sk);
1272 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1274 /* The maximum number of MSS of available cwnd for which TSO defers
1275 * sending if not using sysctl_tcp_tso_win_divisor.
1277 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1282 /* Returns end sequence number of the receiver's advertised window */
1283 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1285 return tp->snd_una + tp->snd_wnd;
1288 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1289 * flexible approach. The RFC suggests cwnd should not be raised unless
1290 * it was fully used previously. And that's exactly what we do in
1291 * congestion avoidance mode. But in slow start we allow cwnd to grow
1292 * as long as the application has used half the cwnd.
1294 * cwnd is 10 (IW10), but application sends 9 frames.
1295 * We allow cwnd to reach 18 when all frames are ACKed.
1296 * This check is safe because it's as aggressive as slow start which already
1297 * risks 100% overshoot. The advantage is that we discourage application to
1298 * either send more filler packets or data to artificially blow up the cwnd
1299 * usage, and allow application-limited process to probe bw more aggressively.
1301 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1303 const struct tcp_sock *tp = tcp_sk(sk);
1305 if (tp->is_cwnd_limited)
1308 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1309 if (tcp_in_slow_start(tp))
1310 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out;
1315 /* BBR congestion control needs pacing.
1316 * Same remark for SO_MAX_PACING_RATE.
1317 * sch_fq packet scheduler is efficiently handling pacing,
1318 * but is not always installed/used.
1319 * Return true if TCP stack should pace packets itself.
1321 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1323 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1326 /* Estimates in how many jiffies next packet for this flow can be sent.
1327 * Scheduling a retransmit timer too early would be silly.
1329 static inline unsigned long tcp_pacing_delay(const struct sock *sk)
1331 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
1333 return delay > 0 ? nsecs_to_jiffies(delay) : 0;
1336 static inline void tcp_reset_xmit_timer(struct sock *sk,
1339 const unsigned long max_when)
1341 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk),
1345 /* Something is really bad, we could not queue an additional packet,
1346 * because qdisc is full or receiver sent a 0 window, or we are paced.
1347 * We do not want to add fuel to the fire, or abort too early,
1348 * so make sure the timer we arm now is at least 200ms in the future,
1349 * regardless of current icsk_rto value (as it could be ~2ms)
1351 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1353 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1356 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1357 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1358 unsigned long max_when)
1360 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1,
1361 inet_csk(sk)->icsk_backoff);
1362 u64 when = (u64)tcp_probe0_base(sk) << backoff;
1364 return (unsigned long)min_t(u64, when, max_when);
1367 static inline void tcp_check_probe_timer(struct sock *sk)
1369 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1370 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1371 tcp_probe0_base(sk), TCP_RTO_MAX);
1374 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1379 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1385 * Calculate(/check) TCP checksum
1387 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1388 __be32 daddr, __wsum base)
1390 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1393 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1395 return !skb_csum_unnecessary(skb) &&
1396 __skb_checksum_complete(skb);
1399 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb,
1400 enum skb_drop_reason *reason);
1403 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1404 void tcp_set_state(struct sock *sk, int state);
1405 void tcp_done(struct sock *sk);
1406 int tcp_abort(struct sock *sk, int err);
1408 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1411 rx_opt->num_sacks = 0;
1414 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1416 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1418 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1419 struct tcp_sock *tp = tcp_sk(sk);
1422 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) ||
1423 tp->packets_out || ca_ops->cong_control)
1425 delta = tcp_jiffies32 - tp->lsndtime;
1426 if (delta > inet_csk(sk)->icsk_rto)
1427 tcp_cwnd_restart(sk, delta);
1430 /* Determine a window scaling and initial window to offer. */
1431 void tcp_select_initial_window(const struct sock *sk, int __space,
1432 __u32 mss, __u32 *rcv_wnd,
1433 __u32 *window_clamp, int wscale_ok,
1434 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1436 static inline int tcp_win_from_space(const struct sock *sk, int space)
1438 int tcp_adv_win_scale = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale);
1440 return tcp_adv_win_scale <= 0 ?
1441 (space>>(-tcp_adv_win_scale)) :
1442 space - (space>>tcp_adv_win_scale);
1445 /* Note: caller must be prepared to deal with negative returns */
1446 static inline int tcp_space(const struct sock *sk)
1448 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1449 READ_ONCE(sk->sk_backlog.len) -
1450 atomic_read(&sk->sk_rmem_alloc));
1453 static inline int tcp_full_space(const struct sock *sk)
1455 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1458 static inline void tcp_adjust_rcv_ssthresh(struct sock *sk)
1460 int unused_mem = sk_unused_reserved_mem(sk);
1461 struct tcp_sock *tp = tcp_sk(sk);
1463 tp->rcv_ssthresh = min(tp->rcv_ssthresh, 4U * tp->advmss);
1465 tp->rcv_ssthresh = max_t(u32, tp->rcv_ssthresh,
1466 tcp_win_from_space(sk, unused_mem));
1469 void tcp_cleanup_rbuf(struct sock *sk, int copied);
1471 /* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1472 * If 87.5 % (7/8) of the space has been consumed, we want to override
1473 * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1474 * len/truesize ratio.
1476 static inline bool tcp_rmem_pressure(const struct sock *sk)
1478 int rcvbuf, threshold;
1480 if (tcp_under_memory_pressure(sk))
1483 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1484 threshold = rcvbuf - (rcvbuf >> 3);
1486 return atomic_read(&sk->sk_rmem_alloc) > threshold;
1489 static inline bool tcp_epollin_ready(const struct sock *sk, int target)
1491 const struct tcp_sock *tp = tcp_sk(sk);
1492 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq);
1497 return (avail >= target) || tcp_rmem_pressure(sk) ||
1498 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss);
1501 extern void tcp_openreq_init_rwin(struct request_sock *req,
1502 const struct sock *sk_listener,
1503 const struct dst_entry *dst);
1505 void tcp_enter_memory_pressure(struct sock *sk);
1506 void tcp_leave_memory_pressure(struct sock *sk);
1508 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1510 struct net *net = sock_net((struct sock *)tp);
1512 return tp->keepalive_intvl ? :
1513 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl);
1516 static inline int keepalive_time_when(const struct tcp_sock *tp)
1518 struct net *net = sock_net((struct sock *)tp);
1520 return tp->keepalive_time ? :
1521 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time);
1524 static inline int keepalive_probes(const struct tcp_sock *tp)
1526 struct net *net = sock_net((struct sock *)tp);
1528 return tp->keepalive_probes ? :
1529 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes);
1532 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1534 const struct inet_connection_sock *icsk = &tp->inet_conn;
1536 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1537 tcp_jiffies32 - tp->rcv_tstamp);
1540 static inline int tcp_fin_time(const struct sock *sk)
1542 int fin_timeout = tcp_sk(sk)->linger2 ? :
1543 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout);
1544 const int rto = inet_csk(sk)->icsk_rto;
1546 if (fin_timeout < (rto << 2) - (rto >> 1))
1547 fin_timeout = (rto << 2) - (rto >> 1);
1552 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1555 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1557 if (unlikely(!time_before32(ktime_get_seconds(),
1558 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1561 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1562 * then following tcp messages have valid values. Ignore 0 value,
1563 * or else 'negative' tsval might forbid us to accept their packets.
1565 if (!rx_opt->ts_recent)
1570 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1573 if (tcp_paws_check(rx_opt, 0))
1576 /* RST segments are not recommended to carry timestamp,
1577 and, if they do, it is recommended to ignore PAWS because
1578 "their cleanup function should take precedence over timestamps."
1579 Certainly, it is mistake. It is necessary to understand the reasons
1580 of this constraint to relax it: if peer reboots, clock may go
1581 out-of-sync and half-open connections will not be reset.
1582 Actually, the problem would be not existing if all
1583 the implementations followed draft about maintaining clock
1584 via reboots. Linux-2.2 DOES NOT!
1586 However, we can relax time bounds for RST segments to MSL.
1588 if (rst && !time_before32(ktime_get_seconds(),
1589 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1594 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1595 int mib_idx, u32 *last_oow_ack_time);
1597 static inline void tcp_mib_init(struct net *net)
1600 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1601 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1602 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1603 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1607 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1609 tp->lost_skb_hint = NULL;
1612 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1614 tcp_clear_retrans_hints_partial(tp);
1615 tp->retransmit_skb_hint = NULL;
1618 union tcp_md5_addr {
1620 #if IS_ENABLED(CONFIG_IPV6)
1625 /* - key database */
1626 struct tcp_md5sig_key {
1627 struct hlist_node node;
1629 u8 family; /* AF_INET or AF_INET6 */
1632 union tcp_md5_addr addr;
1633 int l3index; /* set if key added with L3 scope */
1634 u8 key[TCP_MD5SIG_MAXKEYLEN];
1635 struct rcu_head rcu;
1639 struct tcp_md5sig_info {
1640 struct hlist_head head;
1641 struct rcu_head rcu;
1644 /* - pseudo header */
1645 struct tcp4_pseudohdr {
1653 struct tcp6_pseudohdr {
1654 struct in6_addr saddr;
1655 struct in6_addr daddr;
1657 __be32 protocol; /* including padding */
1660 union tcp_md5sum_block {
1661 struct tcp4_pseudohdr ip4;
1662 #if IS_ENABLED(CONFIG_IPV6)
1663 struct tcp6_pseudohdr ip6;
1667 /* - pool: digest algorithm, hash description and scratch buffer */
1668 struct tcp_md5sig_pool {
1669 struct ahash_request *md5_req;
1674 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1675 const struct sock *sk, const struct sk_buff *skb);
1676 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1677 int family, u8 prefixlen, int l3index, u8 flags,
1678 const u8 *newkey, u8 newkeylen);
1679 int tcp_md5_key_copy(struct sock *sk, const union tcp_md5_addr *addr,
1680 int family, u8 prefixlen, int l3index,
1681 struct tcp_md5sig_key *key);
1683 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1684 int family, u8 prefixlen, int l3index, u8 flags);
1685 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1686 const struct sock *addr_sk);
1688 #ifdef CONFIG_TCP_MD5SIG
1689 #include <linux/jump_label.h>
1690 extern struct static_key_false_deferred tcp_md5_needed;
1691 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1692 const union tcp_md5_addr *addr,
1694 static inline struct tcp_md5sig_key *
1695 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1696 const union tcp_md5_addr *addr, int family)
1698 if (!static_branch_unlikely(&tcp_md5_needed.key))
1700 return __tcp_md5_do_lookup(sk, l3index, addr, family);
1703 enum skb_drop_reason
1704 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
1705 const void *saddr, const void *daddr,
1706 int family, int dif, int sdif);
1709 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1711 static inline struct tcp_md5sig_key *
1712 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1713 const union tcp_md5_addr *addr, int family)
1718 static inline enum skb_drop_reason
1719 tcp_inbound_md5_hash(const struct sock *sk, const struct sk_buff *skb,
1720 const void *saddr, const void *daddr,
1721 int family, int dif, int sdif)
1723 return SKB_NOT_DROPPED_YET;
1725 #define tcp_twsk_md5_key(twsk) NULL
1728 bool tcp_alloc_md5sig_pool(void);
1730 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1731 static inline void tcp_put_md5sig_pool(void)
1736 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1737 unsigned int header_len);
1738 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1739 const struct tcp_md5sig_key *key);
1741 /* From tcp_fastopen.c */
1742 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1743 struct tcp_fastopen_cookie *cookie);
1744 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1745 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1747 struct tcp_fastopen_request {
1748 /* Fast Open cookie. Size 0 means a cookie request */
1749 struct tcp_fastopen_cookie cookie;
1750 struct msghdr *data; /* data in MSG_FASTOPEN */
1752 int copied; /* queued in tcp_connect() */
1753 struct ubuf_info *uarg;
1755 void tcp_free_fastopen_req(struct tcp_sock *tp);
1756 void tcp_fastopen_destroy_cipher(struct sock *sk);
1757 void tcp_fastopen_ctx_destroy(struct net *net);
1758 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1759 void *primary_key, void *backup_key);
1760 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
1762 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1763 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1764 struct request_sock *req,
1765 struct tcp_fastopen_cookie *foc,
1766 const struct dst_entry *dst);
1767 void tcp_fastopen_init_key_once(struct net *net);
1768 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1769 struct tcp_fastopen_cookie *cookie);
1770 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1771 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1772 #define TCP_FASTOPEN_KEY_MAX 2
1773 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1774 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1776 /* Fastopen key context */
1777 struct tcp_fastopen_context {
1778 siphash_key_t key[TCP_FASTOPEN_KEY_MAX];
1780 struct rcu_head rcu;
1783 void tcp_fastopen_active_disable(struct sock *sk);
1784 bool tcp_fastopen_active_should_disable(struct sock *sk);
1785 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1786 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1788 /* Caller needs to wrap with rcu_read_(un)lock() */
1790 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1792 struct tcp_fastopen_context *ctx;
1794 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1796 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1801 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1802 const struct tcp_fastopen_cookie *orig)
1804 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1805 orig->len == foc->len &&
1806 !memcmp(orig->val, foc->val, foc->len))
1812 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1817 /* Latencies incurred by various limits for a sender. They are
1818 * chronograph-like stats that are mutually exclusive.
1822 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1823 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1824 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1828 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1829 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1831 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1832 * the same memory storage than skb->destructor/_skb_refdst
1834 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1836 skb->destructor = NULL;
1837 skb->_skb_refdst = 0UL;
1840 #define tcp_skb_tsorted_save(skb) { \
1841 unsigned long _save = skb->_skb_refdst; \
1842 skb->_skb_refdst = 0UL;
1844 #define tcp_skb_tsorted_restore(skb) \
1845 skb->_skb_refdst = _save; \
1848 void tcp_write_queue_purge(struct sock *sk);
1850 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1852 return skb_rb_first(&sk->tcp_rtx_queue);
1855 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1857 return skb_rb_last(&sk->tcp_rtx_queue);
1860 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1862 return skb_peek_tail(&sk->sk_write_queue);
1865 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1866 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1868 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1870 return skb_peek(&sk->sk_write_queue);
1873 static inline bool tcp_skb_is_last(const struct sock *sk,
1874 const struct sk_buff *skb)
1876 return skb_queue_is_last(&sk->sk_write_queue, skb);
1880 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
1883 * Since the write queue can have a temporary empty skb in it,
1884 * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
1886 static inline bool tcp_write_queue_empty(const struct sock *sk)
1888 const struct tcp_sock *tp = tcp_sk(sk);
1890 return tp->write_seq == tp->snd_nxt;
1893 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1895 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1898 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1900 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1903 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1905 __skb_queue_tail(&sk->sk_write_queue, skb);
1907 /* Queue it, remembering where we must start sending. */
1908 if (sk->sk_write_queue.next == skb)
1909 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1912 /* Insert new before skb on the write queue of sk. */
1913 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1914 struct sk_buff *skb,
1917 __skb_queue_before(&sk->sk_write_queue, skb, new);
1920 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1922 tcp_skb_tsorted_anchor_cleanup(skb);
1923 __skb_unlink(skb, &sk->sk_write_queue);
1926 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1928 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1930 tcp_skb_tsorted_anchor_cleanup(skb);
1931 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1934 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1936 list_del(&skb->tcp_tsorted_anchor);
1937 tcp_rtx_queue_unlink(skb, sk);
1938 tcp_wmem_free_skb(sk, skb);
1941 static inline void tcp_push_pending_frames(struct sock *sk)
1943 if (tcp_send_head(sk)) {
1944 struct tcp_sock *tp = tcp_sk(sk);
1946 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1950 /* Start sequence of the skb just after the highest skb with SACKed
1951 * bit, valid only if sacked_out > 0 or when the caller has ensured
1952 * validity by itself.
1954 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1956 if (!tp->sacked_out)
1959 if (tp->highest_sack == NULL)
1962 return TCP_SKB_CB(tp->highest_sack)->seq;
1965 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1967 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1970 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1972 return tcp_sk(sk)->highest_sack;
1975 static inline void tcp_highest_sack_reset(struct sock *sk)
1977 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1980 /* Called when old skb is about to be deleted and replaced by new skb */
1981 static inline void tcp_highest_sack_replace(struct sock *sk,
1982 struct sk_buff *old,
1983 struct sk_buff *new)
1985 if (old == tcp_highest_sack(sk))
1986 tcp_sk(sk)->highest_sack = new;
1989 /* This helper checks if socket has IP_TRANSPARENT set */
1990 static inline bool inet_sk_transparent(const struct sock *sk)
1992 switch (sk->sk_state) {
1994 return inet_twsk(sk)->tw_transparent;
1995 case TCP_NEW_SYN_RECV:
1996 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1998 return inet_sk(sk)->transparent;
2001 /* Determines whether this is a thin stream (which may suffer from
2002 * increased latency). Used to trigger latency-reducing mechanisms.
2004 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
2006 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
2010 enum tcp_seq_states {
2011 TCP_SEQ_STATE_LISTENING,
2012 TCP_SEQ_STATE_ESTABLISHED,
2015 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
2016 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
2017 void tcp_seq_stop(struct seq_file *seq, void *v);
2019 struct tcp_seq_afinfo {
2023 struct tcp_iter_state {
2024 struct seq_net_private p;
2025 enum tcp_seq_states state;
2026 struct sock *syn_wait_sk;
2027 int bucket, offset, sbucket, num;
2031 extern struct request_sock_ops tcp_request_sock_ops;
2032 extern struct request_sock_ops tcp6_request_sock_ops;
2034 void tcp_v4_destroy_sock(struct sock *sk);
2036 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
2037 netdev_features_t features);
2038 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
2039 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
2040 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
2041 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff));
2042 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb));
2043 int tcp_gro_complete(struct sk_buff *skb);
2045 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
2047 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
2049 struct net *net = sock_net((struct sock *)tp);
2050 return tp->notsent_lowat ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
2053 bool tcp_stream_memory_free(const struct sock *sk, int wake);
2055 #ifdef CONFIG_PROC_FS
2056 int tcp4_proc_init(void);
2057 void tcp4_proc_exit(void);
2060 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
2061 int tcp_conn_request(struct request_sock_ops *rsk_ops,
2062 const struct tcp_request_sock_ops *af_ops,
2063 struct sock *sk, struct sk_buff *skb);
2065 /* TCP af-specific functions */
2066 struct tcp_sock_af_ops {
2067 #ifdef CONFIG_TCP_MD5SIG
2068 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
2069 const struct sock *addr_sk);
2070 int (*calc_md5_hash)(char *location,
2071 const struct tcp_md5sig_key *md5,
2072 const struct sock *sk,
2073 const struct sk_buff *skb);
2074 int (*md5_parse)(struct sock *sk,
2081 struct tcp_request_sock_ops {
2083 #ifdef CONFIG_TCP_MD5SIG
2084 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
2085 const struct sock *addr_sk);
2086 int (*calc_md5_hash) (char *location,
2087 const struct tcp_md5sig_key *md5,
2088 const struct sock *sk,
2089 const struct sk_buff *skb);
2091 #ifdef CONFIG_SYN_COOKIES
2092 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
2095 struct dst_entry *(*route_req)(const struct sock *sk,
2096 struct sk_buff *skb,
2098 struct request_sock *req);
2099 u32 (*init_seq)(const struct sk_buff *skb);
2100 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
2101 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2102 struct flowi *fl, struct request_sock *req,
2103 struct tcp_fastopen_cookie *foc,
2104 enum tcp_synack_type synack_type,
2105 struct sk_buff *syn_skb);
2108 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2109 #if IS_ENABLED(CONFIG_IPV6)
2110 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2113 #ifdef CONFIG_SYN_COOKIES
2114 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2115 const struct sock *sk, struct sk_buff *skb,
2118 tcp_synq_overflow(sk);
2119 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2120 return ops->cookie_init_seq(skb, mss);
2123 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2124 const struct sock *sk, struct sk_buff *skb,
2131 int tcpv4_offload_init(void);
2133 void tcp_v4_init(void);
2134 void tcp_init(void);
2136 /* tcp_recovery.c */
2137 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2138 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2139 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2141 extern bool tcp_rack_mark_lost(struct sock *sk);
2142 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2144 extern void tcp_rack_reo_timeout(struct sock *sk);
2145 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2150 * Scaling factor for fractions in PLB. For example, tcp_plb_update_state
2151 * expects cong_ratio which represents fraction of traffic that experienced
2152 * congestion over a single RTT. In order to avoid floating point operations,
2153 * this fraction should be mapped to (1 << TCP_PLB_SCALE) and passed in.
2155 #define TCP_PLB_SCALE 8
2157 /* State for PLB (Protective Load Balancing) for a single TCP connection. */
2158 struct tcp_plb_state {
2159 u8 consec_cong_rounds:5, /* consecutive congested rounds */
2161 u32 pause_until; /* jiffies32 when PLB can resume rerouting */
2164 static inline void tcp_plb_init(const struct sock *sk,
2165 struct tcp_plb_state *plb)
2167 plb->consec_cong_rounds = 0;
2168 plb->pause_until = 0;
2170 void tcp_plb_update_state(const struct sock *sk, struct tcp_plb_state *plb,
2171 const int cong_ratio);
2172 void tcp_plb_check_rehash(struct sock *sk, struct tcp_plb_state *plb);
2173 void tcp_plb_update_state_upon_rto(struct sock *sk, struct tcp_plb_state *plb);
2175 /* At how many usecs into the future should the RTO fire? */
2176 static inline s64 tcp_rto_delta_us(const struct sock *sk)
2178 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2179 u32 rto = inet_csk(sk)->icsk_rto;
2180 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2182 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2186 * Save and compile IPv4 options, return a pointer to it
2188 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2189 struct sk_buff *skb)
2191 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2192 struct ip_options_rcu *dopt = NULL;
2195 int opt_size = sizeof(*dopt) + opt->optlen;
2197 dopt = kmalloc(opt_size, GFP_ATOMIC);
2198 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2206 /* locally generated TCP pure ACKs have skb->truesize == 2
2207 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2208 * This is much faster than dissecting the packet to find out.
2209 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2211 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2213 return skb->truesize == 2;
2216 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2221 static inline int tcp_inq(struct sock *sk)
2223 struct tcp_sock *tp = tcp_sk(sk);
2226 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2228 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2230 before(tp->urg_seq, tp->copied_seq) ||
2231 !before(tp->urg_seq, tp->rcv_nxt)) {
2233 answ = tp->rcv_nxt - tp->copied_seq;
2235 /* Subtract 1, if FIN was received */
2236 if (answ && sock_flag(sk, SOCK_DONE))
2239 answ = tp->urg_seq - tp->copied_seq;
2245 int tcp_peek_len(struct socket *sock);
2247 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2251 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2253 /* We update these fields while other threads might
2254 * read them from tcp_get_info()
2256 WRITE_ONCE(tp->segs_in, tp->segs_in + segs_in);
2257 if (skb->len > tcp_hdrlen(skb))
2258 WRITE_ONCE(tp->data_segs_in, tp->data_segs_in + segs_in);
2262 * TCP listen path runs lockless.
2263 * We forced "struct sock" to be const qualified to make sure
2264 * we don't modify one of its field by mistake.
2265 * Here, we increment sk_drops which is an atomic_t, so we can safely
2266 * make sock writable again.
2268 static inline void tcp_listendrop(const struct sock *sk)
2270 atomic_inc(&((struct sock *)sk)->sk_drops);
2271 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2274 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2277 * Interface for adding Upper Level Protocols over TCP
2280 #define TCP_ULP_NAME_MAX 16
2281 #define TCP_ULP_MAX 128
2282 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2284 struct tcp_ulp_ops {
2285 struct list_head list;
2287 /* initialize ulp */
2288 int (*init)(struct sock *sk);
2290 void (*update)(struct sock *sk, struct proto *p,
2291 void (*write_space)(struct sock *sk));
2293 void (*release)(struct sock *sk);
2295 int (*get_info)(const struct sock *sk, struct sk_buff *skb);
2296 size_t (*get_info_size)(const struct sock *sk);
2298 void (*clone)(const struct request_sock *req, struct sock *newsk,
2299 const gfp_t priority);
2301 char name[TCP_ULP_NAME_MAX];
2302 struct module *owner;
2304 int tcp_register_ulp(struct tcp_ulp_ops *type);
2305 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2306 int tcp_set_ulp(struct sock *sk, const char *name);
2307 void tcp_get_available_ulp(char *buf, size_t len);
2308 void tcp_cleanup_ulp(struct sock *sk);
2309 void tcp_update_ulp(struct sock *sk, struct proto *p,
2310 void (*write_space)(struct sock *sk));
2312 #define MODULE_ALIAS_TCP_ULP(name) \
2313 __MODULE_INFO(alias, alias_userspace, name); \
2314 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2316 #ifdef CONFIG_NET_SOCK_MSG
2320 #ifdef CONFIG_BPF_SYSCALL
2321 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
2322 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
2323 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2324 #endif /* CONFIG_BPF_SYSCALL */
2326 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress,
2327 struct sk_msg *msg, u32 bytes, int flags);
2328 #endif /* CONFIG_NET_SOCK_MSG */
2330 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG)
2331 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2336 #ifdef CONFIG_CGROUP_BPF
2337 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2338 struct sk_buff *skb,
2339 unsigned int end_offset)
2342 skops->skb_data_end = skb->data + end_offset;
2345 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2346 struct sk_buff *skb,
2347 unsigned int end_offset)
2352 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2353 * is < 0, then the BPF op failed (for example if the loaded BPF
2354 * program does not support the chosen operation or there is no BPF
2358 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2360 struct bpf_sock_ops_kern sock_ops;
2363 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2364 if (sk_fullsock(sk)) {
2365 sock_ops.is_fullsock = 1;
2366 sock_owned_by_me(sk);
2372 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2374 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2376 ret = sock_ops.reply;
2382 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2384 u32 args[2] = {arg1, arg2};
2386 return tcp_call_bpf(sk, op, 2, args);
2389 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2392 u32 args[3] = {arg1, arg2, arg3};
2394 return tcp_call_bpf(sk, op, 3, args);
2398 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2403 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2408 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2416 static inline u32 tcp_timeout_init(struct sock *sk)
2420 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2423 timeout = TCP_TIMEOUT_INIT;
2424 return min_t(int, timeout, TCP_RTO_MAX);
2427 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2431 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2438 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2440 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2443 static inline void tcp_bpf_rtt(struct sock *sk)
2445 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2446 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2449 #if IS_ENABLED(CONFIG_SMC)
2450 extern struct static_key_false tcp_have_smc;
2453 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2454 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2455 void (*cad)(struct sock *sk, u32 ack_seq));
2456 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2457 void clean_acked_data_flush(void);
2460 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2461 static inline void tcp_add_tx_delay(struct sk_buff *skb,
2462 const struct tcp_sock *tp)
2464 if (static_branch_unlikely(&tcp_tx_delay_enabled))
2465 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2468 /* Compute Earliest Departure Time for some control packets
2469 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2471 static inline u64 tcp_transmit_time(const struct sock *sk)
2473 if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2474 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2475 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2477 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;