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 extern struct percpu_counter tcp_sockets_allocated;
257 extern unsigned long tcp_memory_pressure;
259 /* optimized version of sk_under_memory_pressure() for TCP sockets */
260 static inline bool tcp_under_memory_pressure(const struct sock *sk)
262 if (mem_cgroup_sockets_enabled && sk->sk_memcg &&
263 mem_cgroup_under_socket_pressure(sk->sk_memcg))
266 return READ_ONCE(tcp_memory_pressure);
269 * The next routines deal with comparing 32 bit unsigned ints
270 * and worry about wraparound (automatic with unsigned arithmetic).
273 static inline bool before(__u32 seq1, __u32 seq2)
275 return (__s32)(seq1-seq2) < 0;
277 #define after(seq2, seq1) before(seq1, seq2)
279 /* is s2<=s1<=s3 ? */
280 static inline bool between(__u32 seq1, __u32 seq2, __u32 seq3)
282 return seq3 - seq2 >= seq1 - seq2;
285 static inline bool tcp_out_of_memory(struct sock *sk)
287 if (sk->sk_wmem_queued > SOCK_MIN_SNDBUF &&
288 sk_memory_allocated(sk) > sk_prot_mem_limits(sk, 2))
293 void sk_forced_mem_schedule(struct sock *sk, int size);
295 bool tcp_check_oom(struct sock *sk, int shift);
298 extern struct proto tcp_prot;
300 #define TCP_INC_STATS(net, field) SNMP_INC_STATS((net)->mib.tcp_statistics, field)
301 #define __TCP_INC_STATS(net, field) __SNMP_INC_STATS((net)->mib.tcp_statistics, field)
302 #define TCP_DEC_STATS(net, field) SNMP_DEC_STATS((net)->mib.tcp_statistics, field)
303 #define TCP_ADD_STATS(net, field, val) SNMP_ADD_STATS((net)->mib.tcp_statistics, field, val)
305 void tcp_tasklet_init(void);
307 int tcp_v4_err(struct sk_buff *skb, u32);
309 void tcp_shutdown(struct sock *sk, int how);
311 int tcp_v4_early_demux(struct sk_buff *skb);
312 int tcp_v4_rcv(struct sk_buff *skb);
314 void tcp_remove_empty_skb(struct sock *sk, struct sk_buff *skb);
315 int tcp_v4_tw_remember_stamp(struct inet_timewait_sock *tw);
316 int tcp_sendmsg(struct sock *sk, struct msghdr *msg, size_t size);
317 int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size);
318 int tcp_sendpage(struct sock *sk, struct page *page, int offset, size_t size,
320 int tcp_sendpage_locked(struct sock *sk, struct page *page, int offset,
321 size_t size, int flags);
322 struct sk_buff *tcp_build_frag(struct sock *sk, int size_goal, int flags,
323 struct page *page, int offset, size_t *size);
324 ssize_t do_tcp_sendpages(struct sock *sk, struct page *page, int offset,
325 size_t size, int flags);
326 int tcp_send_mss(struct sock *sk, int *size_goal, int flags);
327 void tcp_push(struct sock *sk, int flags, int mss_now, int nonagle,
329 void tcp_release_cb(struct sock *sk);
330 void tcp_wfree(struct sk_buff *skb);
331 void tcp_write_timer_handler(struct sock *sk);
332 void tcp_delack_timer_handler(struct sock *sk);
333 int tcp_ioctl(struct sock *sk, int cmd, unsigned long arg);
334 int tcp_rcv_state_process(struct sock *sk, struct sk_buff *skb);
335 void tcp_rcv_established(struct sock *sk, struct sk_buff *skb);
336 void tcp_rcv_space_adjust(struct sock *sk);
337 int tcp_twsk_unique(struct sock *sk, struct sock *sktw, void *twp);
338 void tcp_twsk_destructor(struct sock *sk);
339 ssize_t tcp_splice_read(struct socket *sk, loff_t *ppos,
340 struct pipe_inode_info *pipe, size_t len,
343 void tcp_enter_quickack_mode(struct sock *sk, unsigned int max_quickacks);
344 static inline void tcp_dec_quickack_mode(struct sock *sk,
345 const unsigned int pkts)
347 struct inet_connection_sock *icsk = inet_csk(sk);
349 if (icsk->icsk_ack.quick) {
350 if (pkts >= icsk->icsk_ack.quick) {
351 icsk->icsk_ack.quick = 0;
352 /* Leaving quickack mode we deflate ATO. */
353 icsk->icsk_ack.ato = TCP_ATO_MIN;
355 icsk->icsk_ack.quick -= pkts;
360 #define TCP_ECN_QUEUE_CWR 2
361 #define TCP_ECN_DEMAND_CWR 4
362 #define TCP_ECN_SEEN 8
372 enum tcp_tw_status tcp_timewait_state_process(struct inet_timewait_sock *tw,
374 const struct tcphdr *th);
375 struct sock *tcp_check_req(struct sock *sk, struct sk_buff *skb,
376 struct request_sock *req, bool fastopen,
378 int tcp_child_process(struct sock *parent, struct sock *child,
379 struct sk_buff *skb);
380 void tcp_enter_loss(struct sock *sk);
381 void tcp_cwnd_reduction(struct sock *sk, int newly_acked_sacked, int newly_lost, int flag);
382 void tcp_clear_retrans(struct tcp_sock *tp);
383 void tcp_update_metrics(struct sock *sk);
384 void tcp_init_metrics(struct sock *sk);
385 void tcp_metrics_init(void);
386 bool tcp_peer_is_proven(struct request_sock *req, struct dst_entry *dst);
387 void __tcp_close(struct sock *sk, long timeout);
388 void tcp_close(struct sock *sk, long timeout);
389 void tcp_init_sock(struct sock *sk);
390 void tcp_init_transfer(struct sock *sk, int bpf_op, struct sk_buff *skb);
391 __poll_t tcp_poll(struct file *file, struct socket *sock,
392 struct poll_table_struct *wait);
393 int tcp_getsockopt(struct sock *sk, int level, int optname,
394 char __user *optval, int __user *optlen);
395 bool tcp_bpf_bypass_getsockopt(int level, int optname);
396 int tcp_setsockopt(struct sock *sk, int level, int optname, sockptr_t optval,
397 unsigned int optlen);
398 void tcp_set_keepalive(struct sock *sk, int val);
399 void tcp_syn_ack_timeout(const struct request_sock *req);
400 int tcp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len, int nonblock,
401 int flags, int *addr_len);
402 int tcp_set_rcvlowat(struct sock *sk, int val);
403 int tcp_set_window_clamp(struct sock *sk, int val);
404 void tcp_update_recv_tstamps(struct sk_buff *skb,
405 struct scm_timestamping_internal *tss);
406 void tcp_recv_timestamp(struct msghdr *msg, const struct sock *sk,
407 struct scm_timestamping_internal *tss);
408 void tcp_data_ready(struct sock *sk);
410 int tcp_mmap(struct file *file, struct socket *sock,
411 struct vm_area_struct *vma);
413 void tcp_parse_options(const struct net *net, const struct sk_buff *skb,
414 struct tcp_options_received *opt_rx,
415 int estab, struct tcp_fastopen_cookie *foc);
416 const u8 *tcp_parse_md5sig_option(const struct tcphdr *th);
419 * BPF SKB-less helpers
421 u16 tcp_v4_get_syncookie(struct sock *sk, struct iphdr *iph,
422 struct tcphdr *th, u32 *cookie);
423 u16 tcp_v6_get_syncookie(struct sock *sk, struct ipv6hdr *iph,
424 struct tcphdr *th, u32 *cookie);
425 u16 tcp_get_syncookie_mss(struct request_sock_ops *rsk_ops,
426 const struct tcp_request_sock_ops *af_ops,
427 struct sock *sk, struct tcphdr *th);
429 * TCP v4 functions exported for the inet6 API
432 void tcp_v4_send_check(struct sock *sk, struct sk_buff *skb);
433 void tcp_v4_mtu_reduced(struct sock *sk);
434 void tcp_req_err(struct sock *sk, u32 seq, bool abort);
435 void tcp_ld_RTO_revert(struct sock *sk, u32 seq);
436 int tcp_v4_conn_request(struct sock *sk, struct sk_buff *skb);
437 struct sock *tcp_create_openreq_child(const struct sock *sk,
438 struct request_sock *req,
439 struct sk_buff *skb);
440 void tcp_ca_openreq_child(struct sock *sk, const struct dst_entry *dst);
441 struct sock *tcp_v4_syn_recv_sock(const struct sock *sk, struct sk_buff *skb,
442 struct request_sock *req,
443 struct dst_entry *dst,
444 struct request_sock *req_unhash,
446 int tcp_v4_do_rcv(struct sock *sk, struct sk_buff *skb);
447 int tcp_v4_connect(struct sock *sk, struct sockaddr *uaddr, int addr_len);
448 int tcp_connect(struct sock *sk);
449 enum tcp_synack_type {
454 struct sk_buff *tcp_make_synack(const struct sock *sk, struct dst_entry *dst,
455 struct request_sock *req,
456 struct tcp_fastopen_cookie *foc,
457 enum tcp_synack_type synack_type,
458 struct sk_buff *syn_skb);
459 int tcp_disconnect(struct sock *sk, int flags);
461 void tcp_finish_connect(struct sock *sk, struct sk_buff *skb);
462 int tcp_send_rcvq(struct sock *sk, struct msghdr *msg, size_t size);
463 void inet_sk_rx_dst_set(struct sock *sk, const struct sk_buff *skb);
465 /* From syncookies.c */
466 struct sock *tcp_get_cookie_sock(struct sock *sk, struct sk_buff *skb,
467 struct request_sock *req,
468 struct dst_entry *dst, u32 tsoff);
469 int __cookie_v4_check(const struct iphdr *iph, const struct tcphdr *th,
471 struct sock *cookie_v4_check(struct sock *sk, struct sk_buff *skb);
472 struct request_sock *cookie_tcp_reqsk_alloc(const struct request_sock_ops *ops,
473 const struct tcp_request_sock_ops *af_ops,
474 struct sock *sk, struct sk_buff *skb);
475 #ifdef CONFIG_SYN_COOKIES
477 /* Syncookies use a monotonic timer which increments every 60 seconds.
478 * This counter is used both as a hash input and partially encoded into
479 * the cookie value. A cookie is only validated further if the delta
480 * between the current counter value and the encoded one is less than this,
481 * i.e. a sent cookie is valid only at most for 2*60 seconds (or less if
482 * the counter advances immediately after a cookie is generated).
484 #define MAX_SYNCOOKIE_AGE 2
485 #define TCP_SYNCOOKIE_PERIOD (60 * HZ)
486 #define TCP_SYNCOOKIE_VALID (MAX_SYNCOOKIE_AGE * TCP_SYNCOOKIE_PERIOD)
488 /* syncookies: remember time of last synqueue overflow
489 * But do not dirty this field too often (once per second is enough)
490 * It is racy as we do not hold a lock, but race is very minor.
492 static inline void tcp_synq_overflow(const struct sock *sk)
494 unsigned int last_overflow;
495 unsigned int now = jiffies;
497 if (sk->sk_reuseport) {
498 struct sock_reuseport *reuse;
500 reuse = rcu_dereference(sk->sk_reuseport_cb);
502 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
503 if (!time_between32(now, last_overflow,
505 WRITE_ONCE(reuse->synq_overflow_ts, now);
510 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
511 if (!time_between32(now, last_overflow, last_overflow + HZ))
512 WRITE_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp, now);
515 /* syncookies: no recent synqueue overflow on this listening socket? */
516 static inline bool tcp_synq_no_recent_overflow(const struct sock *sk)
518 unsigned int last_overflow;
519 unsigned int now = jiffies;
521 if (sk->sk_reuseport) {
522 struct sock_reuseport *reuse;
524 reuse = rcu_dereference(sk->sk_reuseport_cb);
526 last_overflow = READ_ONCE(reuse->synq_overflow_ts);
527 return !time_between32(now, last_overflow - HZ,
529 TCP_SYNCOOKIE_VALID);
533 last_overflow = READ_ONCE(tcp_sk(sk)->rx_opt.ts_recent_stamp);
535 /* If last_overflow <= jiffies <= last_overflow + TCP_SYNCOOKIE_VALID,
536 * then we're under synflood. However, we have to use
537 * 'last_overflow - HZ' as lower bound. That's because a concurrent
538 * tcp_synq_overflow() could update .ts_recent_stamp after we read
539 * jiffies but before we store .ts_recent_stamp into last_overflow,
540 * which could lead to rejecting a valid syncookie.
542 return !time_between32(now, last_overflow - HZ,
543 last_overflow + TCP_SYNCOOKIE_VALID);
546 static inline u32 tcp_cookie_time(void)
548 u64 val = get_jiffies_64();
550 do_div(val, TCP_SYNCOOKIE_PERIOD);
554 u32 __cookie_v4_init_sequence(const struct iphdr *iph, const struct tcphdr *th,
556 __u32 cookie_v4_init_sequence(const struct sk_buff *skb, __u16 *mss);
557 u64 cookie_init_timestamp(struct request_sock *req, u64 now);
558 bool cookie_timestamp_decode(const struct net *net,
559 struct tcp_options_received *opt);
560 bool cookie_ecn_ok(const struct tcp_options_received *opt,
561 const struct net *net, const struct dst_entry *dst);
563 /* From net/ipv6/syncookies.c */
564 int __cookie_v6_check(const struct ipv6hdr *iph, const struct tcphdr *th,
566 struct sock *cookie_v6_check(struct sock *sk, struct sk_buff *skb);
568 u32 __cookie_v6_init_sequence(const struct ipv6hdr *iph,
569 const struct tcphdr *th, u16 *mssp);
570 __u32 cookie_v6_init_sequence(const struct sk_buff *skb, __u16 *mss);
574 void tcp_skb_entail(struct sock *sk, struct sk_buff *skb);
575 void tcp_mark_push(struct tcp_sock *tp, struct sk_buff *skb);
576 void __tcp_push_pending_frames(struct sock *sk, unsigned int cur_mss,
578 int __tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
579 int tcp_retransmit_skb(struct sock *sk, struct sk_buff *skb, int segs);
580 void tcp_retransmit_timer(struct sock *sk);
581 void tcp_xmit_retransmit_queue(struct sock *);
582 void tcp_simple_retransmit(struct sock *);
583 void tcp_enter_recovery(struct sock *sk, bool ece_ack);
584 int tcp_trim_head(struct sock *, struct sk_buff *, u32);
586 TCP_FRAG_IN_WRITE_QUEUE,
587 TCP_FRAG_IN_RTX_QUEUE,
589 int tcp_fragment(struct sock *sk, enum tcp_queue tcp_queue,
590 struct sk_buff *skb, u32 len,
591 unsigned int mss_now, gfp_t gfp);
593 void tcp_send_probe0(struct sock *);
594 void tcp_send_partial(struct sock *);
595 int tcp_write_wakeup(struct sock *, int mib);
596 void tcp_send_fin(struct sock *sk);
597 void tcp_send_active_reset(struct sock *sk, gfp_t priority);
598 int tcp_send_synack(struct sock *);
599 void tcp_push_one(struct sock *, unsigned int mss_now);
600 void __tcp_send_ack(struct sock *sk, u32 rcv_nxt);
601 void tcp_send_ack(struct sock *sk);
602 void tcp_send_delayed_ack(struct sock *sk);
603 void tcp_send_loss_probe(struct sock *sk);
604 bool tcp_schedule_loss_probe(struct sock *sk, bool advancing_rto);
605 void tcp_skb_collapse_tstamp(struct sk_buff *skb,
606 const struct sk_buff *next_skb);
609 void tcp_rearm_rto(struct sock *sk);
610 void tcp_synack_rtt_meas(struct sock *sk, struct request_sock *req);
611 void tcp_reset(struct sock *sk, struct sk_buff *skb);
612 void tcp_skb_mark_lost_uncond_verify(struct tcp_sock *tp, struct sk_buff *skb);
613 void tcp_fin(struct sock *sk);
614 void tcp_check_space(struct sock *sk);
617 void tcp_init_xmit_timers(struct sock *);
618 static inline void tcp_clear_xmit_timers(struct sock *sk)
620 if (hrtimer_try_to_cancel(&tcp_sk(sk)->pacing_timer) == 1)
623 if (hrtimer_try_to_cancel(&tcp_sk(sk)->compressed_ack_timer) == 1)
626 inet_csk_clear_xmit_timers(sk);
629 unsigned int tcp_sync_mss(struct sock *sk, u32 pmtu);
630 unsigned int tcp_current_mss(struct sock *sk);
631 u32 tcp_clamp_probe0_to_user_timeout(const struct sock *sk, u32 when);
633 /* Bound MSS / TSO packet size with the half of the window */
634 static inline int tcp_bound_to_half_wnd(struct tcp_sock *tp, int pktsize)
638 /* When peer uses tiny windows, there is no use in packetizing
639 * to sub-MSS pieces for the sake of SWS or making sure there
640 * are enough packets in the pipe for fast recovery.
642 * On the other hand, for extremely large MSS devices, handling
643 * smaller than MSS windows in this way does make sense.
645 if (tp->max_window > TCP_MSS_DEFAULT)
646 cutoff = (tp->max_window >> 1);
648 cutoff = tp->max_window;
650 if (cutoff && pktsize > cutoff)
651 return max_t(int, cutoff, 68U - tp->tcp_header_len);
657 void tcp_get_info(struct sock *, struct tcp_info *);
659 /* Read 'sendfile()'-style from a TCP socket */
660 int tcp_read_sock(struct sock *sk, read_descriptor_t *desc,
661 sk_read_actor_t recv_actor);
663 void tcp_initialize_rcv_mss(struct sock *sk);
665 int tcp_mtu_to_mss(struct sock *sk, int pmtu);
666 int tcp_mss_to_mtu(struct sock *sk, int mss);
667 void tcp_mtup_init(struct sock *sk);
669 static inline void tcp_bound_rto(const struct sock *sk)
671 if (inet_csk(sk)->icsk_rto > TCP_RTO_MAX)
672 inet_csk(sk)->icsk_rto = TCP_RTO_MAX;
675 static inline u32 __tcp_set_rto(const struct tcp_sock *tp)
677 return usecs_to_jiffies((tp->srtt_us >> 3) + tp->rttvar_us);
680 static inline void __tcp_fast_path_on(struct tcp_sock *tp, u32 snd_wnd)
682 /* mptcp hooks are only on the slow path */
683 if (sk_is_mptcp((struct sock *)tp))
686 tp->pred_flags = htonl((tp->tcp_header_len << 26) |
687 ntohl(TCP_FLAG_ACK) |
691 static inline void tcp_fast_path_on(struct tcp_sock *tp)
693 __tcp_fast_path_on(tp, tp->snd_wnd >> tp->rx_opt.snd_wscale);
696 static inline void tcp_fast_path_check(struct sock *sk)
698 struct tcp_sock *tp = tcp_sk(sk);
700 if (RB_EMPTY_ROOT(&tp->out_of_order_queue) &&
702 atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf &&
704 tcp_fast_path_on(tp);
707 /* Compute the actual rto_min value */
708 static inline u32 tcp_rto_min(struct sock *sk)
710 const struct dst_entry *dst = __sk_dst_get(sk);
711 u32 rto_min = inet_csk(sk)->icsk_rto_min;
713 if (dst && dst_metric_locked(dst, RTAX_RTO_MIN))
714 rto_min = dst_metric_rtt(dst, RTAX_RTO_MIN);
718 static inline u32 tcp_rto_min_us(struct sock *sk)
720 return jiffies_to_usecs(tcp_rto_min(sk));
723 static inline bool tcp_ca_dst_locked(const struct dst_entry *dst)
725 return dst_metric_locked(dst, RTAX_CC_ALGO);
728 /* Minimum RTT in usec. ~0 means not available. */
729 static inline u32 tcp_min_rtt(const struct tcp_sock *tp)
731 return minmax_get(&tp->rtt_min);
734 /* Compute the actual receive window we are currently advertising.
735 * Rcv_nxt can be after the window if our peer push more data
736 * than the offered window.
738 static inline u32 tcp_receive_window(const struct tcp_sock *tp)
740 s32 win = tp->rcv_wup + tp->rcv_wnd - tp->rcv_nxt;
747 /* Choose a new window, without checks for shrinking, and without
748 * scaling applied to the result. The caller does these things
749 * if necessary. This is a "raw" window selection.
751 u32 __tcp_select_window(struct sock *sk);
753 void tcp_send_window_probe(struct sock *sk);
755 /* TCP uses 32bit jiffies to save some space.
756 * Note that this is different from tcp_time_stamp, which
757 * historically has been the same until linux-4.13.
759 #define tcp_jiffies32 ((u32)jiffies)
762 * Deliver a 32bit value for TCP timestamp option (RFC 7323)
763 * It is no longer tied to jiffies, but to 1 ms clock.
764 * Note: double check if you want to use tcp_jiffies32 instead of this.
766 #define TCP_TS_HZ 1000
768 static inline u64 tcp_clock_ns(void)
770 return ktime_get_ns();
773 static inline u64 tcp_clock_us(void)
775 return div_u64(tcp_clock_ns(), NSEC_PER_USEC);
778 /* This should only be used in contexts where tp->tcp_mstamp is up to date */
779 static inline u32 tcp_time_stamp(const struct tcp_sock *tp)
781 return div_u64(tp->tcp_mstamp, USEC_PER_SEC / TCP_TS_HZ);
784 /* Convert a nsec timestamp into TCP TSval timestamp (ms based currently) */
785 static inline u32 tcp_ns_to_ts(u64 ns)
787 return div_u64(ns, NSEC_PER_SEC / TCP_TS_HZ);
790 /* Could use tcp_clock_us() / 1000, but this version uses a single divide */
791 static inline u32 tcp_time_stamp_raw(void)
793 return tcp_ns_to_ts(tcp_clock_ns());
796 void tcp_mstamp_refresh(struct tcp_sock *tp);
798 static inline u32 tcp_stamp_us_delta(u64 t1, u64 t0)
800 return max_t(s64, t1 - t0, 0);
803 static inline u32 tcp_skb_timestamp(const struct sk_buff *skb)
805 return tcp_ns_to_ts(skb->skb_mstamp_ns);
808 /* provide the departure time in us unit */
809 static inline u64 tcp_skb_timestamp_us(const struct sk_buff *skb)
811 return div_u64(skb->skb_mstamp_ns, NSEC_PER_USEC);
815 #define tcp_flag_byte(th) (((u_int8_t *)th)[13])
817 #define TCPHDR_FIN 0x01
818 #define TCPHDR_SYN 0x02
819 #define TCPHDR_RST 0x04
820 #define TCPHDR_PSH 0x08
821 #define TCPHDR_ACK 0x10
822 #define TCPHDR_URG 0x20
823 #define TCPHDR_ECE 0x40
824 #define TCPHDR_CWR 0x80
826 #define TCPHDR_SYN_ECN (TCPHDR_SYN | TCPHDR_ECE | TCPHDR_CWR)
828 /* This is what the send packet queuing engine uses to pass
829 * TCP per-packet control information to the transmission code.
830 * We also store the host-order sequence numbers in here too.
831 * This is 44 bytes if IPV6 is enabled.
832 * If this grows please adjust skbuff.h:skbuff->cb[xxx] size appropriately.
835 __u32 seq; /* Starting sequence number */
836 __u32 end_seq; /* SEQ + FIN + SYN + datalen */
838 /* Note : tcp_tw_isn is used in input path only
839 * (isn chosen by tcp_timewait_state_process())
841 * tcp_gso_segs/size are used in write queue only,
842 * cf tcp_skb_pcount()/tcp_skb_mss()
850 __u8 tcp_flags; /* TCP header flags. (tcp[13]) */
852 __u8 sacked; /* State flags for SACK. */
853 #define TCPCB_SACKED_ACKED 0x01 /* SKB ACK'd by a SACK block */
854 #define TCPCB_SACKED_RETRANS 0x02 /* SKB retransmitted */
855 #define TCPCB_LOST 0x04 /* SKB is lost */
856 #define TCPCB_TAGBITS 0x07 /* All tag bits */
857 #define TCPCB_REPAIRED 0x10 /* SKB repaired (no skb_mstamp_ns) */
858 #define TCPCB_EVER_RETRANS 0x80 /* Ever retransmitted frame */
859 #define TCPCB_RETRANS (TCPCB_SACKED_RETRANS|TCPCB_EVER_RETRANS| \
862 __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
863 __u8 txstamp_ack:1, /* Record TX timestamp for ack? */
864 eor:1, /* Is skb MSG_EOR marked? */
865 has_rxtstamp:1, /* SKB has a RX timestamp */
867 __u32 ack_seq; /* Sequence number ACK'd */
870 /* There is space for up to 24 bytes */
871 __u32 in_flight:30,/* Bytes in flight at transmit */
872 is_app_limited:1, /* cwnd not fully used? */
874 /* pkts S/ACKed so far upon tx of skb, incl retrans: */
876 /* start of send pipeline phase */
878 /* when we reached the "delivered" count */
879 u64 delivered_mstamp;
880 } tx; /* only used for outgoing skbs */
882 struct inet_skb_parm h4;
883 #if IS_ENABLED(CONFIG_IPV6)
884 struct inet6_skb_parm h6;
886 } header; /* For incoming skbs */
890 #define TCP_SKB_CB(__skb) ((struct tcp_skb_cb *)&((__skb)->cb[0]))
892 extern const struct inet_connection_sock_af_ops ipv4_specific;
894 #if IS_ENABLED(CONFIG_IPV6)
895 /* This is the variant of inet6_iif() that must be used by TCP,
896 * as TCP moves IP6CB into a different location in skb->cb[]
898 static inline int tcp_v6_iif(const struct sk_buff *skb)
900 return TCP_SKB_CB(skb)->header.h6.iif;
903 static inline int tcp_v6_iif_l3_slave(const struct sk_buff *skb)
905 bool l3_slave = ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags);
907 return l3_slave ? skb->skb_iif : TCP_SKB_CB(skb)->header.h6.iif;
910 /* TCP_SKB_CB reference means this can not be used from early demux */
911 static inline int tcp_v6_sdif(const struct sk_buff *skb)
913 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
914 if (skb && ipv6_l3mdev_skb(TCP_SKB_CB(skb)->header.h6.flags))
915 return TCP_SKB_CB(skb)->header.h6.iif;
920 extern const struct inet_connection_sock_af_ops ipv6_specific;
922 INDIRECT_CALLABLE_DECLARE(void tcp_v6_send_check(struct sock *sk, struct sk_buff *skb));
923 INDIRECT_CALLABLE_DECLARE(int tcp_v6_rcv(struct sk_buff *skb));
924 void tcp_v6_early_demux(struct sk_buff *skb);
928 /* TCP_SKB_CB reference means this can not be used from early demux */
929 static inline int tcp_v4_sdif(struct sk_buff *skb)
931 #if IS_ENABLED(CONFIG_NET_L3_MASTER_DEV)
932 if (skb && ipv4_l3mdev_skb(TCP_SKB_CB(skb)->header.h4.flags))
933 return TCP_SKB_CB(skb)->header.h4.iif;
938 /* Due to TSO, an SKB can be composed of multiple actual
939 * packets. To keep these tracked properly, we use this.
941 static inline int tcp_skb_pcount(const struct sk_buff *skb)
943 return TCP_SKB_CB(skb)->tcp_gso_segs;
946 static inline void tcp_skb_pcount_set(struct sk_buff *skb, int segs)
948 TCP_SKB_CB(skb)->tcp_gso_segs = segs;
951 static inline void tcp_skb_pcount_add(struct sk_buff *skb, int segs)
953 TCP_SKB_CB(skb)->tcp_gso_segs += segs;
956 /* This is valid iff skb is in write queue and tcp_skb_pcount() > 1. */
957 static inline int tcp_skb_mss(const struct sk_buff *skb)
959 return TCP_SKB_CB(skb)->tcp_gso_size;
962 static inline bool tcp_skb_can_collapse_to(const struct sk_buff *skb)
964 return likely(!TCP_SKB_CB(skb)->eor);
967 static inline bool tcp_skb_can_collapse(const struct sk_buff *to,
968 const struct sk_buff *from)
970 return likely(tcp_skb_can_collapse_to(to) &&
971 mptcp_skb_can_collapse(to, from));
974 /* Events passed to congestion control interface */
976 CA_EVENT_TX_START, /* first transmit when no packets in flight */
977 CA_EVENT_CWND_RESTART, /* congestion window restart */
978 CA_EVENT_COMPLETE_CWR, /* end of congestion recovery */
979 CA_EVENT_LOSS, /* loss timeout */
980 CA_EVENT_ECN_NO_CE, /* ECT set, but not CE marked */
981 CA_EVENT_ECN_IS_CE, /* received CE marked IP packet */
984 /* Information about inbound ACK, passed to cong_ops->in_ack_event() */
985 enum tcp_ca_ack_event_flags {
986 CA_ACK_SLOWPATH = (1 << 0), /* In slow path processing */
987 CA_ACK_WIN_UPDATE = (1 << 1), /* ACK updated window */
988 CA_ACK_ECE = (1 << 2), /* ECE bit is set on ack */
992 * Interface for adding new TCP congestion control handlers
994 #define TCP_CA_NAME_MAX 16
995 #define TCP_CA_MAX 128
996 #define TCP_CA_BUF_MAX (TCP_CA_NAME_MAX*TCP_CA_MAX)
998 #define TCP_CA_UNSPEC 0
1000 /* Algorithm can be set on socket without CAP_NET_ADMIN privileges */
1001 #define TCP_CONG_NON_RESTRICTED 0x1
1002 /* Requires ECN/ECT set on all packets */
1003 #define TCP_CONG_NEEDS_ECN 0x2
1004 #define TCP_CONG_MASK (TCP_CONG_NON_RESTRICTED | TCP_CONG_NEEDS_ECN)
1014 /* A rate sample measures the number of (original/retransmitted) data
1015 * packets delivered "delivered" over an interval of time "interval_us".
1016 * The tcp_rate.c code fills in the rate sample, and congestion
1017 * control modules that define a cong_control function to run at the end
1018 * of ACK processing can optionally chose to consult this sample when
1019 * setting cwnd and pacing rate.
1020 * A sample is invalid if "delivered" or "interval_us" is negative.
1022 struct rate_sample {
1023 u64 prior_mstamp; /* starting timestamp for interval */
1024 u32 prior_delivered; /* tp->delivered at "prior_mstamp" */
1025 s32 delivered; /* number of packets delivered over interval */
1026 long interval_us; /* time for tp->delivered to incr "delivered" */
1027 u32 snd_interval_us; /* snd interval for delivered packets */
1028 u32 rcv_interval_us; /* rcv interval for delivered packets */
1029 long rtt_us; /* RTT of last (S)ACKed packet (or -1) */
1030 int losses; /* number of packets marked lost upon ACK */
1031 u32 acked_sacked; /* number of packets newly (S)ACKed upon ACK */
1032 u32 prior_in_flight; /* in flight before this ACK */
1033 u32 last_end_seq; /* end_seq of most recently ACKed packet */
1034 bool is_app_limited; /* is sample from packet with bubble in pipe? */
1035 bool is_retrans; /* is sample from retransmission? */
1036 bool is_ack_delayed; /* is this (likely) a delayed ACK? */
1039 struct tcp_congestion_ops {
1040 /* fast path fields are put first to fill one cache line */
1042 /* return slow start threshold (required) */
1043 u32 (*ssthresh)(struct sock *sk);
1045 /* do new cwnd calculation (required) */
1046 void (*cong_avoid)(struct sock *sk, u32 ack, u32 acked);
1048 /* call before changing ca_state (optional) */
1049 void (*set_state)(struct sock *sk, u8 new_state);
1051 /* call when cwnd event occurs (optional) */
1052 void (*cwnd_event)(struct sock *sk, enum tcp_ca_event ev);
1054 /* call when ack arrives (optional) */
1055 void (*in_ack_event)(struct sock *sk, u32 flags);
1057 /* hook for packet ack accounting (optional) */
1058 void (*pkts_acked)(struct sock *sk, const struct ack_sample *sample);
1060 /* override sysctl_tcp_min_tso_segs */
1061 u32 (*min_tso_segs)(struct sock *sk);
1063 /* call when packets are delivered to update cwnd and pacing rate,
1064 * after all the ca_state processing. (optional)
1066 void (*cong_control)(struct sock *sk, const struct rate_sample *rs);
1069 /* new value of cwnd after loss (required) */
1070 u32 (*undo_cwnd)(struct sock *sk);
1071 /* returns the multiplier used in tcp_sndbuf_expand (optional) */
1072 u32 (*sndbuf_expand)(struct sock *sk);
1074 /* control/slow paths put last */
1075 /* get info for inet_diag (optional) */
1076 size_t (*get_info)(struct sock *sk, u32 ext, int *attr,
1077 union tcp_cc_info *info);
1079 char name[TCP_CA_NAME_MAX];
1080 struct module *owner;
1081 struct list_head list;
1085 /* initialize private data (optional) */
1086 void (*init)(struct sock *sk);
1087 /* cleanup private data (optional) */
1088 void (*release)(struct sock *sk);
1089 } ____cacheline_aligned_in_smp;
1091 int tcp_register_congestion_control(struct tcp_congestion_ops *type);
1092 void tcp_unregister_congestion_control(struct tcp_congestion_ops *type);
1094 void tcp_assign_congestion_control(struct sock *sk);
1095 void tcp_init_congestion_control(struct sock *sk);
1096 void tcp_cleanup_congestion_control(struct sock *sk);
1097 int tcp_set_default_congestion_control(struct net *net, const char *name);
1098 void tcp_get_default_congestion_control(struct net *net, char *name);
1099 void tcp_get_available_congestion_control(char *buf, size_t len);
1100 void tcp_get_allowed_congestion_control(char *buf, size_t len);
1101 int tcp_set_allowed_congestion_control(char *allowed);
1102 int tcp_set_congestion_control(struct sock *sk, const char *name, bool load,
1103 bool cap_net_admin);
1104 u32 tcp_slow_start(struct tcp_sock *tp, u32 acked);
1105 void tcp_cong_avoid_ai(struct tcp_sock *tp, u32 w, u32 acked);
1107 u32 tcp_reno_ssthresh(struct sock *sk);
1108 u32 tcp_reno_undo_cwnd(struct sock *sk);
1109 void tcp_reno_cong_avoid(struct sock *sk, u32 ack, u32 acked);
1110 extern struct tcp_congestion_ops tcp_reno;
1112 struct tcp_congestion_ops *tcp_ca_find(const char *name);
1113 struct tcp_congestion_ops *tcp_ca_find_key(u32 key);
1114 u32 tcp_ca_get_key_by_name(struct net *net, const char *name, bool *ecn_ca);
1116 char *tcp_ca_get_name_by_key(u32 key, char *buffer);
1118 static inline char *tcp_ca_get_name_by_key(u32 key, char *buffer)
1124 static inline bool tcp_ca_needs_ecn(const struct sock *sk)
1126 const struct inet_connection_sock *icsk = inet_csk(sk);
1128 return icsk->icsk_ca_ops->flags & TCP_CONG_NEEDS_ECN;
1131 static inline void tcp_set_ca_state(struct sock *sk, const u8 ca_state)
1133 struct inet_connection_sock *icsk = inet_csk(sk);
1135 if (icsk->icsk_ca_ops->set_state)
1136 icsk->icsk_ca_ops->set_state(sk, ca_state);
1137 icsk->icsk_ca_state = ca_state;
1140 static inline void tcp_ca_event(struct sock *sk, const enum tcp_ca_event event)
1142 const struct inet_connection_sock *icsk = inet_csk(sk);
1144 if (icsk->icsk_ca_ops->cwnd_event)
1145 icsk->icsk_ca_ops->cwnd_event(sk, event);
1148 /* From tcp_rate.c */
1149 void tcp_rate_skb_sent(struct sock *sk, struct sk_buff *skb);
1150 void tcp_rate_skb_delivered(struct sock *sk, struct sk_buff *skb,
1151 struct rate_sample *rs);
1152 void tcp_rate_gen(struct sock *sk, u32 delivered, u32 lost,
1153 bool is_sack_reneg, struct rate_sample *rs);
1154 void tcp_rate_check_app_limited(struct sock *sk);
1156 static inline bool tcp_skb_sent_after(u64 t1, u64 t2, u32 seq1, u32 seq2)
1158 return t1 > t2 || (t1 == t2 && after(seq1, seq2));
1161 /* These functions determine how the current flow behaves in respect of SACK
1162 * handling. SACK is negotiated with the peer, and therefore it can vary
1163 * between different flows.
1165 * tcp_is_sack - SACK enabled
1166 * tcp_is_reno - No SACK
1168 static inline int tcp_is_sack(const struct tcp_sock *tp)
1170 return likely(tp->rx_opt.sack_ok);
1173 static inline bool tcp_is_reno(const struct tcp_sock *tp)
1175 return !tcp_is_sack(tp);
1178 static inline unsigned int tcp_left_out(const struct tcp_sock *tp)
1180 return tp->sacked_out + tp->lost_out;
1183 /* This determines how many packets are "in the network" to the best
1184 * of our knowledge. In many cases it is conservative, but where
1185 * detailed information is available from the receiver (via SACK
1186 * blocks etc.) we can make more aggressive calculations.
1188 * Use this for decisions involving congestion control, use just
1189 * tp->packets_out to determine if the send queue is empty or not.
1191 * Read this equation as:
1193 * "Packets sent once on transmission queue" MINUS
1194 * "Packets left network, but not honestly ACKed yet" PLUS
1195 * "Packets fast retransmitted"
1197 static inline unsigned int tcp_packets_in_flight(const struct tcp_sock *tp)
1199 return tp->packets_out - tcp_left_out(tp) + tp->retrans_out;
1202 #define TCP_INFINITE_SSTHRESH 0x7fffffff
1204 static inline u32 tcp_snd_cwnd(const struct tcp_sock *tp)
1206 return tp->snd_cwnd;
1209 static inline void tcp_snd_cwnd_set(struct tcp_sock *tp, u32 val)
1211 WARN_ON_ONCE((int)val <= 0);
1215 static inline bool tcp_in_slow_start(const struct tcp_sock *tp)
1217 return tcp_snd_cwnd(tp) < tp->snd_ssthresh;
1220 static inline bool tcp_in_initial_slowstart(const struct tcp_sock *tp)
1222 return tp->snd_ssthresh >= TCP_INFINITE_SSTHRESH;
1225 static inline bool tcp_in_cwnd_reduction(const struct sock *sk)
1227 return (TCPF_CA_CWR | TCPF_CA_Recovery) &
1228 (1 << inet_csk(sk)->icsk_ca_state);
1231 /* If cwnd > ssthresh, we may raise ssthresh to be half-way to cwnd.
1232 * The exception is cwnd reduction phase, when cwnd is decreasing towards
1235 static inline __u32 tcp_current_ssthresh(const struct sock *sk)
1237 const struct tcp_sock *tp = tcp_sk(sk);
1239 if (tcp_in_cwnd_reduction(sk))
1240 return tp->snd_ssthresh;
1242 return max(tp->snd_ssthresh,
1243 ((tcp_snd_cwnd(tp) >> 1) +
1244 (tcp_snd_cwnd(tp) >> 2)));
1247 /* Use define here intentionally to get WARN_ON location shown at the caller */
1248 #define tcp_verify_left_out(tp) WARN_ON(tcp_left_out(tp) > tp->packets_out)
1250 void tcp_enter_cwr(struct sock *sk);
1251 __u32 tcp_init_cwnd(const struct tcp_sock *tp, const struct dst_entry *dst);
1253 /* The maximum number of MSS of available cwnd for which TSO defers
1254 * sending if not using sysctl_tcp_tso_win_divisor.
1256 static inline __u32 tcp_max_tso_deferred_mss(const struct tcp_sock *tp)
1261 /* Returns end sequence number of the receiver's advertised window */
1262 static inline u32 tcp_wnd_end(const struct tcp_sock *tp)
1264 return tp->snd_una + tp->snd_wnd;
1267 /* We follow the spirit of RFC2861 to validate cwnd but implement a more
1268 * flexible approach. The RFC suggests cwnd should not be raised unless
1269 * it was fully used previously. And that's exactly what we do in
1270 * congestion avoidance mode. But in slow start we allow cwnd to grow
1271 * as long as the application has used half the cwnd.
1273 * cwnd is 10 (IW10), but application sends 9 frames.
1274 * We allow cwnd to reach 18 when all frames are ACKed.
1275 * This check is safe because it's as aggressive as slow start which already
1276 * risks 100% overshoot. The advantage is that we discourage application to
1277 * either send more filler packets or data to artificially blow up the cwnd
1278 * usage, and allow application-limited process to probe bw more aggressively.
1280 static inline bool tcp_is_cwnd_limited(const struct sock *sk)
1282 const struct tcp_sock *tp = tcp_sk(sk);
1284 if (tp->is_cwnd_limited)
1287 /* If in slow start, ensure cwnd grows to twice what was ACKed. */
1288 if (tcp_in_slow_start(tp))
1289 return tcp_snd_cwnd(tp) < 2 * tp->max_packets_out;
1294 /* BBR congestion control needs pacing.
1295 * Same remark for SO_MAX_PACING_RATE.
1296 * sch_fq packet scheduler is efficiently handling pacing,
1297 * but is not always installed/used.
1298 * Return true if TCP stack should pace packets itself.
1300 static inline bool tcp_needs_internal_pacing(const struct sock *sk)
1302 return smp_load_acquire(&sk->sk_pacing_status) == SK_PACING_NEEDED;
1305 /* Estimates in how many jiffies next packet for this flow can be sent.
1306 * Scheduling a retransmit timer too early would be silly.
1308 static inline unsigned long tcp_pacing_delay(const struct sock *sk)
1310 s64 delay = tcp_sk(sk)->tcp_wstamp_ns - tcp_sk(sk)->tcp_clock_cache;
1312 return delay > 0 ? nsecs_to_jiffies(delay) : 0;
1315 static inline void tcp_reset_xmit_timer(struct sock *sk,
1318 const unsigned long max_when)
1320 inet_csk_reset_xmit_timer(sk, what, when + tcp_pacing_delay(sk),
1324 /* Something is really bad, we could not queue an additional packet,
1325 * because qdisc is full or receiver sent a 0 window, or we are paced.
1326 * We do not want to add fuel to the fire, or abort too early,
1327 * so make sure the timer we arm now is at least 200ms in the future,
1328 * regardless of current icsk_rto value (as it could be ~2ms)
1330 static inline unsigned long tcp_probe0_base(const struct sock *sk)
1332 return max_t(unsigned long, inet_csk(sk)->icsk_rto, TCP_RTO_MIN);
1335 /* Variant of inet_csk_rto_backoff() used for zero window probes */
1336 static inline unsigned long tcp_probe0_when(const struct sock *sk,
1337 unsigned long max_when)
1339 u8 backoff = min_t(u8, ilog2(TCP_RTO_MAX / TCP_RTO_MIN) + 1,
1340 inet_csk(sk)->icsk_backoff);
1341 u64 when = (u64)tcp_probe0_base(sk) << backoff;
1343 return (unsigned long)min_t(u64, when, max_when);
1346 static inline void tcp_check_probe_timer(struct sock *sk)
1348 if (!tcp_sk(sk)->packets_out && !inet_csk(sk)->icsk_pending)
1349 tcp_reset_xmit_timer(sk, ICSK_TIME_PROBE0,
1350 tcp_probe0_base(sk), TCP_RTO_MAX);
1353 static inline void tcp_init_wl(struct tcp_sock *tp, u32 seq)
1358 static inline void tcp_update_wl(struct tcp_sock *tp, u32 seq)
1364 * Calculate(/check) TCP checksum
1366 static inline __sum16 tcp_v4_check(int len, __be32 saddr,
1367 __be32 daddr, __wsum base)
1369 return csum_tcpudp_magic(saddr, daddr, len, IPPROTO_TCP, base);
1372 static inline bool tcp_checksum_complete(struct sk_buff *skb)
1374 return !skb_csum_unnecessary(skb) &&
1375 __skb_checksum_complete(skb);
1378 bool tcp_add_backlog(struct sock *sk, struct sk_buff *skb);
1379 int tcp_filter(struct sock *sk, struct sk_buff *skb);
1380 void tcp_set_state(struct sock *sk, int state);
1381 void tcp_done(struct sock *sk);
1382 int tcp_abort(struct sock *sk, int err);
1384 static inline void tcp_sack_reset(struct tcp_options_received *rx_opt)
1387 rx_opt->num_sacks = 0;
1390 void tcp_cwnd_restart(struct sock *sk, s32 delta);
1392 static inline void tcp_slow_start_after_idle_check(struct sock *sk)
1394 const struct tcp_congestion_ops *ca_ops = inet_csk(sk)->icsk_ca_ops;
1395 struct tcp_sock *tp = tcp_sk(sk);
1398 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_slow_start_after_idle) ||
1399 tp->packets_out || ca_ops->cong_control)
1401 delta = tcp_jiffies32 - tp->lsndtime;
1402 if (delta > inet_csk(sk)->icsk_rto)
1403 tcp_cwnd_restart(sk, delta);
1406 /* Determine a window scaling and initial window to offer. */
1407 void tcp_select_initial_window(const struct sock *sk, int __space,
1408 __u32 mss, __u32 *rcv_wnd,
1409 __u32 *window_clamp, int wscale_ok,
1410 __u8 *rcv_wscale, __u32 init_rcv_wnd);
1412 static inline int tcp_win_from_space(const struct sock *sk, int space)
1414 int tcp_adv_win_scale = READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_adv_win_scale);
1416 return tcp_adv_win_scale <= 0 ?
1417 (space>>(-tcp_adv_win_scale)) :
1418 space - (space>>tcp_adv_win_scale);
1421 /* Note: caller must be prepared to deal with negative returns */
1422 static inline int tcp_space(const struct sock *sk)
1424 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf) -
1425 READ_ONCE(sk->sk_backlog.len) -
1426 atomic_read(&sk->sk_rmem_alloc));
1429 static inline int tcp_full_space(const struct sock *sk)
1431 return tcp_win_from_space(sk, READ_ONCE(sk->sk_rcvbuf));
1434 void tcp_cleanup_rbuf(struct sock *sk, int copied);
1436 /* We provision sk_rcvbuf around 200% of sk_rcvlowat.
1437 * If 87.5 % (7/8) of the space has been consumed, we want to override
1438 * SO_RCVLOWAT constraint, since we are receiving skbs with too small
1439 * len/truesize ratio.
1441 static inline bool tcp_rmem_pressure(const struct sock *sk)
1443 int rcvbuf, threshold;
1445 if (tcp_under_memory_pressure(sk))
1448 rcvbuf = READ_ONCE(sk->sk_rcvbuf);
1449 threshold = rcvbuf - (rcvbuf >> 3);
1451 return atomic_read(&sk->sk_rmem_alloc) > threshold;
1454 static inline bool tcp_epollin_ready(const struct sock *sk, int target)
1456 const struct tcp_sock *tp = tcp_sk(sk);
1457 int avail = READ_ONCE(tp->rcv_nxt) - READ_ONCE(tp->copied_seq);
1462 return (avail >= target) || tcp_rmem_pressure(sk) ||
1463 (tcp_receive_window(tp) <= inet_csk(sk)->icsk_ack.rcv_mss);
1466 extern void tcp_openreq_init_rwin(struct request_sock *req,
1467 const struct sock *sk_listener,
1468 const struct dst_entry *dst);
1470 void tcp_enter_memory_pressure(struct sock *sk);
1471 void tcp_leave_memory_pressure(struct sock *sk);
1473 static inline int keepalive_intvl_when(const struct tcp_sock *tp)
1475 struct net *net = sock_net((struct sock *)tp);
1477 return tp->keepalive_intvl ? :
1478 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_intvl);
1481 static inline int keepalive_time_when(const struct tcp_sock *tp)
1483 struct net *net = sock_net((struct sock *)tp);
1485 return tp->keepalive_time ? :
1486 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_time);
1489 static inline int keepalive_probes(const struct tcp_sock *tp)
1491 struct net *net = sock_net((struct sock *)tp);
1493 return tp->keepalive_probes ? :
1494 READ_ONCE(net->ipv4.sysctl_tcp_keepalive_probes);
1497 static inline u32 keepalive_time_elapsed(const struct tcp_sock *tp)
1499 const struct inet_connection_sock *icsk = &tp->inet_conn;
1501 return min_t(u32, tcp_jiffies32 - icsk->icsk_ack.lrcvtime,
1502 tcp_jiffies32 - tp->rcv_tstamp);
1505 static inline int tcp_fin_time(const struct sock *sk)
1507 int fin_timeout = tcp_sk(sk)->linger2 ? :
1508 READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_fin_timeout);
1509 const int rto = inet_csk(sk)->icsk_rto;
1511 if (fin_timeout < (rto << 2) - (rto >> 1))
1512 fin_timeout = (rto << 2) - (rto >> 1);
1517 static inline bool tcp_paws_check(const struct tcp_options_received *rx_opt,
1520 if ((s32)(rx_opt->ts_recent - rx_opt->rcv_tsval) <= paws_win)
1522 if (unlikely(!time_before32(ktime_get_seconds(),
1523 rx_opt->ts_recent_stamp + TCP_PAWS_24DAYS)))
1526 * Some OSes send SYN and SYNACK messages with tsval=0 tsecr=0,
1527 * then following tcp messages have valid values. Ignore 0 value,
1528 * or else 'negative' tsval might forbid us to accept their packets.
1530 if (!rx_opt->ts_recent)
1535 static inline bool tcp_paws_reject(const struct tcp_options_received *rx_opt,
1538 if (tcp_paws_check(rx_opt, 0))
1541 /* RST segments are not recommended to carry timestamp,
1542 and, if they do, it is recommended to ignore PAWS because
1543 "their cleanup function should take precedence over timestamps."
1544 Certainly, it is mistake. It is necessary to understand the reasons
1545 of this constraint to relax it: if peer reboots, clock may go
1546 out-of-sync and half-open connections will not be reset.
1547 Actually, the problem would be not existing if all
1548 the implementations followed draft about maintaining clock
1549 via reboots. Linux-2.2 DOES NOT!
1551 However, we can relax time bounds for RST segments to MSL.
1553 if (rst && !time_before32(ktime_get_seconds(),
1554 rx_opt->ts_recent_stamp + TCP_PAWS_MSL))
1559 bool tcp_oow_rate_limited(struct net *net, const struct sk_buff *skb,
1560 int mib_idx, u32 *last_oow_ack_time);
1562 static inline void tcp_mib_init(struct net *net)
1565 TCP_ADD_STATS(net, TCP_MIB_RTOALGORITHM, 1);
1566 TCP_ADD_STATS(net, TCP_MIB_RTOMIN, TCP_RTO_MIN*1000/HZ);
1567 TCP_ADD_STATS(net, TCP_MIB_RTOMAX, TCP_RTO_MAX*1000/HZ);
1568 TCP_ADD_STATS(net, TCP_MIB_MAXCONN, -1);
1572 static inline void tcp_clear_retrans_hints_partial(struct tcp_sock *tp)
1574 tp->lost_skb_hint = NULL;
1577 static inline void tcp_clear_all_retrans_hints(struct tcp_sock *tp)
1579 tcp_clear_retrans_hints_partial(tp);
1580 tp->retransmit_skb_hint = NULL;
1583 union tcp_md5_addr {
1585 #if IS_ENABLED(CONFIG_IPV6)
1590 /* - key database */
1591 struct tcp_md5sig_key {
1592 struct hlist_node node;
1594 u8 family; /* AF_INET or AF_INET6 */
1597 union tcp_md5_addr addr;
1598 int l3index; /* set if key added with L3 scope */
1599 u8 key[TCP_MD5SIG_MAXKEYLEN];
1600 struct rcu_head rcu;
1604 struct tcp_md5sig_info {
1605 struct hlist_head head;
1606 struct rcu_head rcu;
1609 /* - pseudo header */
1610 struct tcp4_pseudohdr {
1618 struct tcp6_pseudohdr {
1619 struct in6_addr saddr;
1620 struct in6_addr daddr;
1622 __be32 protocol; /* including padding */
1625 union tcp_md5sum_block {
1626 struct tcp4_pseudohdr ip4;
1627 #if IS_ENABLED(CONFIG_IPV6)
1628 struct tcp6_pseudohdr ip6;
1632 /* - pool: digest algorithm, hash description and scratch buffer */
1633 struct tcp_md5sig_pool {
1634 struct ahash_request *md5_req;
1639 int tcp_v4_md5_hash_skb(char *md5_hash, const struct tcp_md5sig_key *key,
1640 const struct sock *sk, const struct sk_buff *skb);
1641 int tcp_md5_do_add(struct sock *sk, const union tcp_md5_addr *addr,
1642 int family, u8 prefixlen, int l3index, u8 flags,
1643 const u8 *newkey, u8 newkeylen, gfp_t gfp);
1644 int tcp_md5_do_del(struct sock *sk, const union tcp_md5_addr *addr,
1645 int family, u8 prefixlen, int l3index, u8 flags);
1646 struct tcp_md5sig_key *tcp_v4_md5_lookup(const struct sock *sk,
1647 const struct sock *addr_sk);
1649 #ifdef CONFIG_TCP_MD5SIG
1650 #include <linux/jump_label.h>
1651 extern struct static_key_false tcp_md5_needed;
1652 struct tcp_md5sig_key *__tcp_md5_do_lookup(const struct sock *sk, int l3index,
1653 const union tcp_md5_addr *addr,
1655 static inline struct tcp_md5sig_key *
1656 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1657 const union tcp_md5_addr *addr, int family)
1659 if (!static_branch_unlikely(&tcp_md5_needed))
1661 return __tcp_md5_do_lookup(sk, l3index, addr, family);
1664 #define tcp_twsk_md5_key(twsk) ((twsk)->tw_md5_key)
1666 static inline struct tcp_md5sig_key *
1667 tcp_md5_do_lookup(const struct sock *sk, int l3index,
1668 const union tcp_md5_addr *addr, int family)
1672 #define tcp_twsk_md5_key(twsk) NULL
1675 bool tcp_alloc_md5sig_pool(void);
1677 struct tcp_md5sig_pool *tcp_get_md5sig_pool(void);
1678 static inline void tcp_put_md5sig_pool(void)
1683 int tcp_md5_hash_skb_data(struct tcp_md5sig_pool *, const struct sk_buff *,
1684 unsigned int header_len);
1685 int tcp_md5_hash_key(struct tcp_md5sig_pool *hp,
1686 const struct tcp_md5sig_key *key);
1688 /* From tcp_fastopen.c */
1689 void tcp_fastopen_cache_get(struct sock *sk, u16 *mss,
1690 struct tcp_fastopen_cookie *cookie);
1691 void tcp_fastopen_cache_set(struct sock *sk, u16 mss,
1692 struct tcp_fastopen_cookie *cookie, bool syn_lost,
1694 struct tcp_fastopen_request {
1695 /* Fast Open cookie. Size 0 means a cookie request */
1696 struct tcp_fastopen_cookie cookie;
1697 struct msghdr *data; /* data in MSG_FASTOPEN */
1699 int copied; /* queued in tcp_connect() */
1700 struct ubuf_info *uarg;
1702 void tcp_free_fastopen_req(struct tcp_sock *tp);
1703 void tcp_fastopen_destroy_cipher(struct sock *sk);
1704 void tcp_fastopen_ctx_destroy(struct net *net);
1705 int tcp_fastopen_reset_cipher(struct net *net, struct sock *sk,
1706 void *primary_key, void *backup_key);
1707 int tcp_fastopen_get_cipher(struct net *net, struct inet_connection_sock *icsk,
1709 void tcp_fastopen_add_skb(struct sock *sk, struct sk_buff *skb);
1710 struct sock *tcp_try_fastopen(struct sock *sk, struct sk_buff *skb,
1711 struct request_sock *req,
1712 struct tcp_fastopen_cookie *foc,
1713 const struct dst_entry *dst);
1714 void tcp_fastopen_init_key_once(struct net *net);
1715 bool tcp_fastopen_cookie_check(struct sock *sk, u16 *mss,
1716 struct tcp_fastopen_cookie *cookie);
1717 bool tcp_fastopen_defer_connect(struct sock *sk, int *err);
1718 #define TCP_FASTOPEN_KEY_LENGTH sizeof(siphash_key_t)
1719 #define TCP_FASTOPEN_KEY_MAX 2
1720 #define TCP_FASTOPEN_KEY_BUF_LENGTH \
1721 (TCP_FASTOPEN_KEY_LENGTH * TCP_FASTOPEN_KEY_MAX)
1723 /* Fastopen key context */
1724 struct tcp_fastopen_context {
1725 siphash_key_t key[TCP_FASTOPEN_KEY_MAX];
1727 struct rcu_head rcu;
1730 void tcp_fastopen_active_disable(struct sock *sk);
1731 bool tcp_fastopen_active_should_disable(struct sock *sk);
1732 void tcp_fastopen_active_disable_ofo_check(struct sock *sk);
1733 void tcp_fastopen_active_detect_blackhole(struct sock *sk, bool expired);
1735 /* Caller needs to wrap with rcu_read_(un)lock() */
1737 struct tcp_fastopen_context *tcp_fastopen_get_ctx(const struct sock *sk)
1739 struct tcp_fastopen_context *ctx;
1741 ctx = rcu_dereference(inet_csk(sk)->icsk_accept_queue.fastopenq.ctx);
1743 ctx = rcu_dereference(sock_net(sk)->ipv4.tcp_fastopen_ctx);
1748 bool tcp_fastopen_cookie_match(const struct tcp_fastopen_cookie *foc,
1749 const struct tcp_fastopen_cookie *orig)
1751 if (orig->len == TCP_FASTOPEN_COOKIE_SIZE &&
1752 orig->len == foc->len &&
1753 !memcmp(orig->val, foc->val, foc->len))
1759 int tcp_fastopen_context_len(const struct tcp_fastopen_context *ctx)
1764 /* Latencies incurred by various limits for a sender. They are
1765 * chronograph-like stats that are mutually exclusive.
1769 TCP_CHRONO_BUSY, /* Actively sending data (non-empty write queue) */
1770 TCP_CHRONO_RWND_LIMITED, /* Stalled by insufficient receive window */
1771 TCP_CHRONO_SNDBUF_LIMITED, /* Stalled by insufficient send buffer */
1775 void tcp_chrono_start(struct sock *sk, const enum tcp_chrono type);
1776 void tcp_chrono_stop(struct sock *sk, const enum tcp_chrono type);
1778 /* This helper is needed, because skb->tcp_tsorted_anchor uses
1779 * the same memory storage than skb->destructor/_skb_refdst
1781 static inline void tcp_skb_tsorted_anchor_cleanup(struct sk_buff *skb)
1783 skb->destructor = NULL;
1784 skb->_skb_refdst = 0UL;
1787 #define tcp_skb_tsorted_save(skb) { \
1788 unsigned long _save = skb->_skb_refdst; \
1789 skb->_skb_refdst = 0UL;
1791 #define tcp_skb_tsorted_restore(skb) \
1792 skb->_skb_refdst = _save; \
1795 void tcp_write_queue_purge(struct sock *sk);
1797 static inline struct sk_buff *tcp_rtx_queue_head(const struct sock *sk)
1799 return skb_rb_first(&sk->tcp_rtx_queue);
1802 static inline struct sk_buff *tcp_rtx_queue_tail(const struct sock *sk)
1804 return skb_rb_last(&sk->tcp_rtx_queue);
1807 static inline struct sk_buff *tcp_write_queue_head(const struct sock *sk)
1809 return skb_peek(&sk->sk_write_queue);
1812 static inline struct sk_buff *tcp_write_queue_tail(const struct sock *sk)
1814 return skb_peek_tail(&sk->sk_write_queue);
1817 #define tcp_for_write_queue_from_safe(skb, tmp, sk) \
1818 skb_queue_walk_from_safe(&(sk)->sk_write_queue, skb, tmp)
1820 static inline struct sk_buff *tcp_send_head(const struct sock *sk)
1822 return skb_peek(&sk->sk_write_queue);
1825 static inline bool tcp_skb_is_last(const struct sock *sk,
1826 const struct sk_buff *skb)
1828 return skb_queue_is_last(&sk->sk_write_queue, skb);
1832 * tcp_write_queue_empty - test if any payload (or FIN) is available in write queue
1835 * Since the write queue can have a temporary empty skb in it,
1836 * we must not use "return skb_queue_empty(&sk->sk_write_queue)"
1838 static inline bool tcp_write_queue_empty(const struct sock *sk)
1840 const struct tcp_sock *tp = tcp_sk(sk);
1842 return tp->write_seq == tp->snd_nxt;
1845 static inline bool tcp_rtx_queue_empty(const struct sock *sk)
1847 return RB_EMPTY_ROOT(&sk->tcp_rtx_queue);
1850 static inline bool tcp_rtx_and_write_queues_empty(const struct sock *sk)
1852 return tcp_rtx_queue_empty(sk) && tcp_write_queue_empty(sk);
1855 static inline void tcp_add_write_queue_tail(struct sock *sk, struct sk_buff *skb)
1857 __skb_queue_tail(&sk->sk_write_queue, skb);
1859 /* Queue it, remembering where we must start sending. */
1860 if (sk->sk_write_queue.next == skb)
1861 tcp_chrono_start(sk, TCP_CHRONO_BUSY);
1864 /* Insert new before skb on the write queue of sk. */
1865 static inline void tcp_insert_write_queue_before(struct sk_buff *new,
1866 struct sk_buff *skb,
1869 __skb_queue_before(&sk->sk_write_queue, skb, new);
1872 static inline void tcp_unlink_write_queue(struct sk_buff *skb, struct sock *sk)
1874 tcp_skb_tsorted_anchor_cleanup(skb);
1875 __skb_unlink(skb, &sk->sk_write_queue);
1878 void tcp_rbtree_insert(struct rb_root *root, struct sk_buff *skb);
1880 static inline void tcp_rtx_queue_unlink(struct sk_buff *skb, struct sock *sk)
1882 tcp_skb_tsorted_anchor_cleanup(skb);
1883 rb_erase(&skb->rbnode, &sk->tcp_rtx_queue);
1886 static inline void tcp_rtx_queue_unlink_and_free(struct sk_buff *skb, struct sock *sk)
1888 list_del(&skb->tcp_tsorted_anchor);
1889 tcp_rtx_queue_unlink(skb, sk);
1890 sk_wmem_free_skb(sk, skb);
1893 static inline void tcp_push_pending_frames(struct sock *sk)
1895 if (tcp_send_head(sk)) {
1896 struct tcp_sock *tp = tcp_sk(sk);
1898 __tcp_push_pending_frames(sk, tcp_current_mss(sk), tp->nonagle);
1902 /* Start sequence of the skb just after the highest skb with SACKed
1903 * bit, valid only if sacked_out > 0 or when the caller has ensured
1904 * validity by itself.
1906 static inline u32 tcp_highest_sack_seq(struct tcp_sock *tp)
1908 if (!tp->sacked_out)
1911 if (tp->highest_sack == NULL)
1914 return TCP_SKB_CB(tp->highest_sack)->seq;
1917 static inline void tcp_advance_highest_sack(struct sock *sk, struct sk_buff *skb)
1919 tcp_sk(sk)->highest_sack = skb_rb_next(skb);
1922 static inline struct sk_buff *tcp_highest_sack(struct sock *sk)
1924 return tcp_sk(sk)->highest_sack;
1927 static inline void tcp_highest_sack_reset(struct sock *sk)
1929 tcp_sk(sk)->highest_sack = tcp_rtx_queue_head(sk);
1932 /* Called when old skb is about to be deleted and replaced by new skb */
1933 static inline void tcp_highest_sack_replace(struct sock *sk,
1934 struct sk_buff *old,
1935 struct sk_buff *new)
1937 if (old == tcp_highest_sack(sk))
1938 tcp_sk(sk)->highest_sack = new;
1941 /* This helper checks if socket has IP_TRANSPARENT set */
1942 static inline bool inet_sk_transparent(const struct sock *sk)
1944 switch (sk->sk_state) {
1946 return inet_twsk(sk)->tw_transparent;
1947 case TCP_NEW_SYN_RECV:
1948 return inet_rsk(inet_reqsk(sk))->no_srccheck;
1950 return inet_sk(sk)->transparent;
1953 /* Determines whether this is a thin stream (which may suffer from
1954 * increased latency). Used to trigger latency-reducing mechanisms.
1956 static inline bool tcp_stream_is_thin(struct tcp_sock *tp)
1958 return tp->packets_out < 4 && !tcp_in_initial_slowstart(tp);
1962 enum tcp_seq_states {
1963 TCP_SEQ_STATE_LISTENING,
1964 TCP_SEQ_STATE_ESTABLISHED,
1967 void *tcp_seq_start(struct seq_file *seq, loff_t *pos);
1968 void *tcp_seq_next(struct seq_file *seq, void *v, loff_t *pos);
1969 void tcp_seq_stop(struct seq_file *seq, void *v);
1971 struct tcp_seq_afinfo {
1975 struct tcp_iter_state {
1976 struct seq_net_private p;
1977 enum tcp_seq_states state;
1978 struct sock *syn_wait_sk;
1979 int bucket, offset, sbucket, num;
1983 extern struct request_sock_ops tcp_request_sock_ops;
1984 extern struct request_sock_ops tcp6_request_sock_ops;
1986 void tcp_v4_destroy_sock(struct sock *sk);
1988 struct sk_buff *tcp_gso_segment(struct sk_buff *skb,
1989 netdev_features_t features);
1990 struct sk_buff *tcp_gro_receive(struct list_head *head, struct sk_buff *skb);
1991 INDIRECT_CALLABLE_DECLARE(int tcp4_gro_complete(struct sk_buff *skb, int thoff));
1992 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp4_gro_receive(struct list_head *head, struct sk_buff *skb));
1993 INDIRECT_CALLABLE_DECLARE(int tcp6_gro_complete(struct sk_buff *skb, int thoff));
1994 INDIRECT_CALLABLE_DECLARE(struct sk_buff *tcp6_gro_receive(struct list_head *head, struct sk_buff *skb));
1995 int tcp_gro_complete(struct sk_buff *skb);
1997 void __tcp_v4_send_check(struct sk_buff *skb, __be32 saddr, __be32 daddr);
1999 static inline u32 tcp_notsent_lowat(const struct tcp_sock *tp)
2001 struct net *net = sock_net((struct sock *)tp);
2002 return tp->notsent_lowat ?: READ_ONCE(net->ipv4.sysctl_tcp_notsent_lowat);
2005 bool tcp_stream_memory_free(const struct sock *sk, int wake);
2007 #ifdef CONFIG_PROC_FS
2008 int tcp4_proc_init(void);
2009 void tcp4_proc_exit(void);
2012 int tcp_rtx_synack(const struct sock *sk, struct request_sock *req);
2013 int tcp_conn_request(struct request_sock_ops *rsk_ops,
2014 const struct tcp_request_sock_ops *af_ops,
2015 struct sock *sk, struct sk_buff *skb);
2017 /* TCP af-specific functions */
2018 struct tcp_sock_af_ops {
2019 #ifdef CONFIG_TCP_MD5SIG
2020 struct tcp_md5sig_key *(*md5_lookup) (const struct sock *sk,
2021 const struct sock *addr_sk);
2022 int (*calc_md5_hash)(char *location,
2023 const struct tcp_md5sig_key *md5,
2024 const struct sock *sk,
2025 const struct sk_buff *skb);
2026 int (*md5_parse)(struct sock *sk,
2033 struct tcp_request_sock_ops {
2035 #ifdef CONFIG_TCP_MD5SIG
2036 struct tcp_md5sig_key *(*req_md5_lookup)(const struct sock *sk,
2037 const struct sock *addr_sk);
2038 int (*calc_md5_hash) (char *location,
2039 const struct tcp_md5sig_key *md5,
2040 const struct sock *sk,
2041 const struct sk_buff *skb);
2043 #ifdef CONFIG_SYN_COOKIES
2044 __u32 (*cookie_init_seq)(const struct sk_buff *skb,
2047 struct dst_entry *(*route_req)(const struct sock *sk,
2048 struct sk_buff *skb,
2050 struct request_sock *req);
2051 u32 (*init_seq)(const struct sk_buff *skb);
2052 u32 (*init_ts_off)(const struct net *net, const struct sk_buff *skb);
2053 int (*send_synack)(const struct sock *sk, struct dst_entry *dst,
2054 struct flowi *fl, struct request_sock *req,
2055 struct tcp_fastopen_cookie *foc,
2056 enum tcp_synack_type synack_type,
2057 struct sk_buff *syn_skb);
2060 extern const struct tcp_request_sock_ops tcp_request_sock_ipv4_ops;
2061 #if IS_ENABLED(CONFIG_IPV6)
2062 extern const struct tcp_request_sock_ops tcp_request_sock_ipv6_ops;
2065 #ifdef CONFIG_SYN_COOKIES
2066 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2067 const struct sock *sk, struct sk_buff *skb,
2070 tcp_synq_overflow(sk);
2071 __NET_INC_STATS(sock_net(sk), LINUX_MIB_SYNCOOKIESSENT);
2072 return ops->cookie_init_seq(skb, mss);
2075 static inline __u32 cookie_init_sequence(const struct tcp_request_sock_ops *ops,
2076 const struct sock *sk, struct sk_buff *skb,
2083 int tcpv4_offload_init(void);
2085 void tcp_v4_init(void);
2086 void tcp_init(void);
2088 /* tcp_recovery.c */
2089 void tcp_mark_skb_lost(struct sock *sk, struct sk_buff *skb);
2090 void tcp_newreno_mark_lost(struct sock *sk, bool snd_una_advanced);
2091 extern s32 tcp_rack_skb_timeout(struct tcp_sock *tp, struct sk_buff *skb,
2093 extern bool tcp_rack_mark_lost(struct sock *sk);
2094 extern void tcp_rack_advance(struct tcp_sock *tp, u8 sacked, u32 end_seq,
2096 extern void tcp_rack_reo_timeout(struct sock *sk);
2097 extern void tcp_rack_update_reo_wnd(struct sock *sk, struct rate_sample *rs);
2099 /* At how many usecs into the future should the RTO fire? */
2100 static inline s64 tcp_rto_delta_us(const struct sock *sk)
2102 const struct sk_buff *skb = tcp_rtx_queue_head(sk);
2103 u32 rto = inet_csk(sk)->icsk_rto;
2104 u64 rto_time_stamp_us = tcp_skb_timestamp_us(skb) + jiffies_to_usecs(rto);
2106 return rto_time_stamp_us - tcp_sk(sk)->tcp_mstamp;
2110 * Save and compile IPv4 options, return a pointer to it
2112 static inline struct ip_options_rcu *tcp_v4_save_options(struct net *net,
2113 struct sk_buff *skb)
2115 const struct ip_options *opt = &TCP_SKB_CB(skb)->header.h4.opt;
2116 struct ip_options_rcu *dopt = NULL;
2119 int opt_size = sizeof(*dopt) + opt->optlen;
2121 dopt = kmalloc(opt_size, GFP_ATOMIC);
2122 if (dopt && __ip_options_echo(net, &dopt->opt, skb, opt)) {
2130 /* locally generated TCP pure ACKs have skb->truesize == 2
2131 * (check tcp_send_ack() in net/ipv4/tcp_output.c )
2132 * This is much faster than dissecting the packet to find out.
2133 * (Think of GRE encapsulations, IPv4, IPv6, ...)
2135 static inline bool skb_is_tcp_pure_ack(const struct sk_buff *skb)
2137 return skb->truesize == 2;
2140 static inline void skb_set_tcp_pure_ack(struct sk_buff *skb)
2145 static inline int tcp_inq(struct sock *sk)
2147 struct tcp_sock *tp = tcp_sk(sk);
2150 if ((1 << sk->sk_state) & (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
2152 } else if (sock_flag(sk, SOCK_URGINLINE) ||
2154 before(tp->urg_seq, tp->copied_seq) ||
2155 !before(tp->urg_seq, tp->rcv_nxt)) {
2157 answ = tp->rcv_nxt - tp->copied_seq;
2159 /* Subtract 1, if FIN was received */
2160 if (answ && sock_flag(sk, SOCK_DONE))
2163 answ = tp->urg_seq - tp->copied_seq;
2169 int tcp_peek_len(struct socket *sock);
2171 static inline void tcp_segs_in(struct tcp_sock *tp, const struct sk_buff *skb)
2175 segs_in = max_t(u16, 1, skb_shinfo(skb)->gso_segs);
2176 tp->segs_in += segs_in;
2177 if (skb->len > tcp_hdrlen(skb))
2178 tp->data_segs_in += segs_in;
2182 * TCP listen path runs lockless.
2183 * We forced "struct sock" to be const qualified to make sure
2184 * we don't modify one of its field by mistake.
2185 * Here, we increment sk_drops which is an atomic_t, so we can safely
2186 * make sock writable again.
2188 static inline void tcp_listendrop(const struct sock *sk)
2190 atomic_inc(&((struct sock *)sk)->sk_drops);
2191 __NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENDROPS);
2194 enum hrtimer_restart tcp_pace_kick(struct hrtimer *timer);
2197 * Interface for adding Upper Level Protocols over TCP
2200 #define TCP_ULP_NAME_MAX 16
2201 #define TCP_ULP_MAX 128
2202 #define TCP_ULP_BUF_MAX (TCP_ULP_NAME_MAX*TCP_ULP_MAX)
2204 struct tcp_ulp_ops {
2205 struct list_head list;
2207 /* initialize ulp */
2208 int (*init)(struct sock *sk);
2210 void (*update)(struct sock *sk, struct proto *p,
2211 void (*write_space)(struct sock *sk));
2213 void (*release)(struct sock *sk);
2215 int (*get_info)(const struct sock *sk, struct sk_buff *skb);
2216 size_t (*get_info_size)(const struct sock *sk);
2218 void (*clone)(const struct request_sock *req, struct sock *newsk,
2219 const gfp_t priority);
2221 char name[TCP_ULP_NAME_MAX];
2222 struct module *owner;
2224 int tcp_register_ulp(struct tcp_ulp_ops *type);
2225 void tcp_unregister_ulp(struct tcp_ulp_ops *type);
2226 int tcp_set_ulp(struct sock *sk, const char *name);
2227 void tcp_get_available_ulp(char *buf, size_t len);
2228 void tcp_cleanup_ulp(struct sock *sk);
2229 void tcp_update_ulp(struct sock *sk, struct proto *p,
2230 void (*write_space)(struct sock *sk));
2232 #define MODULE_ALIAS_TCP_ULP(name) \
2233 __MODULE_INFO(alias, alias_userspace, name); \
2234 __MODULE_INFO(alias, alias_tcp_ulp, "tcp-ulp-" name)
2236 #ifdef CONFIG_NET_SOCK_MSG
2240 #ifdef CONFIG_BPF_SYSCALL
2241 struct proto *tcp_bpf_get_proto(struct sock *sk, struct sk_psock *psock);
2242 int tcp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore);
2243 void tcp_bpf_clone(const struct sock *sk, struct sock *newsk);
2244 #endif /* CONFIG_BPF_SYSCALL */
2246 int tcp_bpf_sendmsg_redir(struct sock *sk, bool ingress,
2247 struct sk_msg *msg, u32 bytes, int flags);
2248 #endif /* CONFIG_NET_SOCK_MSG */
2250 #if !defined(CONFIG_BPF_SYSCALL) || !defined(CONFIG_NET_SOCK_MSG)
2251 static inline void tcp_bpf_clone(const struct sock *sk, struct sock *newsk)
2256 #ifdef CONFIG_CGROUP_BPF
2257 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2258 struct sk_buff *skb,
2259 unsigned int end_offset)
2262 skops->skb_data_end = skb->data + end_offset;
2265 static inline void bpf_skops_init_skb(struct bpf_sock_ops_kern *skops,
2266 struct sk_buff *skb,
2267 unsigned int end_offset)
2272 /* Call BPF_SOCK_OPS program that returns an int. If the return value
2273 * is < 0, then the BPF op failed (for example if the loaded BPF
2274 * program does not support the chosen operation or there is no BPF
2278 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2280 struct bpf_sock_ops_kern sock_ops;
2283 memset(&sock_ops, 0, offsetof(struct bpf_sock_ops_kern, temp));
2284 if (sk_fullsock(sk)) {
2285 sock_ops.is_fullsock = 1;
2286 sock_owned_by_me(sk);
2292 memcpy(sock_ops.args, args, nargs * sizeof(*args));
2294 ret = BPF_CGROUP_RUN_PROG_SOCK_OPS(&sock_ops);
2296 ret = sock_ops.reply;
2302 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2304 u32 args[2] = {arg1, arg2};
2306 return tcp_call_bpf(sk, op, 2, args);
2309 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2312 u32 args[3] = {arg1, arg2, arg3};
2314 return tcp_call_bpf(sk, op, 3, args);
2318 static inline int tcp_call_bpf(struct sock *sk, int op, u32 nargs, u32 *args)
2323 static inline int tcp_call_bpf_2arg(struct sock *sk, int op, u32 arg1, u32 arg2)
2328 static inline int tcp_call_bpf_3arg(struct sock *sk, int op, u32 arg1, u32 arg2,
2336 static inline u32 tcp_timeout_init(struct sock *sk)
2340 timeout = tcp_call_bpf(sk, BPF_SOCK_OPS_TIMEOUT_INIT, 0, NULL);
2343 timeout = TCP_TIMEOUT_INIT;
2347 static inline u32 tcp_rwnd_init_bpf(struct sock *sk)
2351 rwnd = tcp_call_bpf(sk, BPF_SOCK_OPS_RWND_INIT, 0, NULL);
2358 static inline bool tcp_bpf_ca_needs_ecn(struct sock *sk)
2360 return (tcp_call_bpf(sk, BPF_SOCK_OPS_NEEDS_ECN, 0, NULL) == 1);
2363 static inline void tcp_bpf_rtt(struct sock *sk)
2365 if (BPF_SOCK_OPS_TEST_FLAG(tcp_sk(sk), BPF_SOCK_OPS_RTT_CB_FLAG))
2366 tcp_call_bpf(sk, BPF_SOCK_OPS_RTT_CB, 0, NULL);
2369 #if IS_ENABLED(CONFIG_SMC)
2370 extern struct static_key_false tcp_have_smc;
2373 #if IS_ENABLED(CONFIG_TLS_DEVICE)
2374 void clean_acked_data_enable(struct inet_connection_sock *icsk,
2375 void (*cad)(struct sock *sk, u32 ack_seq));
2376 void clean_acked_data_disable(struct inet_connection_sock *icsk);
2377 void clean_acked_data_flush(void);
2380 DECLARE_STATIC_KEY_FALSE(tcp_tx_delay_enabled);
2381 static inline void tcp_add_tx_delay(struct sk_buff *skb,
2382 const struct tcp_sock *tp)
2384 if (static_branch_unlikely(&tcp_tx_delay_enabled))
2385 skb->skb_mstamp_ns += (u64)tp->tcp_tx_delay * NSEC_PER_USEC;
2388 /* Compute Earliest Departure Time for some control packets
2389 * like ACK or RST for TIME_WAIT or non ESTABLISHED sockets.
2391 static inline u64 tcp_transmit_time(const struct sock *sk)
2393 if (static_branch_unlikely(&tcp_tx_delay_enabled)) {
2394 u32 delay = (sk->sk_state == TCP_TIME_WAIT) ?
2395 tcp_twsk(sk)->tw_tx_delay : tcp_sk(sk)->tcp_tx_delay;
2397 return tcp_clock_ns() + (u64)delay * NSEC_PER_USEC;